xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision d0dcfe9824bb3386608834a1595f9154487ef24f)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
4  * All rights reserved.
5  *
6  * This software was jointly developed between OKTET Labs (under contract
7  * for Solarflare) and Solarflare Communications, Inc.
8  */
9 
10 #include <rte_dev.h>
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
16 #include <rte_string_fns.h>
17 
18 #include "efx.h"
19 
20 #include "sfc.h"
21 #include "sfc_debug.h"
22 #include "sfc_log.h"
23 #include "sfc_kvargs.h"
24 #include "sfc_ev.h"
25 #include "sfc_rx.h"
26 #include "sfc_tx.h"
27 #include "sfc_flow.h"
28 #include "sfc_dp.h"
29 #include "sfc_dp_rx.h"
30 
31 uint32_t sfc_logtype_driver;
32 
33 static struct sfc_dp_list sfc_dp_head =
34 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
35 
36 static int
37 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
38 {
39 	struct sfc_adapter *sa = dev->data->dev_private;
40 	efx_nic_fw_info_t enfi;
41 	int ret;
42 	int rc;
43 
44 	/*
45 	 * Return value of the callback is likely supposed to be
46 	 * equal to or greater than 0, nevertheless, if an error
47 	 * occurs, it will be desirable to pass it to the caller
48 	 */
49 	if ((fw_version == NULL) || (fw_size == 0))
50 		return -EINVAL;
51 
52 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
53 	if (rc != 0)
54 		return -rc;
55 
56 	ret = snprintf(fw_version, fw_size,
57 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
58 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
59 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
60 	if (ret < 0)
61 		return ret;
62 
63 	if (enfi.enfi_dpcpu_fw_ids_valid) {
64 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
65 		int ret_extra;
66 
67 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
68 				     fw_size - dpcpu_fw_ids_offset,
69 				     " rx%" PRIx16 " tx%" PRIx16,
70 				     enfi.enfi_rx_dpcpu_fw_id,
71 				     enfi.enfi_tx_dpcpu_fw_id);
72 		if (ret_extra < 0)
73 			return ret_extra;
74 
75 		ret += ret_extra;
76 	}
77 
78 	if (fw_size < (size_t)(++ret))
79 		return ret;
80 	else
81 		return 0;
82 }
83 
84 static void
85 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
86 {
87 	struct sfc_adapter *sa = dev->data->dev_private;
88 	struct sfc_rss *rss = &sa->rss;
89 	uint64_t txq_offloads_def = 0;
90 
91 	sfc_log_init(sa, "entry");
92 
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_25000FDX)
102 		dev_info->speed_capa |= ETH_LINK_SPEED_25G;
103 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
104 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
105 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_50000FDX)
106 		dev_info->speed_capa |= ETH_LINK_SPEED_50G;
107 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_100000FDX)
108 		dev_info->speed_capa |= ETH_LINK_SPEED_100G;
109 
110 	dev_info->max_rx_queues = sa->rxq_max;
111 	dev_info->max_tx_queues = sa->txq_max;
112 
113 	/* By default packets are dropped if no descriptors are available */
114 	dev_info->default_rxconf.rx_drop_en = 1;
115 
116 	dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
117 
118 	/*
119 	 * rx_offload_capa includes both device and queue offloads since
120 	 * the latter may be requested on a per device basis which makes
121 	 * sense when some offloads are needed to be set on all queues.
122 	 */
123 	dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
124 				    dev_info->rx_queue_offload_capa;
125 
126 	dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
127 
128 	/*
129 	 * tx_offload_capa includes both device and queue offloads since
130 	 * the latter may be requested on a per device basis which makes
131 	 * sense when some offloads are needed to be set on all queues.
132 	 */
133 	dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
134 				    dev_info->tx_queue_offload_capa;
135 
136 	if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
137 		txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
138 
139 	dev_info->default_txconf.offloads |= txq_offloads_def;
140 
141 	if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
142 		uint64_t rte_hf = 0;
143 		unsigned int i;
144 
145 		for (i = 0; i < rss->hf_map_nb_entries; ++i)
146 			rte_hf |= rss->hf_map[i].rte;
147 
148 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
149 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
150 		dev_info->flow_type_rss_offloads = rte_hf;
151 	}
152 
153 	/* Initialize to hardware limits */
154 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
155 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
156 	/* The RXQ hardware requires that the descriptor count is a power
157 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
158 	 */
159 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
160 
161 	/* Initialize to hardware limits */
162 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
163 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
164 	/*
165 	 * The TXQ hardware requires that the descriptor count is a power
166 	 * of 2, but tx_desc_lim cannot properly describe that constraint
167 	 */
168 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
169 
170 	if (sa->dp_rx->get_dev_info != NULL)
171 		sa->dp_rx->get_dev_info(dev_info);
172 	if (sa->dp_tx->get_dev_info != NULL)
173 		sa->dp_tx->get_dev_info(dev_info);
174 }
175 
176 static const uint32_t *
177 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
178 {
179 	struct sfc_adapter *sa = dev->data->dev_private;
180 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
181 	uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
182 
183 	return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
184 }
185 
186 static int
187 sfc_dev_configure(struct rte_eth_dev *dev)
188 {
189 	struct rte_eth_dev_data *dev_data = dev->data;
190 	struct sfc_adapter *sa = dev_data->dev_private;
191 	int rc;
192 
193 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
194 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
195 
196 	sfc_adapter_lock(sa);
197 	switch (sa->state) {
198 	case SFC_ADAPTER_CONFIGURED:
199 		/* FALLTHROUGH */
200 	case SFC_ADAPTER_INITIALIZED:
201 		rc = sfc_configure(sa);
202 		break;
203 	default:
204 		sfc_err(sa, "unexpected adapter state %u to configure",
205 			sa->state);
206 		rc = EINVAL;
207 		break;
208 	}
209 	sfc_adapter_unlock(sa);
210 
211 	sfc_log_init(sa, "done %d", rc);
212 	SFC_ASSERT(rc >= 0);
213 	return -rc;
214 }
215 
216 static int
217 sfc_dev_start(struct rte_eth_dev *dev)
218 {
219 	struct sfc_adapter *sa = dev->data->dev_private;
220 	int rc;
221 
222 	sfc_log_init(sa, "entry");
223 
224 	sfc_adapter_lock(sa);
225 	rc = sfc_start(sa);
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_link_update(struct rte_eth_dev *dev, int wait_to_complete)
235 {
236 	struct sfc_adapter *sa = dev->data->dev_private;
237 	struct rte_eth_link current_link;
238 	int ret;
239 
240 	sfc_log_init(sa, "entry");
241 
242 	if (sa->state != SFC_ADAPTER_STARTED) {
243 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
244 	} else if (wait_to_complete) {
245 		efx_link_mode_t link_mode;
246 
247 		if (efx_port_poll(sa->nic, &link_mode) != 0)
248 			link_mode = EFX_LINK_UNKNOWN;
249 		sfc_port_link_mode_to_info(link_mode, &current_link);
250 
251 	} else {
252 		sfc_ev_mgmt_qpoll(sa);
253 		rte_eth_linkstatus_get(dev, &current_link);
254 	}
255 
256 	ret = rte_eth_linkstatus_set(dev, &current_link);
257 	if (ret == 0)
258 		sfc_notice(sa, "Link status is %s",
259 			   current_link.link_status ? "UP" : "DOWN");
260 
261 	return ret;
262 }
263 
264 static void
265 sfc_dev_stop(struct rte_eth_dev *dev)
