xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision f3129efdb4974575fa003284d76469b432817b0c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
4  * All rights reserved.
5  *
6  * This software was jointly developed between OKTET Labs (under contract
7  * for Solarflare) and Solarflare Communications, Inc.
8  */
9 
10 #include <rte_dev.h>
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
16 
17 #include "efx.h"
18 
19 #include "sfc.h"
20 #include "sfc_debug.h"
21 #include "sfc_log.h"
22 #include "sfc_kvargs.h"
23 #include "sfc_ev.h"
24 #include "sfc_rx.h"
25 #include "sfc_tx.h"
26 #include "sfc_flow.h"
27 #include "sfc_dp.h"
28 #include "sfc_dp_rx.h"
29 
30 uint32_t sfc_logtype_driver;
31 
32 static struct sfc_dp_list sfc_dp_head =
33 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
34 
35 static int
36 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
37 {
38 	struct sfc_adapter *sa = dev->data->dev_private;
39 	efx_nic_fw_info_t enfi;
40 	int ret;
41 	int rc;
42 
43 	/*
44 	 * Return value of the callback is likely supposed to be
45 	 * equal to or greater than 0, nevertheless, if an error
46 	 * occurs, it will be desirable to pass it to the caller
47 	 */
48 	if ((fw_version == NULL) || (fw_size == 0))
49 		return -EINVAL;
50 
51 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
52 	if (rc != 0)
53 		return -rc;
54 
55 	ret = snprintf(fw_version, fw_size,
56 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
57 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
58 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
59 	if (ret < 0)
60 		return ret;
61 
62 	if (enfi.enfi_dpcpu_fw_ids_valid) {
63 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
64 		int ret_extra;
65 
66 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
67 				     fw_size - dpcpu_fw_ids_offset,
68 				     " rx%" PRIx16 " tx%" PRIx16,
69 				     enfi.enfi_rx_dpcpu_fw_id,
70 				     enfi.enfi_tx_dpcpu_fw_id);
71 		if (ret_extra < 0)
72 			return ret_extra;
73 
74 		ret += ret_extra;
75 	}
76 
77 	if (fw_size < (size_t)(++ret))
78 		return ret;
79 	else
80 		return 0;
81 }
82 
83 static void
84 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
85 {
86 	struct sfc_adapter *sa = dev->data->dev_private;
87 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
88 	uint64_t txq_offloads_def = 0;
89 
90 	sfc_log_init(sa, "entry");
91 
92 	dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
93 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
94 
95 	/* Autonegotiation may be disabled */
96 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
97 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
98 		dev_info->speed_capa |= ETH_LINK_SPEED_1G;
99 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
100 		dev_info->speed_capa |= ETH_LINK_SPEED_10G;
101 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_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 EFSYS_OPT_RX_SCALE
156 	if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
157 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
158 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
159 		dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
160 	}
161 #endif
162 
163 	/* Initialize to hardware limits */
164 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
165 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
166 	/* The RXQ hardware requires that the descriptor count is a power
167 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
168 	 */
169 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
170 
171 	/* Initialize to hardware limits */
172 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
173 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
174 	/*
175 	 * The TXQ hardware requires that the descriptor count is a power
176 	 * of 2, but tx_desc_lim cannot properly describe that constraint
177 	 */
178 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
179 
180 	if (sa->dp_rx->get_dev_info != NULL)
181 		sa->dp_rx->get_dev_info(dev_info);
182 	if (sa->dp_tx->get_dev_info != NULL)
183 		sa->dp_tx->get_dev_info(dev_info);
184 }
185 
186 static const uint32_t *
187 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
188 {
189 	struct sfc_adapter *sa = dev->data->dev_private;
190 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
191 	uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
192 
193 	return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
194 }
195 
196 static int
197 sfc_dev_configure(struct rte_eth_dev *dev)
198 {
199 	struct rte_eth_dev_data *dev_data = dev->data;
200 	struct sfc_adapter *sa = dev_data->dev_private;
201 	int rc;
202 
203 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
204 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
205 
206 	sfc_adapter_lock(sa);
207 	switch (sa->state) {
208 	case SFC_ADAPTER_CONFIGURED:
209 		/* FALLTHROUGH */
210 	case SFC_ADAPTER_INITIALIZED:
211 		rc = sfc_configure(sa);
212 		break;
213 	default:
214 		sfc_err(sa, "unexpected adapter state %u to configure",
215 			sa->state);
216 		rc = EINVAL;
217 		break;
218 	}
219 	sfc_adapter_unlock(sa);
220 
221 	sfc_log_init(sa, "done %d", rc);
222 	SFC_ASSERT(rc >= 0);
223 	return -rc;
224 }
225 
226 static int
227 sfc_dev_start(struct rte_eth_dev *dev)
228 {
229 	struct sfc_adapter *sa = dev->data->dev_private;
230 	int rc;
231 
232 	sfc_log_init(sa, "entry");
233 
234 	sfc_adapter_lock(sa);
235 	rc = sfc_start(sa);
236 	sfc_adapter_unlock(sa);
237 
238 	sfc_log_init(sa, "done %d", rc);
239 	SFC_ASSERT(rc >= 0);
240 	return -rc;
241 }
242 
243 static int
244 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
245 {
246 	struct sfc_adapter *sa = dev->data->dev_private;
247 	struct rte_eth_link current_link;
248 	int ret;
249 
250 	sfc_log_init(sa, "entry");
251 
252 	if (sa->state != SFC_ADAPTER_STARTED) {
253 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
254 	} else if (wait_to_complete) {
255 		efx_link_mode_t link_mode;
256 
257 		if (efx_port_poll(sa->nic, &link_mode) != 0)
258 			link_mode = EFX_LINK_UNKNOWN;
259 		sfc_port_link_mode_to_info(link_mode, &current_link);
260 
261 	} else {
262 		sfc_ev_mgmt_qpoll(sa);
263 		rte_eth_linkstatus_get(dev, &current_link);
264 	}
265 
266 	ret = rte_eth_linkstatus_set(dev, &current_link);
267 	if (ret == 0)
268 		sfc_notice(sa, "Link status is %s",
269 			   current_link.link_status ? "UP" : "DOWN");
270 
271 	return ret;
272 }
273 
274 static void
275 sfc_dev_stop(struct rte_eth_dev *dev)
276 {
277 	struct sfc_adapter *sa = dev->data->dev_private;
278 
279 	sfc_log_init(sa, "entry");
280 
281 	sfc_adapter_lock(sa);
282 	sfc_stop(sa);
283 	sfc_adapter_unlock(sa);
284 
285 	sfc_log_init(sa, "done");
286 }
287 
288 static int
289 sfc_dev_set_link_up(struct rte_eth_dev *dev)
290 {
291 	struct sfc_adapter *sa = dev->data->dev_private;
292 	int rc;
293 
294 	sfc_log_init(sa, "entry");
295 
296 	sfc_adapter_lock(sa);
297 	rc = sfc_start(sa);
298 	sfc_adapter_unlock(sa);
299 
300 	SFC_ASSERT(rc >= 0);
301 	return -rc;
302 }
303 
304 static int
305 sfc_dev_set_link_down(struct rte_eth_dev *dev)
306 {
307 	struct sfc_adapter *sa = dev->data->dev_private;
