xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 113a14a63e64ee042898a10d2262f1fa8beb198f)
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_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
406 	struct sfc_adapter *sa = dev->data->dev_private;
407 	int rc;
408 
409 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
410 		     rx_queue_id, nb_rx_desc, socket_id);
411 
412 	sfc_adapter_lock(sa);
413 
414 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
415 			  rx_conf, mb_pool);
416 	if (rc != 0)
417 		goto fail_rx_qinit;
418 
419 	dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
420 
421 	sfc_adapter_unlock(sa);
422 
423 	return 0;
424 
425 fail_rx_qinit:
426 	sfc_adapter_unlock(sa);
427 	SFC_ASSERT(rc > 0);
428 	return -rc;
429 }
430 
431 static void
432 sfc_rx_queue_release(void *queue)
433 {
434 	struct sfc_dp_rxq *dp_rxq = queue;
435 	struct sfc_rxq *rxq;
436 	struct sfc_adapter *sa;
437 	unsigned int sw_index;
438 
439 	if (dp_rxq == NULL)
440 		return;
441 
442 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
443 	sa = rxq->evq->sa;
444 	sfc_adapter_lock(sa);
445 
446 	sw_index = dp_rxq->dpq.queue_id;
447 
448 	sfc_log_init(sa, "RxQ=%u", sw_index);
449 
450 	sfc_rx_qfini(sa, sw_index);
451 
452 	sfc_adapter_unlock(sa);
453 }
454 
455 static int
456 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
457 		   uint16_t nb_tx_desc, unsigned int socket_id,
458 		   const struct rte_eth_txconf *tx_conf)
459 {
460 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
461 	struct sfc_adapter *sa = dev->data->dev_private;
462 	int rc;
463 
464 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
465 		     tx_queue_id, nb_tx_desc, socket_id);
466 
467 	sfc_adapter_lock(sa);
468 
469 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
470 	if (rc != 0)
471 		goto fail_tx_qinit;
472 
473 	dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
474 
475 	sfc_adapter_unlock(sa);
476 	return 0;
477 
478 fail_tx_qinit:
479 	sfc_adapter_unlock(sa);
480 	SFC_ASSERT(rc > 0);
481 	return -rc;
482 }
483 
484 static void
485 sfc_tx_queue_release(void *queue)
486 {
487 	struct sfc_dp_txq *dp_txq = queue;
488 	struct sfc_txq *txq;
489 	unsigned int sw_index;
490 	struct sfc_adapter *sa;
491 
492 	if (dp_txq == NULL)
493 		return;
494 
495 	txq = sfc_txq_by_dp_txq(dp_txq);
496 	sw_index = dp_txq->dpq.queue_id;
497 
498 	SFC_ASSERT(txq->evq != NULL);
499 	sa = txq->evq->sa;
500 
501 	sfc_log_init(sa, "TxQ = %u", sw_index);
502 
503 	sfc_adapter_lock(sa);
504 
505 	sfc_tx_qfini(sa, sw_index);
506 
507 	sfc_adapter_unlock(sa);
508 }
509 
510 /*
511  * Some statistics are computed as A - B where A and B each increase
512  * monotonically with some hardware counter(s) and the counters are read
513  * asynchronously.
514  *
515  * If packet X is counted in A, but not counted in B yet, computed value is
516  * greater than real.
517  *
518  * If packet X is not counted in A at the moment of reading the counter,
519  * but counted in B at the moment of reading the counter, computed value
520  * is less than real.
521  *
522  * However, counter which grows backward is worse evil than slightly wrong
523  * value. So, let's try to guarantee that it never happens except may be
524  * the case when the MAC stats are zeroed as a result of a NIC reset.
525  */
526 static void
527 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
528 {
529 	if ((int64_t)(newval - *stat) > 0 || newval == 0)
530 		*stat = newval;
531 }
532 
533 static int
534 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
535 {
536 	struct sfc_adapter *sa = dev->data->dev_private;
537 	struct sfc_port *port = &sa->port;
538 	uint64_t *mac_stats;
539 	int ret;
540 
541 	rte_spinlock_lock(&port->mac_stats_lock);
542 
543 	ret = sfc_port_update_mac_stats(sa);
544 	if (ret != 0)
545 		goto unlock;
546 
547 	mac_stats = port->mac_stats_buf;
548 
549 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
550 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
551 		stats->ipackets =
552 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
553 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
554 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
555 		stats->opackets =
556 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
557 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
558 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
559 		stats->ibytes =
560 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
561 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
562 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
563 		stats->obytes =
564 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
565 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
566 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
567 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
568 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
569 	} else {
570 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
571 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
572 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
573 		/*
574 		 * Take into account stats which are whenever supported
575 		 * on EF10. If some stat is not supported by current
576 		 * firmware variant or HW revision, it is guaranteed
577 		 * to be zero in mac_stats.