266 {
267 	struct sfc_adapter *sa = dev->data->dev_private;
268 
269 	sfc_log_init(sa, "entry");
270 
271 	sfc_adapter_lock(sa);
272 	sfc_stop(sa);
273 	sfc_adapter_unlock(sa);
274 
275 	sfc_log_init(sa, "done");
276 }
277 
278 static int
279 sfc_dev_set_link_up(struct rte_eth_dev *dev)
280 {
281 	struct sfc_adapter *sa = dev->data->dev_private;
282 	int rc;
283 
284 	sfc_log_init(sa, "entry");
285 
286 	sfc_adapter_lock(sa);
287 	rc = sfc_start(sa);
288 	sfc_adapter_unlock(sa);
289 
290 	SFC_ASSERT(rc >= 0);
291 	return -rc;
292 }
293 
294 static int
295 sfc_dev_set_link_down(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 	return 0;
306 }
307 
308 static void
309 sfc_dev_close(struct rte_eth_dev *dev)
310 {
311 	struct sfc_adapter *sa = dev->data->dev_private;
312 
313 	sfc_log_init(sa, "entry");
314 
315 	sfc_adapter_lock(sa);
316 	switch (sa->state) {
317 	case SFC_ADAPTER_STARTED:
318 		sfc_stop(sa);
319 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
320 		/* FALLTHROUGH */
321 	case SFC_ADAPTER_CONFIGURED:
322 		sfc_close(sa);
323 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
324 		/* FALLTHROUGH */
325 	case SFC_ADAPTER_INITIALIZED:
326 		break;
327 	default:
328 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
329 		break;
330 	}
331 	sfc_adapter_unlock(sa);
332 
333 	sfc_log_init(sa, "done");
334 }
335 
336 static void
337 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
338 		   boolean_t enabled)
339 {
340 	struct sfc_port *port;
341 	boolean_t *toggle;
342 	struct sfc_adapter *sa = dev->data->dev_private;
343 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
344 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
345 
346 	sfc_adapter_lock(sa);
347 
348 	port = &sa->port;
349 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
350 
351 	if (*toggle != enabled) {
352 		*toggle = enabled;
353 
354 		if (port->isolated) {
355 			sfc_warn(sa, "isolated mode is active on the port");
356 			sfc_warn(sa, "the change is to be applied on the next "
357 				     "start provided that isolated mode is "
358 				     "disabled prior the next start");
359 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
360 			   (sfc_set_rx_mode(sa) != 0)) {
361 			*toggle = !(enabled);
362 			sfc_warn(sa, "Failed to %s %s mode",
363 				 ((enabled) ? "enable" : "disable"), desc);
364 		}
365 	}
366 
367 	sfc_adapter_unlock(sa);
368 }
369 
370 static void
371 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
372 {
373 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
374 }
375 
376 static void
377 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
378 {
379 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
380 }
381 
382 static void
383 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
384 {
385 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
386 }
387 
388 static void
389 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
390 {
391 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
392 }
393 
394 static int
395 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
396 		   uint16_t nb_rx_desc, unsigned int socket_id,
397 		   const struct rte_eth_rxconf *rx_conf,
398 		   struct rte_mempool *mb_pool)
399 {
400 	struct sfc_adapter *sa = dev->data->dev_private;
401 	int rc;
402 
403 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
404 		     rx_queue_id, nb_rx_desc, socket_id);
405 
406 	sfc_adapter_lock(sa);
407 
408 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
409 			  rx_conf, mb_pool);
410 	if (rc != 0)
411 		goto fail_rx_qinit;
412 
413 	dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
414 
415 	sfc_adapter_unlock(sa);
416 
417 	return 0;
418 
419 fail_rx_qinit:
420 	sfc_adapter_unlock(sa);
421 	SFC_ASSERT(rc > 0);
422 	return -rc;
423 }
424 
425 static void
426 sfc_rx_queue_release(void *queue)
427 {
428 	struct sfc_dp_rxq *dp_rxq = queue;
429 	struct sfc_rxq *rxq;
430 	struct sfc_adapter *sa;
431 	unsigned int sw_index;
432 
433 	if (dp_rxq == NULL)
434 		return;
435 
436 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
437 	sa = rxq->evq->sa;
438 	sfc_adapter_lock(sa);
439 
440 	sw_index = sfc_rxq_sw_index(rxq);
441 
442 	sfc_log_init(sa, "RxQ=%u", sw_index);
443 
444 	sa->eth_dev->data->rx_queues[sw_index] = NULL;
445 
446 	sfc_rx_qfini(sa, sw_index);
447 
448 	sfc_adapter_unlock(sa);
449 }
450 
451 static int
452 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
453 		   uint16_t nb_tx_desc, unsigned int socket_id,
454 		   const struct rte_eth_txconf *tx_conf)
455 {
456 	struct sfc_adapter *sa = dev->data->dev_private;
457 	int rc;
458 
459 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
460 		     tx_queue_id, nb_tx_desc, socket_id);
461 
462 	sfc_adapter_lock(sa);
463 
464 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
465 	if (rc != 0)
466 		goto fail_tx_qinit;
467 
468 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
469 
470 	sfc_adapter_unlock(sa);
471 	return 0;
472 
473 fail_tx_qinit:
474 	sfc_adapter_unlock(sa);
475 	SFC_ASSERT(rc > 0);
476 	return -rc;
477 }
478 
479 static void
480 sfc_tx_queue_release(void *queue)
481 {
482 	struct sfc_dp_txq *dp_txq = queue;
483 	struct sfc_txq *txq;
484 	unsigned int sw_index;
485 	struct sfc_adapter *sa;
486 
487 	if (dp_txq == NULL)
488 		return;
489 
490 	txq = sfc_txq_by_dp_txq(dp_txq);
491 	sw_index = sfc_txq_sw_index(txq);
492 
493 	SFC_ASSERT(txq->evq != NULL);
494 	sa = txq->evq->sa;
495 
496 	sfc_log_init(sa, "TxQ = %u", sw_index);
497 
498 	sfc_adapter_lock(sa);
499 
500 	SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
501 	sa->eth_dev->data->tx_queues[sw_index] = NULL;
502 
503 	sfc_tx_qfini(sa, sw_index);
504 
505 	sfc_adapter_unlock(sa);
506 }
507 
508 static int
509 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
510 {
511 	struct sfc_adapter *sa = dev->data->dev_private;
512 	struct sfc_port *port = &sa->port;
513 	uint64_t *mac_stats;
514 	int ret;
515 
516 	rte_spinlock_lock(&port->mac_stats_lock);
517 
518 	ret = sfc_port_update_mac_stats(sa);
519 	if (ret != 0)
520 		goto unlock;
521 
522 	mac_stats = port->mac_stats_buf;
523 
524 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
525 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
526 		stats->ipackets =
527 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
528 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
529 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
530 		stats->opackets =
531 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
532 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
533 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
534 		stats->ibytes =
535 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
536 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
537 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
538 		stats->obytes =
539 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
540 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
541 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
542 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
543 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
544 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
545 	} else {
546 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
547 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
548 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
549 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
550 		/*
551 		 * Take into account stats which are whenever supported
552 		 * on EF10. If some stat is not supported by current
553 		 * firmware variant or HW revision, it is guaranteed
554 		 * to be zero in mac_stats.
