xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 5ecb687a5698d2d8ec1f3b3b5a7a16bceca3e29c)
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 #include <rte_ether.h>
18 
19 #include "efx.h"
20 
21 #include "sfc.h"
22 #include "sfc_debug.h"
23 #include "sfc_log.h"
24 #include "sfc_kvargs.h"
25 #include "sfc_ev.h"
26 #include "sfc_rx.h"
27 #include "sfc_tx.h"
28 #include "sfc_flow.h"
29 #include "sfc_dp.h"
30 #include "sfc_dp_rx.h"
31 
32 uint32_t sfc_logtype_driver;
33 
34 static struct sfc_dp_list sfc_dp_head =
35 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
36 
37 static int
38 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
39 {
40 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
41 	efx_nic_fw_info_t enfi;
42 	int ret;
43 	int rc;
44 
45 	/*
46 	 * Return value of the callback is likely supposed to be
47 	 * equal to or greater than 0, nevertheless, if an error
48 	 * occurs, it will be desirable to pass it to the caller
49 	 */
50 	if ((fw_version == NULL) || (fw_size == 0))
51 		return -EINVAL;
52 
53 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
54 	if (rc != 0)
55 		return -rc;
56 
57 	ret = snprintf(fw_version, fw_size,
58 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
59 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
60 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
61 	if (ret < 0)
62 		return ret;
63 
64 	if (enfi.enfi_dpcpu_fw_ids_valid) {
65 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
66 		int ret_extra;
67 
68 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
69 				     fw_size - dpcpu_fw_ids_offset,
70 				     " rx%" PRIx16 " tx%" PRIx16,
71 				     enfi.enfi_rx_dpcpu_fw_id,
72 				     enfi.enfi_tx_dpcpu_fw_id);
73 		if (ret_extra < 0)
74 			return ret_extra;
75 
76 		ret += ret_extra;
77 	}
78 
79 	if (fw_size < (size_t)(++ret))
80 		return ret;
81 	else
82 		return 0;
83 }
84 
85 static void
86 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
87 {
88 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
89 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
90 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
91 	struct sfc_rss *rss = &sas->rss;
92 	uint64_t txq_offloads_def = 0;
93 
94 	sfc_log_init(sa, "entry");
95 
96 	dev_info->min_mtu = ETHER_MIN_MTU;
97 	dev_info->max_mtu = EFX_MAC_SDU_MAX;
98 
99 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
100 
101 	/* Autonegotiation may be disabled */
102 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
103 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
104 		dev_info->speed_capa |= ETH_LINK_SPEED_1G;
105 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
106 		dev_info->speed_capa |= ETH_LINK_SPEED_10G;
107 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
108 		dev_info->speed_capa |= ETH_LINK_SPEED_25G;
109 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
110 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
111 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
112 		dev_info->speed_capa |= ETH_LINK_SPEED_50G;
113 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
114 		dev_info->speed_capa |= ETH_LINK_SPEED_100G;
115 
116 	dev_info->max_rx_queues = sa->rxq_max;
117 	dev_info->max_tx_queues = sa->txq_max;
118 
119 	/* By default packets are dropped if no descriptors are available */
120 	dev_info->default_rxconf.rx_drop_en = 1;
121 
122 	dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
123 
124 	/*
125 	 * rx_offload_capa includes both device and queue offloads since
126 	 * the latter may be requested on a per device basis which makes
127 	 * sense when some offloads are needed to be set on all queues.
128 	 */
129 	dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
130 				    dev_info->rx_queue_offload_capa;
131 
132 	dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
133 
134 	/*
135 	 * tx_offload_capa includes both device and queue offloads since
136 	 * the latter may be requested on a per device basis which makes
137 	 * sense when some offloads are needed to be set on all queues.
138 	 */
139 	dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
140 				    dev_info->tx_queue_offload_capa;
141 
142 	if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
143 		txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
144 
145 	dev_info->default_txconf.offloads |= txq_offloads_def;
146 
147 	if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
148 		uint64_t rte_hf = 0;
149 		unsigned int i;
150 
151 		for (i = 0; i < rss->hf_map_nb_entries; ++i)
152 			rte_hf |= rss->hf_map[i].rte;
153 
154 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
155 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
156 		dev_info->flow_type_rss_offloads = rte_hf;
157 	}
158 
159 	/* Initialize to hardware limits */
160 	dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
161 	dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
162 	/* The RXQ hardware requires that the descriptor count is a power
163 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
164 	 */
165 	dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
166 
167 	/* Initialize to hardware limits */
168 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
169 	dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
170 	/*
171 	 * The TXQ hardware requires that the descriptor count is a power
172 	 * of 2, but tx_desc_lim cannot properly describe that constraint
173 	 */
174 	dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
175 
176 	if (sap->dp_rx->get_dev_info != NULL)
177 		sap->dp_rx->get_dev_info(dev_info);
178 	if (sap->dp_tx->get_dev_info != NULL)
179 		sap->dp_tx->get_dev_info(dev_info);
180 
181 	dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
182 			     RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
183 }
184 
185 static const uint32_t *
186 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
187 {
188 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
189 
190 	return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
191 }
192 
193 static int
194 sfc_dev_configure(struct rte_eth_dev *dev)
195 {
196 	struct rte_eth_dev_data *dev_data = dev->data;
197 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
198 	int rc;
199 
200 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
201 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
202 
203 	sfc_adapter_lock(sa);
204 	switch (sa->state) {
205 	case SFC_ADAPTER_CONFIGURED:
206 		/* FALLTHROUGH */
207 	case SFC_ADAPTER_INITIALIZED:
208 		rc = sfc_configure(sa);
209 		break;
210 	default:
211 		sfc_err(sa, "unexpected adapter state %u to configure",
212 			sa->state);
213 		rc = EINVAL;
214 		break;
215 	}
216 	sfc_adapter_unlock(sa);
217 
218 	sfc_log_init(sa, "done %d", rc);
219 	SFC_ASSERT(rc >= 0);
220 	return -rc;
221 }
222 
223 static int
224 sfc_dev_start(struct rte_eth_dev *dev)
225 {
226 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
227 	int rc;
228 
229 	sfc_log_init(sa, "entry");
230 
231 	sfc_adapter_lock(sa);
232 	rc = sfc_start(sa);
233 	sfc_adapter_unlock(sa);
234 
235 	sfc_log_init(sa, "done %d", rc);
236 	SFC_ASSERT(rc >= 0);
237 	return -rc;
238 }
239 
240 static int
241 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
242 {
243 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
244 	struct rte_eth_link current_link;
245 	int ret;
246 
247 	sfc_log_init(sa, "entry");
248 
249 	if (sa->state != SFC_ADAPTER_STARTED) {
250 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
251 	} else if (wait_to_complete) {
252 		efx_link_mode_t link_mode;
253 
254 		if (efx_port_poll(sa->nic, &link_mode) != 0)
255 			link_mode = EFX_LINK_UNKNOWN;
256 		sfc_port_link_mode_to_info(link_mode, &current_link);
257 
258 	} else {
259 		sfc_ev_mgmt_qpoll(sa);
260 		rte_eth_linkstatus_get(dev, &current_link);
261 	}
262 
263 	ret = rte_eth_linkstatus_set(dev, &current_link);
264 	if (ret == 0)
265 		sfc_notice(sa, "Link status is %s",
266 			   current_link.link_status ? "UP" : "DOWN");
267 
268 	return ret;
269 }
270 
271 static void
272 sfc_dev_stop(struct rte_eth_dev *dev)
273 {
274 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
275 
276 	sfc_log_init(sa, "entry");
277 
278 	sfc_adapter_lock(sa);
279 	sfc_stop(sa);
280 	sfc_adapter_unlock(sa);
281 
282 	sfc_log_init(sa, "done");
283 }
284 
285 static int