578 		 */
579 		stats->imissed =
580 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
581 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
582 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
583 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
584 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
585 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
586 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
587 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
588 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
589 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
590 		stats->ierrors =
591 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
592 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
593 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
594 		/* no oerrors counters supported on EF10 */
595 
596 		/* Exclude missed, errors and pauses from Rx packets */
597 		sfc_update_diff_stat(&port->ipackets,
598 			mac_stats[EFX_MAC_RX_PKTS] -
599 			mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
600 			stats->imissed - stats->ierrors);
601 		stats->ipackets = port->ipackets;
602 	}
603 
604 unlock:
605 	rte_spinlock_unlock(&port->mac_stats_lock);
606 	SFC_ASSERT(ret >= 0);
607 	return -ret;
608 }
609 
610 static void
611 sfc_stats_reset(struct rte_eth_dev *dev)
612 {
613 	struct sfc_adapter *sa = dev->data->dev_private;
614 	struct sfc_port *port = &sa->port;
615 	int rc;
616 
617 	if (sa->state != SFC_ADAPTER_STARTED) {
618 		/*
619 		 * The operation cannot be done if port is not started; it
620 		 * will be scheduled to be done during the next port start
621 		 */
622 		port->mac_stats_reset_pending = B_TRUE;
623 		return;
624 	}
625 
626 	rc = sfc_port_reset_mac_stats(sa);
627 	if (rc != 0)
628 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
629 }
630 
631 static int
632 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
633 	       unsigned int xstats_count)
634 {
635 	struct sfc_adapter *sa = dev->data->dev_private;
636 	struct sfc_port *port = &sa->port;
637 	uint64_t *mac_stats;
638 	int rc;
639 	unsigned int i;
640 	int nstats = 0;
641 
642 	rte_spinlock_lock(&port->mac_stats_lock);
643 
644 	rc = sfc_port_update_mac_stats(sa);
645 	if (rc != 0) {
646 		SFC_ASSERT(rc > 0);
647 		nstats = -rc;
648 		goto unlock;
649 	}
650 
651 	mac_stats = port->mac_stats_buf;
652 
653 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
654 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
655 			if (xstats != NULL && nstats < (int)xstats_count) {
656 				xstats[nstats].id = nstats;
657 				xstats[nstats].value = mac_stats[i];
658 			}
659 			nstats++;
660 		}
661 	}
662 
663 unlock:
664 	rte_spinlock_unlock(&port->mac_stats_lock);
665 
666 	return nstats;
667 }
668 
669 static int
670 sfc_xstats_get_names(struct rte_eth_dev *dev,
671 		     struct rte_eth_xstat_name *xstats_names,
672 		     unsigned int xstats_count)
673 {
674 	struct sfc_adapter *sa = dev->data->dev_private;
675 	struct sfc_port *port = &sa->port;
676 	unsigned int i;
677 	unsigned int nstats = 0;
678 
679 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
680 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
681 			if (xstats_names != NULL && nstats < xstats_count)
682 				strlcpy(xstats_names[nstats].name,
683 					efx_mac_stat_name(sa->nic, i),
684 					sizeof(xstats_names[0].name));
685 			nstats++;
686 		}
687 	}
688 
689 	return nstats;
690 }
691 
692 static int
693 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
694 		     uint64_t *values, unsigned int n)
695 {
696 	struct sfc_adapter *sa = dev->data->dev_private;
697 	struct sfc_port *port = &sa->port;
698 	uint64_t *mac_stats;
699 	unsigned int nb_supported = 0;
700 	unsigned int nb_written = 0;
701 	unsigned int i;
702 	int ret;
703 	int rc;
704 
705 	if (unlikely(values == NULL) ||
706 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
707 		return port->mac_stats_nb_supported;
708 
709 	rte_spinlock_lock(&port->mac_stats_lock);
710 
711 	rc = sfc_port_update_mac_stats(sa);
712 	if (rc != 0) {
713 		SFC_ASSERT(rc > 0);
714 		ret = -rc;
715 		goto unlock;
716 	}
717 
718 	mac_stats = port->mac_stats_buf;
719 
720 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
721 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
722 			continue;
723 
724 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
725 			values[nb_written++] = mac_stats[i];
726 
727 		++nb_supported;
728 	}
729 
730 	ret = nb_written;
731 
732 unlock:
733 	rte_spinlock_unlock(&port->mac_stats_lock);
734 
735 	return ret;
736 }
737 
738 static int
739 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
740 			   struct rte_eth_xstat_name *xstats_names,
741 			   const uint64_t *ids, unsigned int size)
742 {
743 	struct sfc_adapter *sa = dev->data->dev_private;
744 	struct sfc_port *port = &sa->port;
745 	unsigned int nb_supported = 0;
746 	unsigned int nb_written = 0;
747 	unsigned int i;
748 
749 	if (unlikely(xstats_names == NULL) ||
750 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
751 		return port->mac_stats_nb_supported;
752 
753 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
754 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
755 			continue;
756 
757 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
758 			char *name = xstats_names[nb_written++].name;
759 
760 			strlcpy(name, efx_mac_stat_name(sa->nic, i),
761 				sizeof(xstats_names[0].name));
762 		}
763 
764 		++nb_supported;
765 	}
766 
767 	return nb_written;
768 }
769 
770 static int
771 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
772 {
773 	struct sfc_adapter *sa = dev->data->dev_private;
774 	unsigned int wanted_fc, link_fc;
775 
776 	memset(fc_conf, 0, sizeof(*fc_conf));
777 
778 	sfc_adapter_lock(sa);
779 
780 	if (sa->state == SFC_ADAPTER_STARTED)
781 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
782 	else
783 		link_fc = sa->port.flow_ctrl;
784 
785 	switch (link_fc) {
786 	case 0:
787 		fc_conf->mode = RTE_FC_NONE;
788 		break;
789 	case EFX_FCNTL_RESPOND:
790 		fc_conf->mode = RTE_FC_RX_PAUSE;
791 		break;
792 	case EFX_FCNTL_GENERATE:
793 		fc_conf->mode = RTE_FC_TX_PAUSE;
794 		break;
795 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
796 		fc_conf->mode = RTE_FC_FULL;
797 		break;
798 	default:
799 		sfc_err(sa, "%s: unexpected flow control value %#x",
800 			__func__, link_fc);
801 	}
802 
803 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
804 
805 	sfc_adapter_unlock(sa);
806 
807 	return 0;
808 }
809 
810 static int
811 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
812 {
813 	struct sfc_adapter *sa = dev->data->dev_private;
814 	struct sfc_port *port = &sa->port;
815 	unsigned int fcntl;
816 	int rc;
817 
818 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
819 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
820 	    fc_conf->mac_ctrl_frame_fwd != 0) {
821 		sfc_err(sa, "unsupported flow control settings specified");
822 		rc = EINVAL;
823 		goto fail_inval;
824 	}
825 
826 	switch (fc_conf->mode) {
827 	case RTE_FC_NONE:
828 		fcntl = 0;
829 		break;
830 	case RTE_FC_RX_PAUSE:
831 		fcntl = EFX_FCNTL_RESPOND;
832 		break;
833 	case RTE_FC_TX_PAUSE:
834 		fcntl = EFX_FCNTL_GENERATE;
835 		break;
836 	case RTE_FC_FULL:
837 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
838 		break;
839 	default:
840 		rc = EINVAL;
841 		goto fail_inval;
842 	}
843 
844 	sfc_adapter_lock(sa);
845 
846 	if (sa->state == SFC_ADAPTER_STARTED) {
847 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
848 		if (rc != 0)
849 			goto fail_mac_fcntl_set;
850 	}
851 
852 	port->flow_ctrl = fcntl;
853 	port->flow_ctrl_autoneg = fc_conf->autoneg;
854 
855 	sfc_adapter_unlock(sa);
856 
857 	return 0;
858 
859 fail_mac_fcntl_set:
860 	sfc_adapter_unlock(sa);
861 fail_inval:
862 	SFC_ASSERT(rc > 0);
863 	return -rc;
864 }
865 
866 static int
867 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
868 {
869 	struct sfc_adapter *sa = dev->data->dev_private;
870 	size_t pdu = EFX_MAC_PDU(mtu);
871 	size_t old_pdu;
872 	int rc;
873 
874 	sfc_log_init(sa, "mtu=%u", mtu);
875 
876 	rc = EINVAL;
877 	if (pdu < EFX_MAC_PDU_MIN) {
878 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
879 			(unsigned int)mtu, (unsigned int)pdu,
880 			EFX_MAC_PDU_MIN);
881 		goto fail_inval;
882 	}
883 	if (pdu > EFX_MAC_PDU_MAX) {
884 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
885 			(unsigned int)mtu, (unsigned int)pdu,
886 			EFX_MAC_PDU_MAX);
887 		goto fail_inval;
888 	}
889 
890 	sfc_adapter_lock(sa);
891 
892 	if (pdu != sa->port.pdu) {
893 		if (sa->state == SFC_ADAPTER_STARTED) {
894 			sfc_stop(sa);
895 
896 			old_pdu = sa->port.pdu;
897 			sa->port.pdu = pdu;
898 			rc = sfc_start(sa);
899 			if (rc != 0)
900 				goto fail_start;
901 		} else {
902 			sa->port.pdu = pdu;
903 		}
904 	}
905 
906 	/*
907 	 * The driver does not use it, but other PMDs update jumbo frame
908 	 * flag and max_rx_pkt_len when MTU is set.
