xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 5e23c24988f2df5b7797e57eff93fd8af8ceeca7)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
4  * All rights reserved.
5  *
6  * This software was jointly developed between OKTET Labs (under contract
7  * for Solarflare) and Solarflare Communications, Inc.
8  */
9 
10 #include <rte_dev.h>
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
16 
17 #include "efx.h"
18 
19 #include "sfc.h"
20 #include "sfc_debug.h"
21 #include "sfc_log.h"
22 #include "sfc_kvargs.h"
23 #include "sfc_ev.h"
24 #include "sfc_rx.h"
25 #include "sfc_tx.h"
26 #include "sfc_flow.h"
27 #include "sfc_dp.h"
28 #include "sfc_dp_rx.h"
29 
30 uint32_t sfc_logtype_driver;
31 
32 static struct sfc_dp_list sfc_dp_head =
33 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
34 
35 static int
36 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
37 {
38 	struct sfc_adapter *sa = dev->data->dev_private;
39 	efx_nic_fw_info_t enfi;
40 	int ret;
41 	int rc;
42 
43 	/*
44 	 * Return value of the callback is likely supposed to be
45 	 * equal to or greater than 0, nevertheless, if an error
46 	 * occurs, it will be desirable to pass it to the caller
47 	 */
48 	if ((fw_version == NULL) || (fw_size == 0))
49 		return -EINVAL;
50 
51 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
52 	if (rc != 0)
53 		return -rc;
54 
55 	ret = snprintf(fw_version, fw_size,
56 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
57 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
58 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
59 	if (ret < 0)
60 		return ret;
61 
62 	if (enfi.enfi_dpcpu_fw_ids_valid) {
63 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
64 		int ret_extra;
65 
66 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
67 				     fw_size - dpcpu_fw_ids_offset,
68 				     " rx%" PRIx16 " tx%" PRIx16,
69 				     enfi.enfi_rx_dpcpu_fw_id,
70 				     enfi.enfi_tx_dpcpu_fw_id);
71 		if (ret_extra < 0)
72 			return ret_extra;
73 
74 		ret += ret_extra;
75 	}
76 
77 	if (fw_size < (size_t)(++ret))
78 		return ret;
79 	else
80 		return 0;
81 }
82 
83 static void
84 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
85 {
86 	struct sfc_adapter *sa = dev->data->dev_private;
87 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
88 	uint64_t txq_offloads_def = 0;
89 
90 	sfc_log_init(sa, "entry");
91 
92 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
93 
94 	/* Autonegotiation may be disabled */
95 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
96 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
97 		dev_info->speed_capa |= ETH_LINK_SPEED_1G;
98 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
99 		dev_info->speed_capa |= ETH_LINK_SPEED_10G;
100 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_25000FDX)
101 		dev_info->speed_capa |= ETH_LINK_SPEED_25G;
102 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
103 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
104 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_50000FDX)
105 		dev_info->speed_capa |= ETH_LINK_SPEED_50G;
106 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_100000FDX)
107 		dev_info->speed_capa |= ETH_LINK_SPEED_100G;
108 
109 	dev_info->max_rx_queues = sa->rxq_max;
110 	dev_info->max_tx_queues = sa->txq_max;
111 
112 	/* By default packets are dropped if no descriptors are available */
113 	dev_info->default_rxconf.rx_drop_en = 1;
114 
115 	dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
116 
117 	/*
118 	 * rx_offload_capa includes both device and queue offloads since
119 	 * the latter may be requested on a per device basis which makes
120 	 * sense when some offloads are needed to be set on all queues.
121 	 */
122 	dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
123 				    dev_info->rx_queue_offload_capa;
124 
125 	dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
126 
127 	/*
128 	 * tx_offload_capa includes both device and queue offloads since
129 	 * the latter may be requested on a per device basis which makes
130 	 * sense when some offloads are needed to be set on all queues.
131 	 */
132 	dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
133 				    dev_info->tx_queue_offload_capa;
134 
135 	if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
136 		txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
137 
138 	dev_info->default_txconf.offloads |= txq_offloads_def;
139 
140 	dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
141 	if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
142 	    !encp->enc_hw_tx_insert_vlan_enabled)
143 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
144 
145 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
146 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
147 
148 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
149 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
150 
151 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
152 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
153 
154 	if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
155 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
156 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
157 		dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
158 	}
159 
160 	/* Initialize to hardware limits */
161 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
162 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
163 	/* The RXQ hardware requires that the descriptor count is a power
164 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
165 	 */
166 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
167 
168 	/* Initialize to hardware limits */
169 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
170 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
171 	/*
172 	 * The TXQ hardware requires that the descriptor count is a power
173 	 * of 2, but tx_desc_lim cannot properly describe that constraint
174 	 */
175 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
176 
177 	if (sa->dp_rx->get_dev_info != NULL)
178 		sa->dp_rx->get_dev_info(dev_info);
179 	if (sa->dp_tx->get_dev_info != NULL)
180 		sa->dp_tx->get_dev_info(dev_info);
181 }
182 
183 static const uint32_t *
184 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
185 {
186 	struct sfc_adapter *sa = dev->data->dev_private;
187 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
188 	uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
189 
190 	return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
191 }
192 
193 static int
194 sfc_dev_configure(struct rte_eth_dev *dev)
195 {
196 	struct rte_eth_dev_data *dev_data = dev->data;
197 	struct sfc_adapter *sa = dev_data->dev_private;
198 	int rc;
199 
200 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
201 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
202 
203 	sfc_adapter_lock(sa);
204 	switch (sa->state) {
205 	case SFC_ADAPTER_CONFIGURED:
206 		/* FALLTHROUGH */
207 	case SFC_ADAPTER_INITIALIZED:
208 		rc = sfc_configure(sa);
209 		break;
210 	default:
211 		sfc_err(sa, "unexpected adapter state %u to configure",
212 			sa->state);
213 		rc = EINVAL;
214 		break;
215 	}
216 	sfc_adapter_unlock(sa);
217 
218 	sfc_log_init(sa, "done %d", rc);
219 	SFC_ASSERT(rc >= 0);
220 	return -rc;
221 }
222 
223 static int
224 sfc_dev_start(struct rte_eth_dev *dev)
225 {
226 	struct sfc_adapter *sa = dev->data->dev_private;
227 	int rc;
228 
229 	sfc_log_init(sa, "entry");
230 
231 	sfc_adapter_lock(sa);
232 	rc = sfc_start(sa);
233 	sfc_adapter_unlock(sa);
234 
235 	sfc_log_init(sa, "done %d", rc);
236 	SFC_ASSERT(rc >= 0);
237 	return -rc;
238 }
239 
240 static int
241 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
242 {
243 	struct sfc_adapter *sa = dev->data->dev_private;
244 	struct rte_eth_link current_link;
245 	int ret;
246 
247 	sfc_log_init(sa, "entry");
248 
249 	if (sa->state != SFC_ADAPTER_STARTED) {
250 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
251 	} else if (wait_to_complete) {
252 		efx_link_mode_t link_mode;
253 
254 		if (efx_port_poll(sa->nic, &link_mode) != 0)
255 			link_mode = EFX_LINK_UNKNOWN;
256 		sfc_port_link_mode_to_info(link_mode, &current_link);
