xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision a6fae8f906ef82ca3caf49344efa3132b0a02b4c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
4  * All rights reserved.
5  *
6  * This software was jointly developed between OKTET Labs (under contract
7  * for Solarflare) and Solarflare Communications, Inc.
8  */
9 
10 #include <rte_dev.h>
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
16 
17 #include "efx.h"
18 
19 #include "sfc.h"
20 #include "sfc_debug.h"
21 #include "sfc_log.h"
22 #include "sfc_kvargs.h"
23 #include "sfc_ev.h"
24 #include "sfc_rx.h"
25 #include "sfc_tx.h"
26 #include "sfc_flow.h"
27 #include "sfc_dp.h"
28 #include "sfc_dp_rx.h"
29 
30 uint32_t sfc_logtype_driver;
31 
32 static struct sfc_dp_list sfc_dp_head =
33 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
34 
35 static int
36 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
37 {
38 	struct sfc_adapter *sa = dev->data->dev_private;
39 	efx_nic_fw_info_t enfi;
40 	int ret;
41 	int rc;
42 
43 	/*
44 	 * Return value of the callback is likely supposed to be
45 	 * equal to or greater than 0, nevertheless, if an error
46 	 * occurs, it will be desirable to pass it to the caller
47 	 */
48 	if ((fw_version == NULL) || (fw_size == 0))
49 		return -EINVAL;
50 
51 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
52 	if (rc != 0)
53 		return -rc;
54 
55 	ret = snprintf(fw_version, fw_size,
56 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
57 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
58 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
59 	if (ret < 0)
60 		return ret;
61 
62 	if (enfi.enfi_dpcpu_fw_ids_valid) {
63 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
64 		int ret_extra;
65 
66 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
67 				     fw_size - dpcpu_fw_ids_offset,
68 				     " rx%" PRIx16 " tx%" PRIx16,
69 				     enfi.enfi_rx_dpcpu_fw_id,
70 				     enfi.enfi_tx_dpcpu_fw_id);
71 		if (ret_extra < 0)
72 			return ret_extra;
73 
74 		ret += ret_extra;
75 	}
76 
77 	if (fw_size < (size_t)(++ret))
78 		return ret;
79 	else
80 		return 0;
81 }
82 
83 static void
84 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
85 {
86 	struct sfc_adapter *sa = dev->data->dev_private;
87 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
88 	uint64_t txq_offloads_def = 0;
89 
90 	sfc_log_init(sa, "entry");
91 
92 	dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
93 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
94 
95 	/* Autonegotiation may be disabled */
96 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
97 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
98 		dev_info->speed_capa |= ETH_LINK_SPEED_1G;
99 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
100 		dev_info->speed_capa |= ETH_LINK_SPEED_10G;
101 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
102 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
103 
104 	dev_info->max_rx_queues = sa->rxq_max;
105 	dev_info->max_tx_queues = sa->txq_max;
106 
107 	/* By default packets are dropped if no descriptors are available */
108 	dev_info->default_rxconf.rx_drop_en = 1;
109 
110 	dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
111 
112 	/*
113 	 * rx_offload_capa includes both device and queue offloads since
114 	 * the latter may be requested on a per device basis which makes
115 	 * sense when some offloads are needed to be set on all queues.
116 	 */
117 	dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
118 				    dev_info->rx_queue_offload_capa;
119 
120 	dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
121 
122 	/*
123 	 * tx_offload_capa includes both device and queue offloads since
124 	 * the latter may be requested on a per device basis which makes
125 	 * sense when some offloads are needed to be set on all queues.
126 	 */
127 	dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
128 				    dev_info->tx_queue_offload_capa;
129 
130 	if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
131 		txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
132 
133 	dev_info->default_txconf.offloads |= txq_offloads_def;
134 
135 	dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
136 	if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
137 	    !encp->enc_hw_tx_insert_vlan_enabled)
138 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
139 
140 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
141 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
142 
143 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
144 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
145 
146 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
147 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
148 
149 #if EFSYS_OPT_RX_SCALE
150 	if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
151 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
152 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
153 		dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
154 	}
155 #endif
156 
157 	/* Initialize to hardware limits */
158 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
159 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
160 	/* The RXQ hardware requires that the descriptor count is a power
161 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
162 	 */
163 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
164 
165 	/* Initialize to hardware limits */
166 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
167 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
168 	/*
169 	 * The TXQ hardware requires that the descriptor count is a power
170 	 * of 2, but tx_desc_lim cannot properly describe that constraint
171 	 */
172 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
173 
174 	if (sa->dp_rx->get_dev_info != NULL)
175 		sa->dp_rx->get_dev_info(dev_info);
176 	if (sa->dp_tx->get_dev_info != NULL)
177 		sa->dp_tx->get_dev_info(dev_info);
178 }
179 
180 static const uint32_t *
181 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
182 {
183 	struct sfc_adapter *sa = dev->data->dev_private;
184 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
185 	uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
186 
187 	return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
188 }
189 
190 static int
191 sfc_dev_configure(struct rte_eth_dev *dev)
192 {
193 	struct rte_eth_dev_data *dev_data = dev->data;
194 	struct sfc_adapter *sa = dev_data->dev_private;
195 	int rc;
196 
197 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
198 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
199 
200 	sfc_adapter_lock(sa);
201 	switch (sa->state) {
202 	case SFC_ADAPTER_CONFIGURED:
203 		/* FALLTHROUGH */
204 	case SFC_ADAPTER_INITIALIZED:
205 		rc = sfc_configure(sa);
206 		break;
207 	default:
208 		sfc_err(sa, "unexpected adapter state %u to configure",
209 			sa->state);
210 		rc = EINVAL;
211 		break;
212 	}
213 	sfc_adapter_unlock(sa);
214 
215 	sfc_log_init(sa, "done %d", rc);
216 	SFC_ASSERT(rc >= 0);
217 	return -rc;
218 }
219 
220 static int
221 sfc_dev_start(struct rte_eth_dev *dev)
222 {
223 	struct sfc_adapter *sa = dev->data->dev_private;
224 	int rc;
225 
226 	sfc_log_init(sa, "entry");
227 
228 	sfc_adapter_lock(sa);
229 	rc = sfc_start(sa);
230 	sfc_adapter_unlock(sa);
231 
232 	sfc_log_init(sa, "done %d", rc);
233 	SFC_ASSERT(rc >= 0);
234 	return -rc;
235 }
236 
237 static int
238 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
239 {
240 	struct sfc_adapter *sa = dev->data->dev_private;
241 	struct rte_eth_link current_link;
242 	int ret;
243 
244 	sfc_log_init(sa, "entry");
245 
246 	if (sa->state != SFC_ADAPTER_STARTED) {
247 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
248 	} else if (wait_to_complete) {
249 		efx_link_mode_t link_mode;
250 
251 		if (efx_port_poll(sa->nic, &link_mode) != 0)
252 			link_mode = EFX_LINK_UNKNOWN;
253 		sfc_port_link_mode_to_info(link_mode, &current_link);
254 
255 	} else {
256 		sfc_ev_mgmt_qpoll(sa);
257 		rte_eth_linkstatus_get(dev, &current_link);
258 	}
259 
