xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision d1482e21f0bc995366b59d18e95fbaeae9b1d3fe)
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 	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 	dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
142 	if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
143 	    !encp->enc_hw_tx_insert_vlan_enabled)
144 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
145 
146 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
147 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
148 
149 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
150 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
151 
152 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
153 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
154 
155 	if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
156 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
157 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
158 		dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
159 	}
160 
161 	/* Initialize to hardware limits */
162 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
163 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
164 	/* The RXQ hardware requires that the descriptor count is a power
165 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
166 	 */
167 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
168 
169 	/* Initialize to hardware limits */
170 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
171 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
172 	/*
173 	 * The TXQ hardware requires that the descriptor count is a power
174 	 * of 2, but tx_desc_lim cannot properly describe that constraint
175 	 */
176 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
177 
178 	if (sa->dp_rx->get_dev_info != NULL)
179 		sa->dp_rx->get_dev_info(dev_info);
180 	if (sa->dp_tx->get_dev_info != NULL)
181 		sa->dp_tx->get_dev_info(dev_info);
182 }
183 
184 static const uint32_t *
185 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
186 {
187 	struct sfc_adapter *sa = dev->data->dev_private;
188 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
189 	uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
190 
191 	return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
192 }
193 
194 static int
195 sfc_dev_configure(struct rte_eth_dev *dev)
196 {
197 	struct rte_eth_dev_data *dev_data = dev->data;
198 	struct sfc_adapter *sa = dev_data->dev_private;
199 	int rc;
200 
201 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
202 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
203 
204 	sfc_adapter_lock(sa);
205 	switch (sa->state) {
206 	case SFC_ADAPTER_CONFIGURED:
207 		/* FALLTHROUGH */
208 	case SFC_ADAPTER_INITIALIZED:
209 		rc = sfc_configure(sa);
210 		break;
211 	default:
212 		sfc_err(sa, "unexpected adapter state %u to configure",
213 			sa->state);
214 		rc = EINVAL;
215 		break;
216 	}
217 	sfc_adapter_unlock(sa);
218 
219 	sfc_log_init(sa, "done %d", rc);
220 	SFC_ASSERT(rc >= 0);
221 	return -rc;
222 }
223 
224 static int
225 sfc_dev_start(struct rte_eth_dev *dev)
226 {
227 	struct sfc_adapter *sa = dev->data->dev_private;
228 	int rc;
229 
230 	sfc_log_init(sa, "entry");
231 
232 	sfc_adapter_lock(sa);
233 	rc = sfc_start(sa);
234 	sfc_adapter_unlock(sa);
235 
236 	sfc_log_init(sa, "done %d", rc);
237 	SFC_ASSERT(rc >= 0);
238 	return -rc;
239 }
240 
241 static int
242 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
243 {
244 	struct sfc_adapter *sa = dev->data->dev_private;
245 	struct rte_eth_link current_link;
246 	int ret;
247 
248 	sfc_log_init(sa, "entry");
249 
250 	if (sa->state != SFC_ADAPTER_STARTED) {
251 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
252 	} else if (wait_to_complete) {
253 		efx_link_mode_t link_mode;
254 
255 		if (efx_port_poll(sa->nic, &link_mode) != 0)
256 			link_mode = EFX_LINK_UNKNOWN;
257 		sfc_port_link_mode_to_info(link_mode, &current_link);
258 
259 	} else {
260 		sfc_ev_mgmt_qpoll(sa);
261 		rte_eth_linkstatus_get(dev, &current_link);
262 	}
263 
264 	ret = rte_eth_linkstatus_set(dev, &current_link);
265 	if (ret == 0)
266 		sfc_notice(sa, "Link status is %s",
267 			   current_link.link_status ? "UP" : "DOWN");
268 
269 	return ret;
270 }
271 
272 static void
273 sfc_dev_stop(struct rte_eth_dev *dev)
274 {
275 	struct sfc_adapter *sa = dev->data->dev_private;
276 
277 	sfc_log_init(sa, "entry");
278 
279 	sfc_adapter_lock(sa);
280 	sfc_stop(sa);
281 	sfc_adapter_unlock(sa);
282 
283 	sfc_log_init(sa, "done");
284 }
285 
286 static int
287 sfc_dev_set_link_up(struct rte_eth_dev *dev)
288 {
289 	struct sfc_adapter *sa = dev->data->dev_private;
290 	int rc;
291 
292 	sfc_log_init(sa, "entry");
293 
294 	sfc_adapter_lock(sa);
295 	rc = sfc_start(sa);
296 	sfc_adapter_unlock(sa);
297 
298 	SFC_ASSERT(rc >= 0);
299 	return -rc;
300 }
301 
302 static int
303 sfc_dev_set_link_down(struct rte_eth_dev *dev)
304 {
305 	struct sfc_adapter *sa = dev->data->dev_private;
306 
307 	sfc_log_init(sa, "entry");
308 
309 	sfc_adapter_lock(sa);
310 	sfc_stop(sa);
311 	sfc_adapter_unlock(sa);
312 
313 	return 0;
314 }
315 
316 static void
317 sfc_dev_close(struct rte_eth_dev *dev)
318 {
319 	struct sfc_adapter *sa = dev->data->dev_private;
320 
321 	sfc_log_init(sa, "entry");
322 
323 	sfc_adapter_lock(sa);
324 	switch (sa->state) {
325 	case SFC_ADAPTER_STARTED:
326 		sfc_stop(sa);
327 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
328 		/* FALLTHROUGH */
329 	case SFC_ADAPTER_CONFIGURED:
330 		sfc_close(sa);
331 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
332 		/* FALLTHROUGH */
333 	case SFC_ADAPTER_INITIALIZED:
334 		break;
335 	default:
336 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
337 		break;
338 	}
339 	sfc_adapter_unlock(sa);
340 
341 	sfc_log_init(sa, "done");
342 }
343 
344 static void
345 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
346 		   boolean_t enabled)
347 {
348 	struct sfc_port *port;
349 	boolean_t *toggle;
350 	struct sfc_adapter *sa = dev->data->dev_private;
351 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
352 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
353 
354 	sfc_adapter_lock(sa);
355 
356 	port = &sa->port;
357 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
358 
359 	if (*toggle != enabled) {
360 		*toggle = enabled;
361 
362 		if (port->isolated) {
363 			sfc_warn(sa, "isolated mode is active on the port");
364 			sfc_warn(sa, "the change is to be applied on the next "
365 				     "start provided that isolated mode is "
366 				     "disabled prior the next start");
367 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
368 			   (sfc_set_rx_mode(sa) != 0)) {
369 			*toggle = !(enabled);
370 			sfc_warn(sa, "Failed to %s %s mode",
371 				 ((enabled) ? "enable" : "disable"), desc);
372 		}
373 	}
374 
375 	sfc_adapter_unlock(sa);
376 }
377 
378 static void
379 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
380 {
381 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
382 }
383 
384 static void
385 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
386 {
387 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
388 }
389 
390 static void
391 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
392 {
393 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
394 }
395 
396 static void
