xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 7eb7588c99ab5fd0f31032b29cac74b57e1de2ff)
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 	struct sfc_adapter *sa = dev->data->dev_private;
88 	struct sfc_rss *rss = &sa->rss;
89 	uint64_t txq_offloads_def = 0;
90 
91 	sfc_log_init(sa, "entry");
92 
93 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
94 
95 	/* Autonegotiation may be disabled */
96 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
97 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
98 		dev_info->speed_capa |= ETH_LINK_SPEED_1G;
99 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
100 		dev_info->speed_capa |= ETH_LINK_SPEED_10G;
101 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_25000FDX)
102 		dev_info->speed_capa |= ETH_LINK_SPEED_25G;
103 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
104 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
105 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_50000FDX)
106 		dev_info->speed_capa |= ETH_LINK_SPEED_50G;
107 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_100000FDX)
108 		dev_info->speed_capa |= ETH_LINK_SPEED_100G;
109 
110 	dev_info->max_rx_queues = sa->rxq_max;
111 	dev_info->max_tx_queues = sa->txq_max;
112 
113 	/* By default packets are dropped if no descriptors are available */
114 	dev_info->default_rxconf.rx_drop_en = 1;
115 
116 	dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
117 
118 	/*
119 	 * rx_offload_capa includes both device and queue offloads since
120 	 * the latter may be requested on a per device basis which makes
121 	 * sense when some offloads are needed to be set on all queues.
122 	 */
123 	dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
124 				    dev_info->rx_queue_offload_capa;
125 
126 	dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
127 
128 	/*
129 	 * tx_offload_capa includes both device and queue offloads since
130 	 * the latter may be requested on a per device basis which makes
131 	 * sense when some offloads are needed to be set on all queues.
132 	 */
133 	dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
134 				    dev_info->tx_queue_offload_capa;
135 
136 	if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
137 		txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
138 
139 	dev_info->default_txconf.offloads |= txq_offloads_def;
140 
141 	if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
142 		uint64_t rte_hf = 0;
143 		unsigned int i;
144 
145 		for (i = 0; i < rss->hf_map_nb_entries; ++i)
146 			rte_hf |= rss->hf_map[i].rte;
147 
148 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
149 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
150 		dev_info->flow_type_rss_offloads = rte_hf;
151 	}
152 
153 	/* Initialize to hardware limits */
154 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
155 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
156 	/* The RXQ hardware requires that the descriptor count is a power
157 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
158 	 */
159 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
160 
161 	/* Initialize to hardware limits */
162 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
163 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
164 	/*
165 	 * The TXQ hardware requires that the descriptor count is a power
166 	 * of 2, but tx_desc_lim cannot properly describe that constraint
167 	 */
168 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
169 
170 	if (sa->dp_rx->get_dev_info != NULL)
171 		sa->dp_rx->get_dev_info(dev_info);
172 	if (sa->dp_tx->get_dev_info != NULL)
173 		sa->dp_tx->get_dev_info(dev_info);
174 
175 	dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
176 			     RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
177 }
178 
179 static const uint32_t *
180 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
181 {
182 	struct sfc_adapter *sa = dev->data->dev_private;
183 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
184 	uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
185 
186 	return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
187 }
188 
189 static int
190 sfc_dev_configure(struct rte_eth_dev *dev)
191 {
192 	struct rte_eth_dev_data *dev_data = dev->data;
193 	struct sfc_adapter *sa = dev_data->dev_private;
194 	int rc;
195 
196 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
197 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
198 
199 	sfc_adapter_lock(sa);
200 	switch (sa->state) {
201 	case SFC_ADAPTER_CONFIGURED:
202 		/* FALLTHROUGH */
203 	case SFC_ADAPTER_INITIALIZED:
204 		rc = sfc_configure(sa);
205 		break;
206 	default:
207 		sfc_err(sa, "unexpected adapter state %u to configure",
208 			sa->state);
209 		rc = EINVAL;
210 		break;
211 	}
212 	sfc_adapter_unlock(sa);
213 
214 	sfc_log_init(sa, "done %d", rc);
215 	SFC_ASSERT(rc >= 0);
216 	return -rc;
217 }
218 
219 static int
220 sfc_dev_start(struct rte_eth_dev *dev)
221 {
222 	struct sfc_adapter *sa = dev->data->dev_private;
223 	int rc;
224 
225 	sfc_log_init(sa, "entry");
226 
227 	sfc_adapter_lock(sa);
228 	rc = sfc_start(sa);
229 	sfc_adapter_unlock(sa);
230 
231 	sfc_log_init(sa, "done %d", rc);
232 	SFC_ASSERT(rc >= 0);
233 	return -rc;
234 }
235 
236 static int
237 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
238 {
239 	struct sfc_adapter *sa = dev->data->dev_private;
240 	struct rte_eth_link current_link;
241 	int ret;
242 
243 	sfc_log_init(sa, "entry");
244 
245 	if (sa->state != SFC_ADAPTER_STARTED) {
246 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
247 	} else if (wait_to_complete) {
248 		efx_link_mode_t link_mode;
249 
250 		if (efx_port_poll(sa->nic, &link_mode) != 0)
251 			link_mode = EFX_LINK_UNKNOWN;
252 		sfc_port_link_mode_to_info(link_mode, &current_link);
253 
254 	} else {
255 		sfc_ev_mgmt_qpoll(sa);
256 		rte_eth_linkstatus_get(dev, &current_link);
257 	}
258 
259 	ret = rte_eth_linkstatus_set(dev, &current_link);
260 	if (ret == 0)
261 		sfc_notice(sa, "Link status is %s",
262 			   current_link.link_status ? "UP" : "DOWN");
263 
264 	return ret;
265 }
266 
267 static void
268 sfc_dev_stop(struct rte_eth_dev *dev)
269 {
270 	struct sfc_adapter *sa = dev->data->dev_private;
271 
272 	sfc_log_init(sa, "entry");
273 
274 	sfc_adapter_lock(sa);
275 	sfc_stop(sa);
276 	sfc_adapter_unlock(sa);
277 
278 	sfc_log_init(sa, "done");
279 }
280 
281 static int
282 sfc_dev_set_link_up(struct rte_eth_dev *dev)
283 {
284 	struct sfc_adapter *sa = dev->data->dev_private;
285 	int rc;
286 
287 	sfc_log_init(sa, "entry");
288 
289 	sfc_adapter_lock(sa);
290 	rc = sfc_start(sa);
291 	sfc_adapter_unlock(sa);
292 
293 	SFC_ASSERT(rc >= 0);
294 	return -rc;
295 }
296 
297 static int
298 sfc_dev_set_link_down(struct rte_eth_dev *dev)
299 {
300 	struct sfc_adapter *sa = dev->data->dev_private;
301 
302 	sfc_log_init(sa, "entry");
303 
304 	sfc_adapter_lock(sa);
305 	sfc_stop(sa);
306 	sfc_adapter_unlock(sa);
307 