257 
258 	} else {
259 		sfc_ev_mgmt_qpoll(sa);
260 		rte_eth_linkstatus_get(dev, &current_link);
261 	}
262 
263 	ret = rte_eth_linkstatus_set(dev, &current_link);
264 	if (ret == 0)
265 		sfc_notice(sa, "Link status is %s",
266 			   current_link.link_status ? "UP" : "DOWN");
267 
268 	return ret;
269 }
270 
271 static void
272 sfc_dev_stop(struct rte_eth_dev *dev)
273 {
274 	struct sfc_adapter *sa = dev->data->dev_private;
275 
276 	sfc_log_init(sa, "entry");
277 
278 	sfc_adapter_lock(sa);
279 	sfc_stop(sa);
280 	sfc_adapter_unlock(sa);
281 
282 	sfc_log_init(sa, "done");
283 }
284 
285 static int
286 sfc_dev_set_link_up(struct rte_eth_dev *dev)
287 {
288 	struct sfc_adapter *sa = dev->data->dev_private;
289 	int rc;
290 
291 	sfc_log_init(sa, "entry");
292 
293 	sfc_adapter_lock(sa);
294 	rc = sfc_start(sa);
295 	sfc_adapter_unlock(sa);
296 
297 	SFC_ASSERT(rc >= 0);
298 	return -rc;
299 }
300 
301 static int
302 sfc_dev_set_link_down(struct rte_eth_dev *dev)
303 {
304 	struct sfc_adapter *sa = dev->data->dev_private;
305 
306 	sfc_log_init(sa, "entry");
307 
308 	sfc_adapter_lock(sa);
309 	sfc_stop(sa);
310 	sfc_adapter_unlock(sa);
311 
312 	return 0;
313 }
314 
315 static void
316 sfc_dev_close(struct rte_eth_dev *dev)
317 {
318 	struct sfc_adapter *sa = dev->data->dev_private;
319 
320 	sfc_log_init(sa, "entry");
321 
322 	sfc_adapter_lock(sa);
323 	switch (sa->state) {
324 	case SFC_ADAPTER_STARTED:
325 		sfc_stop(sa);
326 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
327 		/* FALLTHROUGH */
328 	case SFC_ADAPTER_CONFIGURED:
329 		sfc_close(sa);
330 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
331 		/* FALLTHROUGH */
332 	case SFC_ADAPTER_INITIALIZED:
333 		break;
334 	default:
335 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
336 		break;
337 	}
338 	sfc_adapter_unlock(sa);
339 
340 	sfc_log_init(sa, "done");
341 }
342 
343 static void
344 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
345 		   boolean_t enabled)
346 {
347 	struct sfc_port *port;
348 	boolean_t *toggle;
349 	struct sfc_adapter *sa = dev->data->dev_private;
350 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
351 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
352 
353 	sfc_adapter_lock(sa);
354 
355 	port = &sa->port;
356 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
357 
358 	if (*toggle != enabled) {
359 		*toggle = enabled;
360 
361 		if (port->isolated) {
362 			sfc_warn(sa, "isolated mode is active on the port");
363 			sfc_warn(sa, "the change is to be applied on the next "
364 				     "start provided that isolated mode is "
365 				     "disabled prior the next start");
366 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
367 			   (sfc_set_rx_mode(sa) != 0)) {
368 			*toggle = !(enabled);
369 			sfc_warn(sa, "Failed to %s %s mode",
370 				 ((enabled) ? "enable" : "disable"), desc);
371 		}
372 	}
373 
374 	sfc_adapter_unlock(sa);
375 }
376 
377 static void
378 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
379 {
380 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
381 }
382 
383 static void
384 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
385 {
386 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
387 }
388 
389 static void
390 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
391 {
392 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
393 }
394 
395 static void
396 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
397 {
398 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
399 }
400 
401 static int
402 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
403 		   uint16_t nb_rx_desc, unsigned int socket_id,
404 		   const struct rte_eth_rxconf *rx_conf,
405 		   struct rte_mempool *mb_pool)
406 {
407 	struct sfc_adapter *sa = dev->data->dev_private;
408 	int rc;
409 
410 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
411 		     rx_queue_id, nb_rx_desc, socket_id);
412 
413 	sfc_adapter_lock(sa);
414 
415 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
416 			  rx_conf, mb_pool);
417 	if (rc != 0)
418 		goto fail_rx_qinit;
419 
420 	dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
421 
422 	sfc_adapter_unlock(sa);
423 
424 	return 0;
425 
426 fail_rx_qinit:
427 	sfc_adapter_unlock(sa);
428 	SFC_ASSERT(rc > 0);
429 	return -rc;
430 }
431 
432 static void
433 sfc_rx_queue_release(void *queue)
434 {
435 	struct sfc_dp_rxq *dp_rxq = queue;
436 	struct sfc_rxq *rxq;
437 	struct sfc_adapter *sa;
438 	unsigned int sw_index;
439 
440 	if (dp_rxq == NULL)
441 		return;
442 
443 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
444 	sa = rxq->evq->sa;
445 	sfc_adapter_lock(sa);
446 
447 	sw_index = sfc_rxq_sw_index(rxq);
448 
449 	sfc_log_init(sa, "RxQ=%u", sw_index);
450 
451 	sa->eth_dev->data->rx_queues[sw_index] = NULL;
452 
453 	sfc_rx_qfini(sa, sw_index);
454 
455 	sfc_adapter_unlock(sa);
456 }
457 
458 static int
459 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
460 		   uint16_t nb_tx_desc, unsigned int socket_id,
461 		   const struct rte_eth_txconf *tx_conf)
462 {
463 	struct sfc_adapter *sa = dev->data->dev_private;
464 	int rc;
465 
466 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
467 		     tx_queue_id, nb_tx_desc, socket_id);
468 
469 	sfc_adapter_lock(sa);
470 
471 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
472 	if (rc != 0)
473 		goto fail_tx_qinit;
474 
475 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
476 
477 	sfc_adapter_unlock(sa);
478 	return 0;
479 
480 fail_tx_qinit:
481 	sfc_adapter_unlock(sa);
482 	SFC_ASSERT(rc > 0);
483 	return -rc;
484 }
485 
486 static void
487 sfc_tx_queue_release(void *queue)
488 {
489 	struct sfc_dp_txq *dp_txq = queue;
490 	struct sfc_txq *txq;
491 	unsigned int sw_index;
492 	struct sfc_adapter *sa;
493 
494 	if (dp_txq == NULL)
495 		return;
496 
497 	txq = sfc_txq_by_dp_txq(dp_txq);
498 	sw_index = sfc_txq_sw_index(txq);
499 
500 	SFC_ASSERT(txq->evq != NULL);
501 	sa = txq->evq->sa;
502 
503 	sfc_log_init(sa, "TxQ = %u", sw_index);
504 
505 	sfc_adapter_lock(sa);
506 
507 	SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
508 	sa->eth_dev->data->tx_queues[sw_index] = NULL;
509 
510 	sfc_tx_qfini(sa, sw_index);
511 
512 	sfc_adapter_unlock(sa);
513 }
514 
515 static int
516 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
517 {
518 	struct sfc_adapter *sa = dev->data->dev_private;
519 	struct sfc_port *port = &sa->port;
520 	uint64_t *mac_stats;
521 	int ret;
522 
523 	rte_spinlock_lock(&port->mac_stats_lock);
524 
525 	ret = sfc_port_update_mac_stats(sa);
526 	if (ret != 0)
527 		goto unlock;
528 
529 	mac_stats = port->mac_stats_buf;
530 
531 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
532 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
533 		stats->ipackets =
534 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
535 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
536 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
537 		stats->opackets =
538 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
539 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
540 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
541 		stats->ibytes =
542 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
543 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
544 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
545 		stats->obytes =
546 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
547 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
548 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
549 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
550 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
551 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
552 	} else {
553 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
554 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
555 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
556 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
557 		/*
558 		 * Take into account stats which are whenever supported
559 		 * on EF10. If some stat is not supported by current
560 		 * firmware variant or HW revision, it is guaranteed
561 		 * to be zero in mac_stats.