909 	 */
910 	if (mtu > ETHER_MAX_LEN) {
911 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
912 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
913 	}
914 
915 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
916 
917 	sfc_adapter_unlock(sa);
918 
919 	sfc_log_init(sa, "done");
920 	return 0;
921 
922 fail_start:
923 	sa->port.pdu = old_pdu;
924 	if (sfc_start(sa) != 0)
925 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
926 			"PDU max size - port is stopped",
927 			(unsigned int)pdu, (unsigned int)old_pdu);
928 	sfc_adapter_unlock(sa);
929 
930 fail_inval:
931 	sfc_log_init(sa, "failed %d", rc);
932 	SFC_ASSERT(rc > 0);
933 	return -rc;
934 }
935 static int
936 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
937 {
938 	struct sfc_adapter *sa = dev->data->dev_private;
939 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
940 	struct sfc_port *port = &sa->port;
941 	struct ether_addr *old_addr = &dev->data->mac_addrs[0];
942 	int rc = 0;
943 
944 	sfc_adapter_lock(sa);
945 
946 	/*
947 	 * Copy the address to the device private data so that
948 	 * it could be recalled in the case of adapter restart.
949 	 */
950 	ether_addr_copy(mac_addr, &port->default_mac_addr);
951 
952 	/*
953 	 * Neither of the two following checks can return
954 	 * an error. The new MAC address is preserved in
955 	 * the device private data and can be activated
956 	 * on the next port start if the user prevents
957 	 * isolated mode from being enabled.
958 	 */
959 	if (port->isolated) {
960 		sfc_warn(sa, "isolated mode is active on the port");
961 		sfc_warn(sa, "will not set MAC address");
962 		goto unlock;
963 	}
964 
965 	if (sa->state != SFC_ADAPTER_STARTED) {
966 		sfc_notice(sa, "the port is not started");
967 		sfc_notice(sa, "the new MAC address will be set on port start");
968 
969 		goto unlock;
970 	}
971 
972 	if (encp->enc_allow_set_mac_with_installed_filters) {
973 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
974 		if (rc != 0) {
975 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
976 			goto unlock;
977 		}
978 
979 		/*
980 		 * Changing the MAC address by means of MCDI request
981 		 * has no effect on received traffic, therefore
982 		 * we also need to update unicast filters
983 		 */
984 		rc = sfc_set_rx_mode(sa);
985 		if (rc != 0) {
986 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
987 			/* Rollback the old address */
988 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
989 			(void)sfc_set_rx_mode(sa);
990 		}
991 	} else {
992 		sfc_warn(sa, "cannot set MAC address with filters installed");
993 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
994 		sfc_warn(sa, "(some traffic may be dropped)");
995 
996 		/*
997 		 * Since setting MAC address with filters installed is not
998 		 * allowed on the adapter, the new MAC address will be set
999 		 * by means of adapter restart. sfc_start() shall retrieve
1000 		 * the new address from the device private data and set it.
1001 		 */
1002 		sfc_stop(sa);
1003 		rc = sfc_start(sa);
1004 		if (rc != 0)
1005 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1006 	}
1007 
1008 unlock:
1009 	if (rc != 0)
1010 		ether_addr_copy(old_addr, &port->default_mac_addr);
1011 
1012 	sfc_adapter_unlock(sa);
1013 
1014 	SFC_ASSERT(rc >= 0);
1015 	return -rc;
1016 }
1017 
1018 
1019 static int
1020 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1021 		     uint32_t nb_mc_addr)
1022 {
1023 	struct sfc_adapter *sa = dev->data->dev_private;
1024 	struct sfc_port *port = &sa->port;
1025 	uint8_t *mc_addrs = port->mcast_addrs;
1026 	int rc;
1027 	unsigned int i;
1028 
1029 	if (port->isolated) {
1030 		sfc_err(sa, "isolated mode is active on the port");
1031 		sfc_err(sa, "will not set multicast address list");
1032 		return -ENOTSUP;
1033 	}
1034 
1035 	if (mc_addrs == NULL)
1036 		return -ENOBUFS;
1037 
1038 	if (nb_mc_addr > port->max_mcast_addrs) {
1039 		sfc_err(sa, "too many multicast addresses: %u > %u",
1040 			 nb_mc_addr, port->max_mcast_addrs);
1041 		return -EINVAL;
1042 	}
1043 
1044 	for (i = 0; i < nb_mc_addr; ++i) {
1045 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1046 				 EFX_MAC_ADDR_LEN);
1047 		mc_addrs += EFX_MAC_ADDR_LEN;
1048 	}
1049 
1050 	port->nb_mcast_addrs = nb_mc_addr;
1051 
1052 	if (sa->state != SFC_ADAPTER_STARTED)
1053 		return 0;
1054 
1055 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1056 					port->nb_mcast_addrs);
1057 	if (rc != 0)
1058 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1059 
1060 	SFC_ASSERT(rc >= 0);
1061 	return -rc;
1062 }
1063 
1064 /*
1065  * The function is used by the secondary process as well. It must not
1066  * use any process-local pointers from the adapter data.
1067  */
1068 static void
1069 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1070 		      struct rte_eth_rxq_info *qinfo)
1071 {
1072 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1073 	struct sfc_adapter *sa = dev->data->dev_private;
1074 	struct sfc_rxq_info *rxq_info;
1075 
1076 	sfc_adapter_lock(sa);
1077 
1078 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1079 
1080 	rxq_info = &sas->rxq_info[rx_queue_id];
1081 
1082 	qinfo->mp = rxq_info->refill_mb_pool;
1083 	qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1084 	qinfo->conf.rx_drop_en = 1;
1085 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1086 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1087 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1088 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1089 		qinfo->scattered_rx = 1;
1090 	}
1091 	qinfo->nb_desc = rxq_info->entries;
1092 
1093 	sfc_adapter_unlock(sa);
1094 }
1095 
1096 /*
1097  * The function is used by the secondary process as well. It must not
1098  * use any process-local pointers from the adapter data.
