xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 9f25da76b9f7b664965e1afaebbc8b3257a21dfa)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
4  * All rights reserved.
5  *
6  * This software was jointly developed between OKTET Labs (under contract
7  * for Solarflare) and Solarflare Communications, Inc.
8  */
9 
10 #include <rte_dev.h>
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
16 #include <rte_string_fns.h>
17 
18 #include "efx.h"
19 
20 #include "sfc.h"
21 #include "sfc_debug.h"
22 #include "sfc_log.h"
23 #include "sfc_kvargs.h"
24 #include "sfc_ev.h"
25 #include "sfc_rx.h"
26 #include "sfc_tx.h"
27 #include "sfc_flow.h"
28 #include "sfc_dp.h"
29 #include "sfc_dp_rx.h"
30 
31 uint32_t sfc_logtype_driver;
32 
33 static struct sfc_dp_list sfc_dp_head =
34 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
35 
36 static int
37 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
38 {
39 	struct sfc_adapter *sa = dev->data->dev_private;
40 	efx_nic_fw_info_t enfi;
41 	int ret;
42 	int rc;
43 
44 	/*
45 	 * Return value of the callback is likely supposed to be
46 	 * equal to or greater than 0, nevertheless, if an error
47 	 * occurs, it will be desirable to pass it to the caller
48 	 */
49 	if ((fw_version == NULL) || (fw_size == 0))
50 		return -EINVAL;
51 
52 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
53 	if (rc != 0)
54 		return -rc;
55 
56 	ret = snprintf(fw_version, fw_size,
57 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
58 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
59 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
60 	if (ret < 0)
61 		return ret;
62 
63 	if (enfi.enfi_dpcpu_fw_ids_valid) {
64 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
65 		int ret_extra;
66 
67 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
68 				     fw_size - dpcpu_fw_ids_offset,
69 				     " rx%" PRIx16 " tx%" PRIx16,
70 				     enfi.enfi_rx_dpcpu_fw_id,
71 				     enfi.enfi_tx_dpcpu_fw_id);
72 		if (ret_extra < 0)
73 			return ret_extra;
74 
75 		ret += ret_extra;
76 	}
77 
78 	if (fw_size < (size_t)(++ret))
79 		return ret;
80 	else
81 		return 0;
82 }
83 
84 static void
85 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
86 {
87 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
88 	struct sfc_adapter *sa = dev->data->dev_private;
89 	struct sfc_rss *rss = &sa->rss;
90 	uint64_t txq_offloads_def = 0;
91 
92 	sfc_log_init(sa, "entry");
93 
94 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
95 
96 	/* Autonegotiation may be disabled */
97 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
98 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
99 		dev_info->speed_capa |= ETH_LINK_SPEED_1G;
100 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
101 		dev_info->speed_capa |= ETH_LINK_SPEED_10G;
102 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_25000FDX)
103 		dev_info->speed_capa |= ETH_LINK_SPEED_25G;
104 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
105 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
106 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_50000FDX)
107 		dev_info->speed_capa |= ETH_LINK_SPEED_50G;
108 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_100000FDX)
109 		dev_info->speed_capa |= ETH_LINK_SPEED_100G;
110 
111 	dev_info->max_rx_queues = sa->rxq_max;
112 	dev_info->max_tx_queues = sa->txq_max;
113 
114 	/* By default packets are dropped if no descriptors are available */
115 	dev_info->default_rxconf.rx_drop_en = 1;
116 
117 	dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
118 
119 	/*
120 	 * rx_offload_capa includes both device and queue offloads since
121 	 * the latter may be requested on a per device basis which makes
122 	 * sense when some offloads are needed to be set on all queues.
123 	 */
124 	dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
125 				    dev_info->rx_queue_offload_capa;
126 
127 	dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
128 
129 	/*
130 	 * tx_offload_capa includes both device and queue offloads since
131 	 * the latter may be requested on a per device basis which makes
132 	 * sense when some offloads are needed to be set on all queues.
133 	 */
134 	dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
135 				    dev_info->tx_queue_offload_capa;
136 
137 	if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
138 		txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
139 
140 	dev_info->default_txconf.offloads |= txq_offloads_def;
141 
142 	if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
143 		uint64_t rte_hf = 0;
144 		unsigned int i;
145 
146 		for (i = 0; i < rss->hf_map_nb_entries; ++i)
147 			rte_hf |= rss->hf_map[i].rte;
148 
149 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
150 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
151 		dev_info->flow_type_rss_offloads = rte_hf;
152 	}
153 
154 	/* Initialize to hardware limits */
155 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
156 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
157 	/* The RXQ hardware requires that the descriptor count is a power
158 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
159 	 */
160 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
161 
162 	/* Initialize to hardware limits */
163 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
164 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
165 	/*
166 	 * The TXQ hardware requires that the descriptor count is a power
167 	 * of 2, but tx_desc_lim cannot properly describe that constraint
168 	 */
169 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
170 
171 	if (sap->dp_rx->get_dev_info != NULL)
172 		sap->dp_rx->get_dev_info(dev_info);
173 	if (sap->dp_tx->get_dev_info != NULL)
174 		sap->dp_tx->get_dev_info(dev_info);
175 
176 	dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
177 			     RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
178 }
179 
180 static const uint32_t *
181 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
182 {
183 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
184 	struct sfc_adapter *sa = dev->data->dev_private;
185 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
186 	uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
187 
188 	return sap->dp_rx->supported_ptypes_get(tunnel_encaps);
189 }
190 
191 static int
192 sfc_dev_configure(struct rte_eth_dev *dev)
193 {
194 	struct rte_eth_dev_data *dev_data = dev->data;
195 	struct sfc_adapter *sa = dev_data->dev_private;
196 	int rc;
197 
198 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
199 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
200 
201 	sfc_adapter_lock(sa);
202 	switch (sa->state) {
203 	case SFC_ADAPTER_CONFIGURED:
204 		/* FALLTHROUGH */
205 	case SFC_ADAPTER_INITIALIZED:
206 		rc = sfc_configure(sa);
207 		break;
208 	default:
209 		sfc_err(sa, "unexpected adapter state %u to configure",
210 			sa->state);
211 		rc = EINVAL;
212 		break;
213 	}
214 	sfc_adapter_unlock(sa);
215 
216 	sfc_log_init(sa, "done %d", rc);
217 	SFC_ASSERT(rc >= 0);
218 	return -rc;
219 }
220 
221 static int
222 sfc_dev_start(struct rte_eth_dev *dev)
223 {
224 	struct sfc_adapter *sa = dev->data->dev_private;
225 	int rc;
226 
227 	sfc_log_init(sa, "entry");
228 
229 	sfc_adapter_lock(sa);
230 	rc = sfc_start(sa);
231 	sfc_adapter_unlock(sa);
232 
233 	sfc_log_init(sa, "done %d", rc);
234 	SFC_ASSERT(rc >= 0);
235 	return -rc;
236 }
237 
238 static int
239 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
240 {
241 	struct sfc_adapter *sa = dev->data->dev_private;
242 	struct rte_eth_link current_link;
243 	int ret;
244 
245 	sfc_log_init(sa, "entry");
246 
247 	if (sa->state != SFC_ADAPTER_STARTED) {
248 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
249 	} else if (wait_to_complete) {
250 		efx_link_mode_t link_mode;
251 
252 		if (efx_port_poll(sa->nic, &link_mode) != 0)
253 			link_mode = EFX_LINK_UNKNOWN;
254 		sfc_port_link_mode_to_info(link_mode, &current_link);
255 
256 	} else {
257 		sfc_ev_mgmt_qpoll(sa);
258 		rte_eth_linkstatus_get(dev, &current_link);
259 	}
260 
261 	ret = rte_eth_linkstatus_set(dev, &current_link);
262 	if (ret == 0)
263 		sfc_notice(sa, "Link status is %s",
264 			   current_link.link_status ? "UP" : "DOWN");
265 
266 	return ret;
267 }
268 
269 static void
270 sfc_dev_stop(struct rte_eth_dev *dev)
271 {