286 sfc_dev_set_link_up(struct rte_eth_dev *dev)
287 {
288 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
289 	int rc;
290 
291 	sfc_log_init(sa, "entry");
292 
293 	sfc_adapter_lock(sa);
294 	rc = sfc_start(sa);
295 	sfc_adapter_unlock(sa);
296 
297 	SFC_ASSERT(rc >= 0);
298 	return -rc;
299 }
300 
301 static int
302 sfc_dev_set_link_down(struct rte_eth_dev *dev)
303 {
304 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
305 
306 	sfc_log_init(sa, "entry");
307 
308 	sfc_adapter_lock(sa);
309 	sfc_stop(sa);
310 	sfc_adapter_unlock(sa);
311 
312 	return 0;
313 }
314 
315 static void
316 sfc_dev_close(struct rte_eth_dev *dev)
317 {
318 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
319 
320 	sfc_log_init(sa, "entry");
321 
322 	sfc_adapter_lock(sa);
323 	switch (sa->state) {
324 	case SFC_ADAPTER_STARTED:
325 		sfc_stop(sa);
326 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
327 		/* FALLTHROUGH */
328 	case SFC_ADAPTER_CONFIGURED:
329 		sfc_close(sa);
330 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
331 		/* FALLTHROUGH */
332 	case SFC_ADAPTER_INITIALIZED:
333 		break;
334 	default:
335 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
336 		break;
337 	}
338 	sfc_adapter_unlock(sa);
339 
340 	sfc_log_init(sa, "done");
341 }
342 
343 static void
344 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
345 		   boolean_t enabled)
346 {
347 	struct sfc_port *port;
348 	boolean_t *toggle;
349 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
350 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
351 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
352 
353 	sfc_adapter_lock(sa);
354 
355 	port = &sa->port;
356 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
357 
358 	if (*toggle != enabled) {
359 		*toggle = enabled;
360 
361 		if (sfc_sa2shared(sa)->isolated) {
362 			sfc_warn(sa, "isolated mode is active on the port");
363 			sfc_warn(sa, "the change is to be applied on the next "
364 				     "start provided that isolated mode is "
365 				     "disabled prior the next start");
366 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
367 			   (sfc_set_rx_mode(sa) != 0)) {
368 			*toggle = !(enabled);
369 			sfc_warn(sa, "Failed to %s %s mode",
370 				 ((enabled) ? "enable" : "disable"), desc);
371 		}
372 	}
373 
374 	sfc_adapter_unlock(sa);
375 }
376 
377 static void
378 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
379 {
380 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
381 }
382 
383 static void
384 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
385 {
386 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
387 }
388 
389 static void
390 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
391 {
392 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
393 }
394 
395 static void
396 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
397 {
398 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
399 }
400 
401 static int
402 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
403 		   uint16_t nb_rx_desc, unsigned int socket_id,
404 		   const struct rte_eth_rxconf *rx_conf,
405 		   struct rte_mempool *mb_pool)
406 {
407 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
408 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
409 	int rc;
410 
411 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
412 		     rx_queue_id, nb_rx_desc, socket_id);
413 
414 	sfc_adapter_lock(sa);
415 
416 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
417 			  rx_conf, mb_pool);
418 	if (rc != 0)
419 		goto fail_rx_qinit;
420 
421 	dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
422 
423 	sfc_adapter_unlock(sa);
424 
425 	return 0;
426 
427 fail_rx_qinit:
428 	sfc_adapter_unlock(sa);
429 	SFC_ASSERT(rc > 0);
430 	return -rc;
431 }
432 
433 static void
434 sfc_rx_queue_release(void *queue)
435 {
436 	struct sfc_dp_rxq *dp_rxq = queue;
437 	struct sfc_rxq *rxq;
438 	struct sfc_adapter *sa;
439 	unsigned int sw_index;
440 
441 	if (dp_rxq == NULL)
442 		return;
443 
444 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
445 	sa = rxq->evq->sa;
446 	sfc_adapter_lock(sa);
447 
448 	sw_index = dp_rxq->dpq.queue_id;
449 
450 	sfc_log_init(sa, "RxQ=%u", sw_index);
451 
452 	sfc_rx_qfini(sa, sw_index);
453 
454 	sfc_adapter_unlock(sa);
455 }
456 
457 static int
458 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
459 		   uint16_t nb_tx_desc, unsigned int socket_id,
460 		   const struct rte_eth_txconf *tx_conf)
461 {
462 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
463 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
464 	int rc;
465 
466 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
467 		     tx_queue_id, nb_tx_desc, socket_id);
468 
469 	sfc_adapter_lock(sa);
470 
471 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
472 	if (rc != 0)
473 		goto fail_tx_qinit;
474 
475 	dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
476 
477 	sfc_adapter_unlock(sa);
478 	return 0;
479 
480 fail_tx_qinit:
481 	sfc_adapter_unlock(sa);
482 	SFC_ASSERT(rc > 0);
483 	return -rc;
484 }
485 
486 static void
487 sfc_tx_queue_release(void *queue)
488 {
489 	struct sfc_dp_txq *dp_txq = queue;
490 	struct sfc_txq *txq;
491 	unsigned int sw_index;
492 	struct sfc_adapter *sa;
493 
494 	if (dp_txq == NULL)
495 		return;
496 
497 	txq = sfc_txq_by_dp_txq(dp_txq);
498 	sw_index = dp_txq->dpq.queue_id;
499 
500 	SFC_ASSERT(txq->evq != NULL);
501 	sa = txq->evq->sa;
502 
503 	sfc_log_init(sa, "TxQ = %u", sw_index);
504 
505 	sfc_adapter_lock(sa);
506 
507 	sfc_tx_qfini(sa, sw_index);
508 
509 	sfc_adapter_unlock(sa);
510 }
511 
512 /*
513  * Some statistics are computed as A - B where A and B each increase
514  * monotonically with some hardware counter(s) and the counters are read
515  * asynchronously.
516  *
517  * If packet X is counted in A, but not counted in B yet, computed value is
518  * greater than real.
519  *
520  * If packet X is not counted in A at the moment of reading the counter,
521  * but counted in B at the moment of reading the counter, computed value
522  * is less than real.
523  *
524  * However, counter which grows backward is worse evil than slightly wrong
525  * value. So, let's try to guarantee that it never happens except may be
526  * the case when the MAC stats are zeroed as a result of a NIC reset.
527  */
528 static void
529 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
530 {
531 	if ((int64_t)(newval - *stat) > 0 || newval == 0)
532 		*stat = newval;
533 }
534 
535 static int
536 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
537 {
538 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
539 	struct sfc_port *port = &sa->port;
540 	uint64_t *mac_stats;
541 	int ret;
542 
543 	rte_spinlock_lock(&port->mac_stats_lock);
544 
545 	ret = sfc_port_update_mac_stats(sa);
546 	if (ret != 0)
547 		goto unlock;
548 
549 	mac_stats = port->mac_stats_buf;
550 
551 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
552 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
553 		stats->ipackets =
554 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
555 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
556 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
557 		stats->opackets =
558 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
559 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
560 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
561 		stats->ibytes =
562 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
563 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
564 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
565 		stats->obytes =
566 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
567 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
568 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
569 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
570 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
571 	} else {
572 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
573 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
574 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
575 		/*
576 		 * Take into account stats which are whenever supported
577 		 * on EF10. If some stat is not supported by current
578 		 * firmware variant or HW revision, it is guaranteed
579 		 * to be zero in mac_stats.