562 		 */
563 		stats->imissed =
564 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
565 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
566 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
567 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
568 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
569 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
570 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
571 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
572 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
573 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
574 		stats->ierrors =
575 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
576 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
577 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
578 		/* no oerrors counters supported on EF10 */
579 	}
580 
581 unlock:
582 	rte_spinlock_unlock(&port->mac_stats_lock);
583 	SFC_ASSERT(ret >= 0);
584 	return -ret;
585 }
586 
587 static void
588 sfc_stats_reset(struct rte_eth_dev *dev)
589 {
590 	struct sfc_adapter *sa = dev->data->dev_private;
591 	struct sfc_port *port = &sa->port;
592 	int rc;
593 
594 	if (sa->state != SFC_ADAPTER_STARTED) {
595 		/*
596 		 * The operation cannot be done if port is not started; it
597 		 * will be scheduled to be done during the next port start
598 		 */
599 		port->mac_stats_reset_pending = B_TRUE;
600 		return;
601 	}
602 
603 	rc = sfc_port_reset_mac_stats(sa);
604 	if (rc != 0)
605 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
606 }
607 
608 static int
609 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
610 	       unsigned int xstats_count)
611 {
612 	struct sfc_adapter *sa = dev->data->dev_private;
613 	struct sfc_port *port = &sa->port;
614 	uint64_t *mac_stats;
615 	int rc;
616 	unsigned int i;
617 	int nstats = 0;
618 
619 	rte_spinlock_lock(&port->mac_stats_lock);
620 
621 	rc = sfc_port_update_mac_stats(sa);
622 	if (rc != 0) {
623 		SFC_ASSERT(rc > 0);
624 		nstats = -rc;
625 		goto unlock;
626 	}
627 
628 	mac_stats = port->mac_stats_buf;
629 
630 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
631 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
632 			if (xstats != NULL && nstats < (int)xstats_count) {
633 				xstats[nstats].id = nstats;
634 				xstats[nstats].value = mac_stats[i];
635 			}
636 			nstats++;
637 		}
638 	}
639 
640 unlock:
641 	rte_spinlock_unlock(&port->mac_stats_lock);
642 
643 	return nstats;
644 }
645 
646 static int
647 sfc_xstats_get_names(struct rte_eth_dev *dev,
648 		     struct rte_eth_xstat_name *xstats_names,
649 		     unsigned int xstats_count)
650 {
651 	struct sfc_adapter *sa = dev->data->dev_private;
652 	struct sfc_port *port = &sa->port;
653 	unsigned int i;
654 	unsigned int nstats = 0;
655 
656 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
657 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
658 			if (xstats_names != NULL && nstats < xstats_count)
659 				strncpy(xstats_names[nstats].name,
660 					efx_mac_stat_name(sa->nic, i),
661 					sizeof(xstats_names[0].name));
662 			nstats++;
663 		}
664 	}
665 
666 	return nstats;
667 }
668 
669 static int
670 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
671 		     uint64_t *values, unsigned int n)
672 {
673 	struct sfc_adapter *sa = dev->data->dev_private;
674 	struct sfc_port *port = &sa->port;
675 	uint64_t *mac_stats;
676 	unsigned int nb_supported = 0;
677 	unsigned int nb_written = 0;
678 	unsigned int i;
679 	int ret;
680 	int rc;
681 
682 	if (unlikely(values == NULL) ||
683 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
684 		return port->mac_stats_nb_supported;
685 
686 	rte_spinlock_lock(&port->mac_stats_lock);
687 
688 	rc = sfc_port_update_mac_stats(sa);
689 	if (rc != 0) {
690 		SFC_ASSERT(rc > 0);
691 		ret = -rc;
692 		goto unlock;
693 	}
694 
695 	mac_stats = port->mac_stats_buf;
696 
697 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
698 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
699 			continue;
700 
701 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
702 			values[nb_written++] = mac_stats[i];
703 
704 		++nb_supported;
705 	}
706 
707 	ret = nb_written;
708 
709 unlock:
710 	rte_spinlock_unlock(&port->mac_stats_lock);
711 
712 	return ret;
713 }
714 
715 static int
716 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
717 			   struct rte_eth_xstat_name *xstats_names,
718 			   const uint64_t *ids, unsigned int size)
719 {
720 	struct sfc_adapter *sa = dev->data->dev_private;
721 	struct sfc_port *port = &sa->port;
722 	unsigned int nb_supported = 0;
723 	unsigned int nb_written = 0;
724 	unsigned int i;
725 
726 	if (unlikely(xstats_names == NULL) ||
727 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
728 		return port->mac_stats_nb_supported;
729 
730 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
731 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
732 			continue;
733 
734 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
735 			char *name = xstats_names[nb_written++].name;
736 
737 			strncpy(name, efx_mac_stat_name(sa->nic, i),
738 				sizeof(xstats_names[0].name));
739 			name[sizeof(xstats_names[0].name) - 1] = '\0';
740 		}
741 
742 		++nb_supported;
743 	}
744 
745 	return nb_written;
746 }
747 
748 static int
749 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
750 {
751 	struct sfc_adapter *sa = dev->data->dev_private;
752 	unsigned int wanted_fc, link_fc;
753 
754 	memset(fc_conf, 0, sizeof(*fc_conf));
755 
756 	sfc_adapter_lock(sa);
757 
758 	if (sa->state == SFC_ADAPTER_STARTED)
759 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
760 	else
761 		link_fc = sa->port.flow_ctrl;
762 
763 	switch (link_fc) {
764 	case 0:
765 		fc_conf->mode = RTE_FC_NONE;
766 		break;
767 	case EFX_FCNTL_RESPOND:
768 		fc_conf->mode = RTE_FC_RX_PAUSE;
769 		break;
770 	case EFX_FCNTL_GENERATE:
771 		fc_conf->mode = RTE_FC_TX_PAUSE;
772 		break;
773 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
774 		fc_conf->mode = RTE_FC_FULL;
775 		break;
776 	default:
777 		sfc_err(sa, "%s: unexpected flow control value %#x",
778 			__func__, link_fc);
779 	}
780 
781 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
782 
783 	sfc_adapter_unlock(sa);
784 
785 	return 0;
786 }
787 
788 static int
789 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
790 {
791 	struct sfc_adapter *sa = dev->data->dev_private;
792 	struct sfc_port *port = &sa->port;
793 	unsigned int fcntl;
794 	int rc;
795 
796 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
797 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
798 	    fc_conf->mac_ctrl_frame_fwd != 0) {
799 		sfc_err(sa, "unsupported flow control settings specified");
800 		rc = EINVAL;
801 		goto fail_inval;
802 	}
803 
804 	switch (fc_conf->mode) {
805 	case RTE_FC_NONE:
806 		fcntl = 0;
807 		break;
808 	case RTE_FC_RX_PAUSE:
809 		fcntl = EFX_FCNTL_RESPOND;
810 		break;
811 	case RTE_FC_TX_PAUSE:
812 		fcntl = EFX_FCNTL_GENERATE;
813 		break;
814 	case RTE_FC_FULL:
815 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
816 		break;
817 	default:
818 		rc = EINVAL;
819 		goto fail_inval;
820 	}
821 
822 	sfc_adapter_lock(sa);
823 
824 	if (sa->state == SFC_ADAPTER_STARTED) {
825 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
826 		if (rc != 0)
827 			goto fail_mac_fcntl_set;
828 	}
829 
830 	port->flow_ctrl = fcntl;
831 	port->flow_ctrl_autoneg = fc_conf->autoneg;
832 
833 	sfc_adapter_unlock(sa);
834 
835 	return 0;
836 
837 fail_mac_fcntl_set:
838 	sfc_adapter_unlock(sa);
839 fail_inval:
840 	SFC_ASSERT(rc > 0);
841 	return -rc;
842 }
843 
844 static int
845 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
846 {
847 	struct sfc_adapter *sa = dev->data->dev_private;
848 	size_t pdu = EFX_MAC_PDU(mtu);
849 	size_t old_pdu;
850 	int rc;
851 
852 	sfc_log_init(sa, "mtu=%u", mtu);
853 
854 	rc = EINVAL;
855 	if (pdu < EFX_MAC_PDU_MIN) {
856 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
857 			(unsigned int)mtu, (unsigned int)pdu,
858 			EFX_MAC_PDU_MIN);
859 		goto fail_inval;
860 	}
861 	if (pdu > EFX_MAC_PDU_MAX) {
862 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
863 			(unsigned int)mtu, (unsigned int)pdu,
864 			EFX_MAC_PDU_MAX);
865 		goto fail_inval;
866 	}
867 
868 	sfc_adapter_lock(sa);
869 
870 	if (pdu != sa->port.pdu) {
871 		if (sa->state == SFC_ADAPTER_STARTED) {
872 			sfc_stop(sa);
873 
874 			old_pdu = sa->port.pdu;
875 			sa->port.pdu = pdu;
876 			rc = sfc_start(sa);
877 			if (rc != 0)
878 				goto fail_start;
879 		} else {
880 			sa->port.pdu = pdu;
881 		}
882 	}
883 
884 	/*
885 	 * The driver does not use it, but other PMDs update jumbo_frame
886 	 * flag and max_rx_pkt_len when MTU is set.