397 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
398 {
399 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
400 }
401 
402 static int
403 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
404 		   uint16_t nb_rx_desc, unsigned int socket_id,
405 		   const struct rte_eth_rxconf *rx_conf,
406 		   struct rte_mempool *mb_pool)
407 {
408 	struct sfc_adapter *sa = dev->data->dev_private;
409 	int rc;
410 
411 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
412 		     rx_queue_id, nb_rx_desc, socket_id);
413 
414 	sfc_adapter_lock(sa);
415 
416 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
417 			  rx_conf, mb_pool);
418 	if (rc != 0)
419 		goto fail_rx_qinit;
420 
421 	dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
422 
423 	sfc_adapter_unlock(sa);
424 
425 	return 0;
426 
427 fail_rx_qinit:
428 	sfc_adapter_unlock(sa);
429 	SFC_ASSERT(rc > 0);
430 	return -rc;
431 }
432 
433 static void
434 sfc_rx_queue_release(void *queue)
435 {
436 	struct sfc_dp_rxq *dp_rxq = queue;
437 	struct sfc_rxq *rxq;
438 	struct sfc_adapter *sa;
439 	unsigned int sw_index;
440 
441 	if (dp_rxq == NULL)
442 		return;
443 
444 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
445 	sa = rxq->evq->sa;
446 	sfc_adapter_lock(sa);
447 
448 	sw_index = sfc_rxq_sw_index(rxq);
449 
450 	sfc_log_init(sa, "RxQ=%u", sw_index);
451 
452 	sa->eth_dev->data->rx_queues[sw_index] = NULL;
453 
454 	sfc_rx_qfini(sa, sw_index);
455 
456 	sfc_adapter_unlock(sa);
457 }
458 
459 static int
460 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
461 		   uint16_t nb_tx_desc, unsigned int socket_id,
462 		   const struct rte_eth_txconf *tx_conf)
463 {
464 	struct sfc_adapter *sa = dev->data->dev_private;
465 	int rc;
466 
467 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
468 		     tx_queue_id, nb_tx_desc, socket_id);
469 
470 	sfc_adapter_lock(sa);
471 
472 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
473 	if (rc != 0)
474 		goto fail_tx_qinit;
475 
476 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
477 
478 	sfc_adapter_unlock(sa);
479 	return 0;
480 
481 fail_tx_qinit:
482 	sfc_adapter_unlock(sa);
483 	SFC_ASSERT(rc > 0);
484 	return -rc;
485 }
486 
487 static void
488 sfc_tx_queue_release(void *queue)
489 {
490 	struct sfc_dp_txq *dp_txq = queue;
491 	struct sfc_txq *txq;
492 	unsigned int sw_index;
493 	struct sfc_adapter *sa;
494 
495 	if (dp_txq == NULL)
496 		return;
497 
498 	txq = sfc_txq_by_dp_txq(dp_txq);
499 	sw_index = sfc_txq_sw_index(txq);
500 
501 	SFC_ASSERT(txq->evq != NULL);
502 	sa = txq->evq->sa;
503 
504 	sfc_log_init(sa, "TxQ = %u", sw_index);
505 
506 	sfc_adapter_lock(sa);
507 
508 	SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
509 	sa->eth_dev->data->tx_queues[sw_index] = NULL;
510 
511 	sfc_tx_qfini(sa, sw_index);
512 
513 	sfc_adapter_unlock(sa);
514 }
515 
516 static int
517 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
518 {
519 	struct sfc_adapter *sa = dev->data->dev_private;
520 	struct sfc_port *port = &sa->port;
521 	uint64_t *mac_stats;
522 	int ret;
523 
524 	rte_spinlock_lock(&port->mac_stats_lock);
525 
526 	ret = sfc_port_update_mac_stats(sa);
527 	if (ret != 0)
528 		goto unlock;
529 
530 	mac_stats = port->mac_stats_buf;
531 
532 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
533 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
534 		stats->ipackets =
535 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
536 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
537 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
538 		stats->opackets =
539 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
540 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
541 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
542 		stats->ibytes =
543 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
544 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
545 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
546 		stats->obytes =
547 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
548 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
549 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
550 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
551 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
552 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
553 	} else {
554 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
555 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
556 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
557 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
558 		/*
559 		 * Take into account stats which are whenever supported
560 		 * on EF10. If some stat is not supported by current
561 		 * firmware variant or HW revision, it is guaranteed
562 		 * to be zero in mac_stats.
563 		 */
564 		stats->imissed =
565 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
566 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
567 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
568 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
569 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
570 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
571 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
572 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
573 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
574 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
575 		stats->ierrors =
576 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
577 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
578 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
579 		/* no oerrors counters supported on EF10 */
580 	}
581 
582 unlock:
583 	rte_spinlock_unlock(&port->mac_stats_lock);
584 	SFC_ASSERT(ret >= 0);
585 	return -ret;
586 }
587 
588 static void
589 sfc_stats_reset(struct rte_eth_dev *dev)
590 {
591 	struct sfc_adapter *sa = dev->data->dev_private;
592 	struct sfc_port *port = &sa->port;
593 	int rc;
594 
595 	if (sa->state != SFC_ADAPTER_STARTED) {
596 		/*
597 		 * The operation cannot be done if port is not started; it
598 		 * will be scheduled to be done during the next port start
599 		 */
600 		port->mac_stats_reset_pending = B_TRUE;
601 		return;
602 	}
603 
604 	rc = sfc_port_reset_mac_stats(sa);
605 	if (rc != 0)
606 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
607 }
608 
609 static int
610 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
611 	       unsigned int xstats_count)
612 {
613 	struct sfc_adapter *sa = dev->data->dev_private;
614 	struct sfc_port *port = &sa->port;
615 	uint64_t *mac_stats;
616 	int rc;
617 	unsigned int i;
618 	int nstats = 0;
619 
620 	rte_spinlock_lock(&port->mac_stats_lock);
621 
622 	rc = sfc_port_update_mac_stats(sa);
623 	if (rc != 0) {
624 		SFC_ASSERT(rc > 0);
625 		nstats = -rc;
626 		goto unlock;
627 	}
628 
629 	mac_stats = port->mac_stats_buf;
630 
631 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
632 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
633 			if (xstats != NULL && nstats < (int)xstats_count) {
634 				xstats[nstats].id = nstats;
635 				xstats[nstats].value = mac_stats[i];
636 			}
637 			nstats++;
638 		}
639 	}
640 
641 unlock:
642 	rte_spinlock_unlock(&port->mac_stats_lock);
643 
644 	return nstats;
645 }
646 
647 static int
648 sfc_xstats_get_names(struct rte_eth_dev *dev,