308 
309 	sfc_log_init(sa, "entry");
310 
311 	sfc_adapter_lock(sa);
312 	sfc_stop(sa);
313 	sfc_adapter_unlock(sa);
314 
315 	return 0;
316 }
317 
318 static void
319 sfc_dev_close(struct rte_eth_dev *dev)
320 {
321 	struct sfc_adapter *sa = dev->data->dev_private;
322 
323 	sfc_log_init(sa, "entry");
324 
325 	sfc_adapter_lock(sa);
326 	switch (sa->state) {
327 	case SFC_ADAPTER_STARTED:
328 		sfc_stop(sa);
329 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
330 		/* FALLTHROUGH */
331 	case SFC_ADAPTER_CONFIGURED:
332 		sfc_close(sa);
333 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
334 		/* FALLTHROUGH */
335 	case SFC_ADAPTER_INITIALIZED:
336 		break;
337 	default:
338 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
339 		break;
340 	}
341 	sfc_adapter_unlock(sa);
342 
343 	sfc_log_init(sa, "done");
344 }
345 
346 static void
347 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
348 		   boolean_t enabled)
349 {
350 	struct sfc_port *port;
351 	boolean_t *toggle;
352 	struct sfc_adapter *sa = dev->data->dev_private;
353 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
354 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
355 
356 	sfc_adapter_lock(sa);
357 
358 	port = &sa->port;
359 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
360 
361 	if (*toggle != enabled) {
362 		*toggle = enabled;
363 
364 		if (port->isolated) {
365 			sfc_warn(sa, "isolated mode is active on the port");
366 			sfc_warn(sa, "the change is to be applied on the next "
367 				     "start provided that isolated mode is "
368 				     "disabled prior the next start");
369 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
370 			   (sfc_set_rx_mode(sa) != 0)) {
371 			*toggle = !(enabled);
372 			sfc_warn(sa, "Failed to %s %s mode",
373 				 ((enabled) ? "enable" : "disable"), desc);
374 		}
375 	}
376 
377 	sfc_adapter_unlock(sa);
378 }
379 
380 static void
381 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
382 {
383 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
384 }
385 
386 static void
387 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
388 {
389 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
390 }
391 
392 static void
393 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
394 {
395 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
396 }
397 
398 static void
399 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
400 {
401 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
402 }
403 
404 static int
405 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
406 		   uint16_t nb_rx_desc, unsigned int socket_id,
407 		   const struct rte_eth_rxconf *rx_conf,
408 		   struct rte_mempool *mb_pool)
409 {
410 	struct sfc_adapter *sa = dev->data->dev_private;
411 	int rc;
412 
413 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
414 		     rx_queue_id, nb_rx_desc, socket_id);
415 
416 	sfc_adapter_lock(sa);
417 
418 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
419 			  rx_conf, mb_pool);
420 	if (rc != 0)
421 		goto fail_rx_qinit;
422 
423 	dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
424 
425 	sfc_adapter_unlock(sa);
426 
427 	return 0;
428 
429 fail_rx_qinit:
430 	sfc_adapter_unlock(sa);
431 	SFC_ASSERT(rc > 0);
432 	return -rc;
433 }
434 
435 static void
436 sfc_rx_queue_release(void *queue)
437 {
438 	struct sfc_dp_rxq *dp_rxq = queue;
439 	struct sfc_rxq *rxq;
440 	struct sfc_adapter *sa;
441 	unsigned int sw_index;
442 
443 	if (dp_rxq == NULL)
444 		return;
445 
446 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
447 	sa = rxq->evq->sa;
448 	sfc_adapter_lock(sa);
449 
450 	sw_index = sfc_rxq_sw_index(rxq);
451 
452 	sfc_log_init(sa, "RxQ=%u", sw_index);
453 
454 	sa->eth_dev->data->rx_queues[sw_index] = NULL;
455 
456 	sfc_rx_qfini(sa, sw_index);
457 
458 	sfc_adapter_unlock(sa);
459 }
460 
461 static int
462 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
463 		   uint16_t nb_tx_desc, unsigned int socket_id,
464 		   const struct rte_eth_txconf *tx_conf)
465 {
466 	struct sfc_adapter *sa = dev->data->dev_private;
467 	int rc;
468 
469 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
470 		     tx_queue_id, nb_tx_desc, socket_id);
471 
472 	sfc_adapter_lock(sa);
473 
474 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
475 	if (rc != 0)
476 		goto fail_tx_qinit;
477 
478 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
479 
480 	sfc_adapter_unlock(sa);
481 	return 0;
482 
483 fail_tx_qinit:
484 	sfc_adapter_unlock(sa);
485 	SFC_ASSERT(rc > 0);
486 	return -rc;
487 }
488 
489 static void
490 sfc_tx_queue_release(void *queue)
491 {
492 	struct sfc_dp_txq *dp_txq = queue;
493 	struct sfc_txq *txq;
494 	unsigned int sw_index;
495 	struct sfc_adapter *sa;
496 
497 	if (dp_txq == NULL)
498 		return;
499 
500 	txq = sfc_txq_by_dp_txq(dp_txq);
501 	sw_index = sfc_txq_sw_index(txq);
502 
503 	SFC_ASSERT(txq->evq != NULL);
504 	sa = txq->evq->sa;
505 
506 	sfc_log_init(sa, "TxQ = %u", sw_index);
507 
508 	sfc_adapter_lock(sa);
509 
510 	SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
511 	sa->eth_dev->data->tx_queues[sw_index] = NULL;
512 
513 	sfc_tx_qfini(sa, sw_index);
514 
515 	sfc_adapter_unlock(sa);
516 }
517 
518 static int
519 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
520 {
521 	struct sfc_adapter *sa = dev->data->dev_private;
522 	struct sfc_port *port = &sa->port;
523 	uint64_t *mac_stats;
524 	int ret;
525 
526 	rte_spinlock_lock(&port->mac_stats_lock);
527 
528 	ret = sfc_port_update_mac_stats(sa);
529 	if (ret != 0)
530 		goto unlock;
531 
532 	mac_stats = port->mac_stats_buf;
533 
534 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
535 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
536 		stats->ipackets =
537 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
538 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
539 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
540 		stats->opackets =
541 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
542 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
543 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
544 		stats->ibytes =
545 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
546 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
547 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
548 		stats->obytes =
549 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
550 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
551 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
552 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
553 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
554 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
555 	} else {
556 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
557 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
558 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
559 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
560 		/*
561 		 * Take into account stats which are whenever supported
562 		 * on EF10. If some stat is not supported by current
563 		 * firmware variant or HW revision, it is guaranteed
564 		 * to be zero in mac_stats.