308 	return 0;
309 }
310 
311 static void
312 sfc_dev_close(struct rte_eth_dev *dev)
313 {
314 	struct sfc_adapter *sa = dev->data->dev_private;
315 
316 	sfc_log_init(sa, "entry");
317 
318 	sfc_adapter_lock(sa);
319 	switch (sa->state) {
320 	case SFC_ADAPTER_STARTED:
321 		sfc_stop(sa);
322 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
323 		/* FALLTHROUGH */
324 	case SFC_ADAPTER_CONFIGURED:
325 		sfc_close(sa);
326 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
327 		/* FALLTHROUGH */
328 	case SFC_ADAPTER_INITIALIZED:
329 		break;
330 	default:
331 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
332 		break;
333 	}
334 	sfc_adapter_unlock(sa);
335 
336 	sfc_log_init(sa, "done");
337 }
338 
339 static void
340 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
341 		   boolean_t enabled)
342 {
343 	struct sfc_port *port;
344 	boolean_t *toggle;
345 	struct sfc_adapter *sa = dev->data->dev_private;
346 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
347 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
348 
349 	sfc_adapter_lock(sa);
350 
351 	port = &sa->port;
352 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
353 
354 	if (*toggle != enabled) {
355 		*toggle = enabled;
356 
357 		if (port->isolated) {
358 			sfc_warn(sa, "isolated mode is active on the port");
359 			sfc_warn(sa, "the change is to be applied on the next "
360 				     "start provided that isolated mode is "
361 				     "disabled prior the next start");
362 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
363 			   (sfc_set_rx_mode(sa) != 0)) {
364 			*toggle = !(enabled);
365 			sfc_warn(sa, "Failed to %s %s mode",
366 				 ((enabled) ? "enable" : "disable"), desc);
367 		}
368 	}
369 
370 	sfc_adapter_unlock(sa);
371 }
372 
373 static void
374 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
375 {
376 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
377 }
378 
379 static void
380 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
381 {
382 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
383 }
384 
385 static void
386 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
387 {
388 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
389 }
390 
391 static void
392 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
393 {
394 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
395 }
396 
397 static int
398 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
399 		   uint16_t nb_rx_desc, unsigned int socket_id,
400 		   const struct rte_eth_rxconf *rx_conf,
401 		   struct rte_mempool *mb_pool)
402 {
403 	struct sfc_adapter *sa = dev->data->dev_private;
404 	int rc;
405 
406 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
407 		     rx_queue_id, nb_rx_desc, socket_id);
408 
409 	sfc_adapter_lock(sa);
410 
411 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
412 			  rx_conf, mb_pool);
413 	if (rc != 0)
414 		goto fail_rx_qinit;
415 
416 	dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
417 
418 	sfc_adapter_unlock(sa);
419 
420 	return 0;
421 
422 fail_rx_qinit:
423 	sfc_adapter_unlock(sa);
424 	SFC_ASSERT(rc > 0);
425 	return -rc;
426 }
427 
428 static void
429 sfc_rx_queue_release(void *queue)
430 {
431 	struct sfc_dp_rxq *dp_rxq = queue;
432 	struct sfc_rxq *rxq;
433 	struct sfc_adapter *sa;
434 	unsigned int sw_index;
435 
436 	if (dp_rxq == NULL)
437 		return;
438 
439 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
440 	sa = rxq->evq->sa;
441 	sfc_adapter_lock(sa);
442 
443 	sw_index = sfc_rxq_sw_index(rxq);
444 
445 	sfc_log_init(sa, "RxQ=%u", sw_index);
446 
447 	sfc_rx_qfini(sa, sw_index);
448 
449 	sfc_adapter_unlock(sa);
450 }
451 
452 static int
453 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
454 		   uint16_t nb_tx_desc, unsigned int socket_id,
455 		   const struct rte_eth_txconf *tx_conf)
456 {
457 	struct sfc_adapter *sa = dev->data->dev_private;
458 	int rc;
459 
460 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
461 		     tx_queue_id, nb_tx_desc, socket_id);
462 
463 	sfc_adapter_lock(sa);
464 
465 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
466 	if (rc != 0)
467 		goto fail_tx_qinit;
468 
469 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
470 
471 	sfc_adapter_unlock(sa);
472 	return 0;
473 
474 fail_tx_qinit:
475 	sfc_adapter_unlock(sa);
476 	SFC_ASSERT(rc > 0);
477 	return -rc;
478 }
479 
480 static void
481 sfc_tx_queue_release(void *queue)
482 {
483 	struct sfc_dp_txq *dp_txq = queue;
484 	struct sfc_txq *txq;
485 	unsigned int sw_index;
486 	struct sfc_adapter *sa;
487 
488 	if (dp_txq == NULL)
489 		return;
490 
491 	txq = sfc_txq_by_dp_txq(dp_txq);
492 	sw_index = sfc_txq_sw_index(txq);
493 
494 	SFC_ASSERT(txq->evq != NULL);
495 	sa = txq->evq->sa;
496 
497 	sfc_log_init(sa, "TxQ = %u", sw_index);
498 
499 	sfc_adapter_lock(sa);
500 
501 	SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
502 	sa->eth_dev->data->tx_queues[sw_index] = NULL;
503 
504 	sfc_tx_qfini(sa, sw_index);
505 
506 	sfc_adapter_unlock(sa);
507 }
508 
509 static int
510 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
511 {
512 	struct sfc_adapter *sa = dev->data->dev_private;
513 	struct sfc_port *port = &sa->port;
514 	uint64_t *mac_stats;
515 	int ret;
516 
517 	rte_spinlock_lock(&port->mac_stats_lock);
518 
519 	ret = sfc_port_update_mac_stats(sa);
520 	if (ret != 0)
521 		goto unlock;
522 
523 	mac_stats = port->mac_stats_buf;
524 
525 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
526 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
527 		stats->ipackets =
528 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
529 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
530 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
531 		stats->opackets =
532 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
533 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
534 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
535 		stats->ibytes =
536 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
537 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
538 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
539 		stats->obytes =
540 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
541 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
542 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
543 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
544 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
545 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
546 	} else {
547 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
548 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
549 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
550 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
551 		/*
552 		 * Take into account stats which are whenever supported
553 		 * on EF10. If some stat is not supported by current
554 		 * firmware variant or HW revision, it is guaranteed
555 		 * to be zero in mac_stats.