649 		     struct rte_eth_xstat_name *xstats_names,
650 		     unsigned int xstats_count)
651 {
652 	struct sfc_adapter *sa = dev->data->dev_private;
653 	struct sfc_port *port = &sa->port;
654 	unsigned int i;
655 	unsigned int nstats = 0;
656 
657 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
658 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
659 			if (xstats_names != NULL && nstats < xstats_count)
660 				strncpy(xstats_names[nstats].name,
661 					efx_mac_stat_name(sa->nic, i),
662 					sizeof(xstats_names[0].name));
663 			nstats++;
664 		}
665 	}
666 
667 	return nstats;
668 }
669 
670 static int
671 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
672 		     uint64_t *values, unsigned int n)
673 {
674 	struct sfc_adapter *sa = dev->data->dev_private;
675 	struct sfc_port *port = &sa->port;
676 	uint64_t *mac_stats;
677 	unsigned int nb_supported = 0;
678 	unsigned int nb_written = 0;
679 	unsigned int i;
680 	int ret;
681 	int rc;
682 
683 	if (unlikely(values == NULL) ||
684 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
685 		return port->mac_stats_nb_supported;
686 
687 	rte_spinlock_lock(&port->mac_stats_lock);
688 
689 	rc = sfc_port_update_mac_stats(sa);
690 	if (rc != 0) {
691 		SFC_ASSERT(rc > 0);
692 		ret = -rc;
693 		goto unlock;
694 	}
695 
696 	mac_stats = port->mac_stats_buf;
697 
698 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
699 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
700 			continue;
701 
702 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
703 			values[nb_written++] = mac_stats[i];
704 
705 		++nb_supported;
706 	}
707 
708 	ret = nb_written;
709 
710 unlock:
711 	rte_spinlock_unlock(&port->mac_stats_lock);
712 
713 	return ret;
714 }
715 
716 static int
717 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
718 			   struct rte_eth_xstat_name *xstats_names,
719 			   const uint64_t *ids, unsigned int size)
720 {
721 	struct sfc_adapter *sa = dev->data->dev_private;
722 	struct sfc_port *port = &sa->port;
723 	unsigned int nb_supported = 0;
724 	unsigned int nb_written = 0;
725 	unsigned int i;
726 
727 	if (unlikely(xstats_names == NULL) ||
728 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
729 		return port->mac_stats_nb_supported;
730 
731 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
732 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
733 			continue;
734 
735 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
736 			char *name = xstats_names[nb_written++].name;
737 
738 			strncpy(name, efx_mac_stat_name(sa->nic, i),
739 				sizeof(xstats_names[0].name));
740 			name[sizeof(xstats_names[0].name) - 1] = '\0';
741 		}
742 
743 		++nb_supported;
744 	}
745 
746 	return nb_written;
747 }
748 
749 static int
750 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
751 {
752 	struct sfc_adapter *sa = dev->data->dev_private;
753 	unsigned int wanted_fc, link_fc;
754 
755 	memset(fc_conf, 0, sizeof(*fc_conf));
756 
757 	sfc_adapter_lock(sa);
758 
759 	if (sa->state == SFC_ADAPTER_STARTED)
760 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
761 	else
762 		link_fc = sa->port.flow_ctrl;
763 
764 	switch (link_fc) {
765 	case 0:
766 		fc_conf->mode = RTE_FC_NONE;
767 		break;
768 	case EFX_FCNTL_RESPOND:
769 		fc_conf->mode = RTE_FC_RX_PAUSE;
770 		break;
771 	case EFX_FCNTL_GENERATE:
772 		fc_conf->mode = RTE_FC_TX_PAUSE;
773 		break;
774 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
775 		fc_conf->mode = RTE_FC_FULL;
776 		break;
777 	default:
778 		sfc_err(sa, "%s: unexpected flow control value %#x",
779 			__func__, link_fc);
780 	}
781 
782 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
783 
784 	sfc_adapter_unlock(sa);
785 
786 	return 0;
787 }
788 
789 static int
790 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
791 {
792 	struct sfc_adapter *sa = dev->data->dev_private;
793 	struct sfc_port *port = &sa->port;
794 	unsigned int fcntl;
795 	int rc;
796 
797 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
798 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
799 	    fc_conf->mac_ctrl_frame_fwd != 0) {
800 		sfc_err(sa, "unsupported flow control settings specified");
801 		rc = EINVAL;
802 		goto fail_inval;
803 	}
804 
805 	switch (fc_conf->mode) {
806 	case RTE_FC_NONE:
807 		fcntl = 0;
808 		break;
809 	case RTE_FC_RX_PAUSE:
810 		fcntl = EFX_FCNTL_RESPOND;
811 		break;
812 	case RTE_FC_TX_PAUSE:
813 		fcntl = EFX_FCNTL_GENERATE;
814 		break;
815 	case RTE_FC_FULL:
816 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
817 		break;
818 	default:
819 		rc = EINVAL;
820 		goto fail_inval;
821 	}
822 
823 	sfc_adapter_lock(sa);
824 
825 	if (sa->state == SFC_ADAPTER_STARTED) {
826 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
827 		if (rc != 0)
828 			goto fail_mac_fcntl_set;
829 	}
830 
831 	port->flow_ctrl = fcntl;
832 	port->flow_ctrl_autoneg = fc_conf->autoneg;
833 
834 	sfc_adapter_unlock(sa);
835 
836 	return 0;
837 
838 fail_mac_fcntl_set:
839 	sfc_adapter_unlock(sa);
840 fail_inval:
841 	SFC_ASSERT(rc > 0);
842 	return -rc;
843 }
844 
845 static int
846 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
847 {
848 	struct sfc_adapter *sa = dev->data->dev_private;
849 	size_t pdu = EFX_MAC_PDU(mtu);
850 	size_t old_pdu;
851 	int rc;
852 
853 	sfc_log_init(sa, "mtu=%u", mtu);
854 
855 	rc = EINVAL;
856 	if (pdu < EFX_MAC_PDU_MIN) {
857 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
858 			(unsigned int)mtu, (unsigned int)pdu,
859 			EFX_MAC_PDU_MIN);
860 		goto fail_inval;
861 	}
862 	if (pdu > EFX_MAC_PDU_MAX) {
863 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
864 			(unsigned int)mtu, (unsigned int)pdu,
865 			EFX_MAC_PDU_MAX);
866 		goto fail_inval;
867 	}
868 
869 	sfc_adapter_lock(sa);
870 
871 	if (pdu != sa->port.pdu) {
872 		if (sa->state == SFC_ADAPTER_STARTED) {
873 			sfc_stop(sa);
874 
875 			old_pdu = sa->port.pdu;
876 			sa->port.pdu = pdu;
877 			rc = sfc_start(sa);
878 			if (rc != 0)
879 				goto fail_start;
880 		} else {
881 			sa->port.pdu = pdu;
882 		}
883 	}
884 
885 	/*
886 	 * The driver does not use it, but other PMDs update jumbo_frame
887 	 * flag and max_rx_pkt_len when MTU is set.
888 	 */
889 	if (mtu > ETHER_MAX_LEN) {
890 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
891 
892 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
893 		rxmode->jumbo_frame = 1;
894 	}
895 
896 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
897 
898 	sfc_adapter_unlock(sa);
899 
900 	sfc_log_init(sa, "done");
901 	return 0;
902 
903 fail_start:
904 	sa->port.pdu = old_pdu;
905 	if (sfc_start(sa) != 0)
906 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
907 			"PDU max size - port is stopped",
908 			(unsigned int)pdu, (unsigned int)old_pdu);
909 	sfc_adapter_unlock(sa);
910 
911 fail_inval:
912 	sfc_log_init(sa, "failed %d", rc);
913 	SFC_ASSERT(rc > 0);
914 	return -rc;
915 }
916 static int
917 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
918 {
919 	struct sfc_adapter *sa = dev->data->dev_private;
920 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
921 	struct sfc_port *port = &sa->port;
922 	struct ether_addr *old_addr = &dev->data->mac_addrs[0];
923 	int rc = 0;
924 
925 	sfc_adapter_lock(sa);
926 
927 	/*
928 	 * Copy the address to the device private data so that
929 	 * it could be recalled in the case of adapter restart.