260 	ret = rte_eth_linkstatus_set(dev, &current_link);
261 	if (ret == 0)
262 		sfc_notice(sa, "Link status is %s",
263 			   current_link.link_status ? "UP" : "DOWN");
264 
265 	return ret;
266 }
267 
268 static void
269 sfc_dev_stop(struct rte_eth_dev *dev)
270 {
271 	struct sfc_adapter *sa = dev->data->dev_private;
272 
273 	sfc_log_init(sa, "entry");
274 
275 	sfc_adapter_lock(sa);
276 	sfc_stop(sa);
277 	sfc_adapter_unlock(sa);
278 
279 	sfc_log_init(sa, "done");
280 }
281 
282 static int
283 sfc_dev_set_link_up(struct rte_eth_dev *dev)
284 {
285 	struct sfc_adapter *sa = dev->data->dev_private;
286 	int rc;
287 
288 	sfc_log_init(sa, "entry");
289 
290 	sfc_adapter_lock(sa);
291 	rc = sfc_start(sa);
292 	sfc_adapter_unlock(sa);
293 
294 	SFC_ASSERT(rc >= 0);
295 	return -rc;
296 }
297 
298 static int
299 sfc_dev_set_link_down(struct rte_eth_dev *dev)
300 {
301 	struct sfc_adapter *sa = dev->data->dev_private;
302 
303 	sfc_log_init(sa, "entry");
304 
305 	sfc_adapter_lock(sa);
306 	sfc_stop(sa);
307 	sfc_adapter_unlock(sa);
308 
309 	return 0;
310 }
311 
312 static void
313 sfc_dev_close(struct rte_eth_dev *dev)
314 {
315 	struct sfc_adapter *sa = dev->data->dev_private;
316 
317 	sfc_log_init(sa, "entry");
318 
319 	sfc_adapter_lock(sa);
320 	switch (sa->state) {
321 	case SFC_ADAPTER_STARTED:
322 		sfc_stop(sa);
323 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
324 		/* FALLTHROUGH */
325 	case SFC_ADAPTER_CONFIGURED:
326 		sfc_close(sa);
327 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
328 		/* FALLTHROUGH */
329 	case SFC_ADAPTER_INITIALIZED:
330 		break;
331 	default:
332 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
333 		break;
334 	}
335 	sfc_adapter_unlock(sa);
336 
337 	sfc_log_init(sa, "done");
338 }
339 
340 static void
341 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
342 		   boolean_t enabled)
343 {
344 	struct sfc_port *port;
345 	boolean_t *toggle;
346 	struct sfc_adapter *sa = dev->data->dev_private;
347 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
348 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
349 
350 	sfc_adapter_lock(sa);
351 
352 	port = &sa->port;
353 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
354 
355 	if (*toggle != enabled) {
356 		*toggle = enabled;
357 
358 		if (port->isolated) {
359 			sfc_warn(sa, "isolated mode is active on the port");
360 			sfc_warn(sa, "the change is to be applied on the next "
361 				     "start provided that isolated mode is "
362 				     "disabled prior the next start");
363 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
364 			   (sfc_set_rx_mode(sa) != 0)) {
365 			*toggle = !(enabled);
366 			sfc_warn(sa, "Failed to %s %s mode",
367 				 ((enabled) ? "enable" : "disable"), desc);
368 		}
369 	}
370 
371 	sfc_adapter_unlock(sa);
372 }
373 
374 static void
375 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
376 {
377 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
378 }
379 
380 static void
381 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
382 {
383 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
384 }
385 
386 static void
387 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
388 {
389 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
390 }
391 
392 static void
393 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
394 {
395 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
396 }
397 
398 static int
399 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
400 		   uint16_t nb_rx_desc, unsigned int socket_id,
401 		   const struct rte_eth_rxconf *rx_conf,
402 		   struct rte_mempool *mb_pool)
403 {
404 	struct sfc_adapter *sa = dev->data->dev_private;
405 	int rc;
406 
407 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
408 		     rx_queue_id, nb_rx_desc, socket_id);
409 
410 	sfc_adapter_lock(sa);
411 
412 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
413 			  rx_conf, mb_pool);
414 	if (rc != 0)
415 		goto fail_rx_qinit;
416 
417 	dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
418 
419 	sfc_adapter_unlock(sa);
420 
421 	return 0;
422 
423 fail_rx_qinit:
424 	sfc_adapter_unlock(sa);
425 	SFC_ASSERT(rc > 0);
426 	return -rc;
427 }
428 
429 static void
430 sfc_rx_queue_release(void *queue)
431 {
432 	struct sfc_dp_rxq *dp_rxq = queue;
433 	struct sfc_rxq *rxq;
434 	struct sfc_adapter *sa;
435 	unsigned int sw_index;
436 
437 	if (dp_rxq == NULL)
438 		return;
439 
440 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
441 	sa = rxq->evq->sa;
442 	sfc_adapter_lock(sa);
443 
444 	sw_index = sfc_rxq_sw_index(rxq);
445 
446 	sfc_log_init(sa, "RxQ=%u", sw_index);
447 
448 	sa->eth_dev->data->rx_queues[sw_index] = NULL;
449 
450 	sfc_rx_qfini(sa, sw_index);
451 
452 	sfc_adapter_unlock(sa);
453 }
454 
455 static int
456 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
457 		   uint16_t nb_tx_desc, unsigned int socket_id,
458 		   const struct rte_eth_txconf *tx_conf)
459 {
460 	struct sfc_adapter *sa = dev->data->dev_private;
461 	int rc;
462 
463 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
464 		     tx_queue_id, nb_tx_desc, socket_id);
465 
466 	sfc_adapter_lock(sa);
467 
468 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
469 	if (rc != 0)
470 		goto fail_tx_qinit;
471 
472 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
473 
474 	sfc_adapter_unlock(sa);
475 	return 0;
476 
477 fail_tx_qinit:
478 	sfc_adapter_unlock(sa);
479 	SFC_ASSERT(rc > 0);
480 	return -rc;
481 }
482 
483 static void
484 sfc_tx_queue_release(void *queue)
485 {
486 	struct sfc_dp_txq *dp_txq = queue;
487 	struct sfc_txq *txq;
488 	unsigned int sw_index;
489 	struct sfc_adapter *sa;
490 
491 	if (dp_txq == NULL)
492 		return;
493 
494 	txq = sfc_txq_by_dp_txq(dp_txq);
495 	sw_index = sfc_txq_sw_index(txq);
496 
497 	SFC_ASSERT(txq->evq != NULL);
498 	sa = txq->evq->sa;
499 
500 	sfc_log_init(sa, "TxQ = %u", sw_index);
501 
502 	sfc_adapter_lock(sa);
503 
504 	SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
505 	sa->eth_dev->data->tx_queues[sw_index] = NULL;
506 
507 	sfc_tx_qfini(sa, sw_index);
508 
509 	sfc_adapter_unlock(sa);
510 }
511 
512 static int
513 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
514 {
515 	struct sfc_adapter *sa = dev->data->dev_private;
516 	struct sfc_port *port = &sa->port;
517 	uint64_t *mac_stats;
518 	int ret;
519 
520 	rte_spinlock_lock(&port->mac_stats_lock);
521 
522 	ret = sfc_port_update_mac_stats(sa);
523 	if (ret != 0)
524 		goto unlock;
525 
526 	mac_stats = port->mac_stats_buf;
527 
528 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
529 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
530 		stats->ipackets =
531 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
532 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
533 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
534 		stats->opackets =
535 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
536 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
537 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
538 		stats->ibytes =
539 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
540 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
541 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
542 		stats->obytes =
543 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
544 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
545 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
546 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
547 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
548 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
549 	} else {
550 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
551 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
552 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
553 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
554 		/*
555 		 * Take into account stats which are whenever supported
556 		 * on EF10. If some stat is not supported by current
557 		 * firmware variant or HW revision, it is guaranteed
558 		 * to be zero in mac_stats.