887 	 */
888 	if (mtu > ETHER_MAX_LEN) {
889 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
890 
891 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
892 		rxmode->jumbo_frame = 1;
893 	}
894 
895 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
896 
897 	sfc_adapter_unlock(sa);
898 
899 	sfc_log_init(sa, "done");
900 	return 0;
901 
902 fail_start:
903 	sa->port.pdu = old_pdu;
904 	if (sfc_start(sa) != 0)
905 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
906 			"PDU max size - port is stopped",
907 			(unsigned int)pdu, (unsigned int)old_pdu);
908 	sfc_adapter_unlock(sa);
909 
910 fail_inval:
911 	sfc_log_init(sa, "failed %d", rc);
912 	SFC_ASSERT(rc > 0);
913 	return -rc;
914 }
915 static int
916 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
917 {
918 	struct sfc_adapter *sa = dev->data->dev_private;
919 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
920 	struct sfc_port *port = &sa->port;
921 	struct ether_addr *old_addr = &dev->data->mac_addrs[0];
922 	int rc = 0;
923 
924 	sfc_adapter_lock(sa);
925 
926 	/*
927 	 * Copy the address to the device private data so that
928 	 * it could be recalled in the case of adapter restart.
929 	 */
930 	ether_addr_copy(mac_addr, &port->default_mac_addr);
931 
932 	/*
933 	 * Neither of the two following checks can return
934 	 * an error. The new MAC address is preserved in
935 	 * the device private data and can be activated
936 	 * on the next port start if the user prevents
937 	 * isolated mode from being enabled.
938 	 */
939 	if (port->isolated) {
940 		sfc_warn(sa, "isolated mode is active on the port");
941 		sfc_warn(sa, "will not set MAC address");
942 		goto unlock;
943 	}
944 
945 	if (sa->state != SFC_ADAPTER_STARTED) {
946 		sfc_notice(sa, "the port is not started");
947 		sfc_notice(sa, "the new MAC address will be set on port start");
948 
949 		goto unlock;
950 	}
951 
952 	if (encp->enc_allow_set_mac_with_installed_filters) {
953 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
954 		if (rc != 0) {
955 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
956 			goto unlock;
957 		}
958 
959 		/*
960 		 * Changing the MAC address by means of MCDI request
961 		 * has no effect on received traffic, therefore
962 		 * we also need to update unicast filters
963 		 */
964 		rc = sfc_set_rx_mode(sa);
965 		if (rc != 0) {
966 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
967 			/* Rollback the old address */
968 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
969 			(void)sfc_set_rx_mode(sa);
970 		}
971 	} else {
972 		sfc_warn(sa, "cannot set MAC address with filters installed");
973 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
974 		sfc_warn(sa, "(some traffic may be dropped)");
975 
976 		/*
977 		 * Since setting MAC address with filters installed is not
978 		 * allowed on the adapter, the new MAC address will be set
979 		 * by means of adapter restart. sfc_start() shall retrieve
980 		 * the new address from the device private data and set it.
981 		 */
982 		sfc_stop(sa);
983 		rc = sfc_start(sa);
984 		if (rc != 0)
985 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
986 	}
987 
988 unlock:
989 	if (rc != 0)
990 		ether_addr_copy(old_addr, &port->default_mac_addr);
991 
992 	sfc_adapter_unlock(sa);
993 
994 	SFC_ASSERT(rc >= 0);
995 	return -rc;
996 }
997 
998 
999 static int
1000 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1001 		     uint32_t nb_mc_addr)
1002 {
1003 	struct sfc_adapter *sa = dev->data->dev_private;
1004 	struct sfc_port *port = &sa->port;
1005 	uint8_t *mc_addrs = port->mcast_addrs;
1006 	int rc;
1007 	unsigned int i;
1008 
1009 	if (port->isolated) {
1010 		sfc_err(sa, "isolated mode is active on the port");
1011 		sfc_err(sa, "will not set multicast address list");
1012 		return -ENOTSUP;
1013 	}
1014 
1015 	if (mc_addrs == NULL)
1016 		return -ENOBUFS;
1017 
1018 	if (nb_mc_addr > port->max_mcast_addrs) {
1019 		sfc_err(sa, "too many multicast addresses: %u > %u",
1020 			 nb_mc_addr, port->max_mcast_addrs);
1021 		return -EINVAL;
1022 	}
1023 
1024 	for (i = 0; i < nb_mc_addr; ++i) {
1025 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1026 				 EFX_MAC_ADDR_LEN);
1027 		mc_addrs += EFX_MAC_ADDR_LEN;
1028 	}
1029 
1030 	port->nb_mcast_addrs = nb_mc_addr;
1031 
1032 	if (sa->state != SFC_ADAPTER_STARTED)
1033 		return 0;
1034 
1035 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1036 					port->nb_mcast_addrs);
1037 	if (rc != 0)
1038 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1039 
1040 	SFC_ASSERT(rc > 0);
1041 	return -rc;
1042 }
1043 
1044 /*
1045  * The function is used by the secondary process as well. It must not
1046  * use any process-local pointers from the adapter data.
1047  */
1048 static void
1049 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1050 		      struct rte_eth_rxq_info *qinfo)
1051 {
1052 	struct sfc_adapter *sa = dev->data->dev_private;
1053 	struct sfc_rxq_info *rxq_info;
1054 	struct sfc_rxq *rxq;
1055 
1056 	sfc_adapter_lock(sa);
1057 
1058 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1059 
1060 	rxq_info = &sa->rxq_info[rx_queue_id];
1061 	rxq = rxq_info->rxq;
1062 	SFC_ASSERT(rxq != NULL);
1063 
1064 	qinfo->mp = rxq->refill_mb_pool;
1065 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1066 	qinfo->conf.rx_drop_en = 1;
1067 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1068 	qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1069 			       DEV_RX_OFFLOAD_UDP_CKSUM |
1070 			       DEV_RX_OFFLOAD_TCP_CKSUM;
1071 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1072 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1073 		qinfo->scattered_rx = 1;
1074 	}
1075 	qinfo->nb_desc = rxq_info->entries;
1076 
1077 	sfc_adapter_unlock(sa);
1078 }
1079 
1080 /*
1081  * The function is used by the secondary process as well. It must not
1082  * use any process-local pointers from the adapter data.