556 		 */
557 		stats->imissed =
558 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
559 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
560 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
561 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
562 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
563 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
564 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
565 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
566 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
567 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
568 		stats->ierrors =
569 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
570 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
571 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
572 		/* no oerrors counters supported on EF10 */
573 	}
574 
575 unlock:
576 	rte_spinlock_unlock(&port->mac_stats_lock);
577 	SFC_ASSERT(ret >= 0);
578 	return -ret;
579 }
580 
581 static void
582 sfc_stats_reset(struct rte_eth_dev *dev)
583 {
584 	struct sfc_adapter *sa = dev->data->dev_private;
585 	struct sfc_port *port = &sa->port;
586 	int rc;
587 
588 	if (sa->state != SFC_ADAPTER_STARTED) {
589 		/*
590 		 * The operation cannot be done if port is not started; it
591 		 * will be scheduled to be done during the next port start
592 		 */
593 		port->mac_stats_reset_pending = B_TRUE;
594 		return;
595 	}
596 
597 	rc = sfc_port_reset_mac_stats(sa);
598 	if (rc != 0)
599 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
600 }
601 
602 static int
603 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
604 	       unsigned int xstats_count)
605 {
606 	struct sfc_adapter *sa = dev->data->dev_private;
607 	struct sfc_port *port = &sa->port;
608 	uint64_t *mac_stats;
609 	int rc;
610 	unsigned int i;
611 	int nstats = 0;
612 
613 	rte_spinlock_lock(&port->mac_stats_lock);
614 
615 	rc = sfc_port_update_mac_stats(sa);
616 	if (rc != 0) {
617 		SFC_ASSERT(rc > 0);
618 		nstats = -rc;
619 		goto unlock;
620 	}
621 
622 	mac_stats = port->mac_stats_buf;
623 
624 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
625 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
626 			if (xstats != NULL && nstats < (int)xstats_count) {
627 				xstats[nstats].id = nstats;
628 				xstats[nstats].value = mac_stats[i];
629 			}
630 			nstats++;
631 		}
632 	}
633 
634 unlock:
635 	rte_spinlock_unlock(&port->mac_stats_lock);
636 
637 	return nstats;
638 }
639 
640 static int
641 sfc_xstats_get_names(struct rte_eth_dev *dev,
642 		     struct rte_eth_xstat_name *xstats_names,
643 		     unsigned int xstats_count)
644 {
645 	struct sfc_adapter *sa = dev->data->dev_private;
646 	struct sfc_port *port = &sa->port;
647 	unsigned int i;
648 	unsigned int nstats = 0;
649 
650 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
651 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
652 			if (xstats_names != NULL && nstats < xstats_count)
653 				strlcpy(xstats_names[nstats].name,
654 					efx_mac_stat_name(sa->nic, i),
655 					sizeof(xstats_names[0].name));
656 			nstats++;
657 		}
658 	}
659 
660 	return nstats;
661 }
662 
663 static int
664 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
665 		     uint64_t *values, unsigned int n)
666 {
667 	struct sfc_adapter *sa = dev->data->dev_private;
668 	struct sfc_port *port = &sa->port;
669 	uint64_t *mac_stats;
670 	unsigned int nb_supported = 0;
671 	unsigned int nb_written = 0;
672 	unsigned int i;
673 	int ret;
674 	int rc;
675 
676 	if (unlikely(values == NULL) ||
677 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
678 		return port->mac_stats_nb_supported;
679 
680 	rte_spinlock_lock(&port->mac_stats_lock);
681 
682 	rc = sfc_port_update_mac_stats(sa);
683 	if (rc != 0) {
684 		SFC_ASSERT(rc > 0);
685 		ret = -rc;
686 		goto unlock;
687 	}
688 
689 	mac_stats = port->mac_stats_buf;
690 
691 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
692 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
693 			continue;
694 
695 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
696 			values[nb_written++] = mac_stats[i];
697 
698 		++nb_supported;
699 	}
700 
701 	ret = nb_written;
702 
703 unlock:
704 	rte_spinlock_unlock(&port->mac_stats_lock);
705 
706 	return ret;
707 }
708 
709 static int
710 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
711 			   struct rte_eth_xstat_name *xstats_names,
712 			   const uint64_t *ids, unsigned int size)
713 {
714 	struct sfc_adapter *sa = dev->data->dev_private;
715 	struct sfc_port *port = &sa->port;
716 	unsigned int nb_supported = 0;
717 	unsigned int nb_written = 0;
718 	unsigned int i;
719 
720 	if (unlikely(xstats_names == NULL) ||
721 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
722 		return port->mac_stats_nb_supported;
723 
724 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
725 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
726 			continue;
727 
728 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
729 			char *name = xstats_names[nb_written++].name;
730 
731 			strlcpy(name, efx_mac_stat_name(sa->nic, i),
732 				sizeof(xstats_names[0].name));
733 		}
734 
735 		++nb_supported;
736 	}
737 
738 	return nb_written;
739 }
740 
741 static int
742 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
743 {
744 	struct sfc_adapter *sa = dev->data->dev_private;
745 	unsigned int wanted_fc, link_fc;
746 
747 	memset(fc_conf, 0, sizeof(*fc_conf));
748 
749 	sfc_adapter_lock(sa);
750 
751 	if (sa->state == SFC_ADAPTER_STARTED)
752 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
753 	else
754 		link_fc = sa->port.flow_ctrl;
755 
756 	switch (link_fc) {
757 	case 0:
758 		fc_conf->mode = RTE_FC_NONE;
759 		break;
760 	case EFX_FCNTL_RESPOND:
761 		fc_conf->mode = RTE_FC_RX_PAUSE;
762 		break;
763 	case EFX_FCNTL_GENERATE:
764 		fc_conf->mode = RTE_FC_TX_PAUSE;
765 		break;
766 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
767 		fc_conf->mode = RTE_FC_FULL;
768 		break;
769 	default:
770 		sfc_err(sa, "%s: unexpected flow control value %#x",
771 			__func__, link_fc);
772 	}
773 
774 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
775 
776 	sfc_adapter_unlock(sa);
777 
778 	return 0;
779 }
780 
781 static int
782 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
783 {
784 	struct sfc_adapter *sa = dev->data->dev_private;
785 	struct sfc_port *port = &sa->port;
786 	unsigned int fcntl;
787 	int rc;
788 
789 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
790 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
791 	    fc_conf->mac_ctrl_frame_fwd != 0) {
792 		sfc_err(sa, "unsupported flow control settings specified");
793 		rc = EINVAL;
794 		goto fail_inval;
795 	}
796 
797 	switch (fc_conf->mode) {
798 	case RTE_FC_NONE:
799 		fcntl = 0;
800 		break;
801 	case RTE_FC_RX_PAUSE:
802 		fcntl = EFX_FCNTL_RESPOND;
803 		break;
804 	case RTE_FC_TX_PAUSE:
805 		fcntl = EFX_FCNTL_GENERATE;
806 		break;
807 	case RTE_FC_FULL:
808 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
809 		break;
810 	default:
811 		rc = EINVAL;
812 		goto fail_inval;
813 	}
814 
815 	sfc_adapter_lock(sa);
816 
817 	if (sa->state == SFC_ADAPTER_STARTED) {
818 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
819 		if (rc != 0)
820 			goto fail_mac_fcntl_set;
821 	}
822 
823 	port->flow_ctrl = fcntl;
824 	port->flow_ctrl_autoneg = fc_conf->autoneg;
825 
826 	sfc_adapter_unlock(sa);
827 
828 	return 0;
829 
830 fail_mac_fcntl_set:
831 	sfc_adapter_unlock(sa);
832 fail_inval:
833 	SFC_ASSERT(rc > 0);
834 	return -rc;
835 }
836 
837 static int
838 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
839 {
840 	struct sfc_adapter *sa = dev->data->dev_private;
841 	size_t pdu = EFX_MAC_PDU(mtu);
842 	size_t old_pdu;
843 	int rc;
844 
845 	sfc_log_init(sa, "mtu=%u", mtu);
846 
847 	rc = EINVAL;
848 	if (pdu < EFX_MAC_PDU_MIN) {
849 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
850 			(unsigned int)mtu, (unsigned int)pdu,
851 			EFX_MAC_PDU_MIN);
852 		goto fail_inval;
853 	}
854 	if (pdu > EFX_MAC_PDU_MAX) {
855 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
856 			(unsigned int)mtu, (unsigned int)pdu,
857 			EFX_MAC_PDU_MAX);
858 		goto fail_inval;
859 	}
860 
861 	sfc_adapter_lock(sa);
862 
863 	if (pdu != sa->port.pdu) {
864 		if (sa->state == SFC_ADAPTER_STARTED) {
865 			sfc_stop(sa);
866 
867 			old_pdu = sa->port.pdu;
868 			sa->port.pdu = pdu;
869 			rc = sfc_start(sa);
870 			if (rc != 0)
871 				goto fail_start;
872 		} else {
873 			sa->port.pdu = pdu;
874 		}
875 	}
876 
877 	/*
878 	 * The driver does not use it, but other PMDs update jumbo frame
879 	 * flag and max_rx_pkt_len when MTU is set.