580 		 */
581 		stats->imissed =
582 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
583 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
584 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
585 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
586 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
587 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
588 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
589 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
590 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
591 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
592 		stats->ierrors =
593 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
594 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
595 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
596 		/* no oerrors counters supported on EF10 */
597 
598 		/* Exclude missed, errors and pauses from Rx packets */
599 		sfc_update_diff_stat(&port->ipackets,
600 			mac_stats[EFX_MAC_RX_PKTS] -
601 			mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
602 			stats->imissed - stats->ierrors);
603 		stats->ipackets = port->ipackets;
604 	}
605 
606 unlock:
607 	rte_spinlock_unlock(&port->mac_stats_lock);
608 	SFC_ASSERT(ret >= 0);
609 	return -ret;
610 }
611 
612 static void
613 sfc_stats_reset(struct rte_eth_dev *dev)
614 {
615 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
616 	struct sfc_port *port = &sa->port;
617 	int rc;
618 
619 	if (sa->state != SFC_ADAPTER_STARTED) {
620 		/*
621 		 * The operation cannot be done if port is not started; it
622 		 * will be scheduled to be done during the next port start
623 		 */
624 		port->mac_stats_reset_pending = B_TRUE;
625 		return;
626 	}
627 
628 	rc = sfc_port_reset_mac_stats(sa);
629 	if (rc != 0)
630 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
631 }
632 
633 static int
634 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
635 	       unsigned int xstats_count)
636 {
637 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
638 	struct sfc_port *port = &sa->port;
639 	uint64_t *mac_stats;
640 	int rc;
641 	unsigned int i;
642 	int nstats = 0;
643 
644 	rte_spinlock_lock(&port->mac_stats_lock);
645 
646 	rc = sfc_port_update_mac_stats(sa);
647 	if (rc != 0) {
648 		SFC_ASSERT(rc > 0);
649 		nstats = -rc;
650 		goto unlock;
651 	}
652 
653 	mac_stats = port->mac_stats_buf;
654 
655 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
656 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
657 			if (xstats != NULL && nstats < (int)xstats_count) {
658 				xstats[nstats].id = nstats;
659 				xstats[nstats].value = mac_stats[i];
660 			}
661 			nstats++;
662 		}
663 	}
664 
665 unlock:
666 	rte_spinlock_unlock(&port->mac_stats_lock);
667 
668 	return nstats;
669 }
670 
671 static int
672 sfc_xstats_get_names(struct rte_eth_dev *dev,
673 		     struct rte_eth_xstat_name *xstats_names,
674 		     unsigned int xstats_count)
675 {
676 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
677 	struct sfc_port *port = &sa->port;
678 	unsigned int i;
679 	unsigned int nstats = 0;
680 
681 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
682 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
683 			if (xstats_names != NULL && nstats < xstats_count)
684 				strlcpy(xstats_names[nstats].name,
685 					efx_mac_stat_name(sa->nic, i),
686 					sizeof(xstats_names[0].name));
687 			nstats++;
688 		}
689 	}
690 
691 	return nstats;
692 }
693 
694 static int
695 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
696 		     uint64_t *values, unsigned int n)
697 {
698 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
699 	struct sfc_port *port = &sa->port;
700 	uint64_t *mac_stats;
701 	unsigned int nb_supported = 0;
702 	unsigned int nb_written = 0;
703 	unsigned int i;
704 	int ret;
705 	int rc;
706 
707 	if (unlikely(values == NULL) ||
708 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
709 		return port->mac_stats_nb_supported;
710 
711 	rte_spinlock_lock(&port->mac_stats_lock);
712 
713 	rc = sfc_port_update_mac_stats(sa);
714 	if (rc != 0) {
715 		SFC_ASSERT(rc > 0);
716 		ret = -rc;
717 		goto unlock;
718 	}
719 
720 	mac_stats = port->mac_stats_buf;
721 
722 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
723 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
724 			continue;
725 
726 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
727 			values[nb_written++] = mac_stats[i];
728 
729 		++nb_supported;
730 	}
731 
732 	ret = nb_written;
733 
734 unlock:
735 	rte_spinlock_unlock(&port->mac_stats_lock);
736 
737 	return ret;
738 }
739 
740 static int
741 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
742 			   struct rte_eth_xstat_name *xstats_names,
743 			   const uint64_t *ids, unsigned int size)
744 {
745 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
746 	struct sfc_port *port = &sa->port;
747 	unsigned int nb_supported = 0;
748 	unsigned int nb_written = 0;
749 	unsigned int i;
750 
751 	if (unlikely(xstats_names == NULL) ||
752 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
753 		return port->mac_stats_nb_supported;
754 
755 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
756 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
757 			continue;
758 
759 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
760 			char *name = xstats_names[nb_written++].name;
761 
762 			strlcpy(name, efx_mac_stat_name(sa->nic, i),
763 				sizeof(xstats_names[0].name));
764 		}
765 
766 		++nb_supported;
767 	}
768 
769 	return nb_written;
770 }
771 
772 static int
773 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
774 {
775 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
776 	unsigned int wanted_fc, link_fc;
777 
778 	memset(fc_conf, 0, sizeof(*fc_conf));
779 
780 	sfc_adapter_lock(sa);
781 
782 	if (sa->state == SFC_ADAPTER_STARTED)
783 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
784 	else
785 		link_fc = sa->port.flow_ctrl;
786 
787 	switch (link_fc) {
788 	case 0:
789 		fc_conf->mode = RTE_FC_NONE;
790 		break;
791 	case EFX_FCNTL_RESPOND:
792 		fc_conf->mode = RTE_FC_RX_PAUSE;
793 		break;
794 	case EFX_FCNTL_GENERATE:
795 		fc_conf->mode = RTE_FC_TX_PAUSE;
796 		break;
797 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
798 		fc_conf->mode = RTE_FC_FULL;
799 		break;
800 	default:
801 		sfc_err(sa, "%s: unexpected flow control value %#x",
802 			__func__, link_fc);
803 	}
804 
805 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
806 
807 	sfc_adapter_unlock(sa);
808 
809 	return 0;
810 }
811 
812 static int
813 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
814 {
815 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
816 	struct sfc_port *port = &sa->port;
817 	unsigned int fcntl;
818 	int rc;
819 
820 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
821 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
822 	    fc_conf->mac_ctrl_frame_fwd != 0) {
823 		sfc_err(sa, "unsupported flow control settings specified");
824 		rc = EINVAL;
825 		goto fail_inval;
826 	}
827 
828 	switch (fc_conf->mode) {
829 	case RTE_FC_NONE:
830 		fcntl = 0;
831 		break;
832 	case RTE_FC_RX_PAUSE:
833 		fcntl = EFX_FCNTL_RESPOND;
834 		break;
835 	case RTE_FC_TX_PAUSE:
836 		fcntl = EFX_FCNTL_GENERATE;
837 		break;
838 	case RTE_FC_FULL:
839 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
840 		break;
841 	default:
842 		rc = EINVAL;
843 		goto fail_inval;
844 	}
845 
846 	sfc_adapter_lock(sa);
847 
848 	if (sa->state == SFC_ADAPTER_STARTED) {
849 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
850 		if (rc != 0)
851 			goto fail_mac_fcntl_set;
852 	}
853 
854 	port->flow_ctrl = fcntl;
855 	port->flow_ctrl_autoneg = fc_conf->autoneg;
856 
857 	sfc_adapter_unlock(sa);
858 
859 	return 0;
860 
861 fail_mac_fcntl_set:
862 	sfc_adapter_unlock(sa);
863 fail_inval:
864 	SFC_ASSERT(rc > 0);
865 	return -rc;
866 }
867 
868 static int
869 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
870 {
871 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
872 	size_t pdu = EFX_MAC_PDU(mtu);
873 	size_t old_pdu;
874 	int rc;
875 
876 	sfc_log_init(sa, "mtu=%u", mtu);
877 
878 	rc = EINVAL;
879 	if (pdu < EFX_MAC_PDU_MIN) {
880 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
881 			(unsigned int)mtu, (unsigned int)pdu,
882 			EFX_MAC_PDU_MIN);
883 		goto fail_inval;
884 	}
885 	if (pdu > EFX_MAC_PDU_MAX) {
886 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
887 			(unsigned int)mtu, (unsigned int)pdu,
888 			EFX_MAC_PDU_MAX);
889 		goto fail_inval;
890 	}
891 
892 	sfc_adapter_lock(sa);
893 
894 	if (pdu != sa->port.pdu) {
895 		if (sa->state == SFC_ADAPTER_STARTED) {
896 			sfc_stop(sa);
897 
898 			old_pdu = sa->port.pdu;
899 			sa->port.pdu = pdu;
900 			rc = sfc_start(sa);
901 			if (rc != 0)
902 				goto fail_start;
903 		} else {
904 			sa->port.pdu = pdu;
905 		}
906 	}
907 
908 	/*
909 	 * The driver does not use it, but other PMDs update jumbo frame
910 	 * flag and max_rx_pkt_len when MTU is set.