559 		 */
560 		stats->imissed =
561 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
562 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
563 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
564 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
565 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
566 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
567 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
568 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
569 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
570 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
571 		stats->ierrors =
572 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
573 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
574 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
575 		/* no oerrors counters supported on EF10 */
576 	}
577 
578 unlock:
579 	rte_spinlock_unlock(&port->mac_stats_lock);
580 	SFC_ASSERT(ret >= 0);
581 	return -ret;
582 }
583 
584 static void
585 sfc_stats_reset(struct rte_eth_dev *dev)
586 {
587 	struct sfc_adapter *sa = dev->data->dev_private;
588 	struct sfc_port *port = &sa->port;
589 	int rc;
590 
591 	if (sa->state != SFC_ADAPTER_STARTED) {
592 		/*
593 		 * The operation cannot be done if port is not started; it
594 		 * will be scheduled to be done during the next port start
595 		 */
596 		port->mac_stats_reset_pending = B_TRUE;
597 		return;
598 	}
599 
600 	rc = sfc_port_reset_mac_stats(sa);
601 	if (rc != 0)
602 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
603 }
604 
605 static int
606 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
607 	       unsigned int xstats_count)
608 {
609 	struct sfc_adapter *sa = dev->data->dev_private;
610 	struct sfc_port *port = &sa->port;
611 	uint64_t *mac_stats;
612 	int rc;
613 	unsigned int i;
614 	int nstats = 0;
615 
616 	rte_spinlock_lock(&port->mac_stats_lock);
617 
618 	rc = sfc_port_update_mac_stats(sa);
619 	if (rc != 0) {
620 		SFC_ASSERT(rc > 0);
621 		nstats = -rc;
622 		goto unlock;
623 	}
624 
625 	mac_stats = port->mac_stats_buf;
626 
627 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
628 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
629 			if (xstats != NULL && nstats < (int)xstats_count) {
630 				xstats[nstats].id = nstats;
631 				xstats[nstats].value = mac_stats[i];
632 			}
633 			nstats++;
634 		}
635 	}
636 
637 unlock:
638 	rte_spinlock_unlock(&port->mac_stats_lock);
639 
640 	return nstats;
641 }
642 
643 static int
644 sfc_xstats_get_names(struct rte_eth_dev *dev,
645 		     struct rte_eth_xstat_name *xstats_names,
646 		     unsigned int xstats_count)
647 {
648 	struct sfc_adapter *sa = dev->data->dev_private;
649 	struct sfc_port *port = &sa->port;
650 	unsigned int i;
651 	unsigned int nstats = 0;
652 
653 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
654 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
655 			if (xstats_names != NULL && nstats < xstats_count)
656 				strncpy(xstats_names[nstats].name,
657 					efx_mac_stat_name(sa->nic, i),
658 					sizeof(xstats_names[0].name));
659 			nstats++;
660 		}
661 	}
662 
663 	return nstats;
664 }
665 
666 static int
667 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
668 		     uint64_t *values, unsigned int n)
669 {
670 	struct sfc_adapter *sa = dev->data->dev_private;
671 	struct sfc_port *port = &sa->port;
672 	uint64_t *mac_stats;
673 	unsigned int nb_supported = 0;
674 	unsigned int nb_written = 0;
675 	unsigned int i;
676 	int ret;
677 	int rc;
678 
679 	if (unlikely(values == NULL) ||
680 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
681 		return port->mac_stats_nb_supported;
682 
683 	rte_spinlock_lock(&port->mac_stats_lock);
684 
685 	rc = sfc_port_update_mac_stats(sa);
686 	if (rc != 0) {
687 		SFC_ASSERT(rc > 0);
688 		ret = -rc;
689 		goto unlock;
690 	}
691 
692 	mac_stats = port->mac_stats_buf;
693 
694 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
695 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
696 			continue;
697 
698 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
699 			values[nb_written++] = mac_stats[i];
700 
701 		++nb_supported;
702 	}
703 
704 	ret = nb_written;
705 
706 unlock:
707 	rte_spinlock_unlock(&port->mac_stats_lock);
708 
709 	return ret;
710 }
711 
712 static int
713 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
714 			   struct rte_eth_xstat_name *xstats_names,
715 			   const uint64_t *ids, unsigned int size)
716 {
717 	struct sfc_adapter *sa = dev->data->dev_private;
718 	struct sfc_port *port = &sa->port;
719 	unsigned int nb_supported = 0;
720 	unsigned int nb_written = 0;
721 	unsigned int i;
722 
723 	if (unlikely(xstats_names == NULL) ||
724 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
725 		return port->mac_stats_nb_supported;
726 
727 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
728 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
729 			continue;
730 
731 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
732 			char *name = xstats_names[nb_written++].name;
733 
734 			strncpy(name, efx_mac_stat_name(sa->nic, i),
735 				sizeof(xstats_names[0].name));
736 			name[sizeof(xstats_names[0].name) - 1] = '\0';
737 		}
738 
739 		++nb_supported;
740 	}
741 
742 	return nb_written;
743 }
744 
745 static int
746 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
747 {
748 	struct sfc_adapter *sa = dev->data->dev_private;
749 	unsigned int wanted_fc, link_fc;
750 
751 	memset(fc_conf, 0, sizeof(*fc_conf));
752 
753 	sfc_adapter_lock(sa);
754 
755 	if (sa->state == SFC_ADAPTER_STARTED)
756 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
757 	else
758 		link_fc = sa->port.flow_ctrl;
759 
760 	switch (link_fc) {
761 	case 0:
762 		fc_conf->mode = RTE_FC_NONE;
763 		break;
764 	case EFX_FCNTL_RESPOND:
765 		fc_conf->mode = RTE_FC_RX_PAUSE;
766 		break;
767 	case EFX_FCNTL_GENERATE:
768 		fc_conf->mode = RTE_FC_TX_PAUSE;
769 		break;
770 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
771 		fc_conf->mode = RTE_FC_FULL;
772 		break;
773 	default:
774 		sfc_err(sa, "%s: unexpected flow control value %#x",
775 			__func__, link_fc);
776 	}
777 
778 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
779 
780 	sfc_adapter_unlock(sa);
781 
782 	return 0;
783 }
784 
785 static int
786 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
787 {
788 	struct sfc_adapter *sa = dev->data->dev_private;
789 	struct sfc_port *port = &sa->port;
790 	unsigned int fcntl;
791 	int rc;
792 
793 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
794 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
795 	    fc_conf->mac_ctrl_frame_fwd != 0) {
796 		sfc_err(sa, "unsupported flow control settings specified");
797 		rc = EINVAL;
798 		goto fail_inval;
799 	}
800 
801 	switch (fc_conf->mode) {
802 	case RTE_FC_NONE:
803 		fcntl = 0;
804 		break;
805 	case RTE_FC_RX_PAUSE:
806 		fcntl = EFX_FCNTL_RESPOND;
807 		break;
808 	case RTE_FC_TX_PAUSE:
809 		fcntl = EFX_FCNTL_GENERATE;
810 		break;
811 	case RTE_FC_FULL:
812 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
813 		break;
814 	default:
815 		rc = EINVAL;
816 		goto fail_inval;
817 	}
818 
819 	sfc_adapter_lock(sa);
820 
821 	if (sa->state == SFC_ADAPTER_STARTED) {
822 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
823 		if (rc != 0)
824 			goto fail_mac_fcntl_set;
825 	}
826 
827 	port->flow_ctrl = fcntl;
828 	port->flow_ctrl_autoneg = fc_conf->autoneg;
829 
830 	sfc_adapter_unlock(sa);
831 
832 	return 0;
833 
834 fail_mac_fcntl_set:
835 	sfc_adapter_unlock(sa);
836 fail_inval:
837 	SFC_ASSERT(rc > 0);
838 	return -rc;
839 }
840 
841 static int
842 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
843 {
844 	struct sfc_adapter *sa = dev->data->dev_private;
845 	size_t pdu = EFX_MAC_PDU(mtu);
846 	size_t old_pdu;
847 	int rc;
848 
849 	sfc_log_init(sa, "mtu=%u", mtu);
850 
851 	rc = EINVAL;
852 	if (pdu < EFX_MAC_PDU_MIN) {
853 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
854 			(unsigned int)mtu, (unsigned int)pdu,
855 			EFX_MAC_PDU_MIN);
856 		goto fail_inval;
857 	}
858 	if (pdu > EFX_MAC_PDU_MAX) {
859 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
860 			(unsigned int)mtu, (unsigned int)pdu,
861 			EFX_MAC_PDU_MAX);
862 		goto fail_inval;
863 	}
864 
865 	sfc_adapter_lock(sa);
866 
867 	if (pdu != sa->port.pdu) {
868 		if (sa->state == SFC_ADAPTER_STARTED) {
869 			sfc_stop(sa);
870 
871 			old_pdu = sa->port.pdu;
872 			sa->port.pdu = pdu;
873 			rc = sfc_start(sa);
874 			if (rc != 0)
875 				goto fail_start;
876 		} else {
877 			sa->port.pdu = pdu;
878 		}
879 	}
880 
881 	/*
882 	 * The driver does not use it, but other PMDs update jumbo_frame
883 	 * flag and max_rx_pkt_len when MTU is set.