1083  */
1084 static void
1085 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1086 		      struct rte_eth_txq_info *qinfo)
1087 {
1088 	struct sfc_adapter *sa = dev->data->dev_private;
1089 	struct sfc_txq_info *txq_info;
1090 
1091 	sfc_adapter_lock(sa);
1092 
1093 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1094 
1095 	txq_info = &sa->txq_info[tx_queue_id];
1096 	SFC_ASSERT(txq_info->txq != NULL);
1097 
1098 	memset(qinfo, 0, sizeof(*qinfo));
1099 
1100 	qinfo->conf.txq_flags = txq_info->txq->flags;
1101 	qinfo->conf.offloads = txq_info->txq->offloads;
1102 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1103 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1104 	qinfo->nb_desc = txq_info->entries;
1105 
1106 	sfc_adapter_unlock(sa);
1107 }
1108 
1109 static uint32_t
1110 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1111 {
1112 	struct sfc_adapter *sa = dev->data->dev_private;
1113 
1114 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1115 
1116 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
1117 }
1118 
1119 static int
1120 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1121 {
1122 	struct sfc_dp_rxq *dp_rxq = queue;
1123 
1124 	return sfc_rx_qdesc_done(dp_rxq, offset);
1125 }
1126 
1127 static int
1128 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1129 {
1130 	struct sfc_dp_rxq *dp_rxq = queue;
1131 	struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1132 
1133 	return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1134 }
1135 
1136 static int
1137 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1138 {
1139 	struct sfc_dp_txq *dp_txq = queue;
1140 	struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1141 
1142 	return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1143 }
1144 
1145 static int
1146 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1147 {
1148 	struct sfc_adapter *sa = dev->data->dev_private;
1149 	int rc;
1150 
1151 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1152 
1153 	sfc_adapter_lock(sa);
1154 
1155 	rc = EINVAL;
1156 	if (sa->state != SFC_ADAPTER_STARTED)
1157 		goto fail_not_started;
1158 
1159 	rc = sfc_rx_qstart(sa, rx_queue_id);
1160 	if (rc != 0)
1161 		goto fail_rx_qstart;
1162 
1163 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1164 
1165 	sfc_adapter_unlock(sa);
1166 
1167 	return 0;
1168 
1169 fail_rx_qstart:
1170 fail_not_started:
1171 	sfc_adapter_unlock(sa);
1172 	SFC_ASSERT(rc > 0);
1173 	return -rc;
1174 }
1175 
1176 static int
1177 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1178 {
1179 	struct sfc_adapter *sa = dev->data->dev_private;
1180 
1181 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1182 
1183 	sfc_adapter_lock(sa);
1184 	sfc_rx_qstop(sa, rx_queue_id);
1185 
1186 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1187 
1188 	sfc_adapter_unlock(sa);
1189 
1190 	return 0;
1191 }
1192 
1193 static int
1194 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1195 {
1196 	struct sfc_adapter *sa = dev->data->dev_private;
1197 	int rc;
1198 
1199 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1200 
1201 	sfc_adapter_lock(sa);
1202 
1203 	rc = EINVAL;
1204 	if (sa->state != SFC_ADAPTER_STARTED)
1205 		goto fail_not_started;
1206 
1207 	rc = sfc_tx_qstart(sa, tx_queue_id);
1208 	if (rc != 0)
1209 		goto fail_tx_qstart;
1210 
1211 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1212 
1213 	sfc_adapter_unlock(sa);
1214 	return 0;
1215 
1216 fail_tx_qstart:
1217 
1218 fail_not_started:
1219 	sfc_adapter_unlock(sa);
1220 	SFC_ASSERT(rc > 0);
1221 	return -rc;
1222 }
1223 
1224 static int
1225 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1226 {
1227 	struct sfc_adapter *sa = dev->data->dev_private;
1228 
1229 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1230 
1231 	sfc_adapter_lock(sa);
1232 
1233 	sfc_tx_qstop(sa, tx_queue_id);
1234 
1235 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1236 
1237 	sfc_adapter_unlock(sa);
1238 	return 0;
1239 }
1240 
1241 static efx_tunnel_protocol_t
1242 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1243 {
1244 	switch (rte_type) {
1245 	case RTE_TUNNEL_TYPE_VXLAN:
1246 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1247 	case RTE_TUNNEL_TYPE_GENEVE:
1248 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1249 	default:
1250 		return EFX_TUNNEL_NPROTOS;
1251 	}
1252 }
1253 
1254 enum sfc_udp_tunnel_op_e {
1255 	SFC_UDP_TUNNEL_ADD_PORT,
1256 	SFC_UDP_TUNNEL_DEL_PORT,
1257 };
1258 
1259 static int
1260 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1261 		      struct rte_eth_udp_tunnel *tunnel_udp,
1262 		      enum sfc_udp_tunnel_op_e op)
1263 {
1264 	struct sfc_adapter *sa = dev->data->dev_private;
1265 	efx_tunnel_protocol_t tunnel_proto;
1266 	int rc;
1267 
1268 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1269 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1270 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1271 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1272 
1273 	tunnel_proto =
1274 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1275 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1276 		rc = ENOTSUP;
1277 		goto fail_bad_proto;
1278 	}
1279 
1280 	sfc_adapter_lock(sa);
1281 
1282 	switch (op) {
1283 	case SFC_UDP_TUNNEL_ADD_PORT:
1284 		rc = efx_tunnel_config_udp_add(sa->nic,
1285 					       tunnel_udp->udp_port,
1286 					       tunnel_proto);
1287 		break;
1288 	case SFC_UDP_TUNNEL_DEL_PORT:
1289 		rc = efx_tunnel_config_udp_remove(sa->nic,
1290 						  tunnel_udp->udp_port,
1291 						  tunnel_proto);
1292 		break;
1293 	default:
1294 		rc = EINVAL;
1295 		goto fail_bad_op;
1296 	}
1297 
1298 	if (rc != 0)
1299 		goto fail_op;
1300 
1301 	if (sa->state == SFC_ADAPTER_STARTED) {
1302 		rc = efx_tunnel_reconfigure(sa->nic);
1303 		if (rc == EAGAIN) {
1304 			/*
1305 			 * Configuration is accepted by FW and MC reboot
1306 			 * is initiated to apply the changes. MC reboot
1307 			 * will be handled in a usual way (MC reboot
1308 			 * event on management event queue and adapter
1309 			 * restart).