565 		 */
566 		stats->imissed =
567 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
568 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
569 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
570 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
571 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
572 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
573 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
574 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
575 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
576 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
577 		stats->ierrors =
578 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
579 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
580 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
581 		/* no oerrors counters supported on EF10 */
582 	}
583 
584 unlock:
585 	rte_spinlock_unlock(&port->mac_stats_lock);
586 	SFC_ASSERT(ret >= 0);
587 	return -ret;
588 }
589 
590 static void
591 sfc_stats_reset(struct rte_eth_dev *dev)
592 {
593 	struct sfc_adapter *sa = dev->data->dev_private;
594 	struct sfc_port *port = &sa->port;
595 	int rc;
596 
597 	if (sa->state != SFC_ADAPTER_STARTED) {
598 		/*
599 		 * The operation cannot be done if port is not started; it
600 		 * will be scheduled to be done during the next port start
601 		 */
602 		port->mac_stats_reset_pending = B_TRUE;
603 		return;
604 	}
605 
606 	rc = sfc_port_reset_mac_stats(sa);
607 	if (rc != 0)
608 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
609 }
610 
611 static int
612 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
613 	       unsigned int xstats_count)
614 {
615 	struct sfc_adapter *sa = dev->data->dev_private;
616 	struct sfc_port *port = &sa->port;
617 	uint64_t *mac_stats;
618 	int rc;
619 	unsigned int i;
620 	int nstats = 0;
621 
622 	rte_spinlock_lock(&port->mac_stats_lock);
623 
624 	rc = sfc_port_update_mac_stats(sa);
625 	if (rc != 0) {
626 		SFC_ASSERT(rc > 0);
627 		nstats = -rc;
628 		goto unlock;
629 	}
630 
631 	mac_stats = port->mac_stats_buf;
632 
633 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
634 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
635 			if (xstats != NULL && nstats < (int)xstats_count) {
636 				xstats[nstats].id = nstats;
637 				xstats[nstats].value = mac_stats[i];
638 			}
639 			nstats++;
640 		}
641 	}
642 
643 unlock:
644 	rte_spinlock_unlock(&port->mac_stats_lock);
645 
646 	return nstats;
647 }
648 
649 static int
650 sfc_xstats_get_names(struct rte_eth_dev *dev,
651 		     struct rte_eth_xstat_name *xstats_names,
652 		     unsigned int xstats_count)
653 {
654 	struct sfc_adapter *sa = dev->data->dev_private;
655 	struct sfc_port *port = &sa->port;
656 	unsigned int i;
657 	unsigned int nstats = 0;
658 
659 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
660 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
661 			if (xstats_names != NULL && nstats < xstats_count)
662 				strncpy(xstats_names[nstats].name,
663 					efx_mac_stat_name(sa->nic, i),
664 					sizeof(xstats_names[0].name));
665 			nstats++;
666 		}
667 	}
668 
669 	return nstats;
670 }
671 
672 static int
673 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
674 		     uint64_t *values, unsigned int n)
675 {
676 	struct sfc_adapter *sa = dev->data->dev_private;
677 	struct sfc_port *port = &sa->port;
678 	uint64_t *mac_stats;
679 	unsigned int nb_supported = 0;
680 	unsigned int nb_written = 0;
681 	unsigned int i;
682 	int ret;
683 	int rc;
684 
685 	if (unlikely(values == NULL) ||
686 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
687 		return port->mac_stats_nb_supported;
688 
689 	rte_spinlock_lock(&port->mac_stats_lock);
690 
691 	rc = sfc_port_update_mac_stats(sa);
692 	if (rc != 0) {
693 		SFC_ASSERT(rc > 0);
694 		ret = -rc;
695 		goto unlock;
696 	}
697 
698 	mac_stats = port->mac_stats_buf;
699 
700 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
701 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
702 			continue;
703 
704 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
705 			values[nb_written++] = mac_stats[i];
706 
707 		++nb_supported;
708 	}
709 
710 	ret = nb_written;
711 
712 unlock:
713 	rte_spinlock_unlock(&port->mac_stats_lock);
714 
715 	return ret;
716 }
717 
718 static int
719 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
720 			   struct rte_eth_xstat_name *xstats_names,
721 			   const uint64_t *ids, unsigned int size)
722 {
723 	struct sfc_adapter *sa = dev->data->dev_private;
724 	struct sfc_port *port = &sa->port;
725 	unsigned int nb_supported = 0;
726 	unsigned int nb_written = 0;
727 	unsigned int i;
728 
729 	if (unlikely(xstats_names == NULL) ||
730 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
731 		return port->mac_stats_nb_supported;
732 
733 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
734 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
735 			continue;
736 
737 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
738 			char *name = xstats_names[nb_written++].name;
739 
740 			strncpy(name, efx_mac_stat_name(sa->nic, i),
741 				sizeof(xstats_names[0].name));
742 			name[sizeof(xstats_names[0].name) - 1] = '\0';
743 		}
744 
745 		++nb_supported;
746 	}
747 
748 	return nb_written;
749 }
750 
751 static int
752 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
753 {
754 	struct sfc_adapter *sa = dev->data->dev_private;
755 	unsigned int wanted_fc, link_fc;
756 
757 	memset(fc_conf, 0, sizeof(*fc_conf));
758 
759 	sfc_adapter_lock(sa);
760 
761 	if (sa->state == SFC_ADAPTER_STARTED)
762 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
763 	else
764 		link_fc = sa->port.flow_ctrl;
765 
766 	switch (link_fc) {
767 	case 0:
768 		fc_conf->mode = RTE_FC_NONE;
769 		break;
770 	case EFX_FCNTL_RESPOND:
771 		fc_conf->mode = RTE_FC_RX_PAUSE;
772 		break;
773 	case EFX_FCNTL_GENERATE:
774 		fc_conf->mode = RTE_FC_TX_PAUSE;
775 		break;
776 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
777 		fc_conf->mode = RTE_FC_FULL;
778 		break;
779 	default:
780 		sfc_err(sa, "%s: unexpected flow control value %#x",
781 			__func__, link_fc);
782 	}
783 
784 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
785 
786 	sfc_adapter_unlock(sa);
787 
788 	return 0;
789 }
790 
791 static int
792 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
793 {
794 	struct sfc_adapter *sa = dev->data->dev_private;
795 	struct sfc_port *port = &sa->port;
796 	unsigned int fcntl;
797 	int rc;
798 
799 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
800 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
801 	    fc_conf->mac_ctrl_frame_fwd != 0) {
802 		sfc_err(sa, "unsupported flow control settings specified");
803 		rc = EINVAL;
804 		goto fail_inval;
805 	}
806 
807 	switch (fc_conf->mode) {
808 	case RTE_FC_NONE:
809 		fcntl = 0;
810 		break;
811 	case RTE_FC_RX_PAUSE:
812 		fcntl = EFX_FCNTL_RESPOND;
813 		break;
814 	case RTE_FC_TX_PAUSE:
815 		fcntl = EFX_FCNTL_GENERATE;
816 		break;
817 	case RTE_FC_FULL:
818 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
819 		break;
820 	default:
821 		rc = EINVAL;
822 		goto fail_inval;
823 	}
824 
825 	sfc_adapter_lock(sa);
826 
827 	if (sa->state == SFC_ADAPTER_STARTED) {
828 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
829 		if (rc != 0)
830 			goto fail_mac_fcntl_set;
831 	}
832 
833 	port->flow_ctrl = fcntl;
834 	port->flow_ctrl_autoneg = fc_conf->autoneg;
835 
836 	sfc_adapter_unlock(sa);
837 
838 	return 0;
839 
840 fail_mac_fcntl_set:
841 	sfc_adapter_unlock(sa);
842 fail_inval:
843 	SFC_ASSERT(rc > 0);
844 	return -rc;
845 }
846 
847 static int
848 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
849 {
850 	struct sfc_adapter *sa = dev->data->dev_private;
851 	size_t pdu = EFX_MAC_PDU(mtu);
852 	size_t old_pdu;
853 	int rc;
854 
855 	sfc_log_init(sa, "mtu=%u", mtu);
856 
857 	rc = EINVAL;
858 	if (pdu < EFX_MAC_PDU_MIN) {
859 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
860 			(unsigned int)mtu, (unsigned int)pdu,
861 			EFX_MAC_PDU_MIN);
862 		goto fail_inval;
863 	}
864 	if (pdu > EFX_MAC_PDU_MAX) {
865 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
866 			(unsigned int)mtu, (unsigned int)pdu,
867 			EFX_MAC_PDU_MAX);
868 		goto fail_inval;
869 	}
870 
871 	sfc_adapter_lock(sa);
872 
873 	if (pdu != sa->port.pdu) {
874 		if (sa->state == SFC_ADAPTER_STARTED) {
875 			sfc_stop(sa);
876 
877 			old_pdu = sa->port.pdu;
878 			sa->port.pdu = pdu;
879 			rc = sfc_start(sa);
880 			if (rc != 0)
881 				goto fail_start;
882 		} else {
883 			sa->port.pdu = pdu;
884 		}
885 	}
886 
887 	/*
888 	 * The driver does not use it, but other PMDs update jumbo_frame
889 	 * flag and max_rx_pkt_len when MTU is set.