884 	 */
885 	if (mtu > ETHER_MAX_LEN) {
886 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
887 
888 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
889 		rxmode->jumbo_frame = 1;
890 	}
891 
892 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
893 
894 	sfc_adapter_unlock(sa);
895 
896 	sfc_log_init(sa, "done");
897 	return 0;
898 
899 fail_start:
900 	sa->port.pdu = old_pdu;
901 	if (sfc_start(sa) != 0)
902 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
903 			"PDU max size - port is stopped",
904 			(unsigned int)pdu, (unsigned int)old_pdu);
905 	sfc_adapter_unlock(sa);
906 
907 fail_inval:
908 	sfc_log_init(sa, "failed %d", rc);
909 	SFC_ASSERT(rc > 0);
910 	return -rc;
911 }
912 static void
913 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
914 {
915 	struct sfc_adapter *sa = dev->data->dev_private;
916 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
917 	struct sfc_port *port = &sa->port;
918 	int rc;
919 
920 	sfc_adapter_lock(sa);
921 
922 	/*
923 	 * Copy the address to the device private data so that
924 	 * it could be recalled in the case of adapter restart.
925 	 */
926 	ether_addr_copy(mac_addr, &port->default_mac_addr);
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_notice(sa, "the port is not started");
936 		sfc_notice(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, the new MAC address will be set
964 		 * by means of adapter restart. sfc_start() shall retrieve
965 		 * the new address from the device private data and set it.
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 	/*
975 	 * In the case of failure sa->port->default_mac_addr does not
976 	 * need rollback since no error code is returned, and the upper
977 	 * API will anyway update the external MAC address storage.
978 	 * To be consistent with that new value it is better to keep
979 	 * the device private value the same.
980 	 */
981 	sfc_adapter_unlock(sa);
982 }
983 
984 
985 static int
986 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
987 		     uint32_t nb_mc_addr)
988 {
989 	struct sfc_adapter *sa = dev->data->dev_private;
990 	struct sfc_port *port = &sa->port;
991 	uint8_t *mc_addrs = port->mcast_addrs;
992 	int rc;
993 	unsigned int i;
994 
995 	if (port->isolated) {
996 		sfc_err(sa, "isolated mode is active on the port");
997 		sfc_err(sa, "will not set multicast address list");
998 		return -ENOTSUP;
999 	}
1000 
1001 	if (mc_addrs == NULL)
1002 		return -ENOBUFS;
1003 
1004 	if (nb_mc_addr > port->max_mcast_addrs) {
1005 		sfc_err(sa, "too many multicast addresses: %u > %u",
1006 			 nb_mc_addr, port->max_mcast_addrs);
1007 		return -EINVAL;
1008 	}
1009 
1010 	for (i = 0; i < nb_mc_addr; ++i) {
1011 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1012 				 EFX_MAC_ADDR_LEN);
1013 		mc_addrs += EFX_MAC_ADDR_LEN;
1014 	}
1015 
1016 	port->nb_mcast_addrs = nb_mc_addr;
1017 
1018 	if (sa->state != SFC_ADAPTER_STARTED)
1019 		return 0;
1020 
1021 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1022 					port->nb_mcast_addrs);
1023 	if (rc != 0)
1024 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1025 
1026 	SFC_ASSERT(rc > 0);
1027 	return -rc;
1028 }
1029 
1030 /*
1031  * The function is used by the secondary process as well. It must not
1032  * use any process-local pointers from the adapter data.
1033  */
1034 static void
1035 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1036 		      struct rte_eth_rxq_info *qinfo)
1037 {
1038 	struct sfc_adapter *sa = dev->data->dev_private;
1039 	struct sfc_rxq_info *rxq_info;
1040 	struct sfc_rxq *rxq;
1041 
1042 	sfc_adapter_lock(sa);
1043 
1044 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1045 
1046 	rxq_info = &sa->rxq_info[rx_queue_id];
1047 	rxq = rxq_info->rxq;
1048 	SFC_ASSERT(rxq != NULL);
1049 
1050 	qinfo->mp = rxq->refill_mb_pool;
1051 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1052 	qinfo->conf.rx_drop_en = 1;
1053 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1054 	qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1055 			       DEV_RX_OFFLOAD_UDP_CKSUM |
1056 			       DEV_RX_OFFLOAD_TCP_CKSUM;
1057 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1058 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1059 		qinfo->scattered_rx = 1;
1060 	}
1061 	qinfo->nb_desc = rxq_info->entries;
1062 
1063 	sfc_adapter_unlock(sa);
1064 }
1065 
1066 /*
1067  * The function is used by the secondary process as well. It must not
1068  * use any process-local pointers from the adapter data.