1310 			 */
1311 			rc = 0;
1312 		} else if (rc != 0) {
1313 			goto fail_reconfigure;
1314 		}
1315 	}
1316 
1317 	sfc_adapter_unlock(sa);
1318 	return 0;
1319 
1320 fail_reconfigure:
1321 	/* Remove/restore entry since the change makes the trouble */
1322 	switch (op) {
1323 	case SFC_UDP_TUNNEL_ADD_PORT:
1324 		(void)efx_tunnel_config_udp_remove(sa->nic,
1325 						   tunnel_udp->udp_port,
1326 						   tunnel_proto);
1327 		break;
1328 	case SFC_UDP_TUNNEL_DEL_PORT:
1329 		(void)efx_tunnel_config_udp_add(sa->nic,
1330 						tunnel_udp->udp_port,
1331 						tunnel_proto);
1332 		break;
1333 	}
1334 
1335 fail_op:
1336 fail_bad_op:
1337 	sfc_adapter_unlock(sa);
1338 
1339 fail_bad_proto:
1340 	SFC_ASSERT(rc > 0);
1341 	return -rc;
1342 }
1343 
1344 static int
1345 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1346 			    struct rte_eth_udp_tunnel *tunnel_udp)
1347 {
1348 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1349 }
1350 
1351 static int
1352 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1353 			    struct rte_eth_udp_tunnel *tunnel_udp)
1354 {
1355 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1356 }
1357 
1358 static int
1359 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1360 			  struct rte_eth_rss_conf *rss_conf)
1361 {
1362 	struct sfc_adapter *sa = dev->data->dev_private;
1363 	struct sfc_port *port = &sa->port;
1364 
1365 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1366 		return -ENOTSUP;
1367 
1368 	if (sa->rss_channels == 0)
1369 		return -EINVAL;
1370 
1371 	sfc_adapter_lock(sa);
1372 
1373 	/*
1374 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1375 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1376 	 * flags which corresponds to the active EFX configuration stored
1377 	 * locally in 'sfc_adapter' and kept up-to-date
1378 	 */
1379 	rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1380 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1381 	if (rss_conf->rss_key != NULL)
1382 		rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1383 
1384 	sfc_adapter_unlock(sa);
1385 
1386 	return 0;
1387 }
1388 
1389 static int
1390 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1391 			struct rte_eth_rss_conf *rss_conf)
1392 {
1393 	struct sfc_adapter *sa = dev->data->dev_private;
1394 	struct sfc_port *port = &sa->port;
1395 	unsigned int efx_hash_types;
1396 	int rc = 0;
1397 
1398 	if (port->isolated)
1399 		return -ENOTSUP;
1400 
1401 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1402 		sfc_err(sa, "RSS is not available");
1403 		return -ENOTSUP;
1404 	}
1405 
1406 	if (sa->rss_channels == 0) {
1407 		sfc_err(sa, "RSS is not configured");
1408 		return -EINVAL;
1409 	}
1410 
1411 	if ((rss_conf->rss_key != NULL) &&
1412 	    (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1413 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1414 			sizeof(sa->rss_key));
1415 		return -EINVAL;
1416 	}
1417 
1418 	if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1419 		sfc_err(sa, "unsupported hash functions requested");
1420 		return -EINVAL;
1421 	}
1422 
1423 	sfc_adapter_lock(sa);
1424 
1425 	efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1426 
1427 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1428 				   EFX_RX_HASHALG_TOEPLITZ,
1429 				   efx_hash_types, B_TRUE);
1430 	if (rc != 0)
1431 		goto fail_scale_mode_set;
1432 
1433 	if (rss_conf->rss_key != NULL) {
1434 		if (sa->state == SFC_ADAPTER_STARTED) {
1435 			rc = efx_rx_scale_key_set(sa->nic,
1436 						  EFX_RSS_CONTEXT_DEFAULT,
1437 						  rss_conf->rss_key,
1438 						  sizeof(sa->rss_key));
1439 			if (rc != 0)
1440 				goto fail_scale_key_set;
1441 		}
1442 
1443 		rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1444 	}
1445 
1446 	sa->rss_hash_types = efx_hash_types;
1447 
1448 	sfc_adapter_unlock(sa);
1449 
1450 	return 0;
1451 
1452 fail_scale_key_set:
1453 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1454 				  EFX_RX_HASHALG_TOEPLITZ,
1455 				  sa->rss_hash_types, B_TRUE) != 0)
1456 		sfc_err(sa, "failed to restore RSS mode");
1457 
1458 fail_scale_mode_set:
1459 	sfc_adapter_unlock(sa);
1460 	return -rc;
1461 }
1462 
1463 static int
1464 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1465 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1466 		       uint16_t reta_size)
1467 {
1468 	struct sfc_adapter *sa = dev->data->dev_private;
1469 	struct sfc_port *port = &sa->port;
1470 	int entry;
1471 
1472 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1473 		return -ENOTSUP;
1474 
1475 	if (sa->rss_channels == 0)
1476 		return -EINVAL;
1477 
1478 	if (reta_size != EFX_RSS_TBL_SIZE)
1479 		return -EINVAL;
1480 
1481 	sfc_adapter_lock(sa);
1482 
1483 	for (entry = 0; entry < reta_size; entry++) {
1484 		int grp = entry / RTE_RETA_GROUP_SIZE;
1485 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1486 
1487 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1488 			reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1489 	}
1490 
1491 	sfc_adapter_unlock(sa);
1492 
1493 	return 0;
1494 }
1495 
1496 static int
1497 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1498 			struct rte_eth_rss_reta_entry64 *reta_conf,
1499 			uint16_t reta_size)
1500 {
1501 	struct sfc_adapter *sa = dev->data->dev_private;
1502 	struct sfc_port *port = &sa->port;
1503 	unsigned int *rss_tbl_new;
1504 	uint16_t entry;
1505 	int rc = 0;
1506 
1507 
1508 	if (port->isolated)
1509 		return -ENOTSUP;
1510 
1511 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1512 		sfc_err(sa, "RSS is not available");
1513 		return -ENOTSUP;
1514 	}
1515 
1516 	if (sa->rss_channels == 0) {
1517 		sfc_err(sa, "RSS is not configured");
1518 		return -EINVAL;
1519 	}
1520 
1521 	if (reta_size != EFX_RSS_TBL_SIZE) {
1522 		sfc_err(sa, "RETA size is wrong (should be %u)",
1523 			EFX_RSS_TBL_SIZE);
1524 		return -EINVAL;
1525 	}
1526 
1527 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1528 	if (rss_tbl_new == NULL)
1529 		return -ENOMEM;
1530 
1531 	sfc_adapter_lock(sa);
1532 
1533 	rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1534 
1535 	for (entry = 0; entry < reta_size; entry++) {
1536 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1537 		struct rte_eth_rss_reta_entry64 *grp;
1538 
1539 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1540 
1541 		if (grp->mask & (1ull << grp_idx)) {
1542 			if (grp->reta[grp_idx] >= sa->rss_channels) {
1543 				rc = EINVAL;
1544 				goto bad_reta_entry;
1545 			}
1546 			rss_tbl_new[entry] = grp->reta[grp_idx];
1547 		}
1548 	}
1549 
1550 	if (sa->state == SFC_ADAPTER_STARTED) {
1551 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1552 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1553 		if (rc != 0)
1554 			goto fail_scale_tbl_set;
1555 	}
1556 
1557 	rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1558 
1559 fail_scale_tbl_set:
1560 bad_reta_entry:
1561 	sfc_adapter_unlock(sa);
1562 
1563 	rte_free(rss_tbl_new);
1564 
1565 	SFC_ASSERT(rc >= 0);
1566 	return -rc;
1567 }
1568 
1569 static int
1570 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1571 		    enum rte_filter_op filter_op,
1572 		    void *arg)
1573 {
1574 	struct sfc_adapter *sa = dev->data->dev_private;
1575 	int rc = ENOTSUP;
1576 
1577 	sfc_log_init(sa, "entry");
1578 
1579 	switch (filter_type) {
1580 	case RTE_ETH_FILTER_NONE:
1581 		sfc_err(sa, "Global filters configuration not supported");
1582 		break;
1583 	case RTE_ETH_FILTER_MACVLAN:
1584 		sfc_err(sa, "MACVLAN filters not supported");
1585 		break;
1586 	case RTE_ETH_FILTER_ETHERTYPE:
1587 		sfc_err(sa, "EtherType filters not supported");
1588 		break;
1589 	case RTE_ETH_FILTER_FLEXIBLE:
1590 		sfc_err(sa, "Flexible filters not supported");