890 	 */
891 	if (mtu > ETHER_MAX_LEN) {
892 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
893 
894 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
895 		rxmode->jumbo_frame = 1;
896 	}
897 
898 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
899 
900 	sfc_adapter_unlock(sa);
901 
902 	sfc_log_init(sa, "done");
903 	return 0;
904 
905 fail_start:
906 	sa->port.pdu = old_pdu;
907 	if (sfc_start(sa) != 0)
908 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
909 			"PDU max size - port is stopped",
910 			(unsigned int)pdu, (unsigned int)old_pdu);
911 	sfc_adapter_unlock(sa);
912 
913 fail_inval:
914 	sfc_log_init(sa, "failed %d", rc);
915 	SFC_ASSERT(rc > 0);
916 	return -rc;
917 }
918 static void
919 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
920 {
921 	struct sfc_adapter *sa = dev->data->dev_private;
922 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
923 	struct sfc_port *port = &sa->port;
924 	int rc;
925 
926 	sfc_adapter_lock(sa);
927 
928 	/*
929 	 * Copy the address to the device private data so that
930 	 * it could be recalled in the case of adapter restart.
931 	 */
932 	ether_addr_copy(mac_addr, &port->default_mac_addr);
933 
934 	if (port->isolated) {
935 		sfc_err(sa, "isolated mode is active on the port");
936 		sfc_err(sa, "will not set MAC address");
937 		goto unlock;
938 	}
939 
940 	if (sa->state != SFC_ADAPTER_STARTED) {
941 		sfc_notice(sa, "the port is not started");
942 		sfc_notice(sa, "the new MAC address will be set on port start");
943 
944 		goto unlock;
945 	}
946 
947 	if (encp->enc_allow_set_mac_with_installed_filters) {
948 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
949 		if (rc != 0) {
950 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
951 			goto unlock;
952 		}
953 
954 		/*
955 		 * Changing the MAC address by means of MCDI request
956 		 * has no effect on received traffic, therefore
957 		 * we also need to update unicast filters
958 		 */
959 		rc = sfc_set_rx_mode(sa);
960 		if (rc != 0)
961 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
962 	} else {
963 		sfc_warn(sa, "cannot set MAC address with filters installed");
964 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
965 		sfc_warn(sa, "(some traffic may be dropped)");
966 
967 		/*
968 		 * Since setting MAC address with filters installed is not
969 		 * allowed on the adapter, the new MAC address will be set
970 		 * by means of adapter restart. sfc_start() shall retrieve
971 		 * the new address from the device private data and set it.
972 		 */
973 		sfc_stop(sa);
974 		rc = sfc_start(sa);
975 		if (rc != 0)
976 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
977 	}
978 
979 unlock:
980 	/*
981 	 * In the case of failure sa->port->default_mac_addr does not
982 	 * need rollback since no error code is returned, and the upper
983 	 * API will anyway update the external MAC address storage.
984 	 * To be consistent with that new value it is better to keep
985 	 * the device private value the same.
986 	 */
987 	sfc_adapter_unlock(sa);
988 }
989 
990 
991 static int
992 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
993 		     uint32_t nb_mc_addr)
994 {
995 	struct sfc_adapter *sa = dev->data->dev_private;
996 	struct sfc_port *port = &sa->port;
997 	uint8_t *mc_addrs = port->mcast_addrs;
998 	int rc;
999 	unsigned int i;
1000 
1001 	if (port->isolated) {
1002 		sfc_err(sa, "isolated mode is active on the port");
1003 		sfc_err(sa, "will not set multicast address list");
1004 		return -ENOTSUP;
1005 	}
1006 
1007 	if (mc_addrs == NULL)
1008 		return -ENOBUFS;
1009 
1010 	if (nb_mc_addr > port->max_mcast_addrs) {
1011 		sfc_err(sa, "too many multicast addresses: %u > %u",
1012 			 nb_mc_addr, port->max_mcast_addrs);
1013 		return -EINVAL;
1014 	}
1015 
1016 	for (i = 0; i < nb_mc_addr; ++i) {
1017 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1018 				 EFX_MAC_ADDR_LEN);
1019 		mc_addrs += EFX_MAC_ADDR_LEN;
1020 	}
1021 
1022 	port->nb_mcast_addrs = nb_mc_addr;
1023 
1024 	if (sa->state != SFC_ADAPTER_STARTED)
1025 		return 0;
1026 
1027 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1028 					port->nb_mcast_addrs);
1029 	if (rc != 0)
1030 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1031 
1032 	SFC_ASSERT(rc > 0);
1033 	return -rc;
1034 }
1035 
1036 /*
1037  * The function is used by the secondary process as well. It must not
1038  * use any process-local pointers from the adapter data.
1039  */
1040 static void
1041 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1042 		      struct rte_eth_rxq_info *qinfo)
1043 {
1044 	struct sfc_adapter *sa = dev->data->dev_private;
1045 	struct sfc_rxq_info *rxq_info;
1046 	struct sfc_rxq *rxq;
1047 
1048 	sfc_adapter_lock(sa);
1049 
1050 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1051 
1052 	rxq_info = &sa->rxq_info[rx_queue_id];
1053 	rxq = rxq_info->rxq;
1054 	SFC_ASSERT(rxq != NULL);
1055 
1056 	qinfo->mp = rxq->refill_mb_pool;
1057 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1058 	qinfo->conf.rx_drop_en = 1;
1059 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1060 	qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1061 			       DEV_RX_OFFLOAD_UDP_CKSUM |
1062 			       DEV_RX_OFFLOAD_TCP_CKSUM;
1063 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1064 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1065 		qinfo->scattered_rx = 1;
1066 	}
1067 	qinfo->nb_desc = rxq_info->entries;
1068 
1069 	sfc_adapter_unlock(sa);
1070 }
1071 
1072 /*
1073  * The function is used by the secondary process as well. It must not
1074  * use any process-local pointers from the adapter data.