930 	 */
931 	ether_addr_copy(mac_addr, &port->default_mac_addr);
932 
933 	/*
934 	 * Neither of the two following checks can return
935 	 * an error. The new MAC address is preserved in
936 	 * the device private data and can be activated
937 	 * on the next port start if the user prevents
938 	 * isolated mode from being enabled.
939 	 */
940 	if (port->isolated) {
941 		sfc_warn(sa, "isolated mode is active on the port");
942 		sfc_warn(sa, "will not set MAC address");
943 		goto unlock;
944 	}
945 
946 	if (sa->state != SFC_ADAPTER_STARTED) {
947 		sfc_notice(sa, "the port is not started");
948 		sfc_notice(sa, "the new MAC address will be set on port start");
949 
950 		goto unlock;
951 	}
952 
953 	if (encp->enc_allow_set_mac_with_installed_filters) {
954 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
955 		if (rc != 0) {
956 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
957 			goto unlock;
958 		}
959 
960 		/*
961 		 * Changing the MAC address by means of MCDI request
962 		 * has no effect on received traffic, therefore
963 		 * we also need to update unicast filters
964 		 */
965 		rc = sfc_set_rx_mode(sa);
966 		if (rc != 0) {
967 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
968 			/* Rollback the old address */
969 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
970 			(void)sfc_set_rx_mode(sa);
971 		}
972 	} else {
973 		sfc_warn(sa, "cannot set MAC address with filters installed");
974 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
975 		sfc_warn(sa, "(some traffic may be dropped)");
976 
977 		/*
978 		 * Since setting MAC address with filters installed is not
979 		 * allowed on the adapter, the new MAC address will be set
980 		 * by means of adapter restart. sfc_start() shall retrieve
981 		 * the new address from the device private data and set it.
982 		 */
983 		sfc_stop(sa);
984 		rc = sfc_start(sa);
985 		if (rc != 0)
986 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
987 	}
988 
989 unlock:
990 	if (rc != 0)
991 		ether_addr_copy(old_addr, &port->default_mac_addr);
992 
993 	sfc_adapter_unlock(sa);
994 
995 	SFC_ASSERT(rc >= 0);
996 	return -rc;
997 }
998 
999 
1000 static int
1001 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1002 		     uint32_t nb_mc_addr)
1003 {
1004 	struct sfc_adapter *sa = dev->data->dev_private;
1005 	struct sfc_port *port = &sa->port;
1006 	uint8_t *mc_addrs = port->mcast_addrs;
1007 	int rc;
1008 	unsigned int i;
1009 
1010 	if (port->isolated) {
1011 		sfc_err(sa, "isolated mode is active on the port");
1012 		sfc_err(sa, "will not set multicast address list");
1013 		return -ENOTSUP;
1014 	}
1015 
1016 	if (mc_addrs == NULL)
1017 		return -ENOBUFS;
1018 
1019 	if (nb_mc_addr > port->max_mcast_addrs) {
1020 		sfc_err(sa, "too many multicast addresses: %u > %u",
1021 			 nb_mc_addr, port->max_mcast_addrs);
1022 		return -EINVAL;
1023 	}
1024 
1025 	for (i = 0; i < nb_mc_addr; ++i) {
1026 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1027 				 EFX_MAC_ADDR_LEN);
1028 		mc_addrs += EFX_MAC_ADDR_LEN;
1029 	}
1030 
1031 	port->nb_mcast_addrs = nb_mc_addr;
1032 
1033 	if (sa->state != SFC_ADAPTER_STARTED)
1034 		return 0;
1035 
1036 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1037 					port->nb_mcast_addrs);
1038 	if (rc != 0)
1039 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1040 
1041 	SFC_ASSERT(rc > 0);
1042 	return -rc;
1043 }
1044 
1045 /*
1046  * The function is used by the secondary process as well. It must not
1047  * use any process-local pointers from the adapter data.
1048  */
1049 static void
1050 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1051 		      struct rte_eth_rxq_info *qinfo)
1052 {
1053 	struct sfc_adapter *sa = dev->data->dev_private;
1054 	struct sfc_rxq_info *rxq_info;
1055 	struct sfc_rxq *rxq;
1056 
1057 	sfc_adapter_lock(sa);
1058 
1059 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1060 
1061 	rxq_info = &sa->rxq_info[rx_queue_id];
1062 	rxq = rxq_info->rxq;
1063 	SFC_ASSERT(rxq != NULL);
1064 
1065 	qinfo->mp = rxq->refill_mb_pool;
1066 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1067 	qinfo->conf.rx_drop_en = 1;
1068 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1069 	qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1070 			       DEV_RX_OFFLOAD_UDP_CKSUM |
1071 			       DEV_RX_OFFLOAD_TCP_CKSUM;
1072 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1073 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1074 		qinfo->scattered_rx = 1;
1075 	}
1076 	qinfo->nb_desc = rxq_info->entries;
1077 
1078 	sfc_adapter_unlock(sa);
1079 }
1080 
1081 /*
1082  * The function is used by the secondary process as well. It must not
1083  * use any process-local pointers from the adapter data.