272 	struct sfc_adapter *sa = dev->data->dev_private;
273 
274 	sfc_log_init(sa, "entry");
275 
276 	sfc_adapter_lock(sa);
277 	sfc_stop(sa);
278 	sfc_adapter_unlock(sa);
279 
280 	sfc_log_init(sa, "done");
281 }
282 
283 static int
284 sfc_dev_set_link_up(struct rte_eth_dev *dev)
285 {
286 	struct sfc_adapter *sa = dev->data->dev_private;
287 	int rc;
288 
289 	sfc_log_init(sa, "entry");
290 
291 	sfc_adapter_lock(sa);
292 	rc = sfc_start(sa);
293 	sfc_adapter_unlock(sa);
294 
295 	SFC_ASSERT(rc >= 0);
296 	return -rc;
297 }
298 
299 static int
300 sfc_dev_set_link_down(struct rte_eth_dev *dev)
301 {
302 	struct sfc_adapter *sa = dev->data->dev_private;
303 
304 	sfc_log_init(sa, "entry");
305 
306 	sfc_adapter_lock(sa);
307 	sfc_stop(sa);
308 	sfc_adapter_unlock(sa);
309 
310 	return 0;
311 }
312 
313 static void
314 sfc_dev_close(struct rte_eth_dev *dev)
315 {
316 	struct sfc_adapter *sa = dev->data->dev_private;
317 
318 	sfc_log_init(sa, "entry");
319 
320 	sfc_adapter_lock(sa);
321 	switch (sa->state) {
322 	case SFC_ADAPTER_STARTED:
323 		sfc_stop(sa);
324 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
325 		/* FALLTHROUGH */
326 	case SFC_ADAPTER_CONFIGURED:
327 		sfc_close(sa);
328 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
329 		/* FALLTHROUGH */
330 	case SFC_ADAPTER_INITIALIZED:
331 		break;
332 	default:
333 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
334 		break;
335 	}
336 	sfc_adapter_unlock(sa);
337 
338 	sfc_log_init(sa, "done");
339 }
340 
341 static void
342 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
343 		   boolean_t enabled)
344 {
345 	struct sfc_port *port;
346 	boolean_t *toggle;
347 	struct sfc_adapter *sa = dev->data->dev_private;
348 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
349 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
350 
351 	sfc_adapter_lock(sa);
352 
353 	port = &sa->port;
354 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
355 
356 	if (*toggle != enabled) {
357 		*toggle = enabled;
358 
359 		if (port->isolated) {
360 			sfc_warn(sa, "isolated mode is active on the port");
361 			sfc_warn(sa, "the change is to be applied on the next "
362 				     "start provided that isolated mode is "
363 				     "disabled prior the next start");
364 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
365 			   (sfc_set_rx_mode(sa) != 0)) {
366 			*toggle = !(enabled);
367 			sfc_warn(sa, "Failed to %s %s mode",
368 				 ((enabled) ? "enable" : "disable"), desc);
369 		}
370 	}
371 
372 	sfc_adapter_unlock(sa);
373 }
374 
375 static void
376 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
377 {
378 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
379 }
380 
381 static void
382 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
383 {
384 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
385 }
386 
387 static void
388 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
389 {
390 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
391 }
392 
393 static void
394 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
395 {
396 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
397 }
398 
399 static int
400 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
401 		   uint16_t nb_rx_desc, unsigned int socket_id,
402 		   const struct rte_eth_rxconf *rx_conf,
403 		   struct rte_mempool *mb_pool)
404 {
405 	struct sfc_adapter *sa = dev->data->dev_private;
406 	int rc;
407 
408 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
409 		     rx_queue_id, nb_rx_desc, socket_id);
410 
411 	sfc_adapter_lock(sa);
412 
413 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
414 			  rx_conf, mb_pool);
415 	if (rc != 0)
416 		goto fail_rx_qinit;
417 
418 	dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].dp;
419 
420 	sfc_adapter_unlock(sa);
421 
422 	return 0;
423 
424 fail_rx_qinit:
425 	sfc_adapter_unlock(sa);
426 	SFC_ASSERT(rc > 0);
427 	return -rc;
428 }
429 
430 static void
431 sfc_rx_queue_release(void *queue)
432 {
433 	struct sfc_dp_rxq *dp_rxq = queue;
434 	struct sfc_rxq *rxq;
435 	struct sfc_adapter *sa;
436 	unsigned int sw_index;
437 
438 	if (dp_rxq == NULL)
439 		return;
440 
441 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
442 	sa = rxq->evq->sa;
443 	sfc_adapter_lock(sa);
444 
445 	sw_index = dp_rxq->dpq.queue_id;
446 
447 	sfc_log_init(sa, "RxQ=%u", sw_index);
448 
449 	sfc_rx_qfini(sa, sw_index);
450 
451 	sfc_adapter_unlock(sa);
452 }
453 
454 static int
455 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
456 		   uint16_t nb_tx_desc, unsigned int socket_id,
457 		   const struct rte_eth_txconf *tx_conf)
458 {
459 	struct sfc_adapter *sa = dev->data->dev_private;
460 	int rc;
461 
462 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
463 		     tx_queue_id, nb_tx_desc, socket_id);
464 
465 	sfc_adapter_lock(sa);
466 
467 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
468 	if (rc != 0)
469 		goto fail_tx_qinit;
470 
471 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].dp;
472 
473 	sfc_adapter_unlock(sa);
474 	return 0;
475 
476 fail_tx_qinit:
477 	sfc_adapter_unlock(sa);
478 	SFC_ASSERT(rc > 0);
479 	return -rc;
480 }
481 
482 static void
483 sfc_tx_queue_release(void *queue)
484 {
485 	struct sfc_dp_txq *dp_txq = queue;
486 	struct sfc_txq *txq;
487 	unsigned int sw_index;
488 	struct sfc_adapter *sa;
489 
490 	if (dp_txq == NULL)
491 		return;
492 
493 	txq = sfc_txq_by_dp_txq(dp_txq);
494 	sw_index = sfc_txq_sw_index(txq);
495 
496 	SFC_ASSERT(txq->evq != NULL);
497 	sa = txq->evq->sa;
498 
499 	sfc_log_init(sa, "TxQ = %u", sw_index);
500 
501 	sfc_adapter_lock(sa);
502 
503 	sfc_tx_qfini(sa, sw_index);
504 
505 	sfc_adapter_unlock(sa);
506 }
507 
508 /*
509  * Some statistics are computed as A - B where A and B each increase
510  * monotonically with some hardware counter(s) and the counters are read
511  * asynchronously.
512  *
513  * If packet X is counted in A, but not counted in B yet, computed value is
514  * greater than real.
515  *
516  * If packet X is not counted in A at the moment of reading the counter,
517  * but counted in B at the moment of reading the counter, computed value
518  * is less than real.
519  *
520  * However, counter which grows backward is worse evil than slightly wrong
521  * value. So, let's try to guarantee that it never happens except may be
522  * the case when the MAC stats are zeroed as a result of a NIC reset.
523  */
524 static void
525 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
526 {
527 	if ((int64_t)(newval - *stat) > 0 || newval == 0)
528 		*stat = newval;
529 }
530 
531 static int
532 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
533 {
534 	struct sfc_adapter *sa = dev->data->dev_private;
535 	struct sfc_port *port = &sa->port;
536 	uint64_t *mac_stats;
537 	int ret;
538 
539 	rte_spinlock_lock(&port->mac_stats_lock);
540 
541 	ret = sfc_port_update_mac_stats(sa);
542 	if (ret != 0)
543 		goto unlock;
544 
545 	mac_stats = port->mac_stats_buf;
546 
547 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
548 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
549 		stats->ipackets =
550 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
551 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
552 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
553 		stats->opackets =
554 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
555 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
556 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
557 		stats->ibytes =
558 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
559 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
560 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
561 		stats->obytes =
562 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
563 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
564 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
565 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
566 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
567 	} else {
568 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
569 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
570 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
571 		/*
572 		 * Take into account stats which are whenever supported
573 		 * on EF10. If some stat is not supported by current
574 		 * firmware variant or HW revision, it is guaranteed
575 		 * to be zero in mac_stats.