555 		 */
556 		stats->imissed =
557 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
558 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
559 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
560 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
561 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
562 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
563 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
564 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
565 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
566 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
567 		stats->ierrors =
568 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
569 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
570 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
571 		/* no oerrors counters supported on EF10 */
572 	}
573 
574 unlock:
575 	rte_spinlock_unlock(&port->mac_stats_lock);
576 	SFC_ASSERT(ret >= 0);
577 	return -ret;
578 }
579 
580 static void
581 sfc_stats_reset(struct rte_eth_dev *dev)
582 {
583 	struct sfc_adapter *sa = dev->data->dev_private;
584 	struct sfc_port *port = &sa->port;
585 	int rc;
586 
587 	if (sa->state != SFC_ADAPTER_STARTED) {
588 		/*
589 		 * The operation cannot be done if port is not started; it
590 		 * will be scheduled to be done during the next port start
591 		 */
592 		port->mac_stats_reset_pending = B_TRUE;
593 		return;
594 	}
595 
596 	rc = sfc_port_reset_mac_stats(sa);
597 	if (rc != 0)
598 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
599 }
600 
601 static int
602 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
603 	       unsigned int xstats_count)
604 {
605 	struct sfc_adapter *sa = dev->data->dev_private;
606 	struct sfc_port *port = &sa->port;
607 	uint64_t *mac_stats;
608 	int rc;
609 	unsigned int i;
610 	int nstats = 0;
611 
612 	rte_spinlock_lock(&port->mac_stats_lock);
613 
614 	rc = sfc_port_update_mac_stats(sa);
615 	if (rc != 0) {
616 		SFC_ASSERT(rc > 0);
617 		nstats = -rc;
618 		goto unlock;
619 	}
620 
621 	mac_stats = port->mac_stats_buf;
622 
623 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
624 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
625 			if (xstats != NULL && nstats < (int)xstats_count) {
626 				xstats[nstats].id = nstats;
627 				xstats[nstats].value = mac_stats[i];
628 			}
629 			nstats++;
630 		}
631 	}
632 
633 unlock:
634 	rte_spinlock_unlock(&port->mac_stats_lock);
635 
636 	return nstats;
637 }
638 
639 static int
640 sfc_xstats_get_names(struct rte_eth_dev *dev,
641 		     struct rte_eth_xstat_name *xstats_names,
642 		     unsigned int xstats_count)
643 {
644 	struct sfc_adapter *sa = dev->data->dev_private;
645 	struct sfc_port *port = &sa->port;
646 	unsigned int i;
647 	unsigned int nstats = 0;
648 
649 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
650 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
651 			if (xstats_names != NULL && nstats < xstats_count)
652 				strlcpy(xstats_names[nstats].name,
653 					efx_mac_stat_name(sa->nic, i),
654 					sizeof(xstats_names[0].name));
655 			nstats++;
656 		}
657 	}
658 
659 	return nstats;
660 }
661 
662 static int
663 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
664 		     uint64_t *values, unsigned int n)
665 {
666 	struct sfc_adapter *sa = dev->data->dev_private;
667 	struct sfc_port *port = &sa->port;
668 	uint64_t *mac_stats;
669 	unsigned int nb_supported = 0;
670 	unsigned int nb_written = 0;
671 	unsigned int i;
672 	int ret;
673 	int rc;
674 
675 	if (unlikely(values == NULL) ||
676 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
677 		return port->mac_stats_nb_supported;
678 
679 	rte_spinlock_lock(&port->mac_stats_lock);
680 
681 	rc = sfc_port_update_mac_stats(sa);
682 	if (rc != 0) {
683 		SFC_ASSERT(rc > 0);
684 		ret = -rc;
685 		goto unlock;
686 	}
687 
688 	mac_stats = port->mac_stats_buf;
689 
690 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
691 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
692 			continue;
693 
694 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
695 			values[nb_written++] = mac_stats[i];
696 
697 		++nb_supported;
698 	}
699 
700 	ret = nb_written;
701 
702 unlock:
703 	rte_spinlock_unlock(&port->mac_stats_lock);
704 
705 	return ret;
706 }
707 
708 static int
709 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
710 			   struct rte_eth_xstat_name *xstats_names,
711 			   const uint64_t *ids, unsigned int size)
712 {
713 	struct sfc_adapter *sa = dev->data->dev_private;
714 	struct sfc_port *port = &sa->port;
715 	unsigned int nb_supported = 0;
716 	unsigned int nb_written = 0;
717 	unsigned int i;
718 
719 	if (unlikely(xstats_names == NULL) ||
720 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
721 		return port->mac_stats_nb_supported;
722 
723 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
724 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
725 			continue;
726 
727 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
728 			char *name = xstats_names[nb_written++].name;
729 
730 			strlcpy(name, efx_mac_stat_name(sa->nic, i),
731 				sizeof(xstats_names[0].name));
732 		}
733 
734 		++nb_supported;
735 	}
736 
737 	return nb_written;
738 }
739 
740 static int
741 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
742 {
743 	struct sfc_adapter *sa = dev->data->dev_private;
744 	unsigned int wanted_fc, link_fc;
745 
746 	memset(fc_conf, 0, sizeof(*fc_conf));
747 
748 	sfc_adapter_lock(sa);
749 
750 	if (sa->state == SFC_ADAPTER_STARTED)
751 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
752 	else
753 		link_fc = sa->port.flow_ctrl;
754 
755 	switch (link_fc) {
756 	case 0:
757 		fc_conf->mode = RTE_FC_NONE;
758 		break;
759 	case EFX_FCNTL_RESPOND:
760 		fc_conf->mode = RTE_FC_RX_PAUSE;
761 		break;
762 	case EFX_FCNTL_GENERATE:
763 		fc_conf->mode = RTE_FC_TX_PAUSE;
764 		break;
765 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
766 		fc_conf->mode = RTE_FC_FULL;
767 		break;
768 	default:
769 		sfc_err(sa, "%s: unexpected flow control value %#x",
770 			__func__, link_fc);
771 	}
772 
773 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
774 
775 	sfc_adapter_unlock(sa);
776 
777 	return 0;
778 }
779 
780 static int
781 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
782 {
783 	struct sfc_adapter *sa = dev->data->dev_private;
784 	struct sfc_port *port = &sa->port;
785 	unsigned int fcntl;
786 	int rc;
787 
788 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
789 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
790 	    fc_conf->mac_ctrl_frame_fwd != 0) {
791 		sfc_err(sa, "unsupported flow control settings specified");
792 		rc = EINVAL;