1075  */
1076 static void
1077 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1078 		      struct rte_eth_txq_info *qinfo)
1079 {
1080 	struct sfc_adapter *sa = dev->data->dev_private;
1081 	struct sfc_txq_info *txq_info;
1082 
1083 	sfc_adapter_lock(sa);
1084 
1085 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1086 
1087 	txq_info = &sa->txq_info[tx_queue_id];
1088 	SFC_ASSERT(txq_info->txq != NULL);
1089 
1090 	memset(qinfo, 0, sizeof(*qinfo));
1091 
1092 	qinfo->conf.txq_flags = txq_info->txq->flags;
1093 	qinfo->conf.offloads = txq_info->txq->offloads;
1094 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1095 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1096 	qinfo->nb_desc = txq_info->entries;
1097 
1098 	sfc_adapter_unlock(sa);
1099 }
1100 
1101 static uint32_t
1102 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1103 {
1104 	struct sfc_adapter *sa = dev->data->dev_private;
1105 
1106 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1107 
1108 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
1109 }
1110 
1111 static int
1112 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1113 {
1114 	struct sfc_dp_rxq *dp_rxq = queue;
1115 
1116 	return sfc_rx_qdesc_done(dp_rxq, offset);
1117 }
1118 
1119 static int
1120 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1121 {
1122 	struct sfc_dp_rxq *dp_rxq = queue;
1123 	struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1124 
1125 	return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1126 }
1127 
1128 static int
1129 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1130 {
1131 	struct sfc_dp_txq *dp_txq = queue;
1132 	struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1133 
1134 	return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1135 }
1136 
1137 static int
1138 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1139 {
1140 	struct sfc_adapter *sa = dev->data->dev_private;
1141 	int rc;
1142 
1143 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1144 
1145 	sfc_adapter_lock(sa);
1146 
1147 	rc = EINVAL;
1148 	if (sa->state != SFC_ADAPTER_STARTED)
1149 		goto fail_not_started;
1150 
1151 	rc = sfc_rx_qstart(sa, rx_queue_id);
1152 	if (rc != 0)
1153 		goto fail_rx_qstart;
1154 
1155 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1156 
1157 	sfc_adapter_unlock(sa);
1158 
1159 	return 0;
1160 
1161 fail_rx_qstart:
1162 fail_not_started:
1163 	sfc_adapter_unlock(sa);
1164 	SFC_ASSERT(rc > 0);
1165 	return -rc;
1166 }
1167 
1168 static int
1169 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1170 {
1171 	struct sfc_adapter *sa = dev->data->dev_private;
1172 
1173 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1174 
1175 	sfc_adapter_lock(sa);
1176 	sfc_rx_qstop(sa, rx_queue_id);
1177 
1178 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1179 
1180 	sfc_adapter_unlock(sa);
1181 
1182 	return 0;
1183 }
1184 
1185 static int
1186 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1187 {
1188 	struct sfc_adapter *sa = dev->data->dev_private;
1189 	int rc;
1190 
1191 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1192 
1193 	sfc_adapter_lock(sa);
1194 
1195 	rc = EINVAL;
1196 	if (sa->state != SFC_ADAPTER_STARTED)
1197 		goto fail_not_started;
1198 
1199 	rc = sfc_tx_qstart(sa, tx_queue_id);
1200 	if (rc != 0)
1201 		goto fail_tx_qstart;
1202 
1203 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1204 
1205 	sfc_adapter_unlock(sa);
1206 	return 0;
1207 
1208 fail_tx_qstart:
1209 
1210 fail_not_started:
1211 	sfc_adapter_unlock(sa);
1212 	SFC_ASSERT(rc > 0);
1213 	return -rc;
1214 }
1215 
1216 static int
1217 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1218 {
1219 	struct sfc_adapter *sa = dev->data->dev_private;
1220 
1221 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1222 
1223 	sfc_adapter_lock(sa);
1224 
1225 	sfc_tx_qstop(sa, tx_queue_id);
1226 
1227 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1228 
1229 	sfc_adapter_unlock(sa);
1230 	return 0;
1231 }
1232 
1233 static efx_tunnel_protocol_t
1234 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1235 {
1236 	switch (rte_type) {
1237 	case RTE_TUNNEL_TYPE_VXLAN:
1238 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1239 	case RTE_TUNNEL_TYPE_GENEVE:
1240 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1241 	default:
1242 		return EFX_TUNNEL_NPROTOS;
1243 	}
1244 }
1245 
1246 enum sfc_udp_tunnel_op_e {
1247 	SFC_UDP_TUNNEL_ADD_PORT,
1248 	SFC_UDP_TUNNEL_DEL_PORT,
1249 };
1250 
1251 static int
1252 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1253 		      struct rte_eth_udp_tunnel *tunnel_udp,
1254 		      enum sfc_udp_tunnel_op_e op)
1255 {
1256 	struct sfc_adapter *sa = dev->data->dev_private;
1257 	efx_tunnel_protocol_t tunnel_proto;
1258 	int rc;
1259 
1260 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1261 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1262 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1263 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1264 
1265 	tunnel_proto =
1266 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1267 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1268 		rc = ENOTSUP;
1269 		goto fail_bad_proto;
1270 	}
1271 
1272 	sfc_adapter_lock(sa);
1273 
1274 	switch (op) {
1275 	case SFC_UDP_TUNNEL_ADD_PORT:
1276 		rc = efx_tunnel_config_udp_add(sa->nic,
1277 					       tunnel_udp->udp_port,
1278 					       tunnel_proto);
1279 		break;
1280 	case SFC_UDP_TUNNEL_DEL_PORT:
1281 		rc = efx_tunnel_config_udp_remove(sa->nic,
1282 						  tunnel_udp->udp_port,
1283 						  tunnel_proto);
1284 		break;
1285 	default:
1286 		rc = EINVAL;
1287 		goto fail_bad_op;
1288 	}
1289 
1290 	if (rc != 0)
1291 		goto fail_op;
1292 
1293 	if (sa->state == SFC_ADAPTER_STARTED) {
1294 		rc = efx_tunnel_reconfigure(sa->nic);
1295 		if (rc == EAGAIN) {
1296 			/*
1297 			 * Configuration is accepted by FW and MC reboot
1298 			 * is initiated to apply the changes. MC reboot
1299 			 * will be handled in a usual way (MC reboot
1300 			 * event on management event queue and adapter
1301 			 * restart).