1084  */
1085 static void
1086 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1087 		      struct rte_eth_txq_info *qinfo)
1088 {
1089 	struct sfc_adapter *sa = dev->data->dev_private;
1090 	struct sfc_txq_info *txq_info;
1091 
1092 	sfc_adapter_lock(sa);
1093 
1094 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1095 
1096 	txq_info = &sa->txq_info[tx_queue_id];
1097 	SFC_ASSERT(txq_info->txq != NULL);
1098 
1099 	memset(qinfo, 0, sizeof(*qinfo));
1100 
1101 	qinfo->conf.txq_flags = txq_info->txq->flags;
1102 	qinfo->conf.offloads = txq_info->txq->offloads;
1103 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1104 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1105 	qinfo->nb_desc = txq_info->entries;
1106 
1107 	sfc_adapter_unlock(sa);
1108 }
1109 
1110 static uint32_t
1111 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1112 {
1113 	struct sfc_adapter *sa = dev->data->dev_private;
1114 
1115 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1116 
1117 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
1118 }
1119 
1120 static int
1121 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1122 {
1123 	struct sfc_dp_rxq *dp_rxq = queue;
1124 
1125 	return sfc_rx_qdesc_done(dp_rxq, offset);
1126 }
1127 
1128 static int
1129 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1130 {
1131 	struct sfc_dp_rxq *dp_rxq = queue;
1132 	struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1133 
1134 	return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1135 }
1136 
1137 static int
1138 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1139 {
1140 	struct sfc_dp_txq *dp_txq = queue;
1141 	struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1142 
1143 	return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1144 }
1145 
1146 static int
1147 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1148 {
1149 	struct sfc_adapter *sa = dev->data->dev_private;
1150 	int rc;
1151 
1152 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1153 
1154 	sfc_adapter_lock(sa);
1155 
1156 	rc = EINVAL;
1157 	if (sa->state != SFC_ADAPTER_STARTED)
1158 		goto fail_not_started;
1159 
1160 	rc = sfc_rx_qstart(sa, rx_queue_id);
1161 	if (rc != 0)
1162 		goto fail_rx_qstart;
1163 
1164 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1165 
1166 	sfc_adapter_unlock(sa);
1167 
1168 	return 0;
1169 
1170 fail_rx_qstart:
1171 fail_not_started:
1172 	sfc_adapter_unlock(sa);
1173 	SFC_ASSERT(rc > 0);
1174 	return -rc;
1175 }
1176 
1177 static int
1178 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1179 {
1180 	struct sfc_adapter *sa = dev->data->dev_private;
1181 
1182 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1183 
1184 	sfc_adapter_lock(sa);
1185 	sfc_rx_qstop(sa, rx_queue_id);
1186 
1187 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1188 
1189 	sfc_adapter_unlock(sa);
1190 
1191 	return 0;
1192 }
1193 
1194 static int
1195 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1196 {
1197 	struct sfc_adapter *sa = dev->data->dev_private;
1198 	int rc;
1199 
1200 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1201 
1202 	sfc_adapter_lock(sa);
1203 
1204 	rc = EINVAL;
1205 	if (sa->state != SFC_ADAPTER_STARTED)
1206 		goto fail_not_started;
1207 
1208 	rc = sfc_tx_qstart(sa, tx_queue_id);
1209 	if (rc != 0)
1210 		goto fail_tx_qstart;
1211 
1212 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1213 
1214 	sfc_adapter_unlock(sa);
1215 	return 0;
1216 
1217 fail_tx_qstart:
1218 
1219 fail_not_started:
1220 	sfc_adapter_unlock(sa);
1221 	SFC_ASSERT(rc > 0);
1222 	return -rc;
1223 }
1224 
1225 static int
1226 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1227 {
1228 	struct sfc_adapter *sa = dev->data->dev_private;
1229 
1230 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1231 
1232 	sfc_adapter_lock(sa);
1233 
1234 	sfc_tx_qstop(sa, tx_queue_id);
1235 
1236 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1237 
1238 	sfc_adapter_unlock(sa);
1239 	return 0;
1240 }
1241 
1242 static efx_tunnel_protocol_t
1243 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1244 {
1245 	switch (rte_type) {
1246 	case RTE_TUNNEL_TYPE_VXLAN:
1247 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1248 	case RTE_TUNNEL_TYPE_GENEVE:
1249 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1250 	default:
1251 		return EFX_TUNNEL_NPROTOS;
1252 	}
1253 }
1254 
1255 enum sfc_udp_tunnel_op_e {
1256 	SFC_UDP_TUNNEL_ADD_PORT,
1257 	SFC_UDP_TUNNEL_DEL_PORT,
1258 };
1259 
1260 static int
1261 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1262 		      struct rte_eth_udp_tunnel *tunnel_udp,
1263 		      enum sfc_udp_tunnel_op_e op)
1264 {
1265 	struct sfc_adapter *sa = dev->data->dev_private;
1266 	efx_tunnel_protocol_t tunnel_proto;
1267 	int rc;
1268 
1269 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1270 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1271 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1272 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1273 
1274 	tunnel_proto =
1275 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1276 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1277 		rc = ENOTSUP;
1278 		goto fail_bad_proto;
1279 	}
1280 
1281 	sfc_adapter_lock(sa);
1282 
1283 	switch (op) {
1284 	case SFC_UDP_TUNNEL_ADD_PORT:
1285 		rc = efx_tunnel_config_udp_add(sa->nic,
1286 					       tunnel_udp->udp_port,
1287 					       tunnel_proto);
1288 		break;
1289 	case SFC_UDP_TUNNEL_DEL_PORT:
1290 		rc = efx_tunnel_config_udp_remove(sa->nic,
1291 						  tunnel_udp->udp_port,
1292 						  tunnel_proto);
1293 		break;
1294 	default:
1295 		rc = EINVAL;
1296 		goto fail_bad_op;
1297 	}
1298 
1299 	if (rc != 0)
1300 		goto fail_op;
1301 
1302 	if (sa->state == SFC_ADAPTER_STARTED) {
1303 		rc = efx_tunnel_reconfigure(sa->nic);
1304 		if (rc == EAGAIN) {
1305 			/*
1306 			 * Configuration is accepted by FW and MC reboot
1307 			 * is initiated to apply the changes. MC reboot
1308 			 * will be handled in a usual way (MC reboot
1309 			 * event on management event queue and adapter
1310 			 * restart).
1311 			 */
1312 			rc = 0;
1313 		} else if (rc != 0) {
1314 			goto fail_reconfigure;
1315 		}
1316 	}
1317 
1318 	sfc_adapter_unlock(sa);
1319 	return 0;
1320 
1321 fail_reconfigure:
1322 	/* Remove/restore entry since the change makes the trouble */
1323 	switch (op) {
1324 	case SFC_UDP_TUNNEL_ADD_PORT:
1325 		(void)efx_tunnel_config_udp_remove(sa->nic,
1326 						   tunnel_udp->udp_port,
1327 						   tunnel_proto);
1328 		break;
1329 	case SFC_UDP_TUNNEL_DEL_PORT:
1330 		(void)efx_tunnel_config_udp_add(sa->nic,
1331 						tunnel_udp->udp_port,
1332 						tunnel_proto);
1333 		break;
1334 	}
1335 
1336 fail_op:
1337 fail_bad_op:
1338 	sfc_adapter_unlock(sa);
1339 
1340 fail_bad_proto:
1341 	SFC_ASSERT(rc > 0);
1342 	return -rc;
1343 }
1344 
1345 static int
1346 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1347 			    struct rte_eth_udp_tunnel *tunnel_udp)
1348 {
1349 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1350 }
1351 
1352 static int
1353 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1354 			    struct rte_eth_udp_tunnel *tunnel_udp)