880 	 */
881 	if (mtu > ETHER_MAX_LEN) {
882 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
883 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
884 	}
885 
886 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
887 
888 	sfc_adapter_unlock(sa);
889 
890 	sfc_log_init(sa, "done");
891 	return 0;
892 
893 fail_start:
894 	sa->port.pdu = old_pdu;
895 	if (sfc_start(sa) != 0)
896 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
897 			"PDU max size - port is stopped",
898 			(unsigned int)pdu, (unsigned int)old_pdu);
899 	sfc_adapter_unlock(sa);
900 
901 fail_inval:
902 	sfc_log_init(sa, "failed %d", rc);
903 	SFC_ASSERT(rc > 0);
904 	return -rc;
905 }
906 static int
907 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
908 {
909 	struct sfc_adapter *sa = dev->data->dev_private;
910 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
911 	struct sfc_port *port = &sa->port;
912 	struct ether_addr *old_addr = &dev->data->mac_addrs[0];
913 	int rc = 0;
914 
915 	sfc_adapter_lock(sa);
916 
917 	/*
918 	 * Copy the address to the device private data so that
919 	 * it could be recalled in the case of adapter restart.
920 	 */
921 	ether_addr_copy(mac_addr, &port->default_mac_addr);
922 
923 	/*
924 	 * Neither of the two following checks can return
925 	 * an error. The new MAC address is preserved in
926 	 * the device private data and can be activated
927 	 * on the next port start if the user prevents
928 	 * isolated mode from being enabled.
929 	 */
930 	if (port->isolated) {
931 		sfc_warn(sa, "isolated mode is active on the port");
932 		sfc_warn(sa, "will not set MAC address");
933 		goto unlock;
934 	}
935 
936 	if (sa->state != SFC_ADAPTER_STARTED) {
937 		sfc_notice(sa, "the port is not started");
938 		sfc_notice(sa, "the new MAC address will be set on port start");
939 
940 		goto unlock;
941 	}
942 
943 	if (encp->enc_allow_set_mac_with_installed_filters) {
944 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
945 		if (rc != 0) {
946 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
947 			goto unlock;
948 		}
949 
950 		/*
951 		 * Changing the MAC address by means of MCDI request
952 		 * has no effect on received traffic, therefore
953 		 * we also need to update unicast filters
954 		 */
955 		rc = sfc_set_rx_mode(sa);
956 		if (rc != 0) {
957 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
958 			/* Rollback the old address */
959 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
960 			(void)sfc_set_rx_mode(sa);
961 		}
962 	} else {
963 		sfc_warn(sa, "cannot set MAC address with filters installed");
964 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
965 		sfc_warn(sa, "(some traffic may be dropped)");
966 
967 		/*
968 		 * Since setting MAC address with filters installed is not
969 		 * allowed on the adapter, the new MAC address will be set
970 		 * by means of adapter restart. sfc_start() shall retrieve
971 		 * the new address from the device private data and set it.
972 		 */
973 		sfc_stop(sa);
974 		rc = sfc_start(sa);
975 		if (rc != 0)
976 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
977 	}
978 
979 unlock:
980 	if (rc != 0)
981 		ether_addr_copy(old_addr, &port->default_mac_addr);
982 
983 	sfc_adapter_unlock(sa);
984 
985 	SFC_ASSERT(rc >= 0);
986 	return -rc;
987 }
988 
989 
990 static int
991 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
992 		     uint32_t nb_mc_addr)
993 {
994 	struct sfc_adapter *sa = dev->data->dev_private;
995 	struct sfc_port *port = &sa->port;
996 	uint8_t *mc_addrs = port->mcast_addrs;
997 	int rc;
998 	unsigned int i;
999 
1000 	if (port->isolated) {
1001 		sfc_err(sa, "isolated mode is active on the port");
1002 		sfc_err(sa, "will not set multicast address list");
1003 		return -ENOTSUP;
1004 	}
1005 
1006 	if (mc_addrs == NULL)
1007 		return -ENOBUFS;
1008 
1009 	if (nb_mc_addr > port->max_mcast_addrs) {
1010 		sfc_err(sa, "too many multicast addresses: %u > %u",
1011 			 nb_mc_addr, port->max_mcast_addrs);
1012 		return -EINVAL;
1013 	}
1014 
1015 	for (i = 0; i < nb_mc_addr; ++i) {
1016 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1017 				 EFX_MAC_ADDR_LEN);
1018 		mc_addrs += EFX_MAC_ADDR_LEN;
1019 	}
1020 
1021 	port->nb_mcast_addrs = nb_mc_addr;
1022 
1023 	if (sa->state != SFC_ADAPTER_STARTED)
1024 		return 0;
1025 
1026 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1027 					port->nb_mcast_addrs);
1028 	if (rc != 0)
1029 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1030 
1031 	SFC_ASSERT(rc >= 0);
1032 	return -rc;
1033 }
1034 
1035 /*
1036  * The function is used by the secondary process as well. It must not
1037  * use any process-local pointers from the adapter data.
1038  */
1039 static void
1040 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1041 		      struct rte_eth_rxq_info *qinfo)
1042 {
1043 	struct sfc_adapter *sa = dev->data->dev_private;
1044 	struct sfc_rxq_info *rxq_info;
1045 	struct sfc_rxq *rxq;
1046 
1047 	sfc_adapter_lock(sa);
1048 
1049 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1050 
1051 	rxq_info = &sa->rxq_info[rx_queue_id];
1052 	rxq = rxq_info->rxq;
1053 	SFC_ASSERT(rxq != NULL);
1054 
1055 	qinfo->mp = rxq->refill_mb_pool;
1056 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1057 	qinfo->conf.rx_drop_en = 1;
1058 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1059 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1060 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1061 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1062 		qinfo->scattered_rx = 1;
1063 	}
1064 	qinfo->nb_desc = rxq_info->entries;
1065 
1066 	sfc_adapter_unlock(sa);
1067 }
1068 
1069 /*
1070  * The function is used by the secondary process as well. It must not
1071  * use any process-local pointers from the adapter data.