1099  */
1100 static void
1101 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1102 		      struct rte_eth_txq_info *qinfo)
1103 {
1104 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1105 	struct sfc_adapter *sa = dev->data->dev_private;
1106 	struct sfc_txq_info *txq_info;
1107 
1108 	sfc_adapter_lock(sa);
1109 
1110 	SFC_ASSERT(tx_queue_id < sas->txq_count);
1111 
1112 	txq_info = &sas->txq_info[tx_queue_id];
1113 
1114 	memset(qinfo, 0, sizeof(*qinfo));
1115 
1116 	qinfo->conf.offloads = txq_info->offloads;
1117 	qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1118 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1119 	qinfo->nb_desc = txq_info->entries;
1120 
1121 	sfc_adapter_unlock(sa);
1122 }
1123 
1124 /*
1125  * The function is used by the secondary process as well. It must not
1126  * use any process-local pointers from the adapter data.
1127  */
1128 static uint32_t
1129 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1130 {
1131 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1132 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1133 	struct sfc_rxq_info *rxq_info;
1134 
1135 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1136 	rxq_info = &sas->rxq_info[rx_queue_id];
1137 
1138 	if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1139 		return 0;
1140 
1141 	return sap->dp_rx->qdesc_npending(rxq_info->dp);
1142 }
1143 
1144 /*
1145  * The function is used by the secondary process as well. It must not
1146  * use any process-local pointers from the adapter data.
1147  */
1148 static int
1149 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1150 {
1151 	struct sfc_dp_rxq *dp_rxq = queue;
1152 	const struct sfc_dp_rx *dp_rx;
1153 
1154 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1155 
1156 	return offset < dp_rx->qdesc_npending(dp_rxq);
1157 }
1158 
1159 /*
1160  * The function is used by the secondary process as well. It must not
1161  * use any process-local pointers from the adapter data.
1162  */
1163 static int
1164 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1165 {
1166 	struct sfc_dp_rxq *dp_rxq = queue;
1167 	const struct sfc_dp_rx *dp_rx;
1168 
1169 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1170 
1171 	return dp_rx->qdesc_status(dp_rxq, offset);
1172 }
1173 
1174 /*
1175  * The function is used by the secondary process as well. It must not
1176  * use any process-local pointers from the adapter data.
1177  */
1178 static int
1179 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1180 {
1181 	struct sfc_dp_txq *dp_txq = queue;
1182 	const struct sfc_dp_tx *dp_tx;
1183 
1184 	dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1185 
1186 	return dp_tx->qdesc_status(dp_txq, offset);
1187 }
1188 
1189 static int
1190 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1191 {
1192 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1193 	struct sfc_adapter *sa = dev->data->dev_private;
1194 	int rc;
1195 
1196 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1197 
1198 	sfc_adapter_lock(sa);
1199 
1200 	rc = EINVAL;
1201 	if (sa->state != SFC_ADAPTER_STARTED)
1202 		goto fail_not_started;
1203 
1204 	if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1205 		goto fail_not_setup;
1206 
1207 	rc = sfc_rx_qstart(sa, rx_queue_id);
1208 	if (rc != 0)
1209 		goto fail_rx_qstart;
1210 
1211 	sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1212 
1213 	sfc_adapter_unlock(sa);
1214 
1215 	return 0;
1216 
1217 fail_rx_qstart:
1218 fail_not_setup:
1219 fail_not_started:
1220 	sfc_adapter_unlock(sa);
1221 	SFC_ASSERT(rc > 0);
1222 	return -rc;
1223 }
1224 
1225 static int
1226 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1227 {
1228 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1229 	struct sfc_adapter *sa = dev->data->dev_private;
1230 
1231 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1232 
1233 	sfc_adapter_lock(sa);
1234 	sfc_rx_qstop(sa, rx_queue_id);
1235 
1236 	sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1237 
1238 	sfc_adapter_unlock(sa);
1239 
1240 	return 0;
1241 }
1242 
1243 static int
1244 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1245 {
1246 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1247 	struct sfc_adapter *sa = dev->data->dev_private;
1248 	int rc;
1249 
1250 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1251 
1252 	sfc_adapter_lock(sa);
1253 
1254 	rc = EINVAL;
1255 	if (sa->state != SFC_ADAPTER_STARTED)
1256 		goto fail_not_started;
1257 
1258 	if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1259 		goto fail_not_setup;
1260 
1261 	rc = sfc_tx_qstart(sa, tx_queue_id);
1262 	if (rc != 0)
1263 		goto fail_tx_qstart;
1264 
1265 	sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1266 
1267 	sfc_adapter_unlock(sa);
1268 	return 0;
1269 
1270 fail_tx_qstart:
1271 
1272 fail_not_setup:
1273 fail_not_started:
1274 	sfc_adapter_unlock(sa);
1275 	SFC_ASSERT(rc > 0);
1276 	return -rc;
1277 }
1278 
1279 static int
1280 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1281 {
1282 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1283 	struct sfc_adapter *sa = dev->data->dev_private;
1284 
1285 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1286 
1287 	sfc_adapter_lock(sa);
1288 
1289 	sfc_tx_qstop(sa, tx_queue_id);
1290 
1291 	sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1292 
1293 	sfc_adapter_unlock(sa);
1294 	return 0;
1295 }
1296 
1297 static efx_tunnel_protocol_t
1298 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1299 {
1300 	switch (rte_type) {
1301 	case RTE_TUNNEL_TYPE_VXLAN:
1302 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1303 	case RTE_TUNNEL_TYPE_GENEVE:
1304 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1305 	default:
1306 		return EFX_TUNNEL_NPROTOS;
1307 	}
1308 }
1309 
1310 enum sfc_udp_tunnel_op_e {
1311 	SFC_UDP_TUNNEL_ADD_PORT,
1312 	SFC_UDP_TUNNEL_DEL_PORT,
1313 };
1314 
1315 static int
1316 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1317 		      struct rte_eth_udp_tunnel *tunnel_udp,
1318 		      enum sfc_udp_tunnel_op_e op)
1319 {
1320 	struct sfc_adapter *sa = dev->data->dev_private;
1321 	efx_tunnel_protocol_t tunnel_proto;
1322 	int rc;
1323 
1324 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1325 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1326 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1327 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1328 
1329 	tunnel_proto =
1330 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1331 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1332 		rc = ENOTSUP;
1333 		goto fail_bad_proto;
1334 	}
1335 
1336 	sfc_adapter_lock(sa);
1337 
1338 	switch (op) {
1339 	case SFC_UDP_TUNNEL_ADD_PORT:
1340 		rc = efx_tunnel_config_udp_add(sa->nic,
1341 					       tunnel_udp->udp_port,
1342 					       tunnel_proto);
1343 		break;
1344 	case SFC_UDP_TUNNEL_DEL_PORT:
1345 		rc = efx_tunnel_config_udp_remove(sa->nic,
1346 						  tunnel_udp->udp_port,
1347 						  tunnel_proto);
1348 		break;
1349 	default:
1350 		rc = EINVAL;
1351 		goto fail_bad_op;
1352 	}
1353 
1354 	if (rc != 0)
1355 		goto fail_op;
1356 
1357 	if (sa->state == SFC_ADAPTER_STARTED) {
1358 		rc = efx_tunnel_reconfigure(sa->nic);
1359 		if (rc == EAGAIN) {
1360 			/*
1361 			 * Configuration is accepted by FW and MC reboot
1362 			 * is initiated to apply the changes. MC reboot
1363 			 * will be handled in a usual way (MC reboot
1364 			 * event on management event queue and adapter
1365 			 * restart).