911 	 */
912 	if (mtu > ETHER_MAX_LEN) {
913 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
914 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
915 	}
916 
917 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
918 
919 	sfc_adapter_unlock(sa);
920 
921 	sfc_log_init(sa, "done");
922 	return 0;
923 
924 fail_start:
925 	sa->port.pdu = old_pdu;
926 	if (sfc_start(sa) != 0)
927 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
928 			"PDU max size - port is stopped",
929 			(unsigned int)pdu, (unsigned int)old_pdu);
930 	sfc_adapter_unlock(sa);
931 
932 fail_inval:
933 	sfc_log_init(sa, "failed %d", rc);
934 	SFC_ASSERT(rc > 0);
935 	return -rc;
936 }
937 static int
938 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
939 {
940 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
941 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
942 	struct sfc_port *port = &sa->port;
943 	struct ether_addr *old_addr = &dev->data->mac_addrs[0];
944 	int rc = 0;
945 
946 	sfc_adapter_lock(sa);
947 
948 	/*
949 	 * Copy the address to the device private data so that
950 	 * it could be recalled in the case of adapter restart.
951 	 */
952 	ether_addr_copy(mac_addr, &port->default_mac_addr);
953 
954 	/*
955 	 * Neither of the two following checks can return
956 	 * an error. The new MAC address is preserved in
957 	 * the device private data and can be activated
958 	 * on the next port start if the user prevents
959 	 * isolated mode from being enabled.
960 	 */
961 	if (sfc_sa2shared(sa)->isolated) {
962 		sfc_warn(sa, "isolated mode is active on the port");
963 		sfc_warn(sa, "will not set MAC address");
964 		goto unlock;
965 	}
966 
967 	if (sa->state != SFC_ADAPTER_STARTED) {
968 		sfc_notice(sa, "the port is not started");
969 		sfc_notice(sa, "the new MAC address will be set on port start");
970 
971 		goto unlock;
972 	}
973 
974 	if (encp->enc_allow_set_mac_with_installed_filters) {
975 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
976 		if (rc != 0) {
977 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
978 			goto unlock;
979 		}
980 
981 		/*
982 		 * Changing the MAC address by means of MCDI request
983 		 * has no effect on received traffic, therefore
984 		 * we also need to update unicast filters
985 		 */
986 		rc = sfc_set_rx_mode(sa);
987 		if (rc != 0) {
988 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
989 			/* Rollback the old address */
990 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
991 			(void)sfc_set_rx_mode(sa);
992 		}
993 	} else {
994 		sfc_warn(sa, "cannot set MAC address with filters installed");
995 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
996 		sfc_warn(sa, "(some traffic may be dropped)");
997 
998 		/*
999 		 * Since setting MAC address with filters installed is not
1000 		 * allowed on the adapter, the new MAC address will be set
1001 		 * by means of adapter restart. sfc_start() shall retrieve
1002 		 * the new address from the device private data and set it.
1003 		 */
1004 		sfc_stop(sa);
1005 		rc = sfc_start(sa);
1006 		if (rc != 0)
1007 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1008 	}
1009 
1010 unlock:
1011 	if (rc != 0)
1012 		ether_addr_copy(old_addr, &port->default_mac_addr);
1013 
1014 	sfc_adapter_unlock(sa);
1015 
1016 	SFC_ASSERT(rc >= 0);
1017 	return -rc;
1018 }
1019 
1020 
1021 static int
1022 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1023 		     uint32_t nb_mc_addr)
1024 {
1025 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1026 	struct sfc_port *port = &sa->port;
1027 	uint8_t *mc_addrs = port->mcast_addrs;
1028 	int rc;
1029 	unsigned int i;
1030 
1031 	if (sfc_sa2shared(sa)->isolated) {
1032 		sfc_err(sa, "isolated mode is active on the port");
1033 		sfc_err(sa, "will not set multicast address list");
1034 		return -ENOTSUP;
1035 	}
1036 
1037 	if (mc_addrs == NULL)
1038 		return -ENOBUFS;
1039 
1040 	if (nb_mc_addr > port->max_mcast_addrs) {
1041 		sfc_err(sa, "too many multicast addresses: %u > %u",
1042 			 nb_mc_addr, port->max_mcast_addrs);
1043 		return -EINVAL;
1044 	}
1045 
1046 	for (i = 0; i < nb_mc_addr; ++i) {
1047 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1048 				 EFX_MAC_ADDR_LEN);
1049 		mc_addrs += EFX_MAC_ADDR_LEN;
1050 	}
1051 
1052 	port->nb_mcast_addrs = nb_mc_addr;
1053 
1054 	if (sa->state != SFC_ADAPTER_STARTED)
1055 		return 0;
1056 
1057 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1058 					port->nb_mcast_addrs);
1059 	if (rc != 0)
1060 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1061 
1062 	SFC_ASSERT(rc >= 0);
1063 	return -rc;
1064 }
1065 
1066 /*
1067  * The function is used by the secondary process as well. It must not
1068  * use any process-local pointers from the adapter data.