1302 			 */
1303 			rc = 0;
1304 		} else if (rc != 0) {
1305 			goto fail_reconfigure;
1306 		}
1307 	}
1308 
1309 	sfc_adapter_unlock(sa);
1310 	return 0;
1311 
1312 fail_reconfigure:
1313 	/* Remove/restore entry since the change makes the trouble */
1314 	switch (op) {
1315 	case SFC_UDP_TUNNEL_ADD_PORT:
1316 		(void)efx_tunnel_config_udp_remove(sa->nic,
1317 						   tunnel_udp->udp_port,
1318 						   tunnel_proto);
1319 		break;
1320 	case SFC_UDP_TUNNEL_DEL_PORT:
1321 		(void)efx_tunnel_config_udp_add(sa->nic,
1322 						tunnel_udp->udp_port,
1323 						tunnel_proto);
1324 		break;
1325 	}
1326 
1327 fail_op:
1328 fail_bad_op:
1329 	sfc_adapter_unlock(sa);
1330 
1331 fail_bad_proto:
1332 	SFC_ASSERT(rc > 0);
1333 	return -rc;
1334 }
1335 
1336 static int
1337 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1338 			    struct rte_eth_udp_tunnel *tunnel_udp)
1339 {
1340 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1341 }
1342 
1343 static int
1344 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1345 			    struct rte_eth_udp_tunnel *tunnel_udp)
1346 {
1347 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1348 }
1349 
1350 #if EFSYS_OPT_RX_SCALE
1351 static int
1352 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1353 			  struct rte_eth_rss_conf *rss_conf)
1354 {
1355 	struct sfc_adapter *sa = dev->data->dev_private;
1356 	struct sfc_port *port = &sa->port;
1357 
1358 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1359 		return -ENOTSUP;
1360 
1361 	if (sa->rss_channels == 0)
1362 		return -EINVAL;
1363 
1364 	sfc_adapter_lock(sa);
1365 
1366 	/*
1367 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1368 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1369 	 * flags which corresponds to the active EFX configuration stored
1370 	 * locally in 'sfc_adapter' and kept up-to-date
1371 	 */
1372 	rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1373 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1374 	if (rss_conf->rss_key != NULL)
1375 		rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1376 
1377 	sfc_adapter_unlock(sa);
1378 
1379 	return 0;
1380 }
1381 
1382 static int
1383 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1384 			struct rte_eth_rss_conf *rss_conf)
1385 {
1386 	struct sfc_adapter *sa = dev->data->dev_private;
1387 	struct sfc_port *port = &sa->port;
1388 	unsigned int efx_hash_types;
1389 	int rc = 0;
1390 
1391 	if (port->isolated)
1392 		return -ENOTSUP;
1393 
1394 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1395 		sfc_err(sa, "RSS is not available");
1396 		return -ENOTSUP;
1397 	}
1398 
1399 	if (sa->rss_channels == 0) {
1400 		sfc_err(sa, "RSS is not configured");
1401 		return -EINVAL;
1402 	}
1403 
1404 	if ((rss_conf->rss_key != NULL) &&
1405 	    (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1406 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1407 			sizeof(sa->rss_key));
1408 		return -EINVAL;
1409 	}
1410 
1411 	if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1412 		sfc_err(sa, "unsupported hash functions requested");
1413 		return -EINVAL;
1414 	}
1415 
1416 	sfc_adapter_lock(sa);
1417 
1418 	efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1419 
1420 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1421 				   EFX_RX_HASHALG_TOEPLITZ,
1422 				   efx_hash_types, B_TRUE);
1423 	if (rc != 0)
1424 		goto fail_scale_mode_set;
1425 
1426 	if (rss_conf->rss_key != NULL) {
1427 		if (sa->state == SFC_ADAPTER_STARTED) {
1428 			rc = efx_rx_scale_key_set(sa->nic,
1429 						  EFX_RSS_CONTEXT_DEFAULT,
1430 						  rss_conf->rss_key,
1431 						  sizeof(sa->rss_key));
1432 			if (rc != 0)
1433 				goto fail_scale_key_set;
1434 		}
1435 
1436 		rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1437 	}
1438 
1439 	sa->rss_hash_types = efx_hash_types;
1440 
1441 	sfc_adapter_unlock(sa);
1442 
1443 	return 0;
1444 
1445 fail_scale_key_set:
1446 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1447 				  EFX_RX_HASHALG_TOEPLITZ,
1448 				  sa->rss_hash_types, B_TRUE) != 0)
1449 		sfc_err(sa, "failed to restore RSS mode");
1450 
1451 fail_scale_mode_set:
1452 	sfc_adapter_unlock(sa);
1453 	return -rc;
1454 }
1455 
1456 static int
1457 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1458 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1459 		       uint16_t reta_size)
1460 {
1461 	struct sfc_adapter *sa = dev->data->dev_private;
1462 	struct sfc_port *port = &sa->port;
1463 	int entry;
1464 
1465 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1466 		return -ENOTSUP;
1467 
1468 	if (sa->rss_channels == 0)
1469 		return -EINVAL;
1470 
1471 	if (reta_size != EFX_RSS_TBL_SIZE)
1472 		return -EINVAL;
1473 
1474 	sfc_adapter_lock(sa);
1475 
1476 	for (entry = 0; entry < reta_size; entry++) {
1477 		int grp = entry / RTE_RETA_GROUP_SIZE;
1478 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1479 
1480 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1481 			reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1482 	}
1483 
1484 	sfc_adapter_unlock(sa);
1485 
1486 	return 0;
1487 }
1488 
1489 static int
1490 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1491 			struct rte_eth_rss_reta_entry64 *reta_conf,
1492 			uint16_t reta_size)
1493 {
1494 	struct sfc_adapter *sa = dev->data->dev_private;
1495 	struct sfc_port *port = &sa->port;
1496 	unsigned int *rss_tbl_new;
1497 	uint16_t entry;
1498 	int rc = 0;
1499 
1500 
1501 	if (port->isolated)
1502 		return -ENOTSUP;
1503 
1504 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1505 		sfc_err(sa, "RSS is not available");
1506 		return -ENOTSUP;
1507 	}
1508 
1509 	if (sa->rss_channels == 0) {
1510 		sfc_err(sa, "RSS is not configured");
1511 		return -EINVAL;
1512 	}
1513 
1514 	if (reta_size != EFX_RSS_TBL_SIZE) {
1515 		sfc_err(sa, "RETA size is wrong (should be %u)",
1516 			EFX_RSS_TBL_SIZE);
1517 		return -EINVAL;
1518 	}
1519 
1520 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1521 	if (rss_tbl_new == NULL)
1522 		return -ENOMEM;
1523 
1524 	sfc_adapter_lock(sa);
1525 
1526 	rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1527 
1528 	for (entry = 0; entry < reta_size; entry++) {
1529 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1530 		struct rte_eth_rss_reta_entry64 *grp;
1531 
1532 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1533 
1534 		if (grp->mask & (1ull << grp_idx)) {
1535 			if (grp->reta[grp_idx] >= sa->rss_channels) {
1536 				rc = EINVAL;
1537 				goto bad_reta_entry;
1538 			}
1539 			rss_tbl_new[entry] = grp->reta[grp_idx];
1540 		}
1541 	}
1542 
1543 	if (sa->state == SFC_ADAPTER_STARTED) {
1544 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1545 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1546 		if (rc != 0)
1547 			goto fail_scale_tbl_set;
1548 	}
1549 
1550 	rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1551 
1552 fail_scale_tbl_set:
1553 bad_reta_entry:
1554 	sfc_adapter_unlock(sa);
1555 
1556 	rte_free(rss_tbl_new);
1557 
1558 	SFC_ASSERT(rc >= 0);
1559 	return -rc;
1560 }
1561 #endif
1562 
1563 static int
1564 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1565 		    enum rte_filter_op filter_op,
1566 		    void *arg)
1567 {
1568 	struct sfc_adapter *sa = dev->data->dev_private;
1569 	int rc = ENOTSUP;
1570 
1571 	sfc_log_init(sa, "entry");
1572 
1573 	switch (filter_type) {
1574 	case RTE_ETH_FILTER_NONE:
1575 		sfc_err(sa, "Global filters configuration not supported");
1576 		break;
1577 	case RTE_ETH_FILTER_MACVLAN:
1578 		sfc_err(sa, "MACVLAN filters not supported");
1579 		break;
1580 	case RTE_ETH_FILTER_ETHERTYPE:
1581 		sfc_err(sa, "EtherType filters not supported");
1582 		break;
1583 	case RTE_ETH_FILTER_FLEXIBLE:
1584 		sfc_err(sa, "Flexible filters not supported");