793 		goto fail_inval;
794 	}
795 
796 	switch (fc_conf->mode) {
797 	case RTE_FC_NONE:
798 		fcntl = 0;
799 		break;
800 	case RTE_FC_RX_PAUSE:
801 		fcntl = EFX_FCNTL_RESPOND;
802 		break;
803 	case RTE_FC_TX_PAUSE:
804 		fcntl = EFX_FCNTL_GENERATE;
805 		break;
806 	case RTE_FC_FULL:
807 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
808 		break;
809 	default:
810 		rc = EINVAL;
811 		goto fail_inval;
812 	}
813 
814 	sfc_adapter_lock(sa);
815 
816 	if (sa->state == SFC_ADAPTER_STARTED) {
817 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
818 		if (rc != 0)
819 			goto fail_mac_fcntl_set;
820 	}
821 
822 	port->flow_ctrl = fcntl;
823 	port->flow_ctrl_autoneg = fc_conf->autoneg;
824 
825 	sfc_adapter_unlock(sa);
826 
827 	return 0;
828 
829 fail_mac_fcntl_set:
830 	sfc_adapter_unlock(sa);
831 fail_inval:
832 	SFC_ASSERT(rc > 0);
833 	return -rc;
834 }
835 
836 static int
837 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
838 {
839 	struct sfc_adapter *sa = dev->data->dev_private;
840 	size_t pdu = EFX_MAC_PDU(mtu);
841 	size_t old_pdu;
842 	int rc;
843 
844 	sfc_log_init(sa, "mtu=%u", mtu);
845 
846 	rc = EINVAL;
847 	if (pdu < EFX_MAC_PDU_MIN) {
848 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
849 			(unsigned int)mtu, (unsigned int)pdu,
850 			EFX_MAC_PDU_MIN);
851 		goto fail_inval;
852 	}
853 	if (pdu > EFX_MAC_PDU_MAX) {
854 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
855 			(unsigned int)mtu, (unsigned int)pdu,
856 			EFX_MAC_PDU_MAX);
857 		goto fail_inval;
858 	}
859 
860 	sfc_adapter_lock(sa);
861 
862 	if (pdu != sa->port.pdu) {
863 		if (sa->state == SFC_ADAPTER_STARTED) {
864 			sfc_stop(sa);
865 
866 			old_pdu = sa->port.pdu;
867 			sa->port.pdu = pdu;
868 			rc = sfc_start(sa);
869 			if (rc != 0)
870 				goto fail_start;
871 		} else {
872 			sa->port.pdu = pdu;
873 		}
874 	}
875 
876 	/*
877 	 * The driver does not use it, but other PMDs update jumbo frame
878 	 * flag and max_rx_pkt_len when MTU is set.
879 	 */
880 	if (mtu > ETHER_MAX_LEN) {
881 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
882 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
883 	}
884 
885 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
886 
887 	sfc_adapter_unlock(sa);
888 
889 	sfc_log_init(sa, "done");
890 	return 0;
891 
892 fail_start:
893 	sa->port.pdu = old_pdu;
894 	if (sfc_start(sa) != 0)
895 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
896 			"PDU max size - port is stopped",
897 			(unsigned int)pdu, (unsigned int)old_pdu);
898 	sfc_adapter_unlock(sa);
899 
900 fail_inval:
901 	sfc_log_init(sa, "failed %d", rc);
902 	SFC_ASSERT(rc > 0);
903 	return -rc;
904 }
905 static int
906 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
907 {
908 	struct sfc_adapter *sa = dev->data->dev_private;
909 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
910 	struct sfc_port *port = &sa->port;
911 	struct ether_addr *old_addr = &dev->data->mac_addrs[0];
912 	int rc = 0;
913 
914 	sfc_adapter_lock(sa);
915 
916 	/*
917 	 * Copy the address to the device private data so that
918 	 * it could be recalled in the case of adapter restart.
919 	 */
920 	ether_addr_copy(mac_addr, &port->default_mac_addr);
921 
922 	/*
923 	 * Neither of the two following checks can return
924 	 * an error. The new MAC address is preserved in
925 	 * the device private data and can be activated
926 	 * on the next port start if the user prevents
927 	 * isolated mode from being enabled.
928 	 */
929 	if (port->isolated) {
930 		sfc_warn(sa, "isolated mode is active on the port");
931 		sfc_warn(sa, "will not set MAC address");
932 		goto unlock;
933 	}
934 
935 	if (sa->state != SFC_ADAPTER_STARTED) {
936 		sfc_notice(sa, "the port is not started");
937 		sfc_notice(sa, "the new MAC address will be set on port start");
938 
939 		goto unlock;
940 	}
941 
942 	if (encp->enc_allow_set_mac_with_installed_filters) {
943 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
944 		if (rc != 0) {
945 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
946 			goto unlock;
947 		}
948 
949 		/*
950 		 * Changing the MAC address by means of MCDI request
951 		 * has no effect on received traffic, therefore
952 		 * we also need to update unicast filters
953 		 */
954 		rc = sfc_set_rx_mode(sa);
955 		if (rc != 0) {
956 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
957 			/* Rollback the old address */
958 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
959 			(void)sfc_set_rx_mode(sa);
960 		}
961 	} else {
962 		sfc_warn(sa, "cannot set MAC address with filters installed");
963 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
964 		sfc_warn(sa, "(some traffic may be dropped)");
965 
966 		/*
967 		 * Since setting MAC address with filters installed is not
968 		 * allowed on the adapter, the new MAC address will be set
969 		 * by means of adapter restart. sfc_start() shall retrieve
970 		 * the new address from the device private data and set it.
971 		 */
972 		sfc_stop(sa);
973 		rc = sfc_start(sa);
974 		if (rc != 0)
975 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
976 	}
977 
978 unlock:
979 	if (rc != 0)
980 		ether_addr_copy(old_addr, &port->default_mac_addr);
981 
982 	sfc_adapter_unlock(sa);
983 
984 	SFC_ASSERT(rc >= 0);
985 	return -rc;
986 }
987 
988 
989 static int
990 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
991 		     uint32_t nb_mc_addr)
992 {
993 	struct sfc_adapter *sa = dev->data->dev_private;
994 	struct sfc_port *port = &sa->port;
995 	uint8_t *mc_addrs = port->mcast_addrs;
996 	int rc;
997 	unsigned int i;
998 
999 	if (port->isolated) {
1000 		sfc_err(sa, "isolated mode is active on the port");
1001 		sfc_err(sa, "will not set multicast address list");
1002 		return -ENOTSUP;
1003 	}
1004 
1005 	if (mc_addrs == NULL)
1006 		return -ENOBUFS;
1007 
1008 	if (nb_mc_addr > port->max_mcast_addrs) {
1009 		sfc_err(sa, "too many multicast addresses: %u > %u",
1010 			 nb_mc_addr, port->max_mcast_addrs);
1011 		return -EINVAL;
1012 	}
1013 
1014 	for (i = 0; i < nb_mc_addr; ++i) {
1015 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1016 				 EFX_MAC_ADDR_LEN);
1017 		mc_addrs += EFX_MAC_ADDR_LEN;
1018 	}
1019 
1020 	port->nb_mcast_addrs = nb_mc_addr;
1021 
1022 	if (sa->state != SFC_ADAPTER_STARTED)
1023 		return 0;
1024 
1025 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1026 					port->nb_mcast_addrs);
1027 	if (rc != 0)
1028 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1029 
1030 	SFC_ASSERT(rc >= 0);
1031 	return -rc;
1032 }
1033 
1034 /*
1035  * The function is used by the secondary process as well. It must not
1036  * use any process-local pointers from the adapter data.