1069  */
1070 static void
1071 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1072 		      struct rte_eth_txq_info *qinfo)
1073 {
1074 	struct sfc_adapter *sa = dev->data->dev_private;
1075 	struct sfc_txq_info *txq_info;
1076 
1077 	sfc_adapter_lock(sa);
1078 
1079 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1080 
1081 	txq_info = &sa->txq_info[tx_queue_id];
1082 	SFC_ASSERT(txq_info->txq != NULL);
1083 
1084 	memset(qinfo, 0, sizeof(*qinfo));
1085 
1086 	qinfo->conf.txq_flags = txq_info->txq->flags;
1087 	qinfo->conf.offloads = txq_info->txq->offloads;
1088 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1089 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1090 	qinfo->nb_desc = txq_info->entries;
1091 
1092 	sfc_adapter_unlock(sa);
1093 }
1094 
1095 static uint32_t
1096 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1097 {
1098 	struct sfc_adapter *sa = dev->data->dev_private;
1099 
1100 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1101 
1102 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
1103 }
1104 
1105 static int
1106 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1107 {
1108 	struct sfc_dp_rxq *dp_rxq = queue;
1109 
1110 	return sfc_rx_qdesc_done(dp_rxq, offset);
1111 }
1112 
1113 static int
1114 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1115 {
1116 	struct sfc_dp_rxq *dp_rxq = queue;
1117 	struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1118 
1119 	return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1120 }
1121 
1122 static int
1123 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1124 {
1125 	struct sfc_dp_txq *dp_txq = queue;
1126 	struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1127 
1128 	return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1129 }
1130 
1131 static int
1132 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1133 {
1134 	struct sfc_adapter *sa = dev->data->dev_private;
1135 	int rc;
1136 
1137 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1138 
1139 	sfc_adapter_lock(sa);
1140 
1141 	rc = EINVAL;
1142 	if (sa->state != SFC_ADAPTER_STARTED)
1143 		goto fail_not_started;
1144 
1145 	rc = sfc_rx_qstart(sa, rx_queue_id);
1146 	if (rc != 0)
1147 		goto fail_rx_qstart;
1148 
1149 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1150 
1151 	sfc_adapter_unlock(sa);
1152 
1153 	return 0;
1154 
1155 fail_rx_qstart:
1156 fail_not_started:
1157 	sfc_adapter_unlock(sa);
1158 	SFC_ASSERT(rc > 0);
1159 	return -rc;
1160 }
1161 
1162 static int
1163 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1164 {
1165 	struct sfc_adapter *sa = dev->data->dev_private;
1166 
1167 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1168 
1169 	sfc_adapter_lock(sa);
1170 	sfc_rx_qstop(sa, rx_queue_id);
1171 
1172 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1173 
1174 	sfc_adapter_unlock(sa);
1175 
1176 	return 0;
1177 }
1178 
1179 static int
1180 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1181 {
1182 	struct sfc_adapter *sa = dev->data->dev_private;
1183 	int rc;
1184 
1185 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1186 
1187 	sfc_adapter_lock(sa);
1188 
1189 	rc = EINVAL;
1190 	if (sa->state != SFC_ADAPTER_STARTED)
1191 		goto fail_not_started;
1192 
1193 	rc = sfc_tx_qstart(sa, tx_queue_id);
1194 	if (rc != 0)
1195 		goto fail_tx_qstart;
1196 
1197 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1198 
1199 	sfc_adapter_unlock(sa);
1200 	return 0;
1201 
1202 fail_tx_qstart:
1203 
1204 fail_not_started:
1205 	sfc_adapter_unlock(sa);
1206 	SFC_ASSERT(rc > 0);
1207 	return -rc;
1208 }
1209 
1210 static int
1211 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1212 {
1213 	struct sfc_adapter *sa = dev->data->dev_private;
1214 
1215 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1216 
1217 	sfc_adapter_lock(sa);
1218 
1219 	sfc_tx_qstop(sa, tx_queue_id);
1220 
1221 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1222 
1223 	sfc_adapter_unlock(sa);
1224 	return 0;
1225 }
1226 
1227 static efx_tunnel_protocol_t
1228 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1229 {
1230 	switch (rte_type) {
1231 	case RTE_TUNNEL_TYPE_VXLAN:
1232 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1233 	case RTE_TUNNEL_TYPE_GENEVE:
1234 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1235 	default:
1236 		return EFX_TUNNEL_NPROTOS;
1237 	}
1238 }
1239 
1240 enum sfc_udp_tunnel_op_e {
1241 	SFC_UDP_TUNNEL_ADD_PORT,
1242 	SFC_UDP_TUNNEL_DEL_PORT,
1243 };
1244 
1245 static int
1246 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1247 		      struct rte_eth_udp_tunnel *tunnel_udp,
1248 		      enum sfc_udp_tunnel_op_e op)
1249 {
1250 	struct sfc_adapter *sa = dev->data->dev_private;
1251 	efx_tunnel_protocol_t tunnel_proto;
1252 	int rc;
1253 
1254 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1255 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1256 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1257 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1258 
1259 	tunnel_proto =
1260 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1261 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1262 		rc = ENOTSUP;
1263 		goto fail_bad_proto;
1264 	}
1265 
1266 	sfc_adapter_lock(sa);
1267 
1268 	switch (op) {
1269 	case SFC_UDP_TUNNEL_ADD_PORT:
1270 		rc = efx_tunnel_config_udp_add(sa->nic,
1271 					       tunnel_udp->udp_port,
1272 					       tunnel_proto);
1273 		break;
1274 	case SFC_UDP_TUNNEL_DEL_PORT:
1275 		rc = efx_tunnel_config_udp_remove(sa->nic,
1276 						  tunnel_udp->udp_port,
1277 						  tunnel_proto);
1278 		break;
1279 	default:
1280 		rc = EINVAL;
1281 		goto fail_bad_op;
1282 	}
1283 
1284 	if (rc != 0)
1285 		goto fail_op;
1286 
1287 	if (sa->state == SFC_ADAPTER_STARTED) {
1288 		rc = efx_tunnel_reconfigure(sa->nic);
1289 		if (rc == EAGAIN) {
1290 			/*
1291 			 * Configuration is accepted by FW and MC reboot
1292 			 * is initiated to apply the changes. MC reboot
1293 			 * will be handled in a usual way (MC reboot
1294 			 * event on management event queue and adapter
1295 			 * restart).