1355 {
1356 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1357 }
1358 
1359 static int
1360 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1361 			  struct rte_eth_rss_conf *rss_conf)
1362 {
1363 	struct sfc_adapter *sa = dev->data->dev_private;
1364 	struct sfc_rss *rss = &sa->rss;
1365 	struct sfc_port *port = &sa->port;
1366 
1367 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1368 		return -ENOTSUP;
1369 
1370 	if (rss->channels == 0)
1371 		return -EINVAL;
1372 
1373 	sfc_adapter_lock(sa);
1374 
1375 	/*
1376 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1377 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1378 	 * flags which corresponds to the active EFX configuration stored
1379 	 * locally in 'sfc_adapter' and kept up-to-date
1380 	 */
1381 	rss_conf->rss_hf = sfc_efx_to_rte_hash_type(rss->hash_types);
1382 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1383 	if (rss_conf->rss_key != NULL)
1384 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1385 
1386 	sfc_adapter_unlock(sa);
1387 
1388 	return 0;
1389 }
1390 
1391 static int
1392 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1393 			struct rte_eth_rss_conf *rss_conf)
1394 {
1395 	struct sfc_adapter *sa = dev->data->dev_private;
1396 	struct sfc_rss *rss = &sa->rss;
1397 	struct sfc_port *port = &sa->port;
1398 	unsigned int efx_hash_types;
1399 	int rc = 0;
1400 
1401 	if (port->isolated)
1402 		return -ENOTSUP;
1403 
1404 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1405 		sfc_err(sa, "RSS is not available");
1406 		return -ENOTSUP;
1407 	}
1408 
1409 	if (rss->channels == 0) {
1410 		sfc_err(sa, "RSS is not configured");
1411 		return -EINVAL;
1412 	}
1413 
1414 	if ((rss_conf->rss_key != NULL) &&
1415 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1416 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1417 			sizeof(rss->key));
1418 		return -EINVAL;
1419 	}
1420 
1421 	if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1422 		sfc_err(sa, "unsupported hash functions requested");
1423 		return -EINVAL;
1424 	}
1425 
1426 	sfc_adapter_lock(sa);
1427 
1428 	efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1429 
1430 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1431 				   EFX_RX_HASHALG_TOEPLITZ,
1432 				   efx_hash_types, B_TRUE);
1433 	if (rc != 0)
1434 		goto fail_scale_mode_set;
1435 
1436 	if (rss_conf->rss_key != NULL) {
1437 		if (sa->state == SFC_ADAPTER_STARTED) {
1438 			rc = efx_rx_scale_key_set(sa->nic,
1439 						  EFX_RSS_CONTEXT_DEFAULT,
1440 						  rss_conf->rss_key,
1441 						  sizeof(rss->key));
1442 			if (rc != 0)
1443 				goto fail_scale_key_set;
1444 		}
1445 
1446 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1447 	}
1448 
1449 	rss->hash_types = efx_hash_types;
1450 
1451 	sfc_adapter_unlock(sa);
1452 
1453 	return 0;
1454 
1455 fail_scale_key_set:
1456 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1457 				  EFX_RX_HASHALG_TOEPLITZ,
1458 				  rss->hash_types, B_TRUE) != 0)
1459 		sfc_err(sa, "failed to restore RSS mode");
1460 
1461 fail_scale_mode_set:
1462 	sfc_adapter_unlock(sa);
1463 	return -rc;
1464 }
1465 
1466 static int
1467 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1468 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1469 		       uint16_t reta_size)
1470 {
1471 	struct sfc_adapter *sa = dev->data->dev_private;
1472 	struct sfc_rss *rss = &sa->rss;
1473 	struct sfc_port *port = &sa->port;
1474 	int entry;
1475 
1476 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1477 		return -ENOTSUP;
1478 
1479 	if (rss->channels == 0)
1480 		return -EINVAL;
1481 
1482 	if (reta_size != EFX_RSS_TBL_SIZE)
1483 		return -EINVAL;
1484 
1485 	sfc_adapter_lock(sa);
1486 
1487 	for (entry = 0; entry < reta_size; entry++) {
1488 		int grp = entry / RTE_RETA_GROUP_SIZE;
1489 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1490 
1491 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1492 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1493 	}
1494 
1495 	sfc_adapter_unlock(sa);
1496 
1497 	return 0;
1498 }
1499 
1500 static int
1501 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1502 			struct rte_eth_rss_reta_entry64 *reta_conf,
1503 			uint16_t reta_size)
1504 {
1505 	struct sfc_adapter *sa = dev->data->dev_private;
1506 	struct sfc_rss *rss = &sa->rss;
1507 	struct sfc_port *port = &sa->port;
1508 	unsigned int *rss_tbl_new;
1509 	uint16_t entry;
1510 	int rc = 0;
1511 
1512 
1513 	if (port->isolated)
1514 		return -ENOTSUP;
1515 
1516 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1517 		sfc_err(sa, "RSS is not available");
1518 		return -ENOTSUP;
1519 	}
1520 
1521 	if (rss->channels == 0) {
1522 		sfc_err(sa, "RSS is not configured");
1523 		return -EINVAL;
1524 	}
1525 
1526 	if (reta_size != EFX_RSS_TBL_SIZE) {
1527 		sfc_err(sa, "RETA size is wrong (should be %u)",
1528 			EFX_RSS_TBL_SIZE);
1529 		return -EINVAL;
1530 	}
1531 
1532 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1533 	if (rss_tbl_new == NULL)
1534 		return -ENOMEM;
1535 
1536 	sfc_adapter_lock(sa);
1537 
1538 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1539 
1540 	for (entry = 0; entry < reta_size; entry++) {
1541 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1542 		struct rte_eth_rss_reta_entry64 *grp;
1543 
1544 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1545 
1546 		if (grp->mask & (1ull << grp_idx)) {
1547 			if (grp->reta[grp_idx] >= rss->channels) {
1548 				rc = EINVAL;
1549 				goto bad_reta_entry;
1550 			}
1551 			rss_tbl_new[entry] = grp->reta[grp_idx];
1552 		}
1553 	}
1554 
1555 	if (sa->state == SFC_ADAPTER_STARTED) {
1556 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1557 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1558 		if (rc != 0)
1559 			goto fail_scale_tbl_set;
1560 	}
1561 
1562 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1563 
1564 fail_scale_tbl_set:
1565 bad_reta_entry:
1566 	sfc_adapter_unlock(sa);
1567 
1568 	rte_free(rss_tbl_new);
1569 
1570 	SFC_ASSERT(rc >= 0);
1571 	return -rc;
1572 }
1573 
1574 static int
1575 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1576 		    enum rte_filter_op filter_op,
1577 		    void *arg)
1578 {
1579 	struct sfc_adapter *sa = dev->data->dev_private;
1580 	int rc = ENOTSUP;
1581 
1582 	sfc_log_init(sa, "entry");
1583 
1584 	switch (filter_type) {
1585 	case RTE_ETH_FILTER_NONE:
1586 		sfc_err(sa, "Global filters configuration not supported");
1587 		break;
1588 	case RTE_ETH_FILTER_MACVLAN:
1589 		sfc_err(sa, "MACVLAN filters not supported");
1590 		break;
1591 	case RTE_ETH_FILTER_ETHERTYPE:
1592 		sfc_err(sa, "EtherType filters not supported");
1593 		break;
1594 	case RTE_ETH_FILTER_FLEXIBLE:
1595 		sfc_err(sa, "Flexible filters not supported");
1596 		break;
1597 	case RTE_ETH_FILTER_SYN:
1598 		sfc_err(sa, "SYN filters not supported");
1599 		break;
1600 	case RTE_ETH_FILTER_NTUPLE:
1601 		sfc_err(sa, "NTUPLE filters not supported");
1602 		break;
1603 	case RTE_ETH_FILTER_TUNNEL:
1604 		sfc_err(sa, "Tunnel filters not supported");