1072  */
1073 static void
1074 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1075 		      struct rte_eth_txq_info *qinfo)
1076 {
1077 	struct sfc_adapter *sa = dev->data->dev_private;
1078 	struct sfc_txq_info *txq_info;
1079 
1080 	sfc_adapter_lock(sa);
1081 
1082 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1083 
1084 	txq_info = &sa->txq_info[tx_queue_id];
1085 	SFC_ASSERT(txq_info->txq != NULL);
1086 
1087 	memset(qinfo, 0, sizeof(*qinfo));
1088 
1089 	qinfo->conf.offloads = txq_info->txq->offloads;
1090 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1091 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1092 	qinfo->nb_desc = txq_info->entries;
1093 
1094 	sfc_adapter_unlock(sa);
1095 }
1096 
1097 static uint32_t
1098 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1099 {
1100 	struct sfc_adapter *sa = dev->data->dev_private;
1101 
1102 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1103 
1104 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
1105 }
1106 
1107 static int
1108 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1109 {
1110 	struct sfc_dp_rxq *dp_rxq = queue;
1111 
1112 	return sfc_rx_qdesc_done(dp_rxq, offset);
1113 }
1114 
1115 static int
1116 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1117 {
1118 	struct sfc_dp_rxq *dp_rxq = queue;
1119 	struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1120 
1121 	return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1122 }
1123 
1124 static int
1125 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1126 {
1127 	struct sfc_dp_txq *dp_txq = queue;
1128 	struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1129 
1130 	return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1131 }
1132 
1133 static int
1134 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1135 {
1136 	struct sfc_adapter *sa = dev->data->dev_private;
1137 	int rc;
1138 
1139 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1140 
1141 	sfc_adapter_lock(sa);
1142 
1143 	rc = EINVAL;
1144 	if (sa->state != SFC_ADAPTER_STARTED)
1145 		goto fail_not_started;
1146 
1147 	if (sa->rxq_info[rx_queue_id].rxq == NULL)
1148 		goto fail_not_setup;
1149 
1150 	rc = sfc_rx_qstart(sa, rx_queue_id);
1151 	if (rc != 0)
1152 		goto fail_rx_qstart;
1153 
1154 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1155 
1156 	sfc_adapter_unlock(sa);
1157 
1158 	return 0;
1159 
1160 fail_rx_qstart:
1161 fail_not_setup:
1162 fail_not_started:
1163 	sfc_adapter_unlock(sa);
1164 	SFC_ASSERT(rc > 0);
1165 	return -rc;
1166 }
1167 
1168 static int
1169 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1170 {
1171 	struct sfc_adapter *sa = dev->data->dev_private;
1172 
1173 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1174 
1175 	sfc_adapter_lock(sa);
1176 	sfc_rx_qstop(sa, rx_queue_id);
1177 
1178 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1179 
1180 	sfc_adapter_unlock(sa);
1181 
1182 	return 0;
1183 }
1184 
1185 static int
1186 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1187 {
1188 	struct sfc_adapter *sa = dev->data->dev_private;
1189 	int rc;
1190 
1191 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1192 
1193 	sfc_adapter_lock(sa);
1194 
1195 	rc = EINVAL;
1196 	if (sa->state != SFC_ADAPTER_STARTED)
1197 		goto fail_not_started;
1198 
1199 	if (sa->txq_info[tx_queue_id].txq == NULL)
1200 		goto fail_not_setup;
1201 
1202 	rc = sfc_tx_qstart(sa, tx_queue_id);
1203 	if (rc != 0)
1204 		goto fail_tx_qstart;
1205 
1206 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1207 
1208 	sfc_adapter_unlock(sa);
1209 	return 0;
1210 
1211 fail_tx_qstart:
1212 
1213 fail_not_setup:
1214 fail_not_started:
1215 	sfc_adapter_unlock(sa);
1216 	SFC_ASSERT(rc > 0);
1217 	return -rc;
1218 }
1219 
1220 static int
1221 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1222 {
1223 	struct sfc_adapter *sa = dev->data->dev_private;
1224 
1225 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1226 
1227 	sfc_adapter_lock(sa);
1228 
1229 	sfc_tx_qstop(sa, tx_queue_id);
1230 
1231 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1232 
1233 	sfc_adapter_unlock(sa);
1234 	return 0;
1235 }
1236 
1237 static efx_tunnel_protocol_t
1238 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1239 {
1240 	switch (rte_type) {
1241 	case RTE_TUNNEL_TYPE_VXLAN:
1242 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1243 	case RTE_TUNNEL_TYPE_GENEVE:
1244 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1245 	default:
1246 		return EFX_TUNNEL_NPROTOS;
1247 	}
1248 }
1249 
1250 enum sfc_udp_tunnel_op_e {
1251 	SFC_UDP_TUNNEL_ADD_PORT,
1252 	SFC_UDP_TUNNEL_DEL_PORT,
1253 };
1254 
1255 static int
1256 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1257 		      struct rte_eth_udp_tunnel *tunnel_udp,
1258 		      enum sfc_udp_tunnel_op_e op)
1259 {
1260 	struct sfc_adapter *sa = dev->data->dev_private;
1261 	efx_tunnel_protocol_t tunnel_proto;
1262 	int rc;
1263 
1264 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1265 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1266 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1267 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1268 
1269 	tunnel_proto =
1270 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1271 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1272 		rc = ENOTSUP;
1273 		goto fail_bad_proto;
1274 	}
1275 
1276 	sfc_adapter_lock(sa);
1277 
1278 	switch (op) {
1279 	case SFC_UDP_TUNNEL_ADD_PORT:
1280 		rc = efx_tunnel_config_udp_add(sa->nic,
1281 					       tunnel_udp->udp_port,
1282 					       tunnel_proto);
1283 		break;
1284 	case SFC_UDP_TUNNEL_DEL_PORT:
1285 		rc = efx_tunnel_config_udp_remove(sa->nic,
1286 						  tunnel_udp->udp_port,
1287 						  tunnel_proto);
1288 		break;
1289 	default:
1290 		rc = EINVAL;
1291 		goto fail_bad_op;
1292 	}
1293 
1294 	if (rc != 0)
1295 		goto fail_op;
1296 
1297 	if (sa->state == SFC_ADAPTER_STARTED) {
1298 		rc = efx_tunnel_reconfigure(sa->nic);
1299 		if (rc == EAGAIN) {
1300 			/*
1301 			 * Configuration is accepted by FW and MC reboot
1302 			 * is initiated to apply the changes. MC reboot
1303 			 * will be handled in a usual way (MC reboot
1304 			 * event on management event queue and adapter
1305 			 * restart).
1306 			 */
1307 			rc = 0;
1308 		} else if (rc != 0) {
1309 			goto fail_reconfigure;
1310 		}
1311 	}
1312 
1313 	sfc_adapter_unlock(sa);
1314 	return 0;
1315 
1316 fail_reconfigure:
1317 	/* Remove/restore entry since the change makes the trouble */
1318 	switch (op) {
1319 	case SFC_UDP_TUNNEL_ADD_PORT:
1320 		(void)efx_tunnel_config_udp_remove(sa->nic,
1321 						   tunnel_udp->udp_port,
1322 						   tunnel_proto);
1323 		break;
1324 	case SFC_UDP_TUNNEL_DEL_PORT:
1325 		(void)efx_tunnel_config_udp_add(sa->nic,
1326 						tunnel_udp->udp_port,
1327 						tunnel_proto);
1328 		break;
1329 	}
1330 
1331 fail_op:
1332 fail_bad_op:
1333 	sfc_adapter_unlock(sa);
1334 
1335 fail_bad_proto:
1336 	SFC_ASSERT(rc > 0);
1337 	return -rc;
1338 }
1339 
1340 static int
1341 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1342 			    struct rte_eth_udp_tunnel *tunnel_udp)
1343 {
1344 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1345 }
1346 
1347 static int
1348 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1349 			    struct rte_eth_udp_tunnel *tunnel_udp)
1350 {
1351 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1352 }
1353 
1354 static int
1355 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1356 			  struct rte_eth_rss_conf *rss_conf)
1357 {
1358 	struct sfc_adapter *sa = dev->data->dev_private;