1296 			 */
1297 			rc = 0;
1298 		} else if (rc != 0) {
1299 			goto fail_reconfigure;
1300 		}
1301 	}
1302 
1303 	sfc_adapter_unlock(sa);
1304 	return 0;
1305 
1306 fail_reconfigure:
1307 	/* Remove/restore entry since the change makes the trouble */
1308 	switch (op) {
1309 	case SFC_UDP_TUNNEL_ADD_PORT:
1310 		(void)efx_tunnel_config_udp_remove(sa->nic,
1311 						   tunnel_udp->udp_port,
1312 						   tunnel_proto);
1313 		break;
1314 	case SFC_UDP_TUNNEL_DEL_PORT:
1315 		(void)efx_tunnel_config_udp_add(sa->nic,
1316 						tunnel_udp->udp_port,
1317 						tunnel_proto);
1318 		break;
1319 	}
1320 
1321 fail_op:
1322 fail_bad_op:
1323 	sfc_adapter_unlock(sa);
1324 
1325 fail_bad_proto:
1326 	SFC_ASSERT(rc > 0);
1327 	return -rc;
1328 }
1329 
1330 static int
1331 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1332 			    struct rte_eth_udp_tunnel *tunnel_udp)
1333 {
1334 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1335 }
1336 
1337 static int
1338 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1339 			    struct rte_eth_udp_tunnel *tunnel_udp)
1340 {
1341 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1342 }
1343 
1344 #if EFSYS_OPT_RX_SCALE
1345 static int
1346 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1347 			  struct rte_eth_rss_conf *rss_conf)
1348 {
1349 	struct sfc_adapter *sa = dev->data->dev_private;
1350 	struct sfc_port *port = &sa->port;
1351 
1352 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1353 		return -ENOTSUP;
1354 
1355 	if (sa->rss_channels == 0)
1356 		return -EINVAL;
1357 
1358 	sfc_adapter_lock(sa);
1359 
1360 	/*
1361 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1362 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1363 	 * flags which corresponds to the active EFX configuration stored
1364 	 * locally in 'sfc_adapter' and kept up-to-date
1365 	 */
1366 	rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1367 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1368 	if (rss_conf->rss_key != NULL)
1369 		rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1370 
1371 	sfc_adapter_unlock(sa);
1372 
1373 	return 0;
1374 }
1375 
1376 static int
1377 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1378 			struct rte_eth_rss_conf *rss_conf)
1379 {
1380 	struct sfc_adapter *sa = dev->data->dev_private;
1381 	struct sfc_port *port = &sa->port;
1382 	unsigned int efx_hash_types;
1383 	int rc = 0;
1384 
1385 	if (port->isolated)
1386 		return -ENOTSUP;
1387 
1388 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1389 		sfc_err(sa, "RSS is not available");
1390 		return -ENOTSUP;
1391 	}
1392 
1393 	if (sa->rss_channels == 0) {
1394 		sfc_err(sa, "RSS is not configured");
1395 		return -EINVAL;
1396 	}
1397 
1398 	if ((rss_conf->rss_key != NULL) &&
1399 	    (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1400 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1401 			sizeof(sa->rss_key));
1402 		return -EINVAL;
1403 	}
1404 
1405 	if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1406 		sfc_err(sa, "unsupported hash functions requested");
1407 		return -EINVAL;
1408 	}
1409 
1410 	sfc_adapter_lock(sa);
1411 
1412 	efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1413 
1414 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1415 				   EFX_RX_HASHALG_TOEPLITZ,
1416 				   efx_hash_types, B_TRUE);
1417 	if (rc != 0)
1418 		goto fail_scale_mode_set;
1419 
1420 	if (rss_conf->rss_key != NULL) {
1421 		if (sa->state == SFC_ADAPTER_STARTED) {
1422 			rc = efx_rx_scale_key_set(sa->nic,
1423 						  EFX_RSS_CONTEXT_DEFAULT,
1424 						  rss_conf->rss_key,
1425 						  sizeof(sa->rss_key));
1426 			if (rc != 0)
1427 				goto fail_scale_key_set;
1428 		}
1429 
1430 		rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1431 	}
1432 
1433 	sa->rss_hash_types = efx_hash_types;
1434 
1435 	sfc_adapter_unlock(sa);
1436 
1437 	return 0;
1438 
1439 fail_scale_key_set:
1440 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1441 				  EFX_RX_HASHALG_TOEPLITZ,
1442 				  sa->rss_hash_types, B_TRUE) != 0)
1443 		sfc_err(sa, "failed to restore RSS mode");
1444 
1445 fail_scale_mode_set:
1446 	sfc_adapter_unlock(sa);
1447 	return -rc;
1448 }
1449 
1450 static int
1451 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1452 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1453 		       uint16_t reta_size)
1454 {
1455 	struct sfc_adapter *sa = dev->data->dev_private;
1456 	struct sfc_port *port = &sa->port;
1457 	int entry;
1458 
1459 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1460 		return -ENOTSUP;
1461 
1462 	if (sa->rss_channels == 0)
1463 		return -EINVAL;
1464 
1465 	if (reta_size != EFX_RSS_TBL_SIZE)
1466 		return -EINVAL;
1467 
1468 	sfc_adapter_lock(sa);
1469 
1470 	for (entry = 0; entry < reta_size; entry++) {
1471 		int grp = entry / RTE_RETA_GROUP_SIZE;
1472 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1473 
1474 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1475 			reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1476 	}
1477 
1478 	sfc_adapter_unlock(sa);
1479 
1480 	return 0;
1481 }
1482 
1483 static int
1484 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1485 			struct rte_eth_rss_reta_entry64 *reta_conf,
1486 			uint16_t reta_size)
1487 {
1488 	struct sfc_adapter *sa = dev->data->dev_private;
1489 	struct sfc_port *port = &sa->port;
1490 	unsigned int *rss_tbl_new;
1491 	uint16_t entry;
1492 	int rc = 0;
1493 
1494 
1495 	if (port->isolated)
1496 		return -ENOTSUP;
1497 
1498 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1499 		sfc_err(sa, "RSS is not available");
1500 		return -ENOTSUP;
1501 	}
1502 
1503 	if (sa->rss_channels == 0) {
1504 		sfc_err(sa, "RSS is not configured");
1505 		return -EINVAL;
1506 	}
1507 
1508 	if (reta_size != EFX_RSS_TBL_SIZE) {
1509 		sfc_err(sa, "RETA size is wrong (should be %u)",
1510 			EFX_RSS_TBL_SIZE);
1511 		return -EINVAL;
1512 	}
1513 
1514 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1515 	if (rss_tbl_new == NULL)
1516 		return -ENOMEM;
1517 
1518 	sfc_adapter_lock(sa);
1519 
1520 	rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1521 
1522 	for (entry = 0; entry < reta_size; entry++) {
1523 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1524 		struct rte_eth_rss_reta_entry64 *grp;
1525 
1526 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1527 
1528 		if (grp->mask & (1ull << grp_idx)) {
1529 			if (grp->reta[grp_idx] >= sa->rss_channels) {
1530 				rc = EINVAL;
1531 				goto bad_reta_entry;
1532 			}
1533 			rss_tbl_new[entry] = grp->reta[grp_idx];
1534 		}
1535 	}
1536 
1537 	if (sa->state == SFC_ADAPTER_STARTED) {
1538 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1539 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1540 		if (rc != 0)
1541 			goto fail_scale_tbl_set;
1542 	}
1543 
1544 	rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1545 
1546 fail_scale_tbl_set:
1547 bad_reta_entry:
1548 	sfc_adapter_unlock(sa);
1549 
1550 	rte_free(rss_tbl_new);
1551 
1552 	SFC_ASSERT(rc >= 0);
1553 	return -rc;
1554 }
1555 #endif
1556 
1557 static int
1558 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1559 		    enum rte_filter_op filter_op,
1560 		    void *arg)
1561 {
1562 	struct sfc_adapter *sa = dev->data->dev_private;
1563 	int rc = ENOTSUP;
1564 
1565 	sfc_log_init(sa, "entry");
1566 
1567 	switch (filter_type) {
1568 	case RTE_ETH_FILTER_NONE:
1569 		sfc_err(sa, "Global filters configuration not supported");
1570 		break;
1571 	case RTE_ETH_FILTER_MACVLAN:
1572 		sfc_err(sa, "MACVLAN filters not supported");
1573 		break;
1574 	case RTE_ETH_FILTER_ETHERTYPE:
1575 		sfc_err(sa, "EtherType filters not supported");
1576 		break;
1577 	case RTE_ETH_FILTER_FLEXIBLE:
1578 		sfc_err(sa, "Flexible filters not supported");
1579 		break;
1580 	case RTE_ETH_FILTER_SYN:
1581 		sfc_err(sa, "SYN filters not supported");
1582 		break;
1583 	case RTE_ETH_FILTER_NTUPLE:
1584 		sfc_err(sa, "NTUPLE filters not supported");
1585 		break;
1586 	case RTE_ETH_FILTER_TUNNEL:
1587 		sfc_err(sa, "Tunnel filters not supported");
1588 		break;
1589 	case RTE_ETH_FILTER_FDIR:
1590 		sfc_err(sa, "Flow Director filters not supported");
1591 		break;
1592 	case RTE_ETH_FILTER_HASH:
1593 		sfc_err(sa, "Hash filters not supported");
1594 		break;
1595 	case RTE_ETH_FILTER_GENERIC:
1596 		if (filter_op != RTE_ETH_FILTER_GET) {
1597 			rc = EINVAL;
1598 		} else {
1599 			*(const void **)arg = &sfc_flow_ops;
1600 			rc = 0;
1601 		}
1602 		break;
1603 	default:
1604 		sfc_err(sa, "Unknown filter type %u", filter_type);
1605 		break;
1606 	}
1607 
1608 	sfc_log_init(sa, "exit: %d", -rc);
1609 	SFC_ASSERT(rc >= 0);
1610 	return -rc;
1611 }
1612 
1613 static const struct eth_dev_ops sfc_eth_dev_ops = {
1614 	.dev_configure			= sfc_dev_configure,
1615 	.dev_start			= sfc_dev_start,
1616 	.dev_stop			= sfc_dev_stop,
1617 	.dev_set_link_up		= sfc_dev_set_link_up,
1618 	.dev_set_link_down		= sfc_dev_set_link_down,
1619 	.dev_close			= sfc_dev_close,
1620 	.promiscuous_enable		= sfc_dev_promisc_enable,
1621 	.promiscuous_disable		= sfc_dev_promisc_disable,
1622 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1623 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1624 	.link_update			= sfc_dev_link_update,
1625 	.stats_get			= sfc_stats_get,
1626 	.stats_reset			= sfc_stats_reset,
1627 	.xstats_get			= sfc_xstats_get,
1628 	.xstats_reset			= sfc_stats_reset,
1629 	.xstats_get_names		= sfc_xstats_get_names,
1630 	.dev_infos_get			= sfc_dev_infos_get,
1631 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1632 	.mtu_set			= sfc_dev_set_mtu,
1633 	.rx_queue_start			= sfc_rx_queue_start,
1634 	.rx_queue_stop			= sfc_rx_queue_stop,
1635 	.tx_queue_start			= sfc_tx_queue_start,
1636 	.tx_queue_stop			= sfc_tx_queue_stop,
1637 	.rx_queue_setup			= sfc_rx_queue_setup,
1638 	.rx_queue_release		= sfc_rx_queue_release,
1639 	.rx_queue_count			= sfc_rx_queue_count,
1640 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1641 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1642 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1643 	.tx_queue_setup			= sfc_tx_queue_setup,
1644 	.tx_queue_release		= sfc_tx_queue_release,
1645 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1646 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1647 	.mac_addr_set			= sfc_mac_addr_set,
1648 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1649 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1650 #if EFSYS_OPT_RX_SCALE
1651 	.reta_update			= sfc_dev_rss_reta_update,
1652 	.reta_query			= sfc_dev_rss_reta_query,
1653 	.rss_hash_update		= sfc_dev_rss_hash_update,
1654 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1655 #endif
1656 	.filter_ctrl			= sfc_dev_filter_ctrl,
1657 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1658 	.rxq_info_get			= sfc_rx_queue_info_get,
1659 	.txq_info_get			= sfc_tx_queue_info_get,
1660 	.fw_version_get			= sfc_fw_version_get,
1661 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1662 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1663 };
1664 
1665 /**
1666  * Duplicate a string in potentially shared memory required for
1667  * multi-process support.
1668  *
1669  * strdup() allocates from process-local heap/memory.
1670  */
1671 static char *
1672 sfc_strdup(const char *str)
1673 {
1674 	size_t size;
1675 	char *copy;
1676 
1677 	if (str == NULL)
1678 		return NULL;
1679 
1680 	size = strlen(str) + 1;
1681 	copy = rte_malloc(__func__, size, 0);
1682 	if (copy != NULL)
1683 		rte_memcpy(copy, str, size);
1684 
1685 	return copy;
1686 }
1687 
1688 static int
1689 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1690 {
1691 	struct sfc_adapter *sa = dev->data->dev_private;
1692 	unsigned int avail_caps = 0;
1693 	const char *rx_name = NULL;
1694 	const char *tx_name = NULL;
1695 	int rc;
1696 
1697 	switch (sa->family) {
1698 	case EFX_FAMILY_HUNTINGTON:
1699 	case EFX_FAMILY_MEDFORD:
1700 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1701 		break;
1702 	default:
1703 		break;
1704 	}
1705 
1706 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1707 				sfc_kvarg_string_handler, &rx_name);
1708 	if (rc != 0)
1709 		goto fail_kvarg_rx_datapath;
1710 
1711 	if (rx_name != NULL) {
1712 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1713 		if (sa->dp_rx == NULL) {
1714 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1715 			rc = ENOENT;
1716 			goto fail_dp_rx;
1717 		}
1718 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1719 			sfc_err(sa,
1720 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1721 				rx_name);
1722 			rc = EINVAL;
1723 			goto fail_dp_rx_caps;
1724 		}
1725 	} else {
1726 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1727 		if (sa->dp_rx == NULL) {
1728 			sfc_err(sa, "Rx datapath by caps %#x not found",
1729 				avail_caps);
1730 			rc = ENOENT;
1731 			goto fail_dp_rx;
1732 		}
1733 	}
1734 
1735 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1736 	if (sa->dp_rx_name == NULL) {
1737 		rc = ENOMEM;
1738 		goto fail_dp_rx_name;
1739 	}
1740 
1741 	sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1742 
1743 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1744 
1745 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1746 				sfc_kvarg_string_handler, &tx_name);
1747 	if (rc != 0)
1748 		goto fail_kvarg_tx_datapath;
1749 
1750 	if (tx_name != NULL) {
1751 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1752 		if (sa->dp_tx == NULL) {
1753 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1754 			rc = ENOENT;
1755 			goto fail_dp_tx;
1756 		}
1757 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1758 			sfc_err(sa,
1759 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1760 				tx_name);
1761 			rc = EINVAL;
1762 			goto fail_dp_tx_caps;
1763 		}
1764 	} else {
1765 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1766 		if (sa->dp_tx == NULL) {
1767 			sfc_err(sa, "Tx datapath by caps %#x not found",
1768 				avail_caps);
1769 			rc = ENOENT;
1770 			goto fail_dp_tx;
1771 		}
1772 	}
1773 
1774 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1775 	if (sa->dp_tx_name == NULL) {
1776 		rc = ENOMEM;
1777 		goto fail_dp_tx_name;
1778 	}
1779 
1780 	sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1781 
1782 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1783 
1784 	dev->dev_ops = &sfc_eth_dev_ops;
1785 
1786 	return 0;
1787 
1788 fail_dp_tx_name:
1789 fail_dp_tx_caps:
1790 	sa->dp_tx = NULL;
1791 
1792 fail_dp_tx:
1793 fail_kvarg_tx_datapath:
1794 	rte_free(sa->dp_rx_name);
1795 	sa->dp_rx_name = NULL;
1796 
1797 fail_dp_rx_name:
1798 fail_dp_rx_caps:
1799 	sa->dp_rx = NULL;
1800 
1801 fail_dp_rx:
1802 fail_kvarg_rx_datapath:
1803 	return rc;
1804 }
1805 
1806 static void
1807 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1808 {
1809 	struct sfc_adapter *sa = dev->data->dev_private;
1810 
1811 	dev->dev_ops = NULL;
1812 	dev->rx_pkt_burst = NULL;
1813 	dev->tx_pkt_burst = NULL;
1814 
1815 	rte_free(sa->dp_tx_name);
1816 	sa->dp_tx_name = NULL;
1817 	sa->dp_tx = NULL;
1818 
1819 	rte_free(sa->dp_rx_name);
1820 	sa->dp_rx_name = NULL;
1821 	sa->dp_rx = NULL;
1822 }
1823 
1824 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1825 	.rxq_info_get			= sfc_rx_queue_info_get,
1826 	.txq_info_get			= sfc_tx_queue_info_get,
1827 };
1828 
1829 static int
1830 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1831 {
1832 	/*
1833 	 * Device private data has really many process-local pointers.