576 		 */
577 		stats->imissed =
578 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
579 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
580 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
581 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
582 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
583 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
584 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
585 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
586 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
587 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
588 		stats->ierrors =
589 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
590 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
591 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
592 		/* no oerrors counters supported on EF10 */
593 
594 		/* Exclude missed, errors and pauses from Rx packets */
595 		sfc_update_diff_stat(&port->ipackets,
596 			mac_stats[EFX_MAC_RX_PKTS] -
597 			mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
598 			stats->imissed - stats->ierrors);
599 		stats->ipackets = port->ipackets;
600 	}
601 
602 unlock:
603 	rte_spinlock_unlock(&port->mac_stats_lock);
604 	SFC_ASSERT(ret >= 0);
605 	return -ret;
606 }
607 
608 static void
609 sfc_stats_reset(struct rte_eth_dev *dev)
610 {
611 	struct sfc_adapter *sa = dev->data->dev_private;
612 	struct sfc_port *port = &sa->port;
613 	int rc;
614 
615 	if (sa->state != SFC_ADAPTER_STARTED) {
616 		/*
617 		 * The operation cannot be done if port is not started; it
618 		 * will be scheduled to be done during the next port start
619 		 */
620 		port->mac_stats_reset_pending = B_TRUE;
621 		return;
622 	}
623 
624 	rc = sfc_port_reset_mac_stats(sa);
625 	if (rc != 0)
626 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
627 }
628 
629 static int
630 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
631 	       unsigned int xstats_count)
632 {
633 	struct sfc_adapter *sa = dev->data->dev_private;
634 	struct sfc_port *port = &sa->port;
635 	uint64_t *mac_stats;
636 	int rc;
637 	unsigned int i;
638 	int nstats = 0;
639 
640 	rte_spinlock_lock(&port->mac_stats_lock);
641 
642 	rc = sfc_port_update_mac_stats(sa);
643 	if (rc != 0) {
644 		SFC_ASSERT(rc > 0);
645 		nstats = -rc;
646 		goto unlock;
647 	}
648 
649 	mac_stats = port->mac_stats_buf;
650 
651 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
652 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
653 			if (xstats != NULL && nstats < (int)xstats_count) {
654 				xstats[nstats].id = nstats;
655 				xstats[nstats].value = mac_stats[i];
656 			}
657 			nstats++;
658 		}
659 	}
660 
661 unlock:
662 	rte_spinlock_unlock(&port->mac_stats_lock);
663 
664 	return nstats;
665 }
666 
667 static int
668 sfc_xstats_get_names(struct rte_eth_dev *dev,
669 		     struct rte_eth_xstat_name *xstats_names,
670 		     unsigned int xstats_count)
671 {
672 	struct sfc_adapter *sa = dev->data->dev_private;
673 	struct sfc_port *port = &sa->port;
674 	unsigned int i;
675 	unsigned int nstats = 0;
676 
677 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
678 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
679 			if (xstats_names != NULL && nstats < xstats_count)
680 				strlcpy(xstats_names[nstats].name,
681 					efx_mac_stat_name(sa->nic, i),
682 					sizeof(xstats_names[0].name));
683 			nstats++;
684 		}
685 	}
686 
687 	return nstats;
688 }
689 
690 static int
691 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
692 		     uint64_t *values, unsigned int n)
693 {
694 	struct sfc_adapter *sa = dev->data->dev_private;
695 	struct sfc_port *port = &sa->port;
696 	uint64_t *mac_stats;
697 	unsigned int nb_supported = 0;
698 	unsigned int nb_written = 0;
699 	unsigned int i;
700 	int ret;
701 	int rc;
702 
703 	if (unlikely(values == NULL) ||
704 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
705 		return port->mac_stats_nb_supported;
706 
707 	rte_spinlock_lock(&port->mac_stats_lock);
708 
709 	rc = sfc_port_update_mac_stats(sa);
710 	if (rc != 0) {
711 		SFC_ASSERT(rc > 0);
712 		ret = -rc;
713 		goto unlock;
714 	}
715 
716 	mac_stats = port->mac_stats_buf;
717 
718 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
719 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
720 			continue;
721 
722 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
723 			values[nb_written++] = mac_stats[i];
724 
725 		++nb_supported;
726 	}
727 
728 	ret = nb_written;
729 
730 unlock:
731 	rte_spinlock_unlock(&port->mac_stats_lock);
732 
733 	return ret;
734 }
735 
736 static int
737 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
738 			   struct rte_eth_xstat_name *xstats_names,
739 			   const uint64_t *ids, unsigned int size)
740 {
741 	struct sfc_adapter *sa = dev->data->dev_private;
742 	struct sfc_port *port = &sa->port;
743 	unsigned int nb_supported = 0;
744 	unsigned int nb_written = 0;
745 	unsigned int i;
746 
747 	if (unlikely(xstats_names == NULL) ||
748 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
749 		return port->mac_stats_nb_supported;
750 
751 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
752 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
753 			continue;
754 
755 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
756 			char *name = xstats_names[nb_written++].name;
757 
758 			strlcpy(name, efx_mac_stat_name(sa->nic, i),
759 				sizeof(xstats_names[0].name));
760 		}
761 
762 		++nb_supported;
763 	}
764 
765 	return nb_written;
766 }
767 
768 static int
769 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
770 {
771 	struct sfc_adapter *sa = dev->data->dev_private;
772 	unsigned int wanted_fc, link_fc;
773 
774 	memset(fc_conf, 0, sizeof(*fc_conf));
775 
776 	sfc_adapter_lock(sa);
777 
778 	if (sa->state == SFC_ADAPTER_STARTED)
779 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
780 	else
781 		link_fc = sa->port.flow_ctrl;
782 
783 	switch (link_fc) {
784 	case 0:
785 		fc_conf->mode = RTE_FC_NONE;
786 		break;
787 	case EFX_FCNTL_RESPOND:
788 		fc_conf->mode = RTE_FC_RX_PAUSE;
789 		break;
790 	case EFX_FCNTL_GENERATE:
791 		fc_conf->mode = RTE_FC_TX_PAUSE;
792 		break;
793 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
794 		fc_conf->mode = RTE_FC_FULL;
795 		break;
796 	default:
797 		sfc_err(sa, "%s: unexpected flow control value %#x",
798 			__func__, link_fc);
799 	}
800 
801 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
802 
803 	sfc_adapter_unlock(sa);
804 
805 	return 0;
806 }
807 
808 static int
809 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
810 {
811 	struct sfc_adapter *sa = dev->data->dev_private;
812 	struct sfc_port *port = &sa->port;
813 	unsigned int fcntl;
814 	int rc;
815 
816 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
817 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
818 	    fc_conf->mac_ctrl_frame_fwd != 0) {
819 		sfc_err(sa, "unsupported flow control settings specified");
820 		rc = EINVAL;
821 		goto fail_inval;
822 	}
823 
824 	switch (fc_conf->mode) {
825 	case RTE_FC_NONE:
826 		fcntl = 0;
827 		break;
828 	case RTE_FC_RX_PAUSE:
829 		fcntl = EFX_FCNTL_RESPOND;
830 		break;
831 	case RTE_FC_TX_PAUSE:
832 		fcntl = EFX_FCNTL_GENERATE;
833 		break;
834 	case RTE_FC_FULL:
835 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
836 		break;
837 	default:
838 		rc = EINVAL;
839 		goto fail_inval;
840 	}
841 
842 	sfc_adapter_lock(sa);
843 
844 	if (sa->state == SFC_ADAPTER_STARTED) {
845 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
846 		if (rc != 0)
847 			goto fail_mac_fcntl_set;
848 	}
849 
850 	port->flow_ctrl = fcntl;
851 	port->flow_ctrl_autoneg = fc_conf->autoneg;
852 
853 	sfc_adapter_unlock(sa);
854 
855 	return 0;
856 
857 fail_mac_fcntl_set:
858 	sfc_adapter_unlock(sa);
859 fail_inval:
860 	SFC_ASSERT(rc > 0);
861 	return -rc;
862 }
863 
864 static int
865 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
866 {
867 	struct sfc_adapter *sa = dev->data->dev_private;
868 	size_t pdu = EFX_MAC_PDU(mtu);
869 	size_t old_pdu;
870 	int rc;
871 
872 	sfc_log_init(sa, "mtu=%u", mtu);
873 
874 	rc = EINVAL;
875 	if (pdu < EFX_MAC_PDU_MIN) {
876 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
877 			(unsigned int)mtu, (unsigned int)pdu,
878 			EFX_MAC_PDU_MIN);
879 		goto fail_inval;
880 	}
881 	if (pdu > EFX_MAC_PDU_MAX) {
882 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
883 			(unsigned int)mtu, (unsigned int)pdu,
884 			EFX_MAC_PDU_MAX);
885 		goto fail_inval;
886 	}
887 
888 	sfc_adapter_lock(sa);
889 
890 	if (pdu != sa->port.pdu) {
891 		if (sa->state == SFC_ADAPTER_STARTED) {
892 			sfc_stop(sa);
893 
894 			old_pdu = sa->port.pdu;
895 			sa->port.pdu = pdu;
896 			rc = sfc_start(sa);
897 			if (rc != 0)
898 				goto fail_start;
899 		} else {
900 			sa->port.pdu = pdu;
901 		}
902 	}
903 
904 	/*
905 	 * The driver does not use it, but other PMDs update jumbo frame
906 	 * flag and max_rx_pkt_len when MTU is set.