1585 		break;
1586 	case RTE_ETH_FILTER_SYN:
1587 		sfc_err(sa, "SYN filters not supported");
1588 		break;
1589 	case RTE_ETH_FILTER_NTUPLE:
1590 		sfc_err(sa, "NTUPLE filters not supported");
1591 		break;
1592 	case RTE_ETH_FILTER_TUNNEL:
1593 		sfc_err(sa, "Tunnel filters not supported");
1594 		break;
1595 	case RTE_ETH_FILTER_FDIR:
1596 		sfc_err(sa, "Flow Director filters not supported");
1597 		break;
1598 	case RTE_ETH_FILTER_HASH:
1599 		sfc_err(sa, "Hash filters not supported");
1600 		break;
1601 	case RTE_ETH_FILTER_GENERIC:
1602 		if (filter_op != RTE_ETH_FILTER_GET) {
1603 			rc = EINVAL;
1604 		} else {
1605 			*(const void **)arg = &sfc_flow_ops;
1606 			rc = 0;
1607 		}
1608 		break;
1609 	default:
1610 		sfc_err(sa, "Unknown filter type %u", filter_type);
1611 		break;
1612 	}
1613 
1614 	sfc_log_init(sa, "exit: %d", -rc);
1615 	SFC_ASSERT(rc >= 0);
1616 	return -rc;
1617 }
1618 
1619 static const struct eth_dev_ops sfc_eth_dev_ops = {
1620 	.dev_configure			= sfc_dev_configure,
1621 	.dev_start			= sfc_dev_start,
1622 	.dev_stop			= sfc_dev_stop,
1623 	.dev_set_link_up		= sfc_dev_set_link_up,
1624 	.dev_set_link_down		= sfc_dev_set_link_down,
1625 	.dev_close			= sfc_dev_close,
1626 	.promiscuous_enable		= sfc_dev_promisc_enable,
1627 	.promiscuous_disable		= sfc_dev_promisc_disable,
1628 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1629 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1630 	.link_update			= sfc_dev_link_update,
1631 	.stats_get			= sfc_stats_get,
1632 	.stats_reset			= sfc_stats_reset,
1633 	.xstats_get			= sfc_xstats_get,
1634 	.xstats_reset			= sfc_stats_reset,
1635 	.xstats_get_names		= sfc_xstats_get_names,
1636 	.dev_infos_get			= sfc_dev_infos_get,
1637 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1638 	.mtu_set			= sfc_dev_set_mtu,
1639 	.rx_queue_start			= sfc_rx_queue_start,
1640 	.rx_queue_stop			= sfc_rx_queue_stop,
1641 	.tx_queue_start			= sfc_tx_queue_start,
1642 	.tx_queue_stop			= sfc_tx_queue_stop,
1643 	.rx_queue_setup			= sfc_rx_queue_setup,
1644 	.rx_queue_release		= sfc_rx_queue_release,
1645 	.rx_queue_count			= sfc_rx_queue_count,
1646 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1647 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1648 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1649 	.tx_queue_setup			= sfc_tx_queue_setup,
1650 	.tx_queue_release		= sfc_tx_queue_release,
1651 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1652 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1653 	.mac_addr_set			= sfc_mac_addr_set,
1654 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1655 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1656 #if EFSYS_OPT_RX_SCALE
1657 	.reta_update			= sfc_dev_rss_reta_update,
1658 	.reta_query			= sfc_dev_rss_reta_query,
1659 	.rss_hash_update		= sfc_dev_rss_hash_update,
1660 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1661 #endif
1662 	.filter_ctrl			= sfc_dev_filter_ctrl,
1663 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1664 	.rxq_info_get			= sfc_rx_queue_info_get,
1665 	.txq_info_get			= sfc_tx_queue_info_get,
1666 	.fw_version_get			= sfc_fw_version_get,
1667 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1668 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1669 };
1670 
1671 /**
1672  * Duplicate a string in potentially shared memory required for
1673  * multi-process support.
1674  *
1675  * strdup() allocates from process-local heap/memory.
1676  */
1677 static char *
1678 sfc_strdup(const char *str)
1679 {
1680 	size_t size;
1681 	char *copy;
1682 
1683 	if (str == NULL)
1684 		return NULL;
1685 
1686 	size = strlen(str) + 1;
1687 	copy = rte_malloc(__func__, size, 0);
1688 	if (copy != NULL)
1689 		rte_memcpy(copy, str, size);
1690 
1691 	return copy;
1692 }
1693 
1694 static int
1695 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1696 {
1697 	struct sfc_adapter *sa = dev->data->dev_private;
1698 	unsigned int avail_caps = 0;
1699 	const char *rx_name = NULL;
1700 	const char *tx_name = NULL;
1701 	int rc;
1702 
1703 	switch (sa->family) {
1704 	case EFX_FAMILY_HUNTINGTON:
1705 	case EFX_FAMILY_MEDFORD:
1706 	case EFX_FAMILY_MEDFORD2:
1707 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1708 		break;
1709 	default:
1710 		break;
1711 	}
1712 
1713 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1714 				sfc_kvarg_string_handler, &rx_name);
1715 	if (rc != 0)
1716 		goto fail_kvarg_rx_datapath;
1717 
1718 	if (rx_name != NULL) {
1719 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1720 		if (sa->dp_rx == NULL) {
1721 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1722 			rc = ENOENT;
1723 			goto fail_dp_rx;
1724 		}
1725 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1726 			sfc_err(sa,
1727 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1728 				rx_name);
1729 			rc = EINVAL;
1730 			goto fail_dp_rx_caps;
1731 		}
1732 	} else {
1733 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1734 		if (sa->dp_rx == NULL) {
1735 			sfc_err(sa, "Rx datapath by caps %#x not found",
1736 				avail_caps);
1737 			rc = ENOENT;
1738 			goto fail_dp_rx;
1739 		}
1740 	}
1741 
1742 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1743 	if (sa->dp_rx_name == NULL) {
1744 		rc = ENOMEM;
1745 		goto fail_dp_rx_name;
1746 	}
1747 
1748 	sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1749 
1750 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1751 
1752 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1753 				sfc_kvarg_string_handler, &tx_name);
1754 	if (rc != 0)
1755 		goto fail_kvarg_tx_datapath;
1756 
1757 	if (tx_name != NULL) {
1758 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1759 		if (sa->dp_tx == NULL) {
1760 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1761 			rc = ENOENT;
1762 			goto fail_dp_tx;
1763 		}
1764 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1765 			sfc_err(sa,
1766 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1767 				tx_name);
1768 			rc = EINVAL;
1769 			goto fail_dp_tx_caps;
1770 		}
1771 	} else {
1772 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1773 		if (sa->dp_tx == NULL) {
1774 			sfc_err(sa, "Tx datapath by caps %#x not found",
1775 				avail_caps);
1776 			rc = ENOENT;
1777 			goto fail_dp_tx;
1778 		}
1779 	}
1780 
1781 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1782 	if (sa->dp_tx_name == NULL) {
1783 		rc = ENOMEM;
1784 		goto fail_dp_tx_name;
1785 	}
1786 
1787 	sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1788 
1789 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1790 
1791 	dev->dev_ops = &sfc_eth_dev_ops;
1792 
1793 	return 0;
1794 
1795 fail_dp_tx_name:
1796 fail_dp_tx_caps:
1797 	sa->dp_tx = NULL;
1798 
1799 fail_dp_tx:
1800 fail_kvarg_tx_datapath:
1801 	rte_free(sa->dp_rx_name);
1802 	sa->dp_rx_name = NULL;
1803 
1804 fail_dp_rx_name:
1805 fail_dp_rx_caps:
1806 	sa->dp_rx = NULL;
1807 
1808 fail_dp_rx:
1809 fail_kvarg_rx_datapath:
1810 	return rc;
1811 }
1812 
1813 static void
1814 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1815 {
1816 	struct sfc_adapter *sa = dev->data->dev_private;
1817 
1818 	dev->dev_ops = NULL;
1819 	dev->rx_pkt_burst = NULL;
1820 	dev->tx_pkt_burst = NULL;
1821 
1822 	rte_free(sa->dp_tx_name);
1823 	sa->dp_tx_name = NULL;
1824 	sa->dp_tx = NULL;
1825 
1826 	rte_free(sa->dp_rx_name);
1827 	sa->dp_rx_name = NULL;
1828 	sa->dp_rx = NULL;
1829 }
1830 
1831 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1832 	.rxq_info_get			= sfc_rx_queue_info_get,
1833 	.txq_info_get			= sfc_tx_queue_info_get,
1834 };
1835 
1836 static int
1837 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1838 {
1839 	/*
1840 	 * Device private data has really many process-local pointers.