1834 	 * Below code should be extremely careful to use data located
1835 	 * in shared memory only.
1836 	 */
1837 	struct sfc_adapter *sa = dev->data->dev_private;
1838 	const struct sfc_dp_rx *dp_rx;
1839 	const struct sfc_dp_tx *dp_tx;
1840 	int rc;
1841 
1842 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1843 	if (dp_rx == NULL) {
1844 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1845 		rc = ENOENT;
1846 		goto fail_dp_rx;
1847 	}
1848 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1849 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1850 			sa->dp_tx_name);
1851 		rc = EINVAL;
1852 		goto fail_dp_rx_multi_process;
1853 	}
1854 
1855 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1856 	if (dp_tx == NULL) {
1857 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1858 		rc = ENOENT;
1859 		goto fail_dp_tx;
1860 	}
1861 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1862 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1863 			sa->dp_tx_name);
1864 		rc = EINVAL;
1865 		goto fail_dp_tx_multi_process;
1866 	}
1867 
1868 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1869 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1870 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1871 
1872 	return 0;
1873 
1874 fail_dp_tx_multi_process:
1875 fail_dp_tx:
1876 fail_dp_rx_multi_process:
1877 fail_dp_rx:
1878 	return rc;
1879 }
1880 
1881 static void
1882 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1883 {
1884 	dev->dev_ops = NULL;
1885 	dev->tx_pkt_burst = NULL;
1886 	dev->rx_pkt_burst = NULL;
1887 }
1888 
1889 static void
1890 sfc_register_dp(void)
1891 {
1892 	/* Register once */
1893 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1894 		/* Prefer EF10 datapath */
1895 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1896 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1897 
1898 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1899 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1900 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1901 	}
1902 }
1903 
1904 static int
1905 sfc_eth_dev_init(struct rte_eth_dev *dev)
1906 {
1907 	struct sfc_adapter *sa = dev->data->dev_private;
1908 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1909 	int rc;
1910 	const efx_nic_cfg_t *encp;
1911 	const struct ether_addr *from;
1912 
1913 	sfc_register_dp();
1914 
1915 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1916 		return -sfc_eth_dev_secondary_set_ops(dev);
1917 
1918 	/* Required for logging */
1919 	sa->pci_addr = pci_dev->addr;
1920 	sa->port_id = dev->data->port_id;
1921 
1922 	sa->eth_dev = dev;
1923 
1924 	/* Copy PCI device info to the dev->data */
1925 	rte_eth_copy_pci_info(dev, pci_dev);
1926 
1927 	sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1928 						RTE_LOG_NOTICE);
1929 
1930 	rc = sfc_kvargs_parse(sa);
1931 	if (rc != 0)
1932 		goto fail_kvargs_parse;
1933 
1934 	sfc_log_init(sa, "entry");
1935 
1936 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1937 	if (dev->data->mac_addrs == NULL) {
1938 		rc = ENOMEM;
1939 		goto fail_mac_addrs;
1940 	}
1941 
1942 	sfc_adapter_lock_init(sa);
1943 	sfc_adapter_lock(sa);
1944 
1945 	sfc_log_init(sa, "probing");
1946 	rc = sfc_probe(sa);
1947 	if (rc != 0)
1948 		goto fail_probe;
1949 
1950 	sfc_log_init(sa, "set device ops");
1951 	rc = sfc_eth_dev_set_ops(dev);
1952 	if (rc != 0)
1953 		goto fail_set_ops;
1954 
1955 	sfc_log_init(sa, "attaching");
1956 	rc = sfc_attach(sa);
1957 	if (rc != 0)
1958 		goto fail_attach;
1959 
1960 	encp = efx_nic_cfg_get(sa->nic);
1961 
1962 	/*
1963 	 * The arguments are really reverse order in comparison to
1964 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1965 	 */
1966 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1967 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1968 
1969 	sfc_adapter_unlock(sa);
1970 
1971 	sfc_log_init(sa, "done");
1972 	return 0;
1973 
1974 fail_attach:
1975 	sfc_eth_dev_clear_ops(dev);
1976 
1977 fail_set_ops:
1978 	sfc_unprobe(sa);
1979 
1980 fail_probe:
1981 	sfc_adapter_unlock(sa);
1982 	sfc_adapter_lock_fini(sa);
1983 	rte_free(dev->data->mac_addrs);
1984 	dev->data->mac_addrs = NULL;
1985 
1986 fail_mac_addrs:
1987 	sfc_kvargs_cleanup(sa);
1988 
1989 fail_kvargs_parse:
1990 	sfc_log_init(sa, "failed %d", rc);
1991 	SFC_ASSERT(rc > 0);
1992 	return -rc;
1993 }
1994 
1995 static int
1996 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1997 {
1998 	struct sfc_adapter *sa;
1999 
2000 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2001 		sfc_eth_dev_secondary_clear_ops(dev);
2002 		return 0;
2003 	}
2004 
2005 	sa = dev->data->dev_private;
2006 	sfc_log_init(sa, "entry");
2007 
2008 	sfc_adapter_lock(sa);
2009 
2010 	sfc_eth_dev_clear_ops(dev);
2011 
2012 	sfc_detach(sa);
2013 	sfc_unprobe(sa);
2014 
2015 	rte_free(dev->data->mac_addrs);
2016 	dev->data->mac_addrs = NULL;
2017 
2018 	sfc_kvargs_cleanup(sa);
2019 
2020 	sfc_adapter_unlock(sa);
2021 	sfc_adapter_lock_fini(sa);
2022 
2023 	sfc_log_init(sa, "done");
2024 
2025 	/* Required for logging, so cleanup last */
2026 	sa->eth_dev = NULL;
2027 	return 0;
2028 }
2029 
2030 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2031 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2032 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2033 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2034 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2035 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2036 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2037 	{ .vendor_id = 0 /* sentinel */ }
2038 };
2039 
2040 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2041 	struct rte_pci_device *pci_dev)
2042 {
2043 	return rte_eth_dev_pci_generic_probe(pci_dev,
2044 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2045 }
2046 
2047 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2048 {
2049 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2050 }
2051 
2052 static struct rte_pci_driver sfc_efx_pmd = {
2053 	.id_table = pci_id_sfc_efx_map,
2054 	.drv_flags =
2055 		RTE_PCI_DRV_INTR_LSC |
2056 		RTE_PCI_DRV_NEED_MAPPING,
2057 	.probe = sfc_eth_dev_pci_probe,
2058 	.remove = sfc_eth_dev_pci_remove,
2059 };
2060 
2061 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2062 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2063 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2064 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2065 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2066 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2067 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2068 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2069 
2070 RTE_INIT(sfc_driver_register_logtype);
2071 static void
2072 sfc_driver_register_logtype(void)
2073 {
2074 	int ret;
2075 
2076 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2077 						   RTE_LOG_NOTICE);
2078 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2079 }
2080