1069  */
1070 static void
1071 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1072 		      struct rte_eth_rxq_info *qinfo)
1073 {
1074 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1075 	struct sfc_rxq_info *rxq_info;
1076 
1077 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1078 
1079 	rxq_info = &sas->rxq_info[rx_queue_id];
1080 
1081 	qinfo->mp = rxq_info->refill_mb_pool;
1082 	qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1083 	qinfo->conf.rx_drop_en = 1;
1084 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1085 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1086 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1087 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1088 		qinfo->scattered_rx = 1;
1089 	}
1090 	qinfo->nb_desc = rxq_info->entries;
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_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1102 	struct sfc_txq_info *txq_info;
1103 
1104 	SFC_ASSERT(tx_queue_id < sas->txq_count);
1105 
1106 	txq_info = &sas->txq_info[tx_queue_id];
1107 
1108 	memset(qinfo, 0, sizeof(*qinfo));
1109 
1110 	qinfo->conf.offloads = txq_info->offloads;
1111 	qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1112 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1113 	qinfo->nb_desc = txq_info->entries;
1114 }
1115 
1116 /*
1117  * The function is used by the secondary process as well. It must not
1118  * use any process-local pointers from the adapter data.
1119  */
1120 static uint32_t
1121 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1122 {
1123 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1124 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1125 	struct sfc_rxq_info *rxq_info;
1126 
1127 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1128 	rxq_info = &sas->rxq_info[rx_queue_id];
1129 
1130 	if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1131 		return 0;
1132 
1133 	return sap->dp_rx->qdesc_npending(rxq_info->dp);
1134 }
1135 
1136 /*
1137  * The function is used by the secondary process as well. It must not
1138  * use any process-local pointers from the adapter data.
1139  */
1140 static int
1141 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1142 {
1143 	struct sfc_dp_rxq *dp_rxq = queue;
1144 	const struct sfc_dp_rx *dp_rx;
1145 
1146 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1147 
1148 	return offset < dp_rx->qdesc_npending(dp_rxq);
1149 }
1150 
1151 /*
1152  * The function is used by the secondary process as well. It must not
1153  * use any process-local pointers from the adapter data.
1154  */
1155 static int
1156 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1157 {
1158 	struct sfc_dp_rxq *dp_rxq = queue;
1159 	const struct sfc_dp_rx *dp_rx;
1160 
1161 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1162 
1163 	return dp_rx->qdesc_status(dp_rxq, offset);
1164 }
1165 
1166 /*
1167  * The function is used by the secondary process as well. It must not
1168  * use any process-local pointers from the adapter data.
1169  */
1170 static int
1171 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1172 {
1173 	struct sfc_dp_txq *dp_txq = queue;
1174 	const struct sfc_dp_tx *dp_tx;
1175 
1176 	dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1177 
1178 	return dp_tx->qdesc_status(dp_txq, offset);
1179 }
1180 
1181 static int
1182 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1183 {
1184 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1185 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1186 	int rc;
1187 
1188 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1189 
1190 	sfc_adapter_lock(sa);
1191 
1192 	rc = EINVAL;
1193 	if (sa->state != SFC_ADAPTER_STARTED)
1194 		goto fail_not_started;
1195 
1196 	if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1197 		goto fail_not_setup;
1198 
1199 	rc = sfc_rx_qstart(sa, rx_queue_id);
1200 	if (rc != 0)
1201 		goto fail_rx_qstart;
1202 
1203 	sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1204 
1205 	sfc_adapter_unlock(sa);
1206 
1207 	return 0;
1208 
1209 fail_rx_qstart:
1210 fail_not_setup:
1211 fail_not_started:
1212 	sfc_adapter_unlock(sa);
1213 	SFC_ASSERT(rc > 0);
1214 	return -rc;
1215 }
1216 
1217 static int
1218 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1219 {
1220 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1221 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1222 
1223 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1224 
1225 	sfc_adapter_lock(sa);
1226 	sfc_rx_qstop(sa, rx_queue_id);
1227 
1228 	sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1229 
1230 	sfc_adapter_unlock(sa);
1231 
1232 	return 0;
1233 }
1234 
1235 static int
1236 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1237 {
1238 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1239 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1240 	int rc;
1241 
1242 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1243 
1244 	sfc_adapter_lock(sa);
1245 
1246 	rc = EINVAL;
1247 	if (sa->state != SFC_ADAPTER_STARTED)
1248 		goto fail_not_started;
1249 
1250 	if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1251 		goto fail_not_setup;
1252 
1253 	rc = sfc_tx_qstart(sa, tx_queue_id);
1254 	if (rc != 0)
1255 		goto fail_tx_qstart;
1256 
1257 	sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1258 
1259 	sfc_adapter_unlock(sa);
1260 	return 0;
1261 
1262 fail_tx_qstart:
1263 
1264 fail_not_setup:
1265 fail_not_started:
1266 	sfc_adapter_unlock(sa);
1267 	SFC_ASSERT(rc > 0);
1268 	return -rc;
1269 }
1270 
1271 static int
1272 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1273 {
1274 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1275 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
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 	sas->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 = sfc_adapter_by_eth_dev(dev);
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_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1415 	struct sfc_rss *rss = &sas->rss;
1416 
1417 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1418 		return -ENOTSUP;
1419 
1420 	/*
1421 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1422 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1423 	 * flags which corresponds to the active EFX configuration stored
1424 	 * locally in 'sfc_adapter' and kept up-to-date
1425 	 */
1426 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1427 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1428 	if (rss_conf->rss_key != NULL)
1429 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1430 
1431 	return 0;
1432 }
1433 
1434 static int
1435 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1436 			struct rte_eth_rss_conf *rss_conf)
1437 {
1438 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1439 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1440 	unsigned int efx_hash_types;
1441 	int rc = 0;
1442 
1443 	if (sfc_sa2shared(sa)->isolated)
1444 		return -ENOTSUP;
1445 
1446 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1447 		sfc_err(sa, "RSS is not available");
1448 		return -ENOTSUP;
1449 	}
1450 
1451 	if (rss->channels == 0) {
1452 		sfc_err(sa, "RSS is not configured");
1453 		return -EINVAL;
1454 	}
1455 
1456 	if ((rss_conf->rss_key != NULL) &&
1457 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1458 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1459 			sizeof(rss->key));
1460 		return -EINVAL;
1461 	}
1462 
1463 	sfc_adapter_lock(sa);
1464 
1465 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1466 	if (rc != 0)
1467 		goto fail_rx_hf_rte_to_efx;
1468 
1469 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1470 				   rss->hash_alg, efx_hash_types, B_TRUE);
1471 	if (rc != 0)
1472 		goto fail_scale_mode_set;
1473 
1474 	if (rss_conf->rss_key != NULL) {
1475 		if (sa->state == SFC_ADAPTER_STARTED) {
1476 			rc = efx_rx_scale_key_set(sa->nic,
1477 						  EFX_RSS_CONTEXT_DEFAULT,
1478 						  rss_conf->rss_key,
1479 						  sizeof(rss->key));
1480 			if (rc != 0)
1481 				goto fail_scale_key_set;
1482 		}
1483 
1484 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1485 	}
1486 
1487 	rss->hash_types = efx_hash_types;
1488 
1489 	sfc_adapter_unlock(sa);
1490 
1491 	return 0;
1492 
1493 fail_scale_key_set:
1494 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1495 				  EFX_RX_HASHALG_TOEPLITZ,
1496 				  rss->hash_types, B_TRUE) != 0)
1497 		sfc_err(sa, "failed to restore RSS mode");
1498 
1499 fail_scale_mode_set:
1500 fail_rx_hf_rte_to_efx:
1501 	sfc_adapter_unlock(sa);
1502 	return -rc;
1503 }
1504 
1505 /*
1506  * The function is used by the secondary process as well. It must not
1507  * use any process-local pointers from the adapter data.