1037  */
1038 static void
1039 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1040 		      struct rte_eth_rxq_info *qinfo)
1041 {
1042 	struct sfc_adapter *sa = dev->data->dev_private;
1043 	struct sfc_rxq_info *rxq_info;
1044 	struct sfc_rxq *rxq;
1045 
1046 	sfc_adapter_lock(sa);
1047 
1048 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1049 
1050 	rxq_info = &sa->rxq_info[rx_queue_id];
1051 	rxq = rxq_info->rxq;
1052 	SFC_ASSERT(rxq != NULL);
1053 
1054 	qinfo->mp = rxq->refill_mb_pool;
1055 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1056 	qinfo->conf.rx_drop_en = 1;
1057 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1058 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1059 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1060 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1061 		qinfo->scattered_rx = 1;
1062 	}
1063 	qinfo->nb_desc = rxq_info->entries;
1064 
1065 	sfc_adapter_unlock(sa);
1066 }
1067 
1068 /*
1069  * The function is used by the secondary process as well. It must not
1070  * use any process-local pointers from the adapter data.
1071  */
1072 static void
1073 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1074 		      struct rte_eth_txq_info *qinfo)
1075 {
1076 	struct sfc_adapter *sa = dev->data->dev_private;
1077 	struct sfc_txq_info *txq_info;
1078 
1079 	sfc_adapter_lock(sa);
1080 
1081 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1082 
1083 	txq_info = &sa->txq_info[tx_queue_id];
1084 	SFC_ASSERT(txq_info->txq != NULL);
1085 
1086 	memset(qinfo, 0, sizeof(*qinfo));
1087 
1088 	qinfo->conf.offloads = txq_info->txq->offloads;
1089 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1090 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1091 	qinfo->nb_desc = txq_info->entries;
1092 
1093 	sfc_adapter_unlock(sa);
1094 }
1095 
1096 static uint32_t
1097 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1098 {
1099 	struct sfc_adapter *sa = dev->data->dev_private;
1100 
1101 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1102 
1103 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
1104 }
1105 
1106 static int
1107 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1108 {
1109 	struct sfc_dp_rxq *dp_rxq = queue;
1110 
1111 	return sfc_rx_qdesc_done(dp_rxq, offset);
1112 }
1113 
1114 static int
1115 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1116 {
1117 	struct sfc_dp_rxq *dp_rxq = queue;
1118 	struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1119 
1120 	return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1121 }
1122 
1123 static int
1124 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1125 {
1126 	struct sfc_dp_txq *dp_txq = queue;
1127 	struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1128 
1129 	return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1130 }
1131 
1132 static int
1133 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1134 {
1135 	struct sfc_adapter *sa = dev->data->dev_private;
1136 	int rc;
1137 
1138 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1139 
1140 	sfc_adapter_lock(sa);
1141 
1142 	rc = EINVAL;
1143 	if (sa->state != SFC_ADAPTER_STARTED)
1144 		goto fail_not_started;
1145 
1146 	rc = sfc_rx_qstart(sa, rx_queue_id);
1147 	if (rc != 0)
1148 		goto fail_rx_qstart;
1149 
1150 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1151 
1152 	sfc_adapter_unlock(sa);
1153 
1154 	return 0;
1155 
1156 fail_rx_qstart:
1157 fail_not_started:
1158 	sfc_adapter_unlock(sa);
1159 	SFC_ASSERT(rc > 0);
1160 	return -rc;
1161 }
1162 
1163 static int
1164 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1165 {
1166 	struct sfc_adapter *sa = dev->data->dev_private;
1167 
1168 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1169 
1170 	sfc_adapter_lock(sa);
1171 	sfc_rx_qstop(sa, rx_queue_id);
1172 
1173 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1174 
1175 	sfc_adapter_unlock(sa);
1176 
1177 	return 0;
1178 }
1179 
1180 static int
1181 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1182 {
1183 	struct sfc_adapter *sa = dev->data->dev_private;
1184 	int rc;
1185 
1186 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1187 
1188 	sfc_adapter_lock(sa);
1189 
1190 	rc = EINVAL;
1191 	if (sa->state != SFC_ADAPTER_STARTED)
1192 		goto fail_not_started;
1193 
1194 	rc = sfc_tx_qstart(sa, tx_queue_id);
1195 	if (rc != 0)
1196 		goto fail_tx_qstart;
1197 
1198 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1199 
1200 	sfc_adapter_unlock(sa);
1201 	return 0;
1202 
1203 fail_tx_qstart:
1204 
1205 fail_not_started:
1206 	sfc_adapter_unlock(sa);
1207 	SFC_ASSERT(rc > 0);
1208 	return -rc;
1209 }
1210 
1211 static int
1212 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1213 {
1214 	struct sfc_adapter *sa = dev->data->dev_private;
1215 
1216 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1217 
1218 	sfc_adapter_lock(sa);
1219 
1220 	sfc_tx_qstop(sa, tx_queue_id);
1221 
1222 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1223 
1224 	sfc_adapter_unlock(sa);
1225 	return 0;
1226 }
1227 
1228 static efx_tunnel_protocol_t
1229 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1230 {
1231 	switch (rte_type) {
1232 	case RTE_TUNNEL_TYPE_VXLAN:
1233 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1234 	case RTE_TUNNEL_TYPE_GENEVE:
1235 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1236 	default:
1237 		return EFX_TUNNEL_NPROTOS;
1238 	}
1239 }
1240 
1241 enum sfc_udp_tunnel_op_e {
1242 	SFC_UDP_TUNNEL_ADD_PORT,
1243 	SFC_UDP_TUNNEL_DEL_PORT,
1244 };
1245 
1246 static int
1247 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1248 		      struct rte_eth_udp_tunnel *tunnel_udp,
1249 		      enum sfc_udp_tunnel_op_e op)
1250 {
1251 	struct sfc_adapter *sa = dev->data->dev_private;
1252 	efx_tunnel_protocol_t tunnel_proto;
1253 	int rc;
1254 
1255 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1256 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1257 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1258 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1259 
1260 	tunnel_proto =
1261 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1262 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1263 		rc = ENOTSUP;
1264 		goto fail_bad_proto;
1265 	}
1266 
1267 	sfc_adapter_lock(sa);
1268 
1269 	switch (op) {
1270 	case SFC_UDP_TUNNEL_ADD_PORT:
1271 		rc = efx_tunnel_config_udp_add(sa->nic,
1272 					       tunnel_udp->udp_port,
1273 					       tunnel_proto);
1274 		break;
1275 	case SFC_UDP_TUNNEL_DEL_PORT:
1276 		rc = efx_tunnel_config_udp_remove(sa->nic,
1277 						  tunnel_udp->udp_port,
1278 						  tunnel_proto);
1279 		break;
1280 	default:
1281 		rc = EINVAL;
1282 		goto fail_bad_op;
1283 	}
1284 
1285 	if (rc != 0)
1286 		goto fail_op;
1287 
1288 	if (sa->state == SFC_ADAPTER_STARTED) {
1289 		rc = efx_tunnel_reconfigure(sa->nic);
1290 		if (rc == EAGAIN) {
1291 			/*
1292 			 * Configuration is accepted by FW and MC reboot
1293 			 * is initiated to apply the changes. MC reboot
1294 			 * will be handled in a usual way (MC reboot
1295 			 * event on management event queue and adapter
1296 			 * restart).