1359 	struct sfc_rss *rss = &sa->rss;
1360 	struct sfc_port *port = &sa->port;
1361 
1362 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1363 		return -ENOTSUP;
1364 
1365 	if (rss->channels == 0)
1366 		return -EINVAL;
1367 
1368 	sfc_adapter_lock(sa);
1369 
1370 	/*
1371 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1372 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1373 	 * flags which corresponds to the active EFX configuration stored
1374 	 * locally in 'sfc_adapter' and kept up-to-date
1375 	 */
1376 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1377 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1378 	if (rss_conf->rss_key != NULL)
1379 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1380 
1381 	sfc_adapter_unlock(sa);
1382 
1383 	return 0;
1384 }
1385 
1386 static int
1387 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1388 			struct rte_eth_rss_conf *rss_conf)
1389 {
1390 	struct sfc_adapter *sa = dev->data->dev_private;
1391 	struct sfc_rss *rss = &sa->rss;
1392 	struct sfc_port *port = &sa->port;
1393 	unsigned int efx_hash_types;
1394 	int rc = 0;
1395 
1396 	if (port->isolated)
1397 		return -ENOTSUP;
1398 
1399 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1400 		sfc_err(sa, "RSS is not available");
1401 		return -ENOTSUP;
1402 	}
1403 
1404 	if (rss->channels == 0) {
1405 		sfc_err(sa, "RSS is not configured");
1406 		return -EINVAL;
1407 	}
1408 
1409 	if ((rss_conf->rss_key != NULL) &&
1410 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1411 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1412 			sizeof(rss->key));
1413 		return -EINVAL;
1414 	}
1415 
1416 	sfc_adapter_lock(sa);
1417 
1418 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1419 	if (rc != 0)
1420 		goto fail_rx_hf_rte_to_efx;
1421 
1422 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1423 				   rss->hash_alg, efx_hash_types, B_TRUE);
1424 	if (rc != 0)
1425 		goto fail_scale_mode_set;
1426 
1427 	if (rss_conf->rss_key != NULL) {
1428 		if (sa->state == SFC_ADAPTER_STARTED) {
1429 			rc = efx_rx_scale_key_set(sa->nic,
1430 						  EFX_RSS_CONTEXT_DEFAULT,
1431 						  rss_conf->rss_key,
1432 						  sizeof(rss->key));
1433 			if (rc != 0)
1434 				goto fail_scale_key_set;
1435 		}
1436 
1437 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1438 	}
1439 
1440 	rss->hash_types = efx_hash_types;
1441 
1442 	sfc_adapter_unlock(sa);
1443 
1444 	return 0;
1445 
1446 fail_scale_key_set:
1447 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1448 				  EFX_RX_HASHALG_TOEPLITZ,
1449 				  rss->hash_types, B_TRUE) != 0)
1450 		sfc_err(sa, "failed to restore RSS mode");
1451 
1452 fail_scale_mode_set:
1453 fail_rx_hf_rte_to_efx:
1454 	sfc_adapter_unlock(sa);
1455 	return -rc;
1456 }
1457 
1458 static int
1459 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1460 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1461 		       uint16_t reta_size)
1462 {
1463 	struct sfc_adapter *sa = dev->data->dev_private;
1464 	struct sfc_rss *rss = &sa->rss;
1465 	struct sfc_port *port = &sa->port;
1466 	int entry;
1467 
1468 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1469 		return -ENOTSUP;
1470 
1471 	if (rss->channels == 0)
1472 		return -EINVAL;
1473 
1474 	if (reta_size != EFX_RSS_TBL_SIZE)
1475 		return -EINVAL;
1476 
1477 	sfc_adapter_lock(sa);
1478 
1479 	for (entry = 0; entry < reta_size; entry++) {
1480 		int grp = entry / RTE_RETA_GROUP_SIZE;
1481 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1482 
1483 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1484 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1485 	}
1486 
1487 	sfc_adapter_unlock(sa);
1488 
1489 	return 0;
1490 }
1491 
1492 static int
1493 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1494 			struct rte_eth_rss_reta_entry64 *reta_conf,
1495 			uint16_t reta_size)
1496 {
1497 	struct sfc_adapter *sa = dev->data->dev_private;
1498 	struct sfc_rss *rss = &sa->rss;
1499 	struct sfc_port *port = &sa->port;
1500 	unsigned int *rss_tbl_new;
1501 	uint16_t entry;
1502 	int rc = 0;
1503 
1504 
1505 	if (port->isolated)
1506 		return -ENOTSUP;
1507 
1508 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1509 		sfc_err(sa, "RSS is not available");
1510 		return -ENOTSUP;
1511 	}
1512 
1513 	if (rss->channels == 0) {
1514 		sfc_err(sa, "RSS is not configured");
1515 		return -EINVAL;
1516 	}
1517 
1518 	if (reta_size != EFX_RSS_TBL_SIZE) {
1519 		sfc_err(sa, "RETA size is wrong (should be %u)",
1520 			EFX_RSS_TBL_SIZE);
1521 		return -EINVAL;
1522 	}
1523 
1524 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1525 	if (rss_tbl_new == NULL)
1526 		return -ENOMEM;
1527 
1528 	sfc_adapter_lock(sa);
1529 
1530 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1531 
1532 	for (entry = 0; entry < reta_size; entry++) {
1533 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1534 		struct rte_eth_rss_reta_entry64 *grp;
1535 
1536 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1537 
1538 		if (grp->mask & (1ull << grp_idx)) {
1539 			if (grp->reta[grp_idx] >= rss->channels) {
1540 				rc = EINVAL;
1541 				goto bad_reta_entry;
1542 			}
1543 			rss_tbl_new[entry] = grp->reta[grp_idx];
1544 		}
1545 	}
1546 
1547 	if (sa->state == SFC_ADAPTER_STARTED) {
1548 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1549 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1550 		if (rc != 0)
1551 			goto fail_scale_tbl_set;
1552 	}
1553 
1554 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1555 
1556 fail_scale_tbl_set:
1557 bad_reta_entry:
1558 	sfc_adapter_unlock(sa);
1559 
1560 	rte_free(rss_tbl_new);
1561 
1562 	SFC_ASSERT(rc >= 0);
1563 	return -rc;
1564 }
1565 
1566 static int
1567 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1568 		    enum rte_filter_op filter_op,
1569 		    void *arg)
1570 {
1571 	struct sfc_adapter *sa = dev->data->dev_private;
1572 	int rc = ENOTSUP;
1573 
1574 	sfc_log_init(sa, "entry");
1575 
1576 	switch (filter_type) {
1577 	case RTE_ETH_FILTER_NONE:
1578 		sfc_err(sa, "Global filters configuration not supported");
1579 		break;
1580 	case RTE_ETH_FILTER_MACVLAN:
1581 		sfc_err(sa, "MACVLAN filters not supported");
1582 		break;
1583 	case RTE_ETH_FILTER_ETHERTYPE:
1584 		sfc_err(sa, "EtherType filters not supported");
1585 		break;
1586 	case RTE_ETH_FILTER_FLEXIBLE:
1587 		sfc_err(sa, "Flexible filters not supported");
1588 		break;
1589 	case RTE_ETH_FILTER_SYN:
1590 		sfc_err(sa, "SYN filters not supported");
1591 		break;
1592 	case RTE_ETH_FILTER_NTUPLE:
1593 		sfc_err(sa, "NTUPLE filters not supported");
1594 		break;
1595 	case RTE_ETH_FILTER_TUNNEL:
1596 		sfc_err(sa, "Tunnel filters not supported");
1597 		break;
1598 	case RTE_ETH_FILTER_FDIR:
1599 		sfc_err(sa, "Flow Director filters not supported");
1600 		break;
1601 	case RTE_ETH_FILTER_HASH:
1602 		sfc_err(sa, "Hash filters not supported");
1603 		break;
1604 	case RTE_ETH_FILTER_GENERIC:
1605 		if (filter_op != RTE_ETH_FILTER_GET) {
1606 			rc = EINVAL;
1607 		} else {
1608 			*(const void **)arg = &sfc_flow_ops;
1609 			rc = 0;
1610 		}
1611 		break;
1612 	default:
1613 		sfc_err(sa, "Unknown filter type %u", filter_type);
1614 		break;
1615 	}
1616 
1617 	sfc_log_init(sa, "exit: %d", -rc);
1618 	SFC_ASSERT(rc >= 0);
1619 	return -rc;
1620 }
1621 
1622 static int
1623 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1624 {
1625 	struct sfc_adapter *sa = dev->data->dev_private;
1626 
1627 	/*
1628 	 * If Rx datapath does not provide callback to check mempool,
1629 	 * all pools are supported.