907 	 */
908 	if (mtu > ETHER_MAX_LEN) {
909 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
910 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
911 	}
912 
913 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
914 
915 	sfc_adapter_unlock(sa);
916 
917 	sfc_log_init(sa, "done");
918 	return 0;
919 
920 fail_start:
921 	sa->port.pdu = old_pdu;
922 	if (sfc_start(sa) != 0)
923 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
924 			"PDU max size - port is stopped",
925 			(unsigned int)pdu, (unsigned int)old_pdu);
926 	sfc_adapter_unlock(sa);
927 
928 fail_inval:
929 	sfc_log_init(sa, "failed %d", rc);
930 	SFC_ASSERT(rc > 0);
931 	return -rc;
932 }
933 static int
934 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
935 {
936 	struct sfc_adapter *sa = dev->data->dev_private;
937 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
938 	struct sfc_port *port = &sa->port;
939 	struct ether_addr *old_addr = &dev->data->mac_addrs[0];
940 	int rc = 0;
941 
942 	sfc_adapter_lock(sa);
943 
944 	/*
945 	 * Copy the address to the device private data so that
946 	 * it could be recalled in the case of adapter restart.
947 	 */
948 	ether_addr_copy(mac_addr, &port->default_mac_addr);
949 
950 	/*
951 	 * Neither of the two following checks can return
952 	 * an error. The new MAC address is preserved in
953 	 * the device private data and can be activated
954 	 * on the next port start if the user prevents
955 	 * isolated mode from being enabled.
956 	 */
957 	if (port->isolated) {
958 		sfc_warn(sa, "isolated mode is active on the port");
959 		sfc_warn(sa, "will not set MAC address");
960 		goto unlock;
961 	}
962 
963 	if (sa->state != SFC_ADAPTER_STARTED) {
964 		sfc_notice(sa, "the port is not started");
965 		sfc_notice(sa, "the new MAC address will be set on port start");
966 
967 		goto unlock;
968 	}
969 
970 	if (encp->enc_allow_set_mac_with_installed_filters) {
971 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
972 		if (rc != 0) {
973 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
974 			goto unlock;
975 		}
976 
977 		/*
978 		 * Changing the MAC address by means of MCDI request
979 		 * has no effect on received traffic, therefore
980 		 * we also need to update unicast filters
981 		 */
982 		rc = sfc_set_rx_mode(sa);
983 		if (rc != 0) {
984 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
985 			/* Rollback the old address */
986 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
987 			(void)sfc_set_rx_mode(sa);
988 		}
989 	} else {
990 		sfc_warn(sa, "cannot set MAC address with filters installed");
991 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
992 		sfc_warn(sa, "(some traffic may be dropped)");
993 
994 		/*
995 		 * Since setting MAC address with filters installed is not
996 		 * allowed on the adapter, the new MAC address will be set
997 		 * by means of adapter restart. sfc_start() shall retrieve
998 		 * the new address from the device private data and set it.
999 		 */
1000 		sfc_stop(sa);
1001 		rc = sfc_start(sa);
1002 		if (rc != 0)
1003 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1004 	}
1005 
1006 unlock:
1007 	if (rc != 0)
1008 		ether_addr_copy(old_addr, &port->default_mac_addr);
1009 
1010 	sfc_adapter_unlock(sa);
1011 
1012 	SFC_ASSERT(rc >= 0);
1013 	return -rc;
1014 }
1015 
1016 
1017 static int
1018 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1019 		     uint32_t nb_mc_addr)
1020 {
1021 	struct sfc_adapter *sa = dev->data->dev_private;
1022 	struct sfc_port *port = &sa->port;
1023 	uint8_t *mc_addrs = port->mcast_addrs;
1024 	int rc;
1025 	unsigned int i;
1026 
1027 	if (port->isolated) {
1028 		sfc_err(sa, "isolated mode is active on the port");
1029 		sfc_err(sa, "will not set multicast address list");
1030 		return -ENOTSUP;
1031 	}
1032 
1033 	if (mc_addrs == NULL)
1034 		return -ENOBUFS;
1035 
1036 	if (nb_mc_addr > port->max_mcast_addrs) {
1037 		sfc_err(sa, "too many multicast addresses: %u > %u",
1038 			 nb_mc_addr, port->max_mcast_addrs);
1039 		return -EINVAL;
1040 	}
1041 
1042 	for (i = 0; i < nb_mc_addr; ++i) {
1043 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1044 				 EFX_MAC_ADDR_LEN);
1045 		mc_addrs += EFX_MAC_ADDR_LEN;
1046 	}
1047 
1048 	port->nb_mcast_addrs = nb_mc_addr;
1049 
1050 	if (sa->state != SFC_ADAPTER_STARTED)
1051 		return 0;
1052 
1053 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1054 					port->nb_mcast_addrs);
1055 	if (rc != 0)
1056 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1057 
1058 	SFC_ASSERT(rc >= 0);
1059 	return -rc;
1060 }
1061 
1062 /*
1063  * The function is used by the secondary process as well. It must not
1064  * use any process-local pointers from the adapter data.
1065  */
1066 static void
1067 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1068 		      struct rte_eth_rxq_info *qinfo)
1069 {
1070 	struct sfc_adapter *sa = dev->data->dev_private;
1071 	struct sfc_rxq_info *rxq_info;
1072 
1073 	sfc_adapter_lock(sa);
1074 
1075 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1076 
1077 	rxq_info = &sa->rxq_info[rx_queue_id];
1078 
1079 	qinfo->mp = rxq_info->refill_mb_pool;
1080 	qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1081 	qinfo->conf.rx_drop_en = 1;
1082 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1083 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1084 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1085 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1086 		qinfo->scattered_rx = 1;
1087 	}
1088 	qinfo->nb_desc = rxq_info->entries;
1089 
1090 	sfc_adapter_unlock(sa);
1091 }
1092 
1093 /*
1094  * The function is used by the secondary process as well. It must not
1095  * use any process-local pointers from the adapter data.
1096  */
1097 static void
1098 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1099 		      struct rte_eth_txq_info *qinfo)
1100 {
1101 	struct sfc_adapter *sa = dev->data->dev_private;
1102 	struct sfc_txq_info *txq_info;
1103 
1104 	sfc_adapter_lock(sa);
1105 
1106 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1107 
1108 	txq_info = &sa->txq_info[tx_queue_id];
1109 
1110 	memset(qinfo, 0, sizeof(*qinfo));
1111 
1112 	qinfo->conf.offloads = txq_info->offloads;
1113 	qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1114 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1115 	qinfo->nb_desc = txq_info->entries;
1116 
1117 	sfc_adapter_unlock(sa);
1118 }
1119 
1120 /*
1121  * The function is used by the secondary process as well. It must not
1122  * use any process-local pointers from the adapter data.
1123  */
1124 static uint32_t
1125 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1126 {
1127 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1128 	struct sfc_adapter *sa = dev->data->dev_private;
1129 	struct sfc_rxq_info *rxq_info;
1130 
1131 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1132 	rxq_info = &sa->rxq_info[rx_queue_id];
1133 
1134 	if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1135 		return 0;
1136 
1137 	return sap->dp_rx->qdesc_npending(rxq_info->dp);
1138 }
1139 
1140 /*
1141  * The function is used by the secondary process as well. It must not
1142  * use any process-local pointers from the adapter data.