1297 			 */
1298 			rc = 0;
1299 		} else if (rc != 0) {
1300 			goto fail_reconfigure;
1301 		}
1302 	}
1303 
1304 	sfc_adapter_unlock(sa);
1305 	return 0;
1306 
1307 fail_reconfigure:
1308 	/* Remove/restore entry since the change makes the trouble */
1309 	switch (op) {
1310 	case SFC_UDP_TUNNEL_ADD_PORT:
1311 		(void)efx_tunnel_config_udp_remove(sa->nic,
1312 						   tunnel_udp->udp_port,
1313 						   tunnel_proto);
1314 		break;
1315 	case SFC_UDP_TUNNEL_DEL_PORT:
1316 		(void)efx_tunnel_config_udp_add(sa->nic,
1317 						tunnel_udp->udp_port,
1318 						tunnel_proto);
1319 		break;
1320 	}
1321 
1322 fail_op:
1323 fail_bad_op:
1324 	sfc_adapter_unlock(sa);
1325 
1326 fail_bad_proto:
1327 	SFC_ASSERT(rc > 0);
1328 	return -rc;
1329 }
1330 
1331 static int
1332 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1333 			    struct rte_eth_udp_tunnel *tunnel_udp)
1334 {
1335 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1336 }
1337 
1338 static int
1339 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1340 			    struct rte_eth_udp_tunnel *tunnel_udp)
1341 {
1342 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1343 }
1344 
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_rss *rss = &sa->rss;
1351 	struct sfc_port *port = &sa->port;
1352 
1353 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1354 		return -ENOTSUP;
1355 
1356 	if (rss->channels == 0)
1357 		return -EINVAL;
1358 
1359 	sfc_adapter_lock(sa);
1360 
1361 	/*
1362 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1363 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1364 	 * flags which corresponds to the active EFX configuration stored
1365 	 * locally in 'sfc_adapter' and kept up-to-date
1366 	 */
1367 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1368 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1369 	if (rss_conf->rss_key != NULL)
1370 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1371 
1372 	sfc_adapter_unlock(sa);
1373 
1374 	return 0;
1375 }
1376 
1377 static int
1378 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1379 			struct rte_eth_rss_conf *rss_conf)
1380 {
1381 	struct sfc_adapter *sa = dev->data->dev_private;
1382 	struct sfc_rss *rss = &sa->rss;
1383 	struct sfc_port *port = &sa->port;
1384 	unsigned int efx_hash_types;
1385 	int rc = 0;
1386 
1387 	if (port->isolated)
1388 		return -ENOTSUP;
1389 
1390 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1391 		sfc_err(sa, "RSS is not available");
1392 		return -ENOTSUP;
1393 	}
1394 
1395 	if (rss->channels == 0) {
1396 		sfc_err(sa, "RSS is not configured");
1397 		return -EINVAL;
1398 	}
1399 
1400 	if ((rss_conf->rss_key != NULL) &&
1401 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1402 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1403 			sizeof(rss->key));
1404 		return -EINVAL;
1405 	}
1406 
1407 	sfc_adapter_lock(sa);
1408 
1409 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1410 	if (rc != 0)
1411 		goto fail_rx_hf_rte_to_efx;
1412 
1413 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1414 				   rss->hash_alg, efx_hash_types, B_TRUE);
1415 	if (rc != 0)
1416 		goto fail_scale_mode_set;
1417 
1418 	if (rss_conf->rss_key != NULL) {
1419 		if (sa->state == SFC_ADAPTER_STARTED) {
1420 			rc = efx_rx_scale_key_set(sa->nic,
1421 						  EFX_RSS_CONTEXT_DEFAULT,
1422 						  rss_conf->rss_key,
1423 						  sizeof(rss->key));
1424 			if (rc != 0)
1425 				goto fail_scale_key_set;
1426 		}
1427 
1428 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1429 	}
1430 
1431 	rss->hash_types = efx_hash_types;
1432 
1433 	sfc_adapter_unlock(sa);
1434 
1435 	return 0;
1436 
1437 fail_scale_key_set:
1438 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1439 				  EFX_RX_HASHALG_TOEPLITZ,
1440 				  rss->hash_types, B_TRUE) != 0)
1441 		sfc_err(sa, "failed to restore RSS mode");
1442 
1443 fail_scale_mode_set:
1444 fail_rx_hf_rte_to_efx:
1445 	sfc_adapter_unlock(sa);
1446 	return -rc;
1447 }
1448 
1449 static int
1450 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1451 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1452 		       uint16_t reta_size)
1453 {
1454 	struct sfc_adapter *sa = dev->data->dev_private;
1455 	struct sfc_rss *rss = &sa->rss;
1456 	struct sfc_port *port = &sa->port;
1457 	int entry;
1458 
1459 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1460 		return -ENOTSUP;
1461 
1462 	if (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] = 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_rss *rss = &sa->rss;
1490 	struct sfc_port *port = &sa->port;
1491 	unsigned int *rss_tbl_new;
1492 	uint16_t entry;
1493 	int rc = 0;
1494 
1495 
1496 	if (port->isolated)
1497 		return -ENOTSUP;
1498 
1499 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1500 		sfc_err(sa, "RSS is not available");
1501 		return -ENOTSUP;
1502 	}
1503 
1504 	if (rss->channels == 0) {
1505 		sfc_err(sa, "RSS is not configured");
1506 		return -EINVAL;
1507 	}
1508 
1509 	if (reta_size != EFX_RSS_TBL_SIZE) {
1510 		sfc_err(sa, "RETA size is wrong (should be %u)",
1511 			EFX_RSS_TBL_SIZE);
1512 		return -EINVAL;
1513 	}
1514 
1515 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1516 	if (rss_tbl_new == NULL)
1517 		return -ENOMEM;
1518 
1519 	sfc_adapter_lock(sa);
1520 
1521 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1522 
1523 	for (entry = 0; entry < reta_size; entry++) {
1524 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1525 		struct rte_eth_rss_reta_entry64 *grp;
1526 
1527 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1528 
1529 		if (grp->mask & (1ull << grp_idx)) {
1530 			if (grp->reta[grp_idx] >= rss->channels) {
1531 				rc = EINVAL;
1532 				goto bad_reta_entry;
1533 			}
1534 			rss_tbl_new[entry] = grp->reta[grp_idx];
1535 		}
1536 	}
1537 
1538 	if (sa->state == SFC_ADAPTER_STARTED) {
1539 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1540 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1541 		if (rc != 0)
1542 			goto fail_scale_tbl_set;
1543 	}
1544 
1545 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1546 
1547 fail_scale_tbl_set:
1548 bad_reta_entry:
1549 	sfc_adapter_unlock(sa);
1550 
1551 	rte_free(rss_tbl_new);
1552 
1553 	SFC_ASSERT(rc >= 0);
1554 	return -rc;
1555 }
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 int
1614 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1615 {
1616 	struct sfc_adapter *sa = dev->data->dev_private;
1617 
1618 	/*
1619 	 * If Rx datapath does not provide callback to check mempool,
1620 	 * all pools are supported.