1630 	 */
1631 	if (sa->dp_rx->pool_ops_supported == NULL)
1632 		return 1;
1633 
1634 	return sa->dp_rx->pool_ops_supported(pool);
1635 }
1636 
1637 static const struct eth_dev_ops sfc_eth_dev_ops = {
1638 	.dev_configure			= sfc_dev_configure,
1639 	.dev_start			= sfc_dev_start,
1640 	.dev_stop			= sfc_dev_stop,
1641 	.dev_set_link_up		= sfc_dev_set_link_up,
1642 	.dev_set_link_down		= sfc_dev_set_link_down,
1643 	.dev_close			= sfc_dev_close,
1644 	.promiscuous_enable		= sfc_dev_promisc_enable,
1645 	.promiscuous_disable		= sfc_dev_promisc_disable,
1646 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1647 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1648 	.link_update			= sfc_dev_link_update,
1649 	.stats_get			= sfc_stats_get,
1650 	.stats_reset			= sfc_stats_reset,
1651 	.xstats_get			= sfc_xstats_get,
1652 	.xstats_reset			= sfc_stats_reset,
1653 	.xstats_get_names		= sfc_xstats_get_names,
1654 	.dev_infos_get			= sfc_dev_infos_get,
1655 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1656 	.mtu_set			= sfc_dev_set_mtu,
1657 	.rx_queue_start			= sfc_rx_queue_start,
1658 	.rx_queue_stop			= sfc_rx_queue_stop,
1659 	.tx_queue_start			= sfc_tx_queue_start,
1660 	.tx_queue_stop			= sfc_tx_queue_stop,
1661 	.rx_queue_setup			= sfc_rx_queue_setup,
1662 	.rx_queue_release		= sfc_rx_queue_release,
1663 	.rx_queue_count			= sfc_rx_queue_count,
1664 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1665 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1666 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1667 	.tx_queue_setup			= sfc_tx_queue_setup,
1668 	.tx_queue_release		= sfc_tx_queue_release,
1669 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1670 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1671 	.mac_addr_set			= sfc_mac_addr_set,
1672 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1673 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1674 	.reta_update			= sfc_dev_rss_reta_update,
1675 	.reta_query			= sfc_dev_rss_reta_query,
1676 	.rss_hash_update		= sfc_dev_rss_hash_update,
1677 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1678 	.filter_ctrl			= sfc_dev_filter_ctrl,
1679 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1680 	.rxq_info_get			= sfc_rx_queue_info_get,
1681 	.txq_info_get			= sfc_tx_queue_info_get,
1682 	.fw_version_get			= sfc_fw_version_get,
1683 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1684 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1685 	.pool_ops_supported		= sfc_pool_ops_supported,
1686 };
1687 
1688 /**
1689  * Duplicate a string in potentially shared memory required for
1690  * multi-process support.
1691  *
1692  * strdup() allocates from process-local heap/memory.
1693  */
1694 static char *
1695 sfc_strdup(const char *str)
1696 {
1697 	size_t size;
1698 	char *copy;
1699 
1700 	if (str == NULL)
1701 		return NULL;
1702 
1703 	size = strlen(str) + 1;
1704 	copy = rte_malloc(__func__, size, 0);
1705 	if (copy != NULL)
1706 		rte_memcpy(copy, str, size);
1707 
1708 	return copy;
1709 }
1710 
1711 static int
1712 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1713 {
1714 	struct sfc_adapter *sa = dev->data->dev_private;
1715 	const efx_nic_cfg_t *encp;
1716 	unsigned int avail_caps = 0;
1717 	const char *rx_name = NULL;
1718 	const char *tx_name = NULL;
1719 	int rc;
1720 
1721 	switch (sa->family) {
1722 	case EFX_FAMILY_HUNTINGTON:
1723 	case EFX_FAMILY_MEDFORD:
1724 	case EFX_FAMILY_MEDFORD2:
1725 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1726 		break;
1727 	default:
1728 		break;
1729 	}
1730 
1731 	encp = efx_nic_cfg_get(sa->nic);
1732 	if (encp->enc_rx_es_super_buffer_supported)
1733 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1734 
1735 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1736 				sfc_kvarg_string_handler, &rx_name);
1737 	if (rc != 0)
1738 		goto fail_kvarg_rx_datapath;
1739 
1740 	if (rx_name != NULL) {
1741 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1742 		if (sa->dp_rx == NULL) {
1743 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1744 			rc = ENOENT;
1745 			goto fail_dp_rx;
1746 		}
1747 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1748 			sfc_err(sa,
1749 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1750 				rx_name);
1751 			rc = EINVAL;
1752 			goto fail_dp_rx_caps;
1753 		}
1754 	} else {
1755 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1756 		if (sa->dp_rx == NULL) {
1757 			sfc_err(sa, "Rx datapath by caps %#x not found",
1758 				avail_caps);
1759 			rc = ENOENT;
1760 			goto fail_dp_rx;
1761 		}
1762 	}
1763 
1764 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1765 	if (sa->dp_rx_name == NULL) {
1766 		rc = ENOMEM;
1767 		goto fail_dp_rx_name;
1768 	}
1769 
1770 	sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1771 
1772 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1773 
1774 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1775 				sfc_kvarg_string_handler, &tx_name);
1776 	if (rc != 0)
1777 		goto fail_kvarg_tx_datapath;
1778 
1779 	if (tx_name != NULL) {
1780 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1781 		if (sa->dp_tx == NULL) {
1782 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1783 			rc = ENOENT;
1784 			goto fail_dp_tx;
1785 		}
1786 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1787 			sfc_err(sa,
1788 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1789 				tx_name);
1790 			rc = EINVAL;
1791 			goto fail_dp_tx_caps;
1792 		}
1793 	} else {
1794 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1795 		if (sa->dp_tx == NULL) {
1796 			sfc_err(sa, "Tx datapath by caps %#x not found",
1797 				avail_caps);
1798 			rc = ENOENT;
1799 			goto fail_dp_tx;
1800 		}
1801 	}
1802 
1803 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1804 	if (sa->dp_tx_name == NULL) {
1805 		rc = ENOMEM;
1806 		goto fail_dp_tx_name;
1807 	}
1808 
1809 	sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1810 
1811 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1812 
1813 	dev->dev_ops = &sfc_eth_dev_ops;
1814 
1815 	return 0;
1816 
1817 fail_dp_tx_name:
1818 fail_dp_tx_caps:
1819 	sa->dp_tx = NULL;
1820 
1821 fail_dp_tx:
1822 fail_kvarg_tx_datapath:
1823 	rte_free(sa->dp_rx_name);
1824 	sa->dp_rx_name = NULL;
1825 
1826 fail_dp_rx_name:
1827 fail_dp_rx_caps:
1828 	sa->dp_rx = NULL;
1829 
1830 fail_dp_rx:
1831 fail_kvarg_rx_datapath:
1832 	return rc;
1833 }
1834 
1835 static void
1836 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1837 {
1838 	struct sfc_adapter *sa = dev->data->dev_private;
1839 
1840 	dev->dev_ops = NULL;
1841 	dev->rx_pkt_burst = NULL;
1842 	dev->tx_pkt_burst = NULL;
1843 
1844 	rte_free(sa->dp_tx_name);
1845 	sa->dp_tx_name = NULL;
1846 	sa->dp_tx = NULL;
1847 
1848 	rte_free(sa->dp_rx_name);
1849 	sa->dp_rx_name = NULL;
1850 	sa->dp_rx = NULL;
1851 }
1852 
1853 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1854 	.rxq_info_get			= sfc_rx_queue_info_get,
1855 	.txq_info_get			= sfc_tx_queue_info_get,
1856 };
1857 
1858 static int
1859 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1860 {
1861 	/*
1862 	 * Device private data has really many process-local pointers.