1366 			 */
1367 			rc = 0;
1368 		} else if (rc != 0) {
1369 			goto fail_reconfigure;
1370 		}
1371 	}
1372 
1373 	sfc_adapter_unlock(sa);
1374 	return 0;
1375 
1376 fail_reconfigure:
1377 	/* Remove/restore entry since the change makes the trouble */
1378 	switch (op) {
1379 	case SFC_UDP_TUNNEL_ADD_PORT:
1380 		(void)efx_tunnel_config_udp_remove(sa->nic,
1381 						   tunnel_udp->udp_port,
1382 						   tunnel_proto);
1383 		break;
1384 	case SFC_UDP_TUNNEL_DEL_PORT:
1385 		(void)efx_tunnel_config_udp_add(sa->nic,
1386 						tunnel_udp->udp_port,
1387 						tunnel_proto);
1388 		break;
1389 	}
1390 
1391 fail_op:
1392 fail_bad_op:
1393 	sfc_adapter_unlock(sa);
1394 
1395 fail_bad_proto:
1396 	SFC_ASSERT(rc > 0);
1397 	return -rc;
1398 }
1399 
1400 static int
1401 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1402 			    struct rte_eth_udp_tunnel *tunnel_udp)
1403 {
1404 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1405 }
1406 
1407 static int
1408 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1409 			    struct rte_eth_udp_tunnel *tunnel_udp)
1410 {
1411 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1412 }
1413 
1414 /*
1415  * The function is used by the secondary process as well. It must not
1416  * use any process-local pointers from the adapter data.
1417  */
1418 static int
1419 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1420 			  struct rte_eth_rss_conf *rss_conf)
1421 {
1422 	struct sfc_adapter *sa = dev->data->dev_private;
1423 	struct sfc_rss *rss = &sa->rss;
1424 
1425 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1426 		return -ENOTSUP;
1427 
1428 	sfc_adapter_lock(sa);
1429 
1430 	/*
1431 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1432 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1433 	 * flags which corresponds to the active EFX configuration stored
1434 	 * locally in 'sfc_adapter' and kept up-to-date
1435 	 */
1436 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1437 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1438 	if (rss_conf->rss_key != NULL)
1439 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1440 
1441 	sfc_adapter_unlock(sa);
1442 
1443 	return 0;
1444 }
1445 
1446 static int
1447 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1448 			struct rte_eth_rss_conf *rss_conf)
1449 {
1450 	struct sfc_adapter *sa = dev->data->dev_private;
1451 	struct sfc_rss *rss = &sa->rss;
1452 	struct sfc_port *port = &sa->port;
1453 	unsigned int efx_hash_types;
1454 	int rc = 0;
1455 
1456 	if (port->isolated)
1457 		return -ENOTSUP;
1458 
1459 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1460 		sfc_err(sa, "RSS is not available");
1461 		return -ENOTSUP;
1462 	}
1463 
1464 	if (rss->channels == 0) {
1465 		sfc_err(sa, "RSS is not configured");
1466 		return -EINVAL;
1467 	}
1468 
1469 	if ((rss_conf->rss_key != NULL) &&
1470 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1471 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1472 			sizeof(rss->key));
1473 		return -EINVAL;
1474 	}
1475 
1476 	sfc_adapter_lock(sa);
1477 
1478 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1479 	if (rc != 0)
1480 		goto fail_rx_hf_rte_to_efx;
1481 
1482 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1483 				   rss->hash_alg, efx_hash_types, B_TRUE);
1484 	if (rc != 0)
1485 		goto fail_scale_mode_set;
1486 
1487 	if (rss_conf->rss_key != NULL) {
1488 		if (sa->state == SFC_ADAPTER_STARTED) {
1489 			rc = efx_rx_scale_key_set(sa->nic,
1490 						  EFX_RSS_CONTEXT_DEFAULT,
1491 						  rss_conf->rss_key,
1492 						  sizeof(rss->key));
1493 			if (rc != 0)
1494 				goto fail_scale_key_set;
1495 		}
1496 
1497 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1498 	}
1499 
1500 	rss->hash_types = efx_hash_types;
1501 
1502 	sfc_adapter_unlock(sa);
1503 
1504 	return 0;
1505 
1506 fail_scale_key_set:
1507 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1508 				  EFX_RX_HASHALG_TOEPLITZ,
1509 				  rss->hash_types, B_TRUE) != 0)
1510 		sfc_err(sa, "failed to restore RSS mode");
1511 
1512 fail_scale_mode_set:
1513 fail_rx_hf_rte_to_efx:
1514 	sfc_adapter_unlock(sa);
1515 	return -rc;
1516 }
1517 
1518 /*
1519  * The function is used by the secondary process as well. It must not
1520  * use any process-local pointers from the adapter data.