1508  */
1509 static int
1510 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1511 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1512 		       uint16_t reta_size)
1513 {
1514 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1515 	struct sfc_rss *rss = &sas->rss;
1516 	int entry;
1517 
1518 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1519 		return -ENOTSUP;
1520 
1521 	if (rss->channels == 0)
1522 		return -EINVAL;
1523 
1524 	if (reta_size != EFX_RSS_TBL_SIZE)
1525 		return -EINVAL;
1526 
1527 	for (entry = 0; entry < reta_size; entry++) {
1528 		int grp = entry / RTE_RETA_GROUP_SIZE;
1529 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1530 
1531 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1532 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1533 	}
1534 
1535 	return 0;
1536 }
1537 
1538 static int
1539 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1540 			struct rte_eth_rss_reta_entry64 *reta_conf,
1541 			uint16_t reta_size)
1542 {
1543 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1544 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1545 	unsigned int *rss_tbl_new;
1546 	uint16_t entry;
1547 	int rc = 0;
1548 
1549 
1550 	if (sfc_sa2shared(sa)->isolated)
1551 		return -ENOTSUP;
1552 
1553 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1554 		sfc_err(sa, "RSS is not available");
1555 		return -ENOTSUP;
1556 	}
1557 
1558 	if (rss->channels == 0) {
1559 		sfc_err(sa, "RSS is not configured");
1560 		return -EINVAL;
1561 	}
1562 
1563 	if (reta_size != EFX_RSS_TBL_SIZE) {
1564 		sfc_err(sa, "RETA size is wrong (should be %u)",
1565 			EFX_RSS_TBL_SIZE);
1566 		return -EINVAL;
1567 	}
1568 
1569 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1570 	if (rss_tbl_new == NULL)
1571 		return -ENOMEM;
1572 
1573 	sfc_adapter_lock(sa);
1574 
1575 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1576 
1577 	for (entry = 0; entry < reta_size; entry++) {
1578 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1579 		struct rte_eth_rss_reta_entry64 *grp;
1580 
1581 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1582 
1583 		if (grp->mask & (1ull << grp_idx)) {
1584 			if (grp->reta[grp_idx] >= rss->channels) {
1585 				rc = EINVAL;
1586 				goto bad_reta_entry;
1587 			}
1588 			rss_tbl_new[entry] = grp->reta[grp_idx];
1589 		}
1590 	}
1591 
1592 	if (sa->state == SFC_ADAPTER_STARTED) {
1593 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1594 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1595 		if (rc != 0)
1596 			goto fail_scale_tbl_set;
1597 	}
1598 
1599 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1600 
1601 fail_scale_tbl_set:
1602 bad_reta_entry:
1603 	sfc_adapter_unlock(sa);
1604 
1605 	rte_free(rss_tbl_new);
1606 
1607 	SFC_ASSERT(rc >= 0);
1608 	return -rc;
1609 }
1610 
1611 static int
1612 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1613 		    enum rte_filter_op filter_op,
1614 		    void *arg)
1615 {
1616 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1617 	int rc = ENOTSUP;
1618 
1619 	sfc_log_init(sa, "entry");
1620 
1621 	switch (filter_type) {
1622 	case RTE_ETH_FILTER_NONE:
1623 		sfc_err(sa, "Global filters configuration not supported");
1624 		break;
1625 	case RTE_ETH_FILTER_MACVLAN:
1626 		sfc_err(sa, "MACVLAN filters not supported");
1627 		break;
1628 	case RTE_ETH_FILTER_ETHERTYPE:
1629 		sfc_err(sa, "EtherType filters not supported");
1630 		break;
1631 	case RTE_ETH_FILTER_FLEXIBLE:
1632 		sfc_err(sa, "Flexible filters not supported");
1633 		break;
1634 	case RTE_ETH_FILTER_SYN:
1635 		sfc_err(sa, "SYN filters not supported");
1636 		break;
1637 	case RTE_ETH_FILTER_NTUPLE:
1638 		sfc_err(sa, "NTUPLE filters not supported");
1639 		break;
1640 	case RTE_ETH_FILTER_TUNNEL:
1641 		sfc_err(sa, "Tunnel filters not supported");
1642 		break;
1643 	case RTE_ETH_FILTER_FDIR:
1644 		sfc_err(sa, "Flow Director filters not supported");
1645 		break;
1646 	case RTE_ETH_FILTER_HASH:
1647 		sfc_err(sa, "Hash filters not supported");
1648 		break;
1649 	case RTE_ETH_FILTER_GENERIC:
1650 		if (filter_op != RTE_ETH_FILTER_GET) {
1651 			rc = EINVAL;
1652 		} else {
1653 			*(const void **)arg = &sfc_flow_ops;
1654 			rc = 0;
1655 		}
1656 		break;
1657 	default:
1658 		sfc_err(sa, "Unknown filter type %u", filter_type);
1659 		break;
1660 	}
1661 
1662 	sfc_log_init(sa, "exit: %d", -rc);
1663 	SFC_ASSERT(rc >= 0);
1664 	return -rc;
1665 }
1666 
1667 static int
1668 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1669 {
1670 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1671 
1672 	/*
1673 	 * If Rx datapath does not provide callback to check mempool,
1674 	 * all pools are supported.
1675 	 */
1676 	if (sap->dp_rx->pool_ops_supported == NULL)
1677 		return 1;
1678 
1679 	return sap->dp_rx->pool_ops_supported(pool);
1680 }
1681 
1682 static const struct eth_dev_ops sfc_eth_dev_ops = {
1683 	.dev_configure			= sfc_dev_configure,
1684 	.dev_start			= sfc_dev_start,
1685 	.dev_stop			= sfc_dev_stop,
1686 	.dev_set_link_up		= sfc_dev_set_link_up,
1687 	.dev_set_link_down		= sfc_dev_set_link_down,
1688 	.dev_close			= sfc_dev_close,
1689 	.promiscuous_enable		= sfc_dev_promisc_enable,
1690 	.promiscuous_disable		= sfc_dev_promisc_disable,
1691 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1692 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1693 	.link_update			= sfc_dev_link_update,
1694 	.stats_get			= sfc_stats_get,
1695 	.stats_reset			= sfc_stats_reset,
1696 	.xstats_get			= sfc_xstats_get,
1697 	.xstats_reset			= sfc_stats_reset,
1698 	.xstats_get_names		= sfc_xstats_get_names,
1699 	.dev_infos_get			= sfc_dev_infos_get,
1700 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1701 	.mtu_set			= sfc_dev_set_mtu,
1702 	.rx_queue_start			= sfc_rx_queue_start,
1703 	.rx_queue_stop			= sfc_rx_queue_stop,
1704 	.tx_queue_start			= sfc_tx_queue_start,
1705 	.tx_queue_stop			= sfc_tx_queue_stop,
1706 	.rx_queue_setup			= sfc_rx_queue_setup,
1707 	.rx_queue_release		= sfc_rx_queue_release,
1708 	.rx_queue_count			= sfc_rx_queue_count,
1709 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1710 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1711 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1712 	.tx_queue_setup			= sfc_tx_queue_setup,
1713 	.tx_queue_release		= sfc_tx_queue_release,
1714 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1715 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1716 	.mac_addr_set			= sfc_mac_addr_set,
1717 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1718 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1719 	.reta_update			= sfc_dev_rss_reta_update,
1720 	.reta_query			= sfc_dev_rss_reta_query,
1721 	.rss_hash_update		= sfc_dev_rss_hash_update,
1722 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1723 	.filter_ctrl			= sfc_dev_filter_ctrl,
1724 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1725 	.rxq_info_get			= sfc_rx_queue_info_get,
1726 	.txq_info_get			= sfc_tx_queue_info_get,
1727 	.fw_version_get			= sfc_fw_version_get,
1728 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1729 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1730 	.pool_ops_supported		= sfc_pool_ops_supported,
1731 };
1732 
1733 /**
1734  * Duplicate a string in potentially shared memory required for
1735  * multi-process support.