1863 	 * Below code should be extremely careful to use data located
1864 	 * in shared memory only.
1865 	 */
1866 	struct sfc_adapter *sa = dev->data->dev_private;
1867 	const struct sfc_dp_rx *dp_rx;
1868 	const struct sfc_dp_tx *dp_tx;
1869 	int rc;
1870 
1871 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1872 	if (dp_rx == NULL) {
1873 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1874 		rc = ENOENT;
1875 		goto fail_dp_rx;
1876 	}
1877 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1878 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1879 			sa->dp_tx_name);
1880 		rc = EINVAL;
1881 		goto fail_dp_rx_multi_process;
1882 	}
1883 
1884 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1885 	if (dp_tx == NULL) {
1886 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1887 		rc = ENOENT;
1888 		goto fail_dp_tx;
1889 	}
1890 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1891 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1892 			sa->dp_tx_name);
1893 		rc = EINVAL;
1894 		goto fail_dp_tx_multi_process;
1895 	}
1896 
1897 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1898 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1899 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1900 
1901 	return 0;
1902 
1903 fail_dp_tx_multi_process:
1904 fail_dp_tx:
1905 fail_dp_rx_multi_process:
1906 fail_dp_rx:
1907 	return rc;
1908 }
1909 
1910 static void
1911 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1912 {
1913 	dev->dev_ops = NULL;
1914 	dev->tx_pkt_burst = NULL;
1915 	dev->rx_pkt_burst = NULL;
1916 }
1917 
1918 static void
1919 sfc_register_dp(void)
1920 {
1921 	/* Register once */
1922 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1923 		/* Prefer EF10 datapath */
1924 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1925 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1926 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1927 
1928 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1929 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1930 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1931 	}
1932 }
1933 
1934 static int
1935 sfc_eth_dev_init(struct rte_eth_dev *dev)
1936 {
1937 	struct sfc_adapter *sa = dev->data->dev_private;
1938 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1939 	int rc;
1940 	const efx_nic_cfg_t *encp;
1941 	const struct ether_addr *from;
1942 
1943 	sfc_register_dp();
1944 
1945 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1946 		return -sfc_eth_dev_secondary_set_ops(dev);
1947 
1948 	/* Required for logging */
1949 	sa->pci_addr = pci_dev->addr;
1950 	sa->port_id = dev->data->port_id;
1951 
1952 	sa->eth_dev = dev;
1953 
1954 	/* Copy PCI device info to the dev->data */
1955 	rte_eth_copy_pci_info(dev, pci_dev);
1956 
1957 	sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1958 						RTE_LOG_NOTICE);
1959 
1960 	rc = sfc_kvargs_parse(sa);
1961 	if (rc != 0)
1962 		goto fail_kvargs_parse;
1963 
1964 	sfc_log_init(sa, "entry");
1965 
1966 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1967 	if (dev->data->mac_addrs == NULL) {
1968 		rc = ENOMEM;
1969 		goto fail_mac_addrs;
1970 	}
1971 
1972 	sfc_adapter_lock_init(sa);
1973 	sfc_adapter_lock(sa);
1974 
1975 	sfc_log_init(sa, "probing");
1976 	rc = sfc_probe(sa);
1977 	if (rc != 0)
1978 		goto fail_probe;
1979 
1980 	sfc_log_init(sa, "set device ops");
1981 	rc = sfc_eth_dev_set_ops(dev);
1982 	if (rc != 0)
1983 		goto fail_set_ops;
1984 
1985 	sfc_log_init(sa, "attaching");
1986 	rc = sfc_attach(sa);
1987 	if (rc != 0)
1988 		goto fail_attach;
1989 
1990 	encp = efx_nic_cfg_get(sa->nic);
1991 
1992 	/*
1993 	 * The arguments are really reverse order in comparison to
1994 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1995 	 */
1996 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1997 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1998 
1999 	sfc_adapter_unlock(sa);
2000 
2001 	sfc_log_init(sa, "done");
2002 	return 0;
2003 
2004 fail_attach:
2005 	sfc_eth_dev_clear_ops(dev);
2006 
2007 fail_set_ops:
2008 	sfc_unprobe(sa);
2009 
2010 fail_probe:
2011 	sfc_adapter_unlock(sa);
2012 	sfc_adapter_lock_fini(sa);
2013 	rte_free(dev->data->mac_addrs);
2014 	dev->data->mac_addrs = NULL;
2015 
2016 fail_mac_addrs:
2017 	sfc_kvargs_cleanup(sa);
2018 
2019 fail_kvargs_parse:
2020 	sfc_log_init(sa, "failed %d", rc);
2021 	SFC_ASSERT(rc > 0);
2022 	return -rc;
2023 }
2024 
2025 static int
2026 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2027 {
2028 	struct sfc_adapter *sa;
2029 
2030 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2031 		sfc_eth_dev_secondary_clear_ops(dev);
2032 		return 0;
2033 	}
2034 
2035 	sa = dev->data->dev_private;
2036 	sfc_log_init(sa, "entry");
2037 
2038 	sfc_adapter_lock(sa);
2039 
2040 	sfc_eth_dev_clear_ops(dev);
2041 
2042 	sfc_detach(sa);
2043 	sfc_unprobe(sa);
2044 
2045 	rte_free(dev->data->mac_addrs);
2046 	dev->data->mac_addrs = NULL;
2047 
2048 	sfc_kvargs_cleanup(sa);
2049 
2050 	sfc_adapter_unlock(sa);
2051 	sfc_adapter_lock_fini(sa);
2052 
2053 	sfc_log_init(sa, "done");
2054 
2055 	/* Required for logging, so cleanup last */
2056 	sa->eth_dev = NULL;
2057 	return 0;
2058 }
2059 
2060 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2061 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2062 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2063 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2064 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2065 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2066 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2067 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2068 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2069 	{ .vendor_id = 0 /* sentinel */ }
2070 };
2071 
2072 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2073 	struct rte_pci_device *pci_dev)
2074 {
2075 	return rte_eth_dev_pci_generic_probe(pci_dev,
2076 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2077 }
2078 
2079 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2080 {
2081 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2082 }
2083 
2084 static struct rte_pci_driver sfc_efx_pmd = {
2085 	.id_table = pci_id_sfc_efx_map,
2086 	.drv_flags =
2087 		RTE_PCI_DRV_INTR_LSC |
2088 		RTE_PCI_DRV_NEED_MAPPING,
2089 	.probe = sfc_eth_dev_pci_probe,
2090 	.remove = sfc_eth_dev_pci_remove,
2091 };
2092 
2093 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2094 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2095 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2096 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2097 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2098 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2099 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2100 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2101 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2102 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2103 
2104 RTE_INIT(sfc_driver_register_logtype)
2105 {
2106 	int ret;
2107 
2108 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2109 						   RTE_LOG_NOTICE);
2110 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2111 }
2112