1143  */
1144 static int
1145 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1146 {
1147 	struct sfc_dp_rxq *dp_rxq = queue;
1148 	const struct sfc_dp_rx *dp_rx;
1149 
1150 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1151 
1152 	return offset < dp_rx->qdesc_npending(dp_rxq);
1153 }
1154 
1155 /*
1156  * The function is used by the secondary process as well. It must not
1157  * use any process-local pointers from the adapter data.
1158  */
1159 static int
1160 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1161 {
1162 	struct sfc_dp_rxq *dp_rxq = queue;
1163 	const struct sfc_dp_rx *dp_rx;
1164 
1165 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1166 
1167 	return dp_rx->qdesc_status(dp_rxq, offset);
1168 }
1169 
1170 /*
1171  * The function is used by the secondary process as well. It must not
1172  * use any process-local pointers from the adapter data.
1173  */
1174 static int
1175 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1176 {
1177 	struct sfc_dp_txq *dp_txq = queue;
1178 	const struct sfc_dp_tx *dp_tx;
1179 
1180 	dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1181 
1182 	return dp_tx->qdesc_status(dp_txq, offset);
1183 }
1184 
1185 static int
1186 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1187 {
1188 	struct sfc_adapter *sa = dev->data->dev_private;
1189 	int rc;
1190 
1191 	sfc_log_init(sa, "RxQ=%u", rx_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 	if (sa->rxq_info[rx_queue_id].rxq == NULL)
1200 		goto fail_not_setup;
1201 
1202 	rc = sfc_rx_qstart(sa, rx_queue_id);
1203 	if (rc != 0)
1204 		goto fail_rx_qstart;
1205 
1206 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1207 
1208 	sfc_adapter_unlock(sa);
1209 
1210 	return 0;
1211 
1212 fail_rx_qstart:
1213 fail_not_setup:
1214 fail_not_started:
1215 	sfc_adapter_unlock(sa);
1216 	SFC_ASSERT(rc > 0);
1217 	return -rc;
1218 }
1219 
1220 static int
1221 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1222 {
1223 	struct sfc_adapter *sa = dev->data->dev_private;
1224 
1225 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1226 
1227 	sfc_adapter_lock(sa);
1228 	sfc_rx_qstop(sa, rx_queue_id);
1229 
1230 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1231 
1232 	sfc_adapter_unlock(sa);
1233 
1234 	return 0;
1235 }
1236 
1237 static int
1238 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1239 {
1240 	struct sfc_adapter *sa = dev->data->dev_private;
1241 	int rc;
1242 
1243 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1244 
1245 	sfc_adapter_lock(sa);
1246 
1247 	rc = EINVAL;
1248 	if (sa->state != SFC_ADAPTER_STARTED)
1249 		goto fail_not_started;
1250 
1251 	if (sa->txq_info[tx_queue_id].txq == NULL)
1252 		goto fail_not_setup;
1253 
1254 	rc = sfc_tx_qstart(sa, tx_queue_id);
1255 	if (rc != 0)
1256 		goto fail_tx_qstart;
1257 
1258 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1259 
1260 	sfc_adapter_unlock(sa);
1261 	return 0;
1262 
1263 fail_tx_qstart:
1264 
1265 fail_not_setup:
1266 fail_not_started:
1267 	sfc_adapter_unlock(sa);
1268 	SFC_ASSERT(rc > 0);
1269 	return -rc;
1270 }
1271 
1272 static int
1273 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1274 {
1275 	struct sfc_adapter *sa = dev->data->dev_private;
1276 
1277 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1278 
1279 	sfc_adapter_lock(sa);
1280 
1281 	sfc_tx_qstop(sa, tx_queue_id);
1282 
1283 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1284 
1285 	sfc_adapter_unlock(sa);
1286 	return 0;
1287 }
1288 
1289 static efx_tunnel_protocol_t
1290 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1291 {
1292 	switch (rte_type) {
1293 	case RTE_TUNNEL_TYPE_VXLAN:
1294 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1295 	case RTE_TUNNEL_TYPE_GENEVE:
1296 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1297 	default:
1298 		return EFX_TUNNEL_NPROTOS;
1299 	}
1300 }
1301 
1302 enum sfc_udp_tunnel_op_e {
1303 	SFC_UDP_TUNNEL_ADD_PORT,
1304 	SFC_UDP_TUNNEL_DEL_PORT,
1305 };
1306 
1307 static int
1308 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1309 		      struct rte_eth_udp_tunnel *tunnel_udp,
1310 		      enum sfc_udp_tunnel_op_e op)
1311 {
1312 	struct sfc_adapter *sa = dev->data->dev_private;
1313 	efx_tunnel_protocol_t tunnel_proto;
1314 	int rc;
1315 
1316 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1317 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1318 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1319 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1320 
1321 	tunnel_proto =
1322 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1323 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1324 		rc = ENOTSUP;
1325 		goto fail_bad_proto;
1326 	}
1327 
1328 	sfc_adapter_lock(sa);
1329 
1330 	switch (op) {
1331 	case SFC_UDP_TUNNEL_ADD_PORT:
1332 		rc = efx_tunnel_config_udp_add(sa->nic,
1333 					       tunnel_udp->udp_port,
1334 					       tunnel_proto);
1335 		break;
1336 	case SFC_UDP_TUNNEL_DEL_PORT:
1337 		rc = efx_tunnel_config_udp_remove(sa->nic,
1338 						  tunnel_udp->udp_port,
1339 						  tunnel_proto);
1340 		break;
1341 	default:
1342 		rc = EINVAL;
1343 		goto fail_bad_op;
1344 	}
1345 
1346 	if (rc != 0)
1347 		goto fail_op;
1348 
1349 	if (sa->state == SFC_ADAPTER_STARTED) {
1350 		rc = efx_tunnel_reconfigure(sa->nic);
1351 		if (rc == EAGAIN) {
1352 			/*
1353 			 * Configuration is accepted by FW and MC reboot
1354 			 * is initiated to apply the changes. MC reboot
1355 			 * will be handled in a usual way (MC reboot
1356 			 * event on management event queue and adapter
1357 			 * restart).
1358 			 */
1359 			rc = 0;
1360 		} else if (rc != 0) {
1361 			goto fail_reconfigure;
1362 		}
1363 	}
1364 
1365 	sfc_adapter_unlock(sa);
1366 	return 0;
1367 
1368 fail_reconfigure:
1369 	/* Remove/restore entry since the change makes the trouble */
1370 	switch (op) {
1371 	case SFC_UDP_TUNNEL_ADD_PORT:
1372 		(void)efx_tunnel_config_udp_remove(sa->nic,
1373 						   tunnel_udp->udp_port,
1374 						   tunnel_proto);
1375 		break;
1376 	case SFC_UDP_TUNNEL_DEL_PORT:
1377 		(void)efx_tunnel_config_udp_add(sa->nic,
1378 						tunnel_udp->udp_port,
1379 						tunnel_proto);
1380 		break;
1381 	}
1382 
1383 fail_op:
1384 fail_bad_op:
1385 	sfc_adapter_unlock(sa);
1386 
1387 fail_bad_proto:
1388 	SFC_ASSERT(rc > 0);
1389 	return -rc;
1390 }
1391 
1392 static int
1393 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1394 			    struct rte_eth_udp_tunnel *tunnel_udp)
1395 {
1396 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1397 }
1398 
1399 static int
1400 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1401 			    struct rte_eth_udp_tunnel *tunnel_udp)
1402 {
1403 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1404 }
1405 
1406 /*
1407  * The function is used by the secondary process as well. It must not
1408  * use any process-local pointers from the adapter data.