1521  */
1522 static int
1523 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1524 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1525 		       uint16_t reta_size)
1526 {
1527 	struct sfc_adapter *sa = dev->data->dev_private;
1528 	struct sfc_rss *rss = &sa->rss;
1529 	struct sfc_port *port = &sa->port;
1530 	int entry;
1531 
1532 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1533 		return -ENOTSUP;
1534 
1535 	if (rss->channels == 0)
1536 		return -EINVAL;
1537 
1538 	if (reta_size != EFX_RSS_TBL_SIZE)
1539 		return -EINVAL;
1540 
1541 	sfc_adapter_lock(sa);
1542 
1543 	for (entry = 0; entry < reta_size; entry++) {
1544 		int grp = entry / RTE_RETA_GROUP_SIZE;
1545 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1546 
1547 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1548 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1549 	}
1550 
1551 	sfc_adapter_unlock(sa);
1552 
1553 	return 0;
1554 }
1555 
1556 static int
1557 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1558 			struct rte_eth_rss_reta_entry64 *reta_conf,
1559 			uint16_t reta_size)
1560 {
1561 	struct sfc_adapter *sa = dev->data->dev_private;
1562 	struct sfc_rss *rss = &sa->rss;
1563 	struct sfc_port *port = &sa->port;
1564 	unsigned int *rss_tbl_new;
1565 	uint16_t entry;
1566 	int rc = 0;
1567 
1568 
1569 	if (port->isolated)
1570 		return -ENOTSUP;
1571 
1572 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1573 		sfc_err(sa, "RSS is not available");
1574 		return -ENOTSUP;
1575 	}
1576 
1577 	if (rss->channels == 0) {
1578 		sfc_err(sa, "RSS is not configured");
1579 		return -EINVAL;
1580 	}
1581 
1582 	if (reta_size != EFX_RSS_TBL_SIZE) {
1583 		sfc_err(sa, "RETA size is wrong (should be %u)",
1584 			EFX_RSS_TBL_SIZE);
1585 		return -EINVAL;
1586 	}
1587 
1588 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1589 	if (rss_tbl_new == NULL)
1590 		return -ENOMEM;
1591 
1592 	sfc_adapter_lock(sa);
1593 
1594 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1595 
1596 	for (entry = 0; entry < reta_size; entry++) {
1597 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1598 		struct rte_eth_rss_reta_entry64 *grp;
1599 
1600 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1601 
1602 		if (grp->mask & (1ull << grp_idx)) {
1603 			if (grp->reta[grp_idx] >= rss->channels) {
1604 				rc = EINVAL;
1605 				goto bad_reta_entry;
1606 			}
1607 			rss_tbl_new[entry] = grp->reta[grp_idx];
1608 		}
1609 	}
1610 
1611 	if (sa->state == SFC_ADAPTER_STARTED) {
1612 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1613 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1614 		if (rc != 0)
1615 			goto fail_scale_tbl_set;
1616 	}
1617 
1618 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1619 
1620 fail_scale_tbl_set:
1621 bad_reta_entry:
1622 	sfc_adapter_unlock(sa);
1623 
1624 	rte_free(rss_tbl_new);
1625 
1626 	SFC_ASSERT(rc >= 0);
1627 	return -rc;
1628 }
1629 
1630 static int
1631 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1632 		    enum rte_filter_op filter_op,
1633 		    void *arg)
1634 {
1635 	struct sfc_adapter *sa = dev->data->dev_private;
1636 	int rc = ENOTSUP;
1637 
1638 	sfc_log_init(sa, "entry");
1639 
1640 	switch (filter_type) {
1641 	case RTE_ETH_FILTER_NONE:
1642 		sfc_err(sa, "Global filters configuration not supported");
1643 		break;
1644 	case RTE_ETH_FILTER_MACVLAN:
1645 		sfc_err(sa, "MACVLAN filters not supported");
1646 		break;
1647 	case RTE_ETH_FILTER_ETHERTYPE:
1648 		sfc_err(sa, "EtherType filters not supported");
1649 		break;
1650 	case RTE_ETH_FILTER_FLEXIBLE:
1651 		sfc_err(sa, "Flexible filters not supported");
1652 		break;
1653 	case RTE_ETH_FILTER_SYN:
1654 		sfc_err(sa, "SYN filters not supported");
1655 		break;
1656 	case RTE_ETH_FILTER_NTUPLE:
1657 		sfc_err(sa, "NTUPLE filters not supported");
1658 		break;
1659 	case RTE_ETH_FILTER_TUNNEL:
1660 		sfc_err(sa, "Tunnel filters not supported");
1661 		break;
1662 	case RTE_ETH_FILTER_FDIR:
1663 		sfc_err(sa, "Flow Director filters not supported");
1664 		break;
1665 	case RTE_ETH_FILTER_HASH:
1666 		sfc_err(sa, "Hash filters not supported");
1667 		break;
1668 	case RTE_ETH_FILTER_GENERIC:
1669 		if (filter_op != RTE_ETH_FILTER_GET) {
1670 			rc = EINVAL;
1671 		} else {
1672 			*(const void **)arg = &sfc_flow_ops;
1673 			rc = 0;
1674 		}
1675 		break;
1676 	default:
1677 		sfc_err(sa, "Unknown filter type %u", filter_type);
1678 		break;
1679 	}
1680 
1681 	sfc_log_init(sa, "exit: %d", -rc);
1682 	SFC_ASSERT(rc >= 0);
1683 	return -rc;
1684 }
1685 
1686 static int
1687 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1688 {
1689 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1690 
1691 	/*
1692 	 * If Rx datapath does not provide callback to check mempool,
1693 	 * all pools are supported.
1694 	 */
1695 	if (sap->dp_rx->pool_ops_supported == NULL)
1696 		return 1;
1697 
1698 	return sap->dp_rx->pool_ops_supported(pool);
1699 }
1700 
1701 static const struct eth_dev_ops sfc_eth_dev_ops = {
1702 	.dev_configure			= sfc_dev_configure,
1703 	.dev_start			= sfc_dev_start,
1704 	.dev_stop			= sfc_dev_stop,
1705 	.dev_set_link_up		= sfc_dev_set_link_up,
1706 	.dev_set_link_down		= sfc_dev_set_link_down,
1707 	.dev_close			= sfc_dev_close,
1708 	.promiscuous_enable		= sfc_dev_promisc_enable,
1709 	.promiscuous_disable		= sfc_dev_promisc_disable,
1710 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1711 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1712 	.link_update			= sfc_dev_link_update,
1713 	.stats_get			= sfc_stats_get,
1714 	.stats_reset			= sfc_stats_reset,
1715 	.xstats_get			= sfc_xstats_get,
1716 	.xstats_reset			= sfc_stats_reset,
1717 	.xstats_get_names		= sfc_xstats_get_names,
1718 	.dev_infos_get			= sfc_dev_infos_get,
1719 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1720 	.mtu_set			= sfc_dev_set_mtu,
1721 	.rx_queue_start			= sfc_rx_queue_start,
1722 	.rx_queue_stop			= sfc_rx_queue_stop,
1723 	.tx_queue_start			= sfc_tx_queue_start,
1724 	.tx_queue_stop			= sfc_tx_queue_stop,
1725 	.rx_queue_setup			= sfc_rx_queue_setup,
1726 	.rx_queue_release		= sfc_rx_queue_release,
1727 	.rx_queue_count			= sfc_rx_queue_count,
1728 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1729 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1730 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1731 	.tx_queue_setup			= sfc_tx_queue_setup,
1732 	.tx_queue_release		= sfc_tx_queue_release,
1733 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1734 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1735 	.mac_addr_set			= sfc_mac_addr_set,
1736 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1737 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1738 	.reta_update			= sfc_dev_rss_reta_update,
1739 	.reta_query			= sfc_dev_rss_reta_query,
1740 	.rss_hash_update		= sfc_dev_rss_hash_update,
1741 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1742 	.filter_ctrl			= sfc_dev_filter_ctrl,
1743 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1744 	.rxq_info_get			= sfc_rx_queue_info_get,
1745 	.txq_info_get			= sfc_tx_queue_info_get,
1746 	.fw_version_get			= sfc_fw_version_get,
1747 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1748 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1749 	.pool_ops_supported		= sfc_pool_ops_supported,
1750 };
1751 
1752 /**
1753  * Duplicate a string in potentially shared memory required for
1754  * multi-process support.
1755  *
1756  * strdup() allocates from process-local heap/memory.