1605 		break;
1606 	case RTE_ETH_FILTER_FDIR:
1607 		sfc_err(sa, "Flow Director filters not supported");
1608 		break;
1609 	case RTE_ETH_FILTER_HASH:
1610 		sfc_err(sa, "Hash filters not supported");
1611 		break;
1612 	case RTE_ETH_FILTER_GENERIC:
1613 		if (filter_op != RTE_ETH_FILTER_GET) {
1614 			rc = EINVAL;
1615 		} else {
1616 			*(const void **)arg = &sfc_flow_ops;
1617 			rc = 0;
1618 		}
1619 		break;
1620 	default:
1621 		sfc_err(sa, "Unknown filter type %u", filter_type);
1622 		break;
1623 	}
1624 
1625 	sfc_log_init(sa, "exit: %d", -rc);
1626 	SFC_ASSERT(rc >= 0);
1627 	return -rc;
1628 }
1629 
1630 static const struct eth_dev_ops sfc_eth_dev_ops = {
1631 	.dev_configure			= sfc_dev_configure,
1632 	.dev_start			= sfc_dev_start,
1633 	.dev_stop			= sfc_dev_stop,
1634 	.dev_set_link_up		= sfc_dev_set_link_up,
1635 	.dev_set_link_down		= sfc_dev_set_link_down,
1636 	.dev_close			= sfc_dev_close,
1637 	.promiscuous_enable		= sfc_dev_promisc_enable,
1638 	.promiscuous_disable		= sfc_dev_promisc_disable,
1639 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1640 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1641 	.link_update			= sfc_dev_link_update,
1642 	.stats_get			= sfc_stats_get,
1643 	.stats_reset			= sfc_stats_reset,
1644 	.xstats_get			= sfc_xstats_get,
1645 	.xstats_reset			= sfc_stats_reset,
1646 	.xstats_get_names		= sfc_xstats_get_names,
1647 	.dev_infos_get			= sfc_dev_infos_get,
1648 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1649 	.mtu_set			= sfc_dev_set_mtu,
1650 	.rx_queue_start			= sfc_rx_queue_start,
1651 	.rx_queue_stop			= sfc_rx_queue_stop,
1652 	.tx_queue_start			= sfc_tx_queue_start,
1653 	.tx_queue_stop			= sfc_tx_queue_stop,
1654 	.rx_queue_setup			= sfc_rx_queue_setup,
1655 	.rx_queue_release		= sfc_rx_queue_release,
1656 	.rx_queue_count			= sfc_rx_queue_count,
1657 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1658 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1659 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1660 	.tx_queue_setup			= sfc_tx_queue_setup,
1661 	.tx_queue_release		= sfc_tx_queue_release,
1662 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1663 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1664 	.mac_addr_set			= sfc_mac_addr_set,
1665 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1666 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1667 	.reta_update			= sfc_dev_rss_reta_update,
1668 	.reta_query			= sfc_dev_rss_reta_query,
1669 	.rss_hash_update		= sfc_dev_rss_hash_update,
1670 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1671 	.filter_ctrl			= sfc_dev_filter_ctrl,
1672 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1673 	.rxq_info_get			= sfc_rx_queue_info_get,
1674 	.txq_info_get			= sfc_tx_queue_info_get,
1675 	.fw_version_get			= sfc_fw_version_get,
1676 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1677 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1678 };
1679 
1680 /**
1681  * Duplicate a string in potentially shared memory required for
1682  * multi-process support.
1683  *
1684  * strdup() allocates from process-local heap/memory.
1685  */
1686 static char *
1687 sfc_strdup(const char *str)
1688 {
1689 	size_t size;
1690 	char *copy;
1691 
1692 	if (str == NULL)
1693 		return NULL;
1694 
1695 	size = strlen(str) + 1;
1696 	copy = rte_malloc(__func__, size, 0);
1697 	if (copy != NULL)
1698 		rte_memcpy(copy, str, size);
1699 
1700 	return copy;
1701 }
1702 
1703 static int
1704 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1705 {
1706 	struct sfc_adapter *sa = dev->data->dev_private;
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 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1723 				sfc_kvarg_string_handler, &rx_name);
1724 	if (rc != 0)
1725 		goto fail_kvarg_rx_datapath;
1726 
1727 	if (rx_name != NULL) {
1728 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1729 		if (sa->dp_rx == NULL) {
1730 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1731 			rc = ENOENT;
1732 			goto fail_dp_rx;
1733 		}
1734 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1735 			sfc_err(sa,
1736 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1737 				rx_name);
1738 			rc = EINVAL;
1739 			goto fail_dp_rx_caps;
1740 		}
1741 	} else {
1742 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1743 		if (sa->dp_rx == NULL) {
1744 			sfc_err(sa, "Rx datapath by caps %#x not found",
1745 				avail_caps);
1746 			rc = ENOENT;
1747 			goto fail_dp_rx;
1748 		}
1749 	}
1750 
1751 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1752 	if (sa->dp_rx_name == NULL) {
1753 		rc = ENOMEM;
1754 		goto fail_dp_rx_name;
1755 	}
1756 
1757 	sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1758 
1759 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1760 
1761 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1762 				sfc_kvarg_string_handler, &tx_name);
1763 	if (rc != 0)
1764 		goto fail_kvarg_tx_datapath;
1765 
1766 	if (tx_name != NULL) {
1767 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1768 		if (sa->dp_tx == NULL) {
1769 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1770 			rc = ENOENT;
1771 			goto fail_dp_tx;
1772 		}
1773 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1774 			sfc_err(sa,
1775 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1776 				tx_name);
1777 			rc = EINVAL;
1778 			goto fail_dp_tx_caps;
1779 		}
1780 	} else {
1781 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1782 		if (sa->dp_tx == NULL) {
1783 			sfc_err(sa, "Tx datapath by caps %#x not found",
1784 				avail_caps);
1785 			rc = ENOENT;
1786 			goto fail_dp_tx;
1787 		}
1788 	}
1789 
1790 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1791 	if (sa->dp_tx_name == NULL) {
1792 		rc = ENOMEM;
1793 		goto fail_dp_tx_name;
1794 	}
1795 
1796 	sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1797 
1798 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1799 
1800 	dev->dev_ops = &sfc_eth_dev_ops;
1801 
1802 	return 0;
1803 
1804 fail_dp_tx_name:
1805 fail_dp_tx_caps:
1806 	sa->dp_tx = NULL;
1807 
1808 fail_dp_tx:
1809 fail_kvarg_tx_datapath:
1810 	rte_free(sa->dp_rx_name);
1811 	sa->dp_rx_name = NULL;
1812 
1813 fail_dp_rx_name:
1814 fail_dp_rx_caps:
1815 	sa->dp_rx = NULL;
1816 
1817 fail_dp_rx:
1818 fail_kvarg_rx_datapath:
1819 	return rc;
1820 }
1821 
1822 static void
1823 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1824 {
1825 	struct sfc_adapter *sa = dev->data->dev_private;
1826 
1827 	dev->dev_ops = NULL;
1828 	dev->rx_pkt_burst = NULL;
1829 	dev->tx_pkt_burst = NULL;
1830 
1831 	rte_free(sa->dp_tx_name);
1832 	sa->dp_tx_name = NULL;
1833 	sa->dp_tx = NULL;
1834 
1835 	rte_free(sa->dp_rx_name);
1836 	sa->dp_rx_name = NULL;
1837 	sa->dp_rx = NULL;
1838 }
1839 
1840 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1841 	.rxq_info_get			= sfc_rx_queue_info_get,
1842 	.txq_info_get			= sfc_tx_queue_info_get,
1843 };
1844 
1845 static int
1846 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1847 {
1848 	/*
1849 	 * Device private data has really many process-local pointers.