1591 		break;
1592 	case RTE_ETH_FILTER_SYN:
1593 		sfc_err(sa, "SYN filters not supported");
1594 		break;
1595 	case RTE_ETH_FILTER_NTUPLE:
1596 		sfc_err(sa, "NTUPLE filters not supported");
1597 		break;
1598 	case RTE_ETH_FILTER_TUNNEL:
1599 		sfc_err(sa, "Tunnel filters not supported");
1600 		break;
1601 	case RTE_ETH_FILTER_FDIR:
1602 		sfc_err(sa, "Flow Director filters not supported");
1603 		break;
1604 	case RTE_ETH_FILTER_HASH:
1605 		sfc_err(sa, "Hash filters not supported");
1606 		break;
1607 	case RTE_ETH_FILTER_GENERIC:
1608 		if (filter_op != RTE_ETH_FILTER_GET) {
1609 			rc = EINVAL;
1610 		} else {
1611 			*(const void **)arg = &sfc_flow_ops;
1612 			rc = 0;
1613 		}
1614 		break;
1615 	default:
1616 		sfc_err(sa, "Unknown filter type %u", filter_type);
1617 		break;
1618 	}
1619 
1620 	sfc_log_init(sa, "exit: %d", -rc);
1621 	SFC_ASSERT(rc >= 0);
1622 	return -rc;
1623 }
1624 
1625 static const struct eth_dev_ops sfc_eth_dev_ops = {
1626 	.dev_configure			= sfc_dev_configure,
1627 	.dev_start			= sfc_dev_start,
1628 	.dev_stop			= sfc_dev_stop,
1629 	.dev_set_link_up		= sfc_dev_set_link_up,
1630 	.dev_set_link_down		= sfc_dev_set_link_down,
1631 	.dev_close			= sfc_dev_close,
1632 	.promiscuous_enable		= sfc_dev_promisc_enable,
1633 	.promiscuous_disable		= sfc_dev_promisc_disable,
1634 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1635 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1636 	.link_update			= sfc_dev_link_update,
1637 	.stats_get			= sfc_stats_get,
1638 	.stats_reset			= sfc_stats_reset,
1639 	.xstats_get			= sfc_xstats_get,
1640 	.xstats_reset			= sfc_stats_reset,
1641 	.xstats_get_names		= sfc_xstats_get_names,
1642 	.dev_infos_get			= sfc_dev_infos_get,
1643 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1644 	.mtu_set			= sfc_dev_set_mtu,
1645 	.rx_queue_start			= sfc_rx_queue_start,
1646 	.rx_queue_stop			= sfc_rx_queue_stop,
1647 	.tx_queue_start			= sfc_tx_queue_start,
1648 	.tx_queue_stop			= sfc_tx_queue_stop,
1649 	.rx_queue_setup			= sfc_rx_queue_setup,
1650 	.rx_queue_release		= sfc_rx_queue_release,
1651 	.rx_queue_count			= sfc_rx_queue_count,
1652 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1653 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1654 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1655 	.tx_queue_setup			= sfc_tx_queue_setup,
1656 	.tx_queue_release		= sfc_tx_queue_release,
1657 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1658 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1659 	.mac_addr_set			= sfc_mac_addr_set,
1660 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1661 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1662 	.reta_update			= sfc_dev_rss_reta_update,
1663 	.reta_query			= sfc_dev_rss_reta_query,
1664 	.rss_hash_update		= sfc_dev_rss_hash_update,
1665 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1666 	.filter_ctrl			= sfc_dev_filter_ctrl,
1667 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1668 	.rxq_info_get			= sfc_rx_queue_info_get,
1669 	.txq_info_get			= sfc_tx_queue_info_get,
1670 	.fw_version_get			= sfc_fw_version_get,
1671 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1672 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1673 };
1674 
1675 /**
1676  * Duplicate a string in potentially shared memory required for
1677  * multi-process support.
1678  *
1679  * strdup() allocates from process-local heap/memory.
1680  */
1681 static char *
1682 sfc_strdup(const char *str)
1683 {
1684 	size_t size;
1685 	char *copy;
1686 
1687 	if (str == NULL)
1688 		return NULL;
1689 
1690 	size = strlen(str) + 1;
1691 	copy = rte_malloc(__func__, size, 0);
1692 	if (copy != NULL)
1693 		rte_memcpy(copy, str, size);
1694 
1695 	return copy;
1696 }
1697 
1698 static int
1699 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1700 {
1701 	struct sfc_adapter *sa = dev->data->dev_private;
1702 	unsigned int avail_caps = 0;
1703 	const char *rx_name = NULL;
1704 	const char *tx_name = NULL;
1705 	int rc;
1706 
1707 	switch (sa->family) {
1708 	case EFX_FAMILY_HUNTINGTON:
1709 	case EFX_FAMILY_MEDFORD:
1710 	case EFX_FAMILY_MEDFORD2:
1711 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1712 		break;
1713 	default:
1714 		break;
1715 	}
1716 
1717 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1718 				sfc_kvarg_string_handler, &rx_name);
1719 	if (rc != 0)
1720 		goto fail_kvarg_rx_datapath;
1721 
1722 	if (rx_name != NULL) {
1723 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1724 		if (sa->dp_rx == NULL) {
1725 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1726 			rc = ENOENT;
1727 			goto fail_dp_rx;
1728 		}
1729 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1730 			sfc_err(sa,
1731 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1732 				rx_name);
1733 			rc = EINVAL;
1734 			goto fail_dp_rx_caps;
1735 		}
1736 	} else {
1737 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1738 		if (sa->dp_rx == NULL) {
1739 			sfc_err(sa, "Rx datapath by caps %#x not found",
1740 				avail_caps);
1741 			rc = ENOENT;
1742 			goto fail_dp_rx;
1743 		}
1744 	}
1745 
1746 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1747 	if (sa->dp_rx_name == NULL) {
1748 		rc = ENOMEM;
1749 		goto fail_dp_rx_name;
1750 	}
1751 
1752 	sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1753 
1754 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1755 
1756 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1757 				sfc_kvarg_string_handler, &tx_name);
1758 	if (rc != 0)
1759 		goto fail_kvarg_tx_datapath;
1760 
1761 	if (tx_name != NULL) {
1762 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1763 		if (sa->dp_tx == NULL) {
1764 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1765 			rc = ENOENT;
1766 			goto fail_dp_tx;
1767 		}
1768 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1769 			sfc_err(sa,
1770 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1771 				tx_name);
1772 			rc = EINVAL;
1773 			goto fail_dp_tx_caps;
1774 		}
1775 	} else {
1776 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1777 		if (sa->dp_tx == NULL) {
1778 			sfc_err(sa, "Tx datapath by caps %#x not found",
1779 				avail_caps);
1780 			rc = ENOENT;
1781 			goto fail_dp_tx;
1782 		}
1783 	}
1784 
1785 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1786 	if (sa->dp_tx_name == NULL) {
1787 		rc = ENOMEM;
1788 		goto fail_dp_tx_name;
1789 	}
1790 
1791 	sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1792 
1793 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1794 
1795 	dev->dev_ops = &sfc_eth_dev_ops;
1796 
1797 	return 0;
1798 
1799 fail_dp_tx_name:
1800 fail_dp_tx_caps:
1801 	sa->dp_tx = NULL;
1802 
1803 fail_dp_tx:
1804 fail_kvarg_tx_datapath:
1805 	rte_free(sa->dp_rx_name);
1806 	sa->dp_rx_name = NULL;
1807 
1808 fail_dp_rx_name:
1809 fail_dp_rx_caps:
1810 	sa->dp_rx = NULL;
1811 
1812 fail_dp_rx:
1813 fail_kvarg_rx_datapath:
1814 	return rc;
1815 }
1816 
1817 static void
1818 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1819 {
1820 	struct sfc_adapter *sa = dev->data->dev_private;
1821 
1822 	dev->dev_ops = NULL;
1823 	dev->rx_pkt_burst = NULL;
1824 	dev->tx_pkt_burst = NULL;
1825 
1826 	rte_free(sa->dp_tx_name);
1827 	sa->dp_tx_name = NULL;
1828 	sa->dp_tx = NULL;
1829 
1830 	rte_free(sa->dp_rx_name);
1831 	sa->dp_rx_name = NULL;
1832 	sa->dp_rx = NULL;
1833 }
1834 
1835 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1836 	.rxq_info_get			= sfc_rx_queue_info_get,
1837 	.txq_info_get			= sfc_tx_queue_info_get,
1838 };
1839 
1840 static int
1841 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1842 {
1843 	/*
1844 	 * Device private data has really many process-local pointers.