1757  */
1758 static char *
1759 sfc_strdup(const char *str)
1760 {
1761 	size_t size;
1762 	char *copy;
1763 
1764 	if (str == NULL)
1765 		return NULL;
1766 
1767 	size = strlen(str) + 1;
1768 	copy = rte_malloc(__func__, size, 0);
1769 	if (copy != NULL)
1770 		rte_memcpy(copy, str, size);
1771 
1772 	return copy;
1773 }
1774 
1775 static int
1776 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1777 {
1778 	struct sfc_adapter *sa = dev->data->dev_private;
1779 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1780 	const struct sfc_dp_rx *dp_rx;
1781 	const struct sfc_dp_tx *dp_tx;
1782 	const efx_nic_cfg_t *encp;
1783 	unsigned int avail_caps = 0;
1784 	const char *rx_name = NULL;
1785 	const char *tx_name = NULL;
1786 	int rc;
1787 
1788 	switch (sa->family) {
1789 	case EFX_FAMILY_HUNTINGTON:
1790 	case EFX_FAMILY_MEDFORD:
1791 	case EFX_FAMILY_MEDFORD2:
1792 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1793 		break;
1794 	default:
1795 		break;
1796 	}
1797 
1798 	encp = efx_nic_cfg_get(sa->nic);
1799 	if (encp->enc_rx_es_super_buffer_supported)
1800 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1801 
1802 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1803 				sfc_kvarg_string_handler, &rx_name);
1804 	if (rc != 0)
1805 		goto fail_kvarg_rx_datapath;
1806 
1807 	if (rx_name != NULL) {
1808 		dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1809 		if (dp_rx == NULL) {
1810 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1811 			rc = ENOENT;
1812 			goto fail_dp_rx;
1813 		}
1814 		if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1815 			sfc_err(sa,
1816 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1817 				rx_name);
1818 			rc = EINVAL;
1819 			goto fail_dp_rx_caps;
1820 		}
1821 	} else {
1822 		dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1823 		if (dp_rx == NULL) {
1824 			sfc_err(sa, "Rx datapath by caps %#x not found",
1825 				avail_caps);
1826 			rc = ENOENT;
1827 			goto fail_dp_rx;
1828 		}
1829 	}
1830 
1831 	sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1832 	if (sas->dp_rx_name == NULL) {
1833 		rc = ENOMEM;
1834 		goto fail_dp_rx_name;
1835 	}
1836 
1837 	sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1838 
1839 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1840 				sfc_kvarg_string_handler, &tx_name);
1841 	if (rc != 0)
1842 		goto fail_kvarg_tx_datapath;
1843 
1844 	if (tx_name != NULL) {
1845 		dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1846 		if (dp_tx == NULL) {
1847 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1848 			rc = ENOENT;
1849 			goto fail_dp_tx;
1850 		}
1851 		if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1852 			sfc_err(sa,
1853 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1854 				tx_name);
1855 			rc = EINVAL;
1856 			goto fail_dp_tx_caps;
1857 		}
1858 	} else {
1859 		dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1860 		if (dp_tx == NULL) {
1861 			sfc_err(sa, "Tx datapath by caps %#x not found",
1862 				avail_caps);
1863 			rc = ENOENT;
1864 			goto fail_dp_tx;
1865 		}
1866 	}
1867 
1868 	sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1869 	if (sas->dp_tx_name == NULL) {
1870 		rc = ENOMEM;
1871 		goto fail_dp_tx_name;
1872 	}
1873 
1874 	sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1875 
1876 	sa->priv.dp_rx = dp_rx;
1877 	sa->priv.dp_tx = dp_tx;
1878 
1879 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1880 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1881 
1882 	dev->dev_ops = &sfc_eth_dev_ops;
1883 
1884 	return 0;
1885 
1886 fail_dp_tx_name:
1887 fail_dp_tx_caps:
1888 fail_dp_tx:
1889 fail_kvarg_tx_datapath:
1890 	rte_free(sas->dp_rx_name);
1891 	sas->dp_rx_name = NULL;
1892 
1893 fail_dp_rx_name:
1894 fail_dp_rx_caps:
1895 fail_dp_rx:
1896 fail_kvarg_rx_datapath:
1897 	return rc;
1898 }
1899 
1900 static void
1901 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1902 {
1903 	struct sfc_adapter *sa = dev->data->dev_private;
1904 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1905 
1906 	dev->dev_ops = NULL;
1907 	dev->rx_pkt_burst = NULL;
1908 	dev->tx_pkt_burst = NULL;
1909 
1910 	rte_free(sas->dp_tx_name);
1911 	sas->dp_tx_name = NULL;
1912 	sa->priv.dp_tx = NULL;
1913 
1914 	rte_free(sas->dp_rx_name);
1915 	sas->dp_rx_name = NULL;
1916 	sa->priv.dp_rx = NULL;
1917 }
1918 
1919 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1920 	.rx_queue_count			= sfc_rx_queue_count,
1921 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1922 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1923 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1924 	.reta_query			= sfc_dev_rss_reta_query,
1925 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1926 	.rxq_info_get			= sfc_rx_queue_info_get,
1927 	.txq_info_get			= sfc_tx_queue_info_get,
1928 };
1929 
1930 static int
1931 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
1932 {
1933 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1934 	struct sfc_adapter_priv *sap;
1935 	const struct sfc_dp_rx *dp_rx;
1936 	const struct sfc_dp_tx *dp_tx;
1937 	int rc;
1938 
1939 	/*
1940 	 * Allocate process private data from heap, since it should not
1941 	 * be located in shared memory allocated using rte_malloc() API.