1621 	 */
1622 	if (sa->dp_rx->pool_ops_supported == NULL)
1623 		return 1;
1624 
1625 	return sa->dp_rx->pool_ops_supported(pool);
1626 }
1627 
1628 static const struct eth_dev_ops sfc_eth_dev_ops = {
1629 	.dev_configure			= sfc_dev_configure,
1630 	.dev_start			= sfc_dev_start,
1631 	.dev_stop			= sfc_dev_stop,
1632 	.dev_set_link_up		= sfc_dev_set_link_up,
1633 	.dev_set_link_down		= sfc_dev_set_link_down,
1634 	.dev_close			= sfc_dev_close,
1635 	.promiscuous_enable		= sfc_dev_promisc_enable,
1636 	.promiscuous_disable		= sfc_dev_promisc_disable,
1637 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1638 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1639 	.link_update			= sfc_dev_link_update,
1640 	.stats_get			= sfc_stats_get,
1641 	.stats_reset			= sfc_stats_reset,
1642 	.xstats_get			= sfc_xstats_get,
1643 	.xstats_reset			= sfc_stats_reset,
1644 	.xstats_get_names		= sfc_xstats_get_names,
1645 	.dev_infos_get			= sfc_dev_infos_get,
1646 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1647 	.mtu_set			= sfc_dev_set_mtu,
1648 	.rx_queue_start			= sfc_rx_queue_start,
1649 	.rx_queue_stop			= sfc_rx_queue_stop,
1650 	.tx_queue_start			= sfc_tx_queue_start,
1651 	.tx_queue_stop			= sfc_tx_queue_stop,
1652 	.rx_queue_setup			= sfc_rx_queue_setup,
1653 	.rx_queue_release		= sfc_rx_queue_release,
1654 	.rx_queue_count			= sfc_rx_queue_count,
1655 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1656 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1657 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1658 	.tx_queue_setup			= sfc_tx_queue_setup,
1659 	.tx_queue_release		= sfc_tx_queue_release,
1660 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1661 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1662 	.mac_addr_set			= sfc_mac_addr_set,
1663 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1664 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
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 	.filter_ctrl			= sfc_dev_filter_ctrl,
1670 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1671 	.rxq_info_get			= sfc_rx_queue_info_get,
1672 	.txq_info_get			= sfc_tx_queue_info_get,
1673 	.fw_version_get			= sfc_fw_version_get,
1674 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1675 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1676 	.pool_ops_supported		= sfc_pool_ops_supported,
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 	const efx_nic_cfg_t *encp;
1707 	unsigned int avail_caps = 0;
1708 	const char *rx_name = NULL;
1709 	const char *tx_name = NULL;
1710 	int rc;
1711 
1712 	switch (sa->family) {
1713 	case EFX_FAMILY_HUNTINGTON:
1714 	case EFX_FAMILY_MEDFORD:
1715 	case EFX_FAMILY_MEDFORD2:
1716 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1717 		break;
1718 	default:
1719 		break;
1720 	}
1721 
1722 	encp = efx_nic_cfg_get(sa->nic);
1723 	if (encp->enc_rx_es_super_buffer_supported)
1724 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1725 
1726 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1727 				sfc_kvarg_string_handler, &rx_name);
1728 	if (rc != 0)
1729 		goto fail_kvarg_rx_datapath;
1730 
1731 	if (rx_name != NULL) {
1732 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1733 		if (sa->dp_rx == NULL) {
1734 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1735 			rc = ENOENT;
1736 			goto fail_dp_rx;
1737 		}
1738 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1739 			sfc_err(sa,
1740 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1741 				rx_name);
1742 			rc = EINVAL;
1743 			goto fail_dp_rx_caps;
1744 		}
1745 	} else {
1746 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1747 		if (sa->dp_rx == NULL) {
1748 			sfc_err(sa, "Rx datapath by caps %#x not found",
1749 				avail_caps);
1750 			rc = ENOENT;
1751 			goto fail_dp_rx;
1752 		}
1753 	}
1754 
1755 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1756 	if (sa->dp_rx_name == NULL) {
1757 		rc = ENOMEM;
1758 		goto fail_dp_rx_name;
1759 	}
1760 
1761 	sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1762 
1763 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1764 
1765 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1766 				sfc_kvarg_string_handler, &tx_name);
1767 	if (rc != 0)
1768 		goto fail_kvarg_tx_datapath;
1769 
1770 	if (tx_name != NULL) {
1771 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1772 		if (sa->dp_tx == NULL) {
1773 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1774 			rc = ENOENT;
1775 			goto fail_dp_tx;
1776 		}
1777 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1778 			sfc_err(sa,
1779 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1780 				tx_name);
1781 			rc = EINVAL;
1782 			goto fail_dp_tx_caps;
1783 		}
1784 	} else {
1785 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1786 		if (sa->dp_tx == NULL) {
1787 			sfc_err(sa, "Tx datapath by caps %#x not found",
1788 				avail_caps);
1789 			rc = ENOENT;
1790 			goto fail_dp_tx;
1791 		}
1792 	}
1793 
1794 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1795 	if (sa->dp_tx_name == NULL) {
1796 		rc = ENOMEM;
1797 		goto fail_dp_tx_name;
1798 	}
1799 
1800 	sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1801 
1802 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1803 
1804 	dev->dev_ops = &sfc_eth_dev_ops;
1805 
1806 	return 0;
1807 
1808 fail_dp_tx_name:
1809 fail_dp_tx_caps:
1810 	sa->dp_tx = NULL;
1811 
1812 fail_dp_tx:
1813 fail_kvarg_tx_datapath:
1814 	rte_free(sa->dp_rx_name);
1815 	sa->dp_rx_name = NULL;
1816 
1817 fail_dp_rx_name:
1818 fail_dp_rx_caps:
1819 	sa->dp_rx = NULL;
1820 
1821 fail_dp_rx:
1822 fail_kvarg_rx_datapath:
1823 	return rc;
1824 }
1825 
1826 static void
1827 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1828 {
1829 	struct sfc_adapter *sa = dev->data->dev_private;
1830 
1831 	dev->dev_ops = NULL;
1832 	dev->rx_pkt_burst = NULL;
1833 	dev->tx_pkt_burst = NULL;
1834 
1835 	rte_free(sa->dp_tx_name);
1836 	sa->dp_tx_name = NULL;
1837 	sa->dp_tx = NULL;
1838 
1839 	rte_free(sa->dp_rx_name);
1840 	sa->dp_rx_name = NULL;
1841 	sa->dp_rx = NULL;
1842 }
1843 
1844 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1845 	.rxq_info_get			= sfc_rx_queue_info_get,
1846 	.txq_info_get			= sfc_tx_queue_info_get,
1847 };
1848 
1849 static int
1850 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1851 {
1852 	/*
1853 	 * Device private data has really many process-local pointers.