1736  *
1737  * strdup() allocates from process-local heap/memory.
1738  */
1739 static char *
1740 sfc_strdup(const char *str)
1741 {
1742 	size_t size;
1743 	char *copy;
1744 
1745 	if (str == NULL)
1746 		return NULL;
1747 
1748 	size = strlen(str) + 1;
1749 	copy = rte_malloc(__func__, size, 0);
1750 	if (copy != NULL)
1751 		rte_memcpy(copy, str, size);
1752 
1753 	return copy;
1754 }
1755 
1756 static int
1757 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1758 {
1759 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1760 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1761 	const struct sfc_dp_rx *dp_rx;
1762 	const struct sfc_dp_tx *dp_tx;
1763 	const efx_nic_cfg_t *encp;
1764 	unsigned int avail_caps = 0;
1765 	const char *rx_name = NULL;
1766 	const char *tx_name = NULL;
1767 	int rc;
1768 
1769 	switch (sa->family) {
1770 	case EFX_FAMILY_HUNTINGTON:
1771 	case EFX_FAMILY_MEDFORD:
1772 	case EFX_FAMILY_MEDFORD2:
1773 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1774 		break;
1775 	default:
1776 		break;
1777 	}
1778 
1779 	encp = efx_nic_cfg_get(sa->nic);
1780 	if (encp->enc_rx_es_super_buffer_supported)
1781 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1782 
1783 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1784 				sfc_kvarg_string_handler, &rx_name);
1785 	if (rc != 0)
1786 		goto fail_kvarg_rx_datapath;
1787 
1788 	if (rx_name != NULL) {
1789 		dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1790 		if (dp_rx == NULL) {
1791 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1792 			rc = ENOENT;
1793 			goto fail_dp_rx;
1794 		}
1795 		if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1796 			sfc_err(sa,
1797 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1798 				rx_name);
1799 			rc = EINVAL;
1800 			goto fail_dp_rx_caps;
1801 		}
1802 	} else {
1803 		dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1804 		if (dp_rx == NULL) {
1805 			sfc_err(sa, "Rx datapath by caps %#x not found",
1806 				avail_caps);
1807 			rc = ENOENT;
1808 			goto fail_dp_rx;
1809 		}
1810 	}
1811 
1812 	sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1813 	if (sas->dp_rx_name == NULL) {
1814 		rc = ENOMEM;
1815 		goto fail_dp_rx_name;
1816 	}
1817 
1818 	sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1819 
1820 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1821 				sfc_kvarg_string_handler, &tx_name);
1822 	if (rc != 0)
1823 		goto fail_kvarg_tx_datapath;
1824 
1825 	if (tx_name != NULL) {
1826 		dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1827 		if (dp_tx == NULL) {
1828 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1829 			rc = ENOENT;
1830 			goto fail_dp_tx;
1831 		}
1832 		if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1833 			sfc_err(sa,
1834 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1835 				tx_name);
1836 			rc = EINVAL;
1837 			goto fail_dp_tx_caps;
1838 		}
1839 	} else {
1840 		dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1841 		if (dp_tx == NULL) {
1842 			sfc_err(sa, "Tx datapath by caps %#x not found",
1843 				avail_caps);
1844 			rc = ENOENT;
1845 			goto fail_dp_tx;
1846 		}
1847 	}
1848 
1849 	sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1850 	if (sas->dp_tx_name == NULL) {
1851 		rc = ENOMEM;
1852 		goto fail_dp_tx_name;
1853 	}
1854 
1855 	sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1856 
1857 	sa->priv.dp_rx = dp_rx;
1858 	sa->priv.dp_tx = dp_tx;
1859 
1860 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1861 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1862 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1863 
1864 	dev->dev_ops = &sfc_eth_dev_ops;
1865 
1866 	return 0;
1867 
1868 fail_dp_tx_name:
1869 fail_dp_tx_caps:
1870 fail_dp_tx:
1871 fail_kvarg_tx_datapath:
1872 	rte_free(sas->dp_rx_name);
1873 	sas->dp_rx_name = NULL;
1874 
1875 fail_dp_rx_name:
1876 fail_dp_rx_caps:
1877 fail_dp_rx:
1878 fail_kvarg_rx_datapath:
1879 	return rc;
1880 }
1881 
1882 static void
1883 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1884 {
1885 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1886 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1887 
1888 	dev->dev_ops = NULL;
1889 	dev->tx_pkt_prepare = NULL;
1890 	dev->rx_pkt_burst = NULL;
1891 	dev->tx_pkt_burst = NULL;
1892 
1893 	rte_free(sas->dp_tx_name);
1894 	sas->dp_tx_name = NULL;
1895 	sa->priv.dp_tx = NULL;
1896 
1897 	rte_free(sas->dp_rx_name);
1898 	sas->dp_rx_name = NULL;
1899 	sa->priv.dp_rx = NULL;
1900 }
1901 
1902 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1903 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1904 	.rx_queue_count			= sfc_rx_queue_count,
1905 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1906 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1907 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1908 	.reta_query			= sfc_dev_rss_reta_query,
1909 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1910 	.rxq_info_get			= sfc_rx_queue_info_get,
1911 	.txq_info_get			= sfc_tx_queue_info_get,
1912 };
1913 
1914 static int
1915 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
1916 {
1917 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1918 	struct sfc_adapter_priv *sap;
1919 	const struct sfc_dp_rx *dp_rx;
1920 	const struct sfc_dp_tx *dp_tx;
1921 	int rc;
1922 
1923 	/*
1924 	 * Allocate process private data from heap, since it should not
1925 	 * be located in shared memory allocated using rte_malloc() API.