1942 	 */
1943 	sap = calloc(1, sizeof(*sap));
1944 	if (sap == NULL) {
1945 		rc = ENOMEM;
1946 		goto fail_alloc_priv;
1947 	}
1948 
1949 	sap->logtype_main = logtype_main;
1950 
1951 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
1952 	if (dp_rx == NULL) {
1953 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1954 			"cannot find %s Rx datapath", sas->dp_rx_name);
1955 		rc = ENOENT;
1956 		goto fail_dp_rx;
1957 	}
1958 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1959 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1960 			"%s Rx datapath does not support multi-process",
1961 			sas->dp_rx_name);
1962 		rc = EINVAL;
1963 		goto fail_dp_rx_multi_process;
1964 	}
1965 
1966 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
1967 	if (dp_tx == NULL) {
1968 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1969 			"cannot find %s Tx datapath", sas->dp_tx_name);
1970 		rc = ENOENT;
1971 		goto fail_dp_tx;
1972 	}
1973 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1974 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1975 			"%s Tx datapath does not support multi-process",
1976 			sas->dp_tx_name);
1977 		rc = EINVAL;
1978 		goto fail_dp_tx_multi_process;
1979 	}
1980 
1981 	sap->dp_rx = dp_rx;
1982 	sap->dp_tx = dp_tx;
1983 
1984 	dev->process_private = sap;
1985 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1986 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1987 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1988 
1989 	return 0;
1990 
1991 fail_dp_tx_multi_process:
1992 fail_dp_tx:
1993 fail_dp_rx_multi_process:
1994 fail_dp_rx:
1995 	free(sap);
1996 
1997 fail_alloc_priv:
1998 	return rc;
1999 }
2000 
2001 static void
2002 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
2003 {
2004 	free(dev->process_private);
2005 	dev->process_private = NULL;
2006 	dev->dev_ops = NULL;
2007 	dev->tx_pkt_burst = NULL;
2008 	dev->rx_pkt_burst = NULL;
2009 }
2010 
2011 static void
2012 sfc_register_dp(void)
2013 {
2014 	/* Register once */
2015 	if (TAILQ_EMPTY(&sfc_dp_head)) {
2016 		/* Prefer EF10 datapath */
2017 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2018 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2019 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2020 
2021 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2022 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2023 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2024 	}
2025 }
2026 
2027 static int
2028 sfc_eth_dev_init(struct rte_eth_dev *dev)
2029 {
2030 	struct sfc_adapter *sa = dev->data->dev_private;
2031 	struct sfc_adapter_shared *sas;
2032 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2033 	uint32_t logtype_main;
2034 	int rc;
2035 	const efx_nic_cfg_t *encp;
2036 	const struct ether_addr *from;
2037 
2038 	sfc_register_dp();
2039 
2040 	logtype_main = sfc_register_logtype(&pci_dev->addr,
2041 					    SFC_LOGTYPE_MAIN_STR,
2042 					    RTE_LOG_NOTICE);
2043 
2044 	sa->priv.shared = &sa->_shared;
2045 	sas = sa->priv.shared;
2046 
2047 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2048 		return -sfc_eth_dev_secondary_init(dev, logtype_main);
2049 
2050 	/*
2051 	 * sfc_adapter is a mixture of shared and process private data.
2052 	 * During transition period use it in both kinds. When the
2053 	 * driver becomes ready to separate it, sfc_adapter will become
2054 	 * primary process private only.
2055 	 */
2056 	dev->process_private = sa;
2057 
2058 	/* Required for logging */
2059 	sas->pci_addr = pci_dev->addr;
2060 	sas->port_id = dev->data->port_id;
2061 	sa->priv.logtype_main = logtype_main;
2062 
2063 	sa->eth_dev = dev;
2064 
2065 	/* Copy PCI device info to the dev->data */
2066 	rte_eth_copy_pci_info(dev, pci_dev);
2067 
2068 	rc = sfc_kvargs_parse(sa);
2069 	if (rc != 0)
2070 		goto fail_kvargs_parse;
2071 
2072 	sfc_log_init(sa, "entry");
2073 
2074 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
2075 	if (dev->data->mac_addrs == NULL) {
2076 		rc = ENOMEM;
2077 		goto fail_mac_addrs;
2078 	}
2079 
2080 	sfc_adapter_lock_init(sa);
2081 	sfc_adapter_lock(sa);
2082 
2083 	sfc_log_init(sa, "probing");
2084 	rc = sfc_probe(sa);
2085 	if (rc != 0)
2086 		goto fail_probe;
2087 
2088 	sfc_log_init(sa, "set device ops");
2089 	rc = sfc_eth_dev_set_ops(dev);
2090 	if (rc != 0)
2091 		goto fail_set_ops;
2092 
2093 	sfc_log_init(sa, "attaching");
2094 	rc = sfc_attach(sa);
2095 	if (rc != 0)
2096 		goto fail_attach;
2097 
2098 	encp = efx_nic_cfg_get(sa->nic);
2099 
2100 	/*
2101 	 * The arguments are really reverse order in comparison to
2102 	 * Linux kernel. Copy from NIC config to Ethernet device data.
2103 	 */
2104 	from = (const struct ether_addr *)(encp->enc_mac_addr);
2105 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
2106 
2107 	sfc_adapter_unlock(sa);
2108 
2109 	sfc_log_init(sa, "done");
2110 	return 0;
2111 
2112 fail_attach:
2113 	sfc_eth_dev_clear_ops(dev);
2114 
2115 fail_set_ops:
2116 	sfc_unprobe(sa);
2117 
2118 fail_probe:
2119 	sfc_adapter_unlock(sa);
2120 	sfc_adapter_lock_fini(sa);
2121 	rte_free(dev->data->mac_addrs);
2122 	dev->data->mac_addrs = NULL;
2123 
2124 fail_mac_addrs:
2125 	sfc_kvargs_cleanup(sa);
2126 
2127 fail_kvargs_parse:
2128 	sfc_log_init(sa, "failed %d", rc);
2129 	dev->process_private = NULL;
2130 	SFC_ASSERT(rc > 0);
2131 	return -rc;
2132 }
2133 
2134 static int
2135 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2136 {
2137 	struct sfc_adapter *sa;
2138 
2139 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2140 		sfc_eth_dev_secondary_clear_ops(dev);
2141 		return 0;
2142 	}
2143 
2144 	sa = dev->data->dev_private;
2145 	sfc_log_init(sa, "entry");
2146 
2147 	sfc_adapter_lock(sa);
2148 
2149 	sfc_eth_dev_clear_ops(dev);
2150 
2151 	sfc_detach(sa);
2152 	sfc_unprobe(sa);
2153 
2154 	sfc_kvargs_cleanup(sa);
2155 
2156 	sfc_adapter_unlock(sa);
2157 	sfc_adapter_lock_fini(sa);
2158 
2159 	sfc_log_init(sa, "done");
2160 
2161 	/* Required for logging, so cleanup last */
2162 	sa->eth_dev = NULL;
2163 	return 0;
2164 }
2165 
2166 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2167 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2168 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2169 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2170 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2171 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2172 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2173 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2174 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2175 	{ .vendor_id = 0 /* sentinel */ }
2176 };
2177 
2178 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2179 	struct rte_pci_device *pci_dev)
2180 {
2181 	return rte_eth_dev_pci_generic_probe(pci_dev,
2182 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2183 }
2184 
2185 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2186 {
2187 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2188 }
2189 
2190 static struct rte_pci_driver sfc_efx_pmd = {
2191 	.id_table = pci_id_sfc_efx_map,
2192 	.drv_flags =
2193 		RTE_PCI_DRV_INTR_LSC |
2194 		RTE_PCI_DRV_NEED_MAPPING,
2195 	.probe = sfc_eth_dev_pci_probe,
2196 	.remove = sfc_eth_dev_pci_remove,
2197 };
2198 
2199 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2200 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2201 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2202 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2203 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2204 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2205 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2206 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2207 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2208 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2209 
2210 RTE_INIT(sfc_driver_register_logtype)
2211 {
2212 	int ret;
2213 
2214 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2215 						   RTE_LOG_NOTICE);
2216 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2217 }
2218