1845 	 * Below code should be extremely careful to use data located
1846 	 * in shared memory only.
1847 	 */
1848 	struct sfc_adapter *sa = dev->data->dev_private;
1849 	const struct sfc_dp_rx *dp_rx;
1850 	const struct sfc_dp_tx *dp_tx;
1851 	int rc;
1852 
1853 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1854 	if (dp_rx == NULL) {
1855 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1856 		rc = ENOENT;
1857 		goto fail_dp_rx;
1858 	}
1859 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1860 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1861 			sa->dp_tx_name);
1862 		rc = EINVAL;
1863 		goto fail_dp_rx_multi_process;
1864 	}
1865 
1866 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1867 	if (dp_tx == NULL) {
1868 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1869 		rc = ENOENT;
1870 		goto fail_dp_tx;
1871 	}
1872 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1873 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1874 			sa->dp_tx_name);
1875 		rc = EINVAL;
1876 		goto fail_dp_tx_multi_process;
1877 	}
1878 
1879 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1880 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1881 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1882 
1883 	return 0;
1884 
1885 fail_dp_tx_multi_process:
1886 fail_dp_tx:
1887 fail_dp_rx_multi_process:
1888 fail_dp_rx:
1889 	return rc;
1890 }
1891 
1892 static void
1893 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1894 {
1895 	dev->dev_ops = NULL;
1896 	dev->tx_pkt_burst = NULL;
1897 	dev->rx_pkt_burst = NULL;
1898 }
1899 
1900 static void
1901 sfc_register_dp(void)
1902 {
1903 	/* Register once */
1904 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1905 		/* Prefer EF10 datapath */
1906 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1907 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1908 
1909 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1910 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1911 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1912 	}
1913 }
1914 
1915 static int
1916 sfc_eth_dev_init(struct rte_eth_dev *dev)
1917 {
1918 	struct sfc_adapter *sa = dev->data->dev_private;
1919 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1920 	int rc;
1921 	const efx_nic_cfg_t *encp;
1922 	const struct ether_addr *from;
1923 
1924 	sfc_register_dp();
1925 
1926 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1927 		return -sfc_eth_dev_secondary_set_ops(dev);
1928 
1929 	/* Required for logging */
1930 	sa->pci_addr = pci_dev->addr;
1931 	sa->port_id = dev->data->port_id;
1932 
1933 	sa->eth_dev = dev;
1934 
1935 	/* Copy PCI device info to the dev->data */
1936 	rte_eth_copy_pci_info(dev, pci_dev);
1937 
1938 	sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1939 						RTE_LOG_NOTICE);
1940 
1941 	rc = sfc_kvargs_parse(sa);
1942 	if (rc != 0)
1943 		goto fail_kvargs_parse;
1944 
1945 	sfc_log_init(sa, "entry");
1946 
1947 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1948 	if (dev->data->mac_addrs == NULL) {
1949 		rc = ENOMEM;
1950 		goto fail_mac_addrs;
1951 	}
1952 
1953 	sfc_adapter_lock_init(sa);
1954 	sfc_adapter_lock(sa);
1955 
1956 	sfc_log_init(sa, "probing");
1957 	rc = sfc_probe(sa);
1958 	if (rc != 0)
1959 		goto fail_probe;
1960 
1961 	sfc_log_init(sa, "set device ops");
1962 	rc = sfc_eth_dev_set_ops(dev);
1963 	if (rc != 0)
1964 		goto fail_set_ops;
1965 
1966 	sfc_log_init(sa, "attaching");
1967 	rc = sfc_attach(sa);
1968 	if (rc != 0)
1969 		goto fail_attach;
1970 
1971 	encp = efx_nic_cfg_get(sa->nic);
1972 
1973 	/*
1974 	 * The arguments are really reverse order in comparison to
1975 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1976 	 */
1977 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1978 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1979 
1980 	sfc_adapter_unlock(sa);
1981 
1982 	sfc_log_init(sa, "done");
1983 	return 0;
1984 
1985 fail_attach:
1986 	sfc_eth_dev_clear_ops(dev);
1987 
1988 fail_set_ops:
1989 	sfc_unprobe(sa);
1990 
1991 fail_probe:
1992 	sfc_adapter_unlock(sa);
1993 	sfc_adapter_lock_fini(sa);
1994 	rte_free(dev->data->mac_addrs);
1995 	dev->data->mac_addrs = NULL;
1996 
1997 fail_mac_addrs:
1998 	sfc_kvargs_cleanup(sa);
1999 
2000 fail_kvargs_parse:
2001 	sfc_log_init(sa, "failed %d", rc);
2002 	SFC_ASSERT(rc > 0);
2003 	return -rc;
2004 }
2005 
2006 static int
2007 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2008 {
2009 	struct sfc_adapter *sa;
2010 
2011 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2012 		sfc_eth_dev_secondary_clear_ops(dev);
2013 		return 0;
2014 	}
2015 
2016 	sa = dev->data->dev_private;
2017 	sfc_log_init(sa, "entry");
2018 
2019 	sfc_adapter_lock(sa);
2020 
2021 	sfc_eth_dev_clear_ops(dev);
2022 
2023 	sfc_detach(sa);
2024 	sfc_unprobe(sa);
2025 
2026 	rte_free(dev->data->mac_addrs);
2027 	dev->data->mac_addrs = NULL;
2028 
2029 	sfc_kvargs_cleanup(sa);
2030 
2031 	sfc_adapter_unlock(sa);
2032 	sfc_adapter_lock_fini(sa);
2033 
2034 	sfc_log_init(sa, "done");
2035 
2036 	/* Required for logging, so cleanup last */
2037 	sa->eth_dev = NULL;
2038 	return 0;
2039 }
2040 
2041 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2042 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2043 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2044 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2045 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2046 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2047 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2048 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2049 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2050 	{ .vendor_id = 0 /* sentinel */ }
2051 };
2052 
2053 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2054 	struct rte_pci_device *pci_dev)
2055 {
2056 	return rte_eth_dev_pci_generic_probe(pci_dev,
2057 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2058 }
2059 
2060 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2061 {
2062 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2063 }
2064 
2065 static struct rte_pci_driver sfc_efx_pmd = {
2066 	.id_table = pci_id_sfc_efx_map,
2067 	.drv_flags =
2068 		RTE_PCI_DRV_INTR_LSC |
2069 		RTE_PCI_DRV_NEED_MAPPING,
2070 	.probe = sfc_eth_dev_pci_probe,
2071 	.remove = sfc_eth_dev_pci_remove,
2072 };
2073 
2074 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2075 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2076 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2077 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2078 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2079 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2080 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2081 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2082 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2083 
2084 RTE_INIT(sfc_driver_register_logtype);
2085 static void
2086 sfc_driver_register_logtype(void)
2087 {
2088 	int ret;
2089 
2090 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2091 						   RTE_LOG_NOTICE);
2092 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2093 }
2094