1409  */
1410 static int
1411 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1412 			  struct rte_eth_rss_conf *rss_conf)
1413 {
1414 	struct sfc_adapter *sa = dev->data->dev_private;
1415 	struct sfc_rss *rss = &sa->rss;
1416 
1417 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1418 		return -ENOTSUP;
1419 
1420 	sfc_adapter_lock(sa);
1421 
1422 	/*
1423 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1424 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1425 	 * flags which corresponds to the active EFX configuration stored
1426 	 * locally in 'sfc_adapter' and kept up-to-date
1427 	 */
1428 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1429 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1430 	if (rss_conf->rss_key != NULL)
1431 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1432 
1433 	sfc_adapter_unlock(sa);
1434 
1435 	return 0;
1436 }
1437 
1438 static int
1439 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1440 			struct rte_eth_rss_conf *rss_conf)
1441 {
1442 	struct sfc_adapter *sa = dev->data->dev_private;
1443 	struct sfc_rss *rss = &sa->rss;
1444 	struct sfc_port *port = &sa->port;
1445 	unsigned int efx_hash_types;
1446 	int rc = 0;
1447 
1448 	if (port->isolated)
1449 		return -ENOTSUP;
1450 
1451 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1452 		sfc_err(sa, "RSS is not available");
1453 		return -ENOTSUP;
1454 	}
1455 
1456 	if (rss->channels == 0) {
1457 		sfc_err(sa, "RSS is not configured");
1458 		return -EINVAL;
1459 	}
1460 
1461 	if ((rss_conf->rss_key != NULL) &&
1462 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1463 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1464 			sizeof(rss->key));
1465 		return -EINVAL;
1466 	}
1467 
1468 	sfc_adapter_lock(sa);
1469 
1470 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1471 	if (rc != 0)
1472 		goto fail_rx_hf_rte_to_efx;
1473 
1474 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1475 				   rss->hash_alg, efx_hash_types, B_TRUE);
1476 	if (rc != 0)
1477 		goto fail_scale_mode_set;
1478 
1479 	if (rss_conf->rss_key != NULL) {
1480 		if (sa->state == SFC_ADAPTER_STARTED) {
1481 			rc = efx_rx_scale_key_set(sa->nic,
1482 						  EFX_RSS_CONTEXT_DEFAULT,
1483 						  rss_conf->rss_key,
1484 						  sizeof(rss->key));
1485 			if (rc != 0)
1486 				goto fail_scale_key_set;
1487 		}
1488 
1489 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1490 	}
1491 
1492 	rss->hash_types = efx_hash_types;
1493 
1494 	sfc_adapter_unlock(sa);
1495 
1496 	return 0;
1497 
1498 fail_scale_key_set:
1499 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1500 				  EFX_RX_HASHALG_TOEPLITZ,
1501 				  rss->hash_types, B_TRUE) != 0)
1502 		sfc_err(sa, "failed to restore RSS mode");
1503 
1504 fail_scale_mode_set:
1505 fail_rx_hf_rte_to_efx:
1506 	sfc_adapter_unlock(sa);
1507 	return -rc;
1508 }
1509 
1510 /*
1511  * The function is used by the secondary process as well. It must not
1512  * use any process-local pointers from the adapter data.
1513  */
1514 static int
1515 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1516 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1517 		       uint16_t reta_size)
1518 {
1519 	struct sfc_adapter *sa = dev->data->dev_private;
1520 	struct sfc_rss *rss = &sa->rss;
1521 	struct sfc_port *port = &sa->port;
1522 	int entry;
1523 
1524 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1525 		return -ENOTSUP;
1526 
1527 	if (rss->channels == 0)
1528 		return -EINVAL;
1529 
1530 	if (reta_size != EFX_RSS_TBL_SIZE)
1531 		return -EINVAL;
1532 
1533 	sfc_adapter_lock(sa);
1534 
1535 	for (entry = 0; entry < reta_size; entry++) {
1536 		int grp = entry / RTE_RETA_GROUP_SIZE;
1537 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1538 
1539 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1540 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1541 	}
1542 
1543 	sfc_adapter_unlock(sa);
1544 
1545 	return 0;
1546 }
1547 
1548 static int
1549 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1550 			struct rte_eth_rss_reta_entry64 *reta_conf,
1551 			uint16_t reta_size)
1552 {
1553 	struct sfc_adapter *sa = dev->data->dev_private;
1554 	struct sfc_rss *rss = &sa->rss;
1555 	struct sfc_port *port = &sa->port;
1556 	unsigned int *rss_tbl_new;
1557 	uint16_t entry;
1558 	int rc = 0;
1559 
1560 
1561 	if (port->isolated)
1562 		return -ENOTSUP;
1563 
1564 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1565 		sfc_err(sa, "RSS is not available");
1566 		return -ENOTSUP;
1567 	}
1568 
1569 	if (rss->channels == 0) {
1570 		sfc_err(sa, "RSS is not configured");
1571 		return -EINVAL;
1572 	}
1573 
1574 	if (reta_size != EFX_RSS_TBL_SIZE) {
1575 		sfc_err(sa, "RETA size is wrong (should be %u)",
1576 			EFX_RSS_TBL_SIZE);
1577 		return -EINVAL;
1578 	}
1579 
1580 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1581 	if (rss_tbl_new == NULL)
1582 		return -ENOMEM;
1583 
1584 	sfc_adapter_lock(sa);
1585 
1586 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1587 
1588 	for (entry = 0; entry < reta_size; entry++) {
1589 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1590 		struct rte_eth_rss_reta_entry64 *grp;
1591 
1592 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1593 
1594 		if (grp->mask & (1ull << grp_idx)) {
1595 			if (grp->reta[grp_idx] >= rss->channels) {
1596 				rc = EINVAL;
1597 				goto bad_reta_entry;
1598 			}
1599 			rss_tbl_new[entry] = grp->reta[grp_idx];
1600 		}
1601 	}
1602 
1603 	if (sa->state == SFC_ADAPTER_STARTED) {
1604 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1605 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1606 		if (rc != 0)
1607 			goto fail_scale_tbl_set;
1608 	}
1609 
1610 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1611 
1612 fail_scale_tbl_set:
1613 bad_reta_entry:
1614 	sfc_adapter_unlock(sa);
1615 
1616 	rte_free(rss_tbl_new);
1617 
1618 	SFC_ASSERT(rc >= 0);
1619 	return -rc;
1620 }
1621 
1622 static int
1623 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1624 		    enum rte_filter_op filter_op,
1625 		    void *arg)
1626 {
1627 	struct sfc_adapter *sa = dev->data->dev_private;
1628 	int rc = ENOTSUP;
1629 
1630 	sfc_log_init(sa, "entry");
1631 
1632 	switch (filter_type) {
1633 	case RTE_ETH_FILTER_NONE:
1634 		sfc_err(sa, "Global filters configuration not supported");
1635 		break;
1636 	case RTE_ETH_FILTER_MACVLAN:
1637 		sfc_err(sa, "MACVLAN filters not supported");
1638 		break;
1639 	case RTE_ETH_FILTER_ETHERTYPE:
1640 		sfc_err(sa, "EtherType filters not supported");
1641 		break;
1642 	case RTE_ETH_FILTER_FLEXIBLE:
1643 		sfc_err(sa, "Flexible filters not supported");
1644 		break;
1645 	case RTE_ETH_FILTER_SYN:
1646 		sfc_err(sa, "SYN filters not supported");
1647 		break;
1648 	case RTE_ETH_FILTER_NTUPLE:
1649 		sfc_err(sa, "NTUPLE filters not supported");
1650 		break;
1651 	case RTE_ETH_FILTER_TUNNEL:
1652 		sfc_err(sa, "Tunnel filters not supported");
1653 		break;
1654 	case RTE_ETH_FILTER_FDIR:
1655 		sfc_err(sa, "Flow Director filters not supported");
1656 		break;
1657 	case RTE_ETH_FILTER_HASH:
1658 		sfc_err(sa, "Hash filters not supported");
1659 		break;
1660 	case RTE_ETH_FILTER_GENERIC:
1661 		if (filter_op != RTE_ETH_FILTER_GET) {
1662 			rc = EINVAL;
1663 		} else {
1664 			*(const void **)arg = &sfc_flow_ops;
1665 			rc = 0;
1666 		}
1667 		break;
1668 	default:
1669 		sfc_err(sa, "Unknown filter type %u", filter_type);
1670 		break;
1671 	}
1672 
1673 	sfc_log_init(sa, "exit: %d", -rc);
1674 	SFC_ASSERT(rc >= 0);
1675 	return -rc;
1676 }
1677 
1678 static int
1679 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1680 {
1681 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1682 
1683 	/*
1684 	 * If Rx datapath does not provide callback to check mempool,
1685 	 * all pools are supported.