1926 	 */
1927 	sap = calloc(1, sizeof(*sap));
1928 	if (sap == NULL) {
1929 		rc = ENOMEM;
1930 		goto fail_alloc_priv;
1931 	}
1932 
1933 	sap->logtype_main = logtype_main;
1934 
1935 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
1936 	if (dp_rx == NULL) {
1937 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1938 			"cannot find %s Rx datapath", sas->dp_rx_name);
1939 		rc = ENOENT;
1940 		goto fail_dp_rx;
1941 	}
1942 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1943 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1944 			"%s Rx datapath does not support multi-process",
1945 			sas->dp_rx_name);
1946 		rc = EINVAL;
1947 		goto fail_dp_rx_multi_process;
1948 	}
1949 
1950 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
1951 	if (dp_tx == NULL) {
1952 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1953 			"cannot find %s Tx datapath", sas->dp_tx_name);
1954 		rc = ENOENT;
1955 		goto fail_dp_tx;
1956 	}
1957 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1958 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1959 			"%s Tx datapath does not support multi-process",
1960 			sas->dp_tx_name);
1961 		rc = EINVAL;
1962 		goto fail_dp_tx_multi_process;
1963 	}
1964 
1965 	sap->dp_rx = dp_rx;
1966 	sap->dp_tx = dp_tx;
1967 
1968 	dev->process_private = sap;
1969 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1970 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1971 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1972 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1973 
1974 	return 0;
1975 
1976 fail_dp_tx_multi_process:
1977 fail_dp_tx:
1978 fail_dp_rx_multi_process:
1979 fail_dp_rx:
1980 	free(sap);
1981 
1982 fail_alloc_priv:
1983 	return rc;
1984 }
1985 
1986 static void
1987 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1988 {
1989 	free(dev->process_private);
1990 	dev->process_private = NULL;
1991 	dev->dev_ops = NULL;
1992 	dev->tx_pkt_prepare = NULL;
1993 	dev->tx_pkt_burst = NULL;
1994 	dev->rx_pkt_burst = NULL;
1995 }
1996 
1997 static void
1998 sfc_register_dp(void)
1999 {
2000 	/* Register once */
2001 	if (TAILQ_EMPTY(&sfc_dp_head)) {
2002 		/* Prefer EF10 datapath */
2003 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2004 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2005 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2006 
2007 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2008 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2009 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2010 	}
2011 }
2012 
2013 static int
2014 sfc_eth_dev_init(struct rte_eth_dev *dev)
2015 {
2016 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2017 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2018 	uint32_t logtype_main;
2019 	struct sfc_adapter *sa;
2020 	int rc;
2021 	const efx_nic_cfg_t *encp;
2022 	const struct ether_addr *from;
2023 
2024 	sfc_register_dp();
2025 
2026 	logtype_main = sfc_register_logtype(&pci_dev->addr,
2027 					    SFC_LOGTYPE_MAIN_STR,
2028 					    RTE_LOG_NOTICE);
2029 
2030 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2031 		return -sfc_eth_dev_secondary_init(dev, logtype_main);
2032 
2033 	/* Required for logging */
2034 	sas->pci_addr = pci_dev->addr;
2035 	sas->port_id = dev->data->port_id;
2036 
2037 	/*
2038 	 * Allocate process private data from heap, since it should not
2039 	 * be located in shared memory allocated using rte_malloc() API.
2040 	 */
2041 	sa = calloc(1, sizeof(*sa));
2042 	if (sa == NULL) {
2043 		rc = ENOMEM;
2044 		goto fail_alloc_sa;
2045 	}
2046 
2047 	dev->process_private = sa;
2048 
2049 	/* Required for logging */
2050 	sa->priv.shared = sas;
2051 	sa->priv.logtype_main = logtype_main;
2052 
2053 	sa->eth_dev = dev;
2054 
2055 	/* Copy PCI device info to the dev->data */
2056 	rte_eth_copy_pci_info(dev, pci_dev);
2057 
2058 	rc = sfc_kvargs_parse(sa);
2059 	if (rc != 0)
2060 		goto fail_kvargs_parse;
2061 
2062 	sfc_log_init(sa, "entry");
2063 
2064 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
2065 	if (dev->data->mac_addrs == NULL) {
2066 		rc = ENOMEM;
2067 		goto fail_mac_addrs;
2068 	}
2069 
2070 	sfc_adapter_lock_init(sa);
2071 	sfc_adapter_lock(sa);
2072 
2073 	sfc_log_init(sa, "probing");
2074 	rc = sfc_probe(sa);
2075 	if (rc != 0)
2076 		goto fail_probe;
2077 
2078 	sfc_log_init(sa, "set device ops");
2079 	rc = sfc_eth_dev_set_ops(dev);
2080 	if (rc != 0)
2081 		goto fail_set_ops;
2082 
2083 	sfc_log_init(sa, "attaching");
2084 	rc = sfc_attach(sa);
2085 	if (rc != 0)
2086 		goto fail_attach;
2087 
2088 	encp = efx_nic_cfg_get(sa->nic);
2089 
2090 	/*
2091 	 * The arguments are really reverse order in comparison to
2092 	 * Linux kernel. Copy from NIC config to Ethernet device data.
2093 	 */
2094 	from = (const struct ether_addr *)(encp->enc_mac_addr);
2095 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
2096 
2097 	sfc_adapter_unlock(sa);
2098 
2099 	sfc_log_init(sa, "done");
2100 	return 0;
2101 
2102 fail_attach:
2103 	sfc_eth_dev_clear_ops(dev);
2104 
2105 fail_set_ops:
2106 	sfc_unprobe(sa);
2107 
2108 fail_probe:
2109 	sfc_adapter_unlock(sa);
2110 	sfc_adapter_lock_fini(sa);
2111 	rte_free(dev->data->mac_addrs);
2112 	dev->data->mac_addrs = NULL;
2113 
2114 fail_mac_addrs:
2115 	sfc_kvargs_cleanup(sa);
2116 
2117 fail_kvargs_parse:
2118 	sfc_log_init(sa, "failed %d", rc);
2119 	dev->process_private = NULL;
2120 	free(sa);
2121 
2122 fail_alloc_sa:
2123 	SFC_ASSERT(rc > 0);
2124 	return -rc;
2125 }
2126 
2127 static int
2128 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2129 {
2130 	struct sfc_adapter *sa;
2131 
2132 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2133 		sfc_eth_dev_secondary_clear_ops(dev);
2134 		return 0;
2135 	}
2136 
2137 	sa = sfc_adapter_by_eth_dev(dev);
2138 	sfc_log_init(sa, "entry");
2139 
2140 	sfc_adapter_lock(sa);
2141 
2142 	sfc_eth_dev_clear_ops(dev);
2143 
2144 	sfc_detach(sa);
2145 	sfc_unprobe(sa);
2146 
2147 	sfc_kvargs_cleanup(sa);
2148 
2149 	sfc_adapter_unlock(sa);
2150 	sfc_adapter_lock_fini(sa);
2151 
2152 	sfc_log_init(sa, "done");
2153 
2154 	/* Required for logging, so cleanup last */
2155 	sa->eth_dev = NULL;
2156 
2157 	dev->process_private = NULL;
2158 	free(sa);
2159 
2160 	return 0;
2161 }
2162 
2163 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2164 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2165 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2166 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2167 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2168 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2169 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2170 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2171 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2172 	{ .vendor_id = 0 /* sentinel */ }
2173 };
2174 
2175 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2176 	struct rte_pci_device *pci_dev)
2177 {
2178 	return rte_eth_dev_pci_generic_probe(pci_dev,
2179 		sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2180 }
2181 
2182 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2183 {
2184 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2185 }
2186 
2187 static struct rte_pci_driver sfc_efx_pmd = {
2188 	.id_table = pci_id_sfc_efx_map,
2189 	.drv_flags =
2190 		RTE_PCI_DRV_INTR_LSC |
2191 		RTE_PCI_DRV_NEED_MAPPING,
2192 	.probe = sfc_eth_dev_pci_probe,
2193 	.remove = sfc_eth_dev_pci_remove,
2194 };
2195 
2196 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2197 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2198 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2199 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2200 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2201 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2202 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2203 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2204 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2205 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2206 
2207 RTE_INIT(sfc_driver_register_logtype)
2208 {
2209 	int ret;
2210 
2211 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2212 						   RTE_LOG_NOTICE);
2213 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2214 }
2215