1854 	 * Below code should be extremely careful to use data located
1855 	 * in shared memory only.
1856 	 */
1857 	struct sfc_adapter *sa = dev->data->dev_private;
1858 	const struct sfc_dp_rx *dp_rx;
1859 	const struct sfc_dp_tx *dp_tx;
1860 	int rc;
1861 
1862 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1863 	if (dp_rx == NULL) {
1864 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1865 		rc = ENOENT;
1866 		goto fail_dp_rx;
1867 	}
1868 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1869 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1870 			sa->dp_tx_name);
1871 		rc = EINVAL;
1872 		goto fail_dp_rx_multi_process;
1873 	}
1874 
1875 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1876 	if (dp_tx == NULL) {
1877 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1878 		rc = ENOENT;
1879 		goto fail_dp_tx;
1880 	}
1881 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1882 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1883 			sa->dp_tx_name);
1884 		rc = EINVAL;
1885 		goto fail_dp_tx_multi_process;
1886 	}
1887 
1888 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1889 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1890 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1891 
1892 	return 0;
1893 
1894 fail_dp_tx_multi_process:
1895 fail_dp_tx:
1896 fail_dp_rx_multi_process:
1897 fail_dp_rx:
1898 	return rc;
1899 }
1900 
1901 static void
1902 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1903 {
1904 	dev->dev_ops = NULL;
1905 	dev->tx_pkt_burst = NULL;
1906 	dev->rx_pkt_burst = NULL;
1907 }
1908 
1909 static void
1910 sfc_register_dp(void)
1911 {
1912 	/* Register once */
1913 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1914 		/* Prefer EF10 datapath */
1915 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1916 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1917 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1918 
1919 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1920 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1921 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1922 	}
1923 }
1924 
1925 static int
1926 sfc_eth_dev_init(struct rte_eth_dev *dev)
1927 {
1928 	struct sfc_adapter *sa = dev->data->dev_private;
1929 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1930 	int rc;
1931 	const efx_nic_cfg_t *encp;
1932 	const struct ether_addr *from;
1933 
1934 	sfc_register_dp();
1935 
1936 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1937 		return -sfc_eth_dev_secondary_set_ops(dev);
1938 
1939 	/* Required for logging */
1940 	sa->pci_addr = pci_dev->addr;
1941 	sa->port_id = dev->data->port_id;
1942 
1943 	sa->eth_dev = dev;
1944 
1945 	/* Copy PCI device info to the dev->data */
1946 	rte_eth_copy_pci_info(dev, pci_dev);
1947 
1948 	sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1949 						RTE_LOG_NOTICE);
1950 
1951 	rc = sfc_kvargs_parse(sa);
1952 	if (rc != 0)
1953 		goto fail_kvargs_parse;
1954 
1955 	sfc_log_init(sa, "entry");
1956 
1957 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1958 	if (dev->data->mac_addrs == NULL) {
1959 		rc = ENOMEM;
1960 		goto fail_mac_addrs;
1961 	}
1962 
1963 	sfc_adapter_lock_init(sa);
1964 	sfc_adapter_lock(sa);
1965 
1966 	sfc_log_init(sa, "probing");
1967 	rc = sfc_probe(sa);
1968 	if (rc != 0)
1969 		goto fail_probe;
1970 
1971 	sfc_log_init(sa, "set device ops");
1972 	rc = sfc_eth_dev_set_ops(dev);
1973 	if (rc != 0)
1974 		goto fail_set_ops;
1975 
1976 	sfc_log_init(sa, "attaching");
1977 	rc = sfc_attach(sa);
1978 	if (rc != 0)
1979 		goto fail_attach;
1980 
1981 	encp = efx_nic_cfg_get(sa->nic);
1982 
1983 	/*
1984 	 * The arguments are really reverse order in comparison to
1985 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1986 	 */
1987 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1988 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1989 
1990 	sfc_adapter_unlock(sa);
1991 
1992 	sfc_log_init(sa, "done");
1993 	return 0;
1994 
1995 fail_attach:
1996 	sfc_eth_dev_clear_ops(dev);
1997 
1998 fail_set_ops:
1999 	sfc_unprobe(sa);
2000 
2001 fail_probe:
2002 	sfc_adapter_unlock(sa);
2003 	sfc_adapter_lock_fini(sa);
2004 	rte_free(dev->data->mac_addrs);
2005 	dev->data->mac_addrs = NULL;
2006 
2007 fail_mac_addrs:
2008 	sfc_kvargs_cleanup(sa);
2009 
2010 fail_kvargs_parse:
2011 	sfc_log_init(sa, "failed %d", rc);
2012 	SFC_ASSERT(rc > 0);
2013 	return -rc;
2014 }
2015 
2016 static int
2017 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2018 {
2019 	struct sfc_adapter *sa;
2020 
2021 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2022 		sfc_eth_dev_secondary_clear_ops(dev);
2023 		return 0;
2024 	}
2025 
2026 	sa = dev->data->dev_private;
2027 	sfc_log_init(sa, "entry");
2028 
2029 	sfc_adapter_lock(sa);
2030 
2031 	sfc_eth_dev_clear_ops(dev);
2032 
2033 	sfc_detach(sa);
2034 	sfc_unprobe(sa);
2035 
2036 	rte_free(dev->data->mac_addrs);
2037 	dev->data->mac_addrs = NULL;
2038 
2039 	sfc_kvargs_cleanup(sa);
2040 
2041 	sfc_adapter_unlock(sa);
2042 	sfc_adapter_lock_fini(sa);
2043 
2044 	sfc_log_init(sa, "done");
2045 
2046 	/* Required for logging, so cleanup last */
2047 	sa->eth_dev = NULL;
2048 	return 0;
2049 }
2050 
2051 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2052 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2053 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2054 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2055 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2056 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2057 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2058 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2059 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2060 	{ .vendor_id = 0 /* sentinel */ }
2061 };
2062 
2063 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2064 	struct rte_pci_device *pci_dev)
2065 {
2066 	return rte_eth_dev_pci_generic_probe(pci_dev,
2067 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2068 }
2069 
2070 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2071 {
2072 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2073 }
2074 
2075 static struct rte_pci_driver sfc_efx_pmd = {
2076 	.id_table = pci_id_sfc_efx_map,
2077 	.drv_flags =
2078 		RTE_PCI_DRV_INTR_LSC |
2079 		RTE_PCI_DRV_NEED_MAPPING,
2080 	.probe = sfc_eth_dev_pci_probe,
2081 	.remove = sfc_eth_dev_pci_remove,
2082 };
2083 
2084 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2085 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2086 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2087 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2088 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2089 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2090 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2091 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2092 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2093 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2094 
2095 RTE_INIT(sfc_driver_register_logtype)
2096 {
2097 	int ret;
2098 
2099 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2100 						   RTE_LOG_NOTICE);
2101 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2102 }
2103