1850 	 * Below code should be extremely careful to use data located
1851 	 * in shared memory only.
1852 	 */
1853 	struct sfc_adapter *sa = dev->data->dev_private;
1854 	const struct sfc_dp_rx *dp_rx;
1855 	const struct sfc_dp_tx *dp_tx;
1856 	int rc;
1857 
1858 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1859 	if (dp_rx == NULL) {
1860 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1861 		rc = ENOENT;
1862 		goto fail_dp_rx;
1863 	}
1864 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1865 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1866 			sa->dp_tx_name);
1867 		rc = EINVAL;
1868 		goto fail_dp_rx_multi_process;
1869 	}
1870 
1871 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1872 	if (dp_tx == NULL) {
1873 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1874 		rc = ENOENT;
1875 		goto fail_dp_tx;
1876 	}
1877 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1878 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1879 			sa->dp_tx_name);
1880 		rc = EINVAL;
1881 		goto fail_dp_tx_multi_process;
1882 	}
1883 
1884 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1885 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1886 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1887 
1888 	return 0;
1889 
1890 fail_dp_tx_multi_process:
1891 fail_dp_tx:
1892 fail_dp_rx_multi_process:
1893 fail_dp_rx:
1894 	return rc;
1895 }
1896 
1897 static void
1898 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1899 {
1900 	dev->dev_ops = NULL;
1901 	dev->tx_pkt_burst = NULL;
1902 	dev->rx_pkt_burst = NULL;
1903 }
1904 
1905 static void
1906 sfc_register_dp(void)
1907 {
1908 	/* Register once */
1909 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1910 		/* Prefer EF10 datapath */
1911 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1912 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1913 
1914 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1915 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1916 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1917 	}
1918 }
1919 
1920 static int
1921 sfc_eth_dev_init(struct rte_eth_dev *dev)
1922 {
1923 	struct sfc_adapter *sa = dev->data->dev_private;
1924 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1925 	int rc;
1926 	const efx_nic_cfg_t *encp;
1927 	const struct ether_addr *from;
1928 
1929 	sfc_register_dp();
1930 
1931 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1932 		return -sfc_eth_dev_secondary_set_ops(dev);
1933 
1934 	/* Required for logging */
1935 	sa->pci_addr = pci_dev->addr;
1936 	sa->port_id = dev->data->port_id;
1937 
1938 	sa->eth_dev = dev;
1939 
1940 	/* Copy PCI device info to the dev->data */
1941 	rte_eth_copy_pci_info(dev, pci_dev);
1942 
1943 	sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1944 						RTE_LOG_NOTICE);
1945 
1946 	rc = sfc_kvargs_parse(sa);
1947 	if (rc != 0)
1948 		goto fail_kvargs_parse;
1949 
1950 	sfc_log_init(sa, "entry");
1951 
1952 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1953 	if (dev->data->mac_addrs == NULL) {
1954 		rc = ENOMEM;
1955 		goto fail_mac_addrs;
1956 	}
1957 
1958 	sfc_adapter_lock_init(sa);
1959 	sfc_adapter_lock(sa);
1960 
1961 	sfc_log_init(sa, "probing");
1962 	rc = sfc_probe(sa);
1963 	if (rc != 0)
1964 		goto fail_probe;
1965 
1966 	sfc_log_init(sa, "set device ops");
1967 	rc = sfc_eth_dev_set_ops(dev);
1968 	if (rc != 0)
1969 		goto fail_set_ops;
1970 
1971 	sfc_log_init(sa, "attaching");
1972 	rc = sfc_attach(sa);
1973 	if (rc != 0)
1974 		goto fail_attach;
1975 
1976 	encp = efx_nic_cfg_get(sa->nic);
1977 
1978 	/*
1979 	 * The arguments are really reverse order in comparison to
1980 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1981 	 */
1982 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1983 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1984 
1985 	sfc_adapter_unlock(sa);
1986 
1987 	sfc_log_init(sa, "done");
1988 	return 0;
1989 
1990 fail_attach:
1991 	sfc_eth_dev_clear_ops(dev);
1992 
1993 fail_set_ops:
1994 	sfc_unprobe(sa);
1995 
1996 fail_probe:
1997 	sfc_adapter_unlock(sa);
1998 	sfc_adapter_lock_fini(sa);
1999 	rte_free(dev->data->mac_addrs);
2000 	dev->data->mac_addrs = NULL;
2001 
2002 fail_mac_addrs:
2003 	sfc_kvargs_cleanup(sa);
2004 
2005 fail_kvargs_parse:
2006 	sfc_log_init(sa, "failed %d", rc);
2007 	SFC_ASSERT(rc > 0);
2008 	return -rc;
2009 }
2010 
2011 static int
2012 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2013 {
2014 	struct sfc_adapter *sa;
2015 
2016 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2017 		sfc_eth_dev_secondary_clear_ops(dev);
2018 		return 0;
2019 	}
2020 
2021 	sa = dev->data->dev_private;
2022 	sfc_log_init(sa, "entry");
2023 
2024 	sfc_adapter_lock(sa);
2025 
2026 	sfc_eth_dev_clear_ops(dev);
2027 
2028 	sfc_detach(sa);
2029 	sfc_unprobe(sa);
2030 
2031 	rte_free(dev->data->mac_addrs);
2032 	dev->data->mac_addrs = NULL;
2033 
2034 	sfc_kvargs_cleanup(sa);
2035 
2036 	sfc_adapter_unlock(sa);
2037 	sfc_adapter_lock_fini(sa);
2038 
2039 	sfc_log_init(sa, "done");
2040 
2041 	/* Required for logging, so cleanup last */
2042 	sa->eth_dev = NULL;
2043 	return 0;
2044 }
2045 
2046 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2047 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2048 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2049 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2050 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2051 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2052 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2053 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2054 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2055 	{ .vendor_id = 0 /* sentinel */ }
2056 };
2057 
2058 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2059 	struct rte_pci_device *pci_dev)
2060 {
2061 	return rte_eth_dev_pci_generic_probe(pci_dev,
2062 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2063 }
2064 
2065 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2066 {
2067 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2068 }
2069 
2070 static struct rte_pci_driver sfc_efx_pmd = {
2071 	.id_table = pci_id_sfc_efx_map,
2072 	.drv_flags =
2073 		RTE_PCI_DRV_INTR_LSC |
2074 		RTE_PCI_DRV_NEED_MAPPING,
2075 	.probe = sfc_eth_dev_pci_probe,
2076 	.remove = sfc_eth_dev_pci_remove,
2077 };
2078 
2079 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2080 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2081 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2082 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2083 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2084 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2085 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2086 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2087 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2088 
2089 RTE_INIT(sfc_driver_register_logtype);
2090 static void
2091 sfc_driver_register_logtype(void)
2092 {
2093 	int ret;
2094 
2095 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2096 						   RTE_LOG_NOTICE);
2097 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2098 }
2099