1841 	 * Below code should be extremely careful to use data located
1842 	 * in shared memory only.
1843 	 */
1844 	struct sfc_adapter *sa = dev->data->dev_private;
1845 	const struct sfc_dp_rx *dp_rx;
1846 	const struct sfc_dp_tx *dp_tx;
1847 	int rc;
1848 
1849 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1850 	if (dp_rx == NULL) {
1851 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1852 		rc = ENOENT;
1853 		goto fail_dp_rx;
1854 	}
1855 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1856 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1857 			sa->dp_tx_name);
1858 		rc = EINVAL;
1859 		goto fail_dp_rx_multi_process;
1860 	}
1861 
1862 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1863 	if (dp_tx == NULL) {
1864 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1865 		rc = ENOENT;
1866 		goto fail_dp_tx;
1867 	}
1868 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1869 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1870 			sa->dp_tx_name);
1871 		rc = EINVAL;
1872 		goto fail_dp_tx_multi_process;
1873 	}
1874 
1875 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1876 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1877 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1878 
1879 	return 0;
1880 
1881 fail_dp_tx_multi_process:
1882 fail_dp_tx:
1883 fail_dp_rx_multi_process:
1884 fail_dp_rx:
1885 	return rc;
1886 }
1887 
1888 static void
1889 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1890 {
1891 	dev->dev_ops = NULL;
1892 	dev->tx_pkt_burst = NULL;
1893 	dev->rx_pkt_burst = NULL;
1894 }
1895 
1896 static void
1897 sfc_register_dp(void)
1898 {
1899 	/* Register once */
1900 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1901 		/* Prefer EF10 datapath */
1902 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1903 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1904 
1905 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1906 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1907 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1908 	}
1909 }
1910 
1911 static int
1912 sfc_eth_dev_init(struct rte_eth_dev *dev)
1913 {
1914 	struct sfc_adapter *sa = dev->data->dev_private;
1915 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1916 	int rc;
1917 	const efx_nic_cfg_t *encp;
1918 	const struct ether_addr *from;
1919 
1920 	sfc_register_dp();
1921 
1922 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1923 		return -sfc_eth_dev_secondary_set_ops(dev);
1924 
1925 	/* Required for logging */
1926 	sa->pci_addr = pci_dev->addr;
1927 	sa->port_id = dev->data->port_id;
1928 
1929 	sa->eth_dev = dev;
1930 
1931 	/* Copy PCI device info to the dev->data */
1932 	rte_eth_copy_pci_info(dev, pci_dev);
1933 
1934 	sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1935 						RTE_LOG_NOTICE);
1936 
1937 	rc = sfc_kvargs_parse(sa);
1938 	if (rc != 0)
1939 		goto fail_kvargs_parse;
1940 
1941 	sfc_log_init(sa, "entry");
1942 
1943 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1944 	if (dev->data->mac_addrs == NULL) {
1945 		rc = ENOMEM;
1946 		goto fail_mac_addrs;
1947 	}
1948 
1949 	sfc_adapter_lock_init(sa);
1950 	sfc_adapter_lock(sa);
1951 
1952 	sfc_log_init(sa, "probing");
1953 	rc = sfc_probe(sa);
1954 	if (rc != 0)
1955 		goto fail_probe;
1956 
1957 	sfc_log_init(sa, "set device ops");
1958 	rc = sfc_eth_dev_set_ops(dev);
1959 	if (rc != 0)
1960 		goto fail_set_ops;
1961 
1962 	sfc_log_init(sa, "attaching");
1963 	rc = sfc_attach(sa);
1964 	if (rc != 0)
1965 		goto fail_attach;
1966 
1967 	encp = efx_nic_cfg_get(sa->nic);
1968 
1969 	/*
1970 	 * The arguments are really reverse order in comparison to
1971 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1972 	 */
1973 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1974 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1975 
1976 	sfc_adapter_unlock(sa);
1977 
1978 	sfc_log_init(sa, "done");
1979 	return 0;
1980 
1981 fail_attach:
1982 	sfc_eth_dev_clear_ops(dev);
1983 
1984 fail_set_ops:
1985 	sfc_unprobe(sa);
1986 
1987 fail_probe:
1988 	sfc_adapter_unlock(sa);
1989 	sfc_adapter_lock_fini(sa);
1990 	rte_free(dev->data->mac_addrs);
1991 	dev->data->mac_addrs = NULL;
1992 
1993 fail_mac_addrs:
1994 	sfc_kvargs_cleanup(sa);
1995 
1996 fail_kvargs_parse:
1997 	sfc_log_init(sa, "failed %d", rc);
1998 	SFC_ASSERT(rc > 0);
1999 	return -rc;
2000 }
2001 
2002 static int
2003 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2004 {
2005 	struct sfc_adapter *sa;
2006 
2007 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2008 		sfc_eth_dev_secondary_clear_ops(dev);
2009 		return 0;
2010 	}
2011 
2012 	sa = dev->data->dev_private;
2013 	sfc_log_init(sa, "entry");
2014 
2015 	sfc_adapter_lock(sa);
2016 
2017 	sfc_eth_dev_clear_ops(dev);
2018 
2019 	sfc_detach(sa);
2020 	sfc_unprobe(sa);
2021 
2022 	rte_free(dev->data->mac_addrs);
2023 	dev->data->mac_addrs = NULL;
2024 
2025 	sfc_kvargs_cleanup(sa);
2026 
2027 	sfc_adapter_unlock(sa);
2028 	sfc_adapter_lock_fini(sa);
2029 
2030 	sfc_log_init(sa, "done");
2031 
2032 	/* Required for logging, so cleanup last */
2033 	sa->eth_dev = NULL;
2034 	return 0;
2035 }
2036 
2037 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2038 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2039 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2040 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2041 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2042 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2043 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2044 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2045 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2046 	{ .vendor_id = 0 /* sentinel */ }
2047 };
2048 
2049 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2050 	struct rte_pci_device *pci_dev)
2051 {
2052 	return rte_eth_dev_pci_generic_probe(pci_dev,
2053 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2054 }
2055 
2056 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2057 {
2058 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2059 }
2060 
2061 static struct rte_pci_driver sfc_efx_pmd = {
2062 	.id_table = pci_id_sfc_efx_map,
2063 	.drv_flags =
2064 		RTE_PCI_DRV_INTR_LSC |
2065 		RTE_PCI_DRV_NEED_MAPPING,
2066 	.probe = sfc_eth_dev_pci_probe,
2067 	.remove = sfc_eth_dev_pci_remove,
2068 };
2069 
2070 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2071 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2072 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2073 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2074 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2075 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2076 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2077 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2078 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2079 
2080 RTE_INIT(sfc_driver_register_logtype);
2081 static void
2082 sfc_driver_register_logtype(void)
2083 {
2084 	int ret;
2085 
2086 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2087 						   RTE_LOG_NOTICE);
2088 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2089 }
2090