1686 	 */
1687 	if (sap->dp_rx->pool_ops_supported == NULL)
1688 		return 1;
1689 
1690 	return sap->dp_rx->pool_ops_supported(pool);
1691 }
1692 
1693 static const struct eth_dev_ops sfc_eth_dev_ops = {
1694 	.dev_configure			= sfc_dev_configure,
1695 	.dev_start			= sfc_dev_start,
1696 	.dev_stop			= sfc_dev_stop,
1697 	.dev_set_link_up		= sfc_dev_set_link_up,
1698 	.dev_set_link_down		= sfc_dev_set_link_down,
1699 	.dev_close			= sfc_dev_close,
1700 	.promiscuous_enable		= sfc_dev_promisc_enable,
1701 	.promiscuous_disable		= sfc_dev_promisc_disable,
1702 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1703 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1704 	.link_update			= sfc_dev_link_update,
1705 	.stats_get			= sfc_stats_get,
1706 	.stats_reset			= sfc_stats_reset,
1707 	.xstats_get			= sfc_xstats_get,
1708 	.xstats_reset			= sfc_stats_reset,
1709 	.xstats_get_names		= sfc_xstats_get_names,
1710 	.dev_infos_get			= sfc_dev_infos_get,
1711 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1712 	.mtu_set			= sfc_dev_set_mtu,
1713 	.rx_queue_start			= sfc_rx_queue_start,
1714 	.rx_queue_stop			= sfc_rx_queue_stop,
1715 	.tx_queue_start			= sfc_tx_queue_start,
1716 	.tx_queue_stop			= sfc_tx_queue_stop,
1717 	.rx_queue_setup			= sfc_rx_queue_setup,
1718 	.rx_queue_release		= sfc_rx_queue_release,
1719 	.rx_queue_count			= sfc_rx_queue_count,
1720 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1721 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1722 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1723 	.tx_queue_setup			= sfc_tx_queue_setup,
1724 	.tx_queue_release		= sfc_tx_queue_release,
1725 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1726 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1727 	.mac_addr_set			= sfc_mac_addr_set,
1728 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1729 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1730 	.reta_update			= sfc_dev_rss_reta_update,
1731 	.reta_query			= sfc_dev_rss_reta_query,
1732 	.rss_hash_update		= sfc_dev_rss_hash_update,
1733 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1734 	.filter_ctrl			= sfc_dev_filter_ctrl,
1735 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1736 	.rxq_info_get			= sfc_rx_queue_info_get,
1737 	.txq_info_get			= sfc_tx_queue_info_get,
1738 	.fw_version_get			= sfc_fw_version_get,
1739 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1740 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1741 	.pool_ops_supported		= sfc_pool_ops_supported,
1742 };
1743 
1744 /**
1745  * Duplicate a string in potentially shared memory required for
1746  * multi-process support.
1747  *
1748  * strdup() allocates from process-local heap/memory.
1749  */
1750 static char *
1751 sfc_strdup(const char *str)
1752 {
1753 	size_t size;
1754 	char *copy;
1755 
1756 	if (str == NULL)
1757 		return NULL;
1758 
1759 	size = strlen(str) + 1;
1760 	copy = rte_malloc(__func__, size, 0);
1761 	if (copy != NULL)
1762 		rte_memcpy(copy, str, size);
1763 
1764 	return copy;
1765 }
1766 
1767 static int
1768 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1769 {
1770 	struct sfc_adapter *sa = dev->data->dev_private;
1771 	const struct sfc_dp_rx *dp_rx;
1772 	const struct sfc_dp_tx *dp_tx;
1773 	const efx_nic_cfg_t *encp;
1774 	unsigned int avail_caps = 0;
1775 	const char *rx_name = NULL;
1776 	const char *tx_name = NULL;
1777 	int rc;
1778 
1779 	switch (sa->family) {
1780 	case EFX_FAMILY_HUNTINGTON:
1781 	case EFX_FAMILY_MEDFORD:
1782 	case EFX_FAMILY_MEDFORD2:
1783 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1784 		break;
1785 	default:
1786 		break;
1787 	}
1788 
1789 	encp = efx_nic_cfg_get(sa->nic);
1790 	if (encp->enc_rx_es_super_buffer_supported)
1791 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1792 
1793 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1794 				sfc_kvarg_string_handler, &rx_name);
1795 	if (rc != 0)
1796 		goto fail_kvarg_rx_datapath;
1797 
1798 	if (rx_name != NULL) {
1799 		dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1800 		if (dp_rx == NULL) {
1801 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1802 			rc = ENOENT;
1803 			goto fail_dp_rx;
1804 		}
1805 		if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1806 			sfc_err(sa,
1807 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1808 				rx_name);
1809 			rc = EINVAL;
1810 			goto fail_dp_rx_caps;
1811 		}
1812 	} else {
1813 		dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1814 		if (dp_rx == NULL) {
1815 			sfc_err(sa, "Rx datapath by caps %#x not found",
1816 				avail_caps);
1817 			rc = ENOENT;
1818 			goto fail_dp_rx;
1819 		}
1820 	}
1821 
1822 	sa->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1823 	if (sa->dp_rx_name == NULL) {
1824 		rc = ENOMEM;
1825 		goto fail_dp_rx_name;
1826 	}
1827 
1828 	sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1829 
1830 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1831 				sfc_kvarg_string_handler, &tx_name);
1832 	if (rc != 0)
1833 		goto fail_kvarg_tx_datapath;
1834 
1835 	if (tx_name != NULL) {
1836 		dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1837 		if (dp_tx == NULL) {
1838 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1839 			rc = ENOENT;
1840 			goto fail_dp_tx;
1841 		}
1842 		if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1843 			sfc_err(sa,
1844 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1845 				tx_name);
1846 			rc = EINVAL;
1847 			goto fail_dp_tx_caps;
1848 		}
1849 	} else {
1850 		dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1851 		if (dp_tx == NULL) {
1852 			sfc_err(sa, "Tx datapath by caps %#x not found",
1853 				avail_caps);
1854 			rc = ENOENT;
1855 			goto fail_dp_tx;
1856 		}
1857 	}
1858 
1859 	sa->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1860 	if (sa->dp_tx_name == NULL) {
1861 		rc = ENOMEM;
1862 		goto fail_dp_tx_name;
1863 	}
1864 
1865 	sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1866 
1867 	sa->priv.dp_rx = dp_rx;
1868 	sa->priv.dp_tx = dp_tx;
1869 
1870 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1871 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1872 
1873 	dev->dev_ops = &sfc_eth_dev_ops;
1874 
1875 	return 0;
1876 
1877 fail_dp_tx_name:
1878 fail_dp_tx_caps:
1879 fail_dp_tx:
1880 fail_kvarg_tx_datapath:
1881 	rte_free(sa->dp_rx_name);
1882 	sa->dp_rx_name = NULL;
1883 
1884 fail_dp_rx_name:
1885 fail_dp_rx_caps:
1886 fail_dp_rx:
1887 fail_kvarg_rx_datapath:
1888 	return rc;
1889 }
1890 
1891 static void
1892 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1893 {
1894 	struct sfc_adapter *sa = dev->data->dev_private;
1895 
1896 	dev->dev_ops = NULL;
1897 	dev->rx_pkt_burst = NULL;
1898 	dev->tx_pkt_burst = NULL;
1899 
1900 	rte_free(sa->dp_tx_name);
1901 	sa->dp_tx_name = NULL;
1902 	sa->priv.dp_tx = NULL;
1903 
1904 	rte_free(sa->dp_rx_name);
1905 	sa->dp_rx_name = NULL;
1906 	sa->priv.dp_rx = NULL;
1907 }
1908 
1909 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1910 	.rx_queue_count			= sfc_rx_queue_count,
1911 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1912 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1913 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1914 	.reta_query			= sfc_dev_rss_reta_query,
1915 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1916 	.rxq_info_get			= sfc_rx_queue_info_get,
1917 	.txq_info_get			= sfc_tx_queue_info_get,
1918 };
1919 
1920 static int
1921 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
1922 {
1923 	/*
1924 	 * Device private data has really many process-local pointers.
1925 	 * Below code should be extremely careful to use data located
1926 	 * in shared memory only.
1927 	 */
1928 	struct sfc_adapter *sa = dev->data->dev_private;
1929 	struct sfc_adapter_priv *sap;
1930 	const struct sfc_dp_rx *dp_rx;
1931 	const struct sfc_dp_tx *dp_tx;
1932 	int rc;
1933 
1934 	/*
1935 	 * Allocate process private data from heap, since it should not
1936 	 * be located in shared memory allocated using rte_malloc() API.
1937 	 */
1938 	sap = calloc(1, sizeof(*sap));
1939 	if (sap == NULL) {
1940 		rc = ENOMEM;
1941 		goto fail_alloc_priv;
1942 	}
1943 
1944 	sap->logtype_main = logtype_main;
1945 
1946 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1947 	if (dp_rx == NULL) {
1948 		SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1949 			"cannot find %s Rx datapath", sa->dp_rx_name);
1950 		rc = ENOENT;
1951 		goto fail_dp_rx;
1952 	}
1953 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1954 		SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1955 			"%s Rx datapath does not support multi-process",
1956 			sa->dp_rx_name);
1957 		rc = EINVAL;
1958 		goto fail_dp_rx_multi_process;
1959 	}
1960 
1961 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1962 	if (dp_tx == NULL) {
1963 		SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1964 			"cannot find %s Tx datapath", sa->dp_tx_name);
1965 		rc = ENOENT;
1966 		goto fail_dp_tx;
1967 	}
1968 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1969 		SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1970 			"%s Tx datapath does not support multi-process",
1971 			sa->dp_tx_name);
1972 		rc = EINVAL;
1973 		goto fail_dp_tx_multi_process;
1974 	}
1975 
1976 	sap->dp_rx = dp_rx;
1977 	sap->dp_tx = dp_tx;
1978 
1979 	dev->process_private = sap;
1980 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1981 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1982 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1983 
1984 	return 0;
1985 
1986 fail_dp_tx_multi_process:
1987 fail_dp_tx:
1988 fail_dp_rx_multi_process:
1989 fail_dp_rx:
1990 	free(sap);
1991 
1992 fail_alloc_priv:
1993 	return rc;
1994 }
1995 
1996 static void
1997 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1998 {
1999 	free(dev->process_private);
2000 	dev->process_private = NULL;
2001 	dev->dev_ops = NULL;
2002 	dev->tx_pkt_burst = NULL;
2003 	dev->rx_pkt_burst = NULL;
2004 }
2005 
2006 static void
2007 sfc_register_dp(void)
2008 {
2009 	/* Register once */
2010 	if (TAILQ_EMPTY(&sfc_dp_head)) {
2011 		/* Prefer EF10 datapath */
2012 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2013 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2014 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2015 
2016 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2017 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2018 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2019 	}
2020 }
2021 
2022 static int
2023 sfc_eth_dev_init(struct rte_eth_dev *dev)
2024 {
2025 	struct sfc_adapter *sa = dev->data->dev_private;
2026 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2027 	uint32_t logtype_main;
2028 	int rc;
2029 	const efx_nic_cfg_t *encp;
2030 	const struct ether_addr *from;
2031 
2032 	sfc_register_dp();
2033 
2034 	logtype_main = sfc_register_logtype(&pci_dev->addr,
2035 					    SFC_LOGTYPE_MAIN_STR,
2036 					    RTE_LOG_NOTICE);
2037 
2038 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2039 		return -sfc_eth_dev_secondary_init(dev, logtype_main);
2040 
2041 	/*
2042 	 * sfc_adapter is a mixture of shared and process private data.
2043 	 * During transition period use it in both kinds. When the
2044 	 * driver becomes ready to separate it, sfc_adapter will become
2045 	 * primary process private only.
2046 	 */
2047 	dev->process_private = sa;
2048 
2049 	/* Required for logging */
2050 	sa->pci_addr = pci_dev->addr;
2051 	sa->port_id = dev->data->port_id;
2052 	sa->priv.logtype_main = logtype_main;
2053 
2054 	sa->eth_dev = dev;
2055 
2056 	/* Copy PCI device info to the dev->data */
2057 	rte_eth_copy_pci_info(dev, pci_dev);
2058 
2059 	rc = sfc_kvargs_parse(sa);
2060 	if (rc != 0)
2061 		goto fail_kvargs_parse;
2062 
2063 	sfc_log_init(sa, "entry");
2064 
2065 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
2066 	if (dev->data->mac_addrs == NULL) {
2067 		rc = ENOMEM;
2068 		goto fail_mac_addrs;
2069 	}
2070 
2071 	sfc_adapter_lock_init(sa);
2072 	sfc_adapter_lock(sa);
2073 
2074 	sfc_log_init(sa, "probing");
2075 	rc = sfc_probe(sa);
2076 	if (rc != 0)
2077 		goto fail_probe;
2078 
2079 	sfc_log_init(sa, "set device ops");
2080 	rc = sfc_eth_dev_set_ops(dev);
2081 	if (rc != 0)
2082 		goto fail_set_ops;
2083 
2084 	sfc_log_init(sa, "attaching");
2085 	rc = sfc_attach(sa);
2086 	if (rc != 0)
2087 		goto fail_attach;
2088 
2089 	encp = efx_nic_cfg_get(sa->nic);
2090 
2091 	/*
2092 	 * The arguments are really reverse order in comparison to
2093 	 * Linux kernel. Copy from NIC config to Ethernet device data.
2094 	 */
2095 	from = (const struct ether_addr *)(encp->enc_mac_addr);
2096 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
2097 
2098 	sfc_adapter_unlock(sa);
2099 
2100 	sfc_log_init(sa, "done");
2101 	return 0;
2102 
2103 fail_attach:
2104 	sfc_eth_dev_clear_ops(dev);
2105 
2106 fail_set_ops:
2107 	sfc_unprobe(sa);
2108 
2109 fail_probe:
2110 	sfc_adapter_unlock(sa);
2111 	sfc_adapter_lock_fini(sa);
2112 	rte_free(dev->data->mac_addrs);
2113 	dev->data->mac_addrs = NULL;
2114 
2115 fail_mac_addrs:
2116 	sfc_kvargs_cleanup(sa);
2117 
2118 fail_kvargs_parse:
2119 	sfc_log_init(sa, "failed %d", rc);
2120 	dev->process_private = NULL;
2121 	SFC_ASSERT(rc > 0);
2122 	return -rc;
2123 }
2124 
2125 static int
2126 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2127 {
2128 	struct sfc_adapter *sa;
2129 
2130 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2131 		sfc_eth_dev_secondary_clear_ops(dev);
2132 		return 0;
2133 	}
2134 
2135 	sa = dev->data->dev_private;
2136 	sfc_log_init(sa, "entry");
2137 
2138 	sfc_adapter_lock(sa);
2139 
2140 	sfc_eth_dev_clear_ops(dev);
2141 
2142 	sfc_detach(sa);
2143 	sfc_unprobe(sa);
2144 
2145 	sfc_kvargs_cleanup(sa);
2146 
2147 	sfc_adapter_unlock(sa);
2148 	sfc_adapter_lock_fini(sa);
2149 
2150 	sfc_log_init(sa, "done");
2151 
2152 	/* Required for logging, so cleanup last */
2153 	sa->eth_dev = NULL;
2154 	return 0;
2155 }
2156 
2157 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2158 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2159 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2160 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2161 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2162 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2163 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2164 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2165 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2166 	{ .vendor_id = 0 /* sentinel */ }
2167 };
2168 
2169 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2170 	struct rte_pci_device *pci_dev)
2171 {
2172 	return rte_eth_dev_pci_generic_probe(pci_dev,
2173 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2174 }
2175 
2176 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2177 {
2178 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2179 }
2180 
2181 static struct rte_pci_driver sfc_efx_pmd = {
2182 	.id_table = pci_id_sfc_efx_map,
2183 	.drv_flags =
2184 		RTE_PCI_DRV_INTR_LSC |
2185 		RTE_PCI_DRV_NEED_MAPPING,
2186 	.probe = sfc_eth_dev_pci_probe,
2187 	.remove = sfc_eth_dev_pci_remove,
2188 };
2189 
2190 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2191 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2192 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2193 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2194 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2195 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2196 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2197 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2198 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2199 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2200 
2201 RTE_INIT(sfc_driver_register_logtype)
2202 {
2203 	int ret;
2204 
2205 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2206 						   RTE_LOG_NOTICE);
2207 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2208 }
2209