xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 1edccebcccdbe600dc0a3a418fae68336648a87e)
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_tx_qfini(sa, sw_index);
502 
503 	sfc_adapter_unlock(sa);
504 }
505 
506 static int
507 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
508 {
509 	struct sfc_adapter *sa = dev->data->dev_private;
510 	struct sfc_port *port = &sa->port;
511 	uint64_t *mac_stats;
512 	int ret;
513 
514 	rte_spinlock_lock(&port->mac_stats_lock);
515 
516 	ret = sfc_port_update_mac_stats(sa);
517 	if (ret != 0)
518 		goto unlock;
519 
520 	mac_stats = port->mac_stats_buf;
521 
522 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
523 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
524 		stats->ipackets =
525 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
526 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
527 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
528 		stats->opackets =
529 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
530 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
531 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
532 		stats->ibytes =
533 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
534 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
535 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
536 		stats->obytes =
537 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
538 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
539 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
540 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
541 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
542 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
543 	} else {
544 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
545 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
546 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
547 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
548 		/*
549 		 * Take into account stats which are whenever supported
550 		 * on EF10. If some stat is not supported by current
551 		 * firmware variant or HW revision, it is guaranteed
552 		 * to be zero in mac_stats.
553 		 */
554 		stats->imissed =
555 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
556 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
557 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
558 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
559 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
560 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
561 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
562 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
563 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
564 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
565 		stats->ierrors =
566 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
567 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
568 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
569 		/* no oerrors counters supported on EF10 */
570 	}
571 
572 unlock:
573 	rte_spinlock_unlock(&port->mac_stats_lock);
574 	SFC_ASSERT(ret >= 0);
575 	return -ret;
576 }
577 
578 static void
579 sfc_stats_reset(struct rte_eth_dev *dev)
580 {
581 	struct sfc_adapter *sa = dev->data->dev_private;
582 	struct sfc_port *port = &sa->port;
583 	int rc;
584 
585 	if (sa->state != SFC_ADAPTER_STARTED) {
586 		/*
587 		 * The operation cannot be done if port is not started; it
588 		 * will be scheduled to be done during the next port start
589 		 */
590 		port->mac_stats_reset_pending = B_TRUE;
591 		return;
592 	}
593 
594 	rc = sfc_port_reset_mac_stats(sa);
595 	if (rc != 0)
596 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
597 }
598 
599 static int
600 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
601 	       unsigned int xstats_count)
602 {
603 	struct sfc_adapter *sa = dev->data->dev_private;
604 	struct sfc_port *port = &sa->port;
605 	uint64_t *mac_stats;
606 	int rc;
607 	unsigned int i;
608 	int nstats = 0;
609 
610 	rte_spinlock_lock(&port->mac_stats_lock);
611 
612 	rc = sfc_port_update_mac_stats(sa);
613 	if (rc != 0) {
614 		SFC_ASSERT(rc > 0);
615 		nstats = -rc;
616 		goto unlock;
617 	}
618 
619 	mac_stats = port->mac_stats_buf;
620 
621 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
622 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
623 			if (xstats != NULL && nstats < (int)xstats_count) {
624 				xstats[nstats].id = nstats;
625 				xstats[nstats].value = mac_stats[i];
626 			}
627 			nstats++;
628 		}
629 	}
630 
631 unlock:
632 	rte_spinlock_unlock(&port->mac_stats_lock);
633 
634 	return nstats;
635 }
636 
637 static int
638 sfc_xstats_get_names(struct rte_eth_dev *dev,
639 		     struct rte_eth_xstat_name *xstats_names,
640 		     unsigned int xstats_count)
641 {
642 	struct sfc_adapter *sa = dev->data->dev_private;
643 	struct sfc_port *port = &sa->port;
644 	unsigned int i;
645 	unsigned int nstats = 0;
646 
647 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
648 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
649 			if (xstats_names != NULL && nstats < xstats_count)
650 				strlcpy(xstats_names[nstats].name,
651 					efx_mac_stat_name(sa->nic, i),
652 					sizeof(xstats_names[0].name));
653 			nstats++;
654 		}
655 	}
656 
657 	return nstats;
658 }
659 
660 static int
661 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
662 		     uint64_t *values, unsigned int n)
663 {
664 	struct sfc_adapter *sa = dev->data->dev_private;
665 	struct sfc_port *port = &sa->port;
666 	uint64_t *mac_stats;
667 	unsigned int nb_supported = 0;
668 	unsigned int nb_written = 0;
669 	unsigned int i;
670 	int ret;
671 	int rc;
672 
673 	if (unlikely(values == NULL) ||
674 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
675 		return port->mac_stats_nb_supported;
676 
677 	rte_spinlock_lock(&port->mac_stats_lock);
678 
679 	rc = sfc_port_update_mac_stats(sa);
680 	if (rc != 0) {
681 		SFC_ASSERT(rc > 0);
682 		ret = -rc;
683 		goto unlock;
684 	}
685 
686 	mac_stats = port->mac_stats_buf;
687 
688 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
689 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
690 			continue;
691 
692 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
693 			values[nb_written++] = mac_stats[i];
694 
695 		++nb_supported;
696 	}
697 
698 	ret = nb_written;
699 
700 unlock:
701 	rte_spinlock_unlock(&port->mac_stats_lock);
702 
703 	return ret;
704 }
705 
706 static int
707 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
708 			   struct rte_eth_xstat_name *xstats_names,
709 			   const uint64_t *ids, unsigned int size)
710 {
711 	struct sfc_adapter *sa = dev->data->dev_private;
712 	struct sfc_port *port = &sa->port;
713 	unsigned int nb_supported = 0;
714 	unsigned int nb_written = 0;
715 	unsigned int i;
716 
717 	if (unlikely(xstats_names == NULL) ||
718 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
719 		return port->mac_stats_nb_supported;
720 
721 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
722 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
723 			continue;
724 
725 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
726 			char *name = xstats_names[nb_written++].name;
727 
728 			strlcpy(name, efx_mac_stat_name(sa->nic, i),
729 				sizeof(xstats_names[0].name));
730 		}
731 
732 		++nb_supported;
733 	}
734 
735 	return nb_written;
736 }
737 
738 static int
739 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
740 {
741 	struct sfc_adapter *sa = dev->data->dev_private;
742 	unsigned int wanted_fc, link_fc;
743 
744 	memset(fc_conf, 0, sizeof(*fc_conf));
745 
746 	sfc_adapter_lock(sa);
747 
748 	if (sa->state == SFC_ADAPTER_STARTED)
749 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
750 	else
751 		link_fc = sa->port.flow_ctrl;
752 
753 	switch (link_fc) {
754 	case 0:
755 		fc_conf->mode = RTE_FC_NONE;
756 		break;
757 	case EFX_FCNTL_RESPOND:
758 		fc_conf->mode = RTE_FC_RX_PAUSE;
759 		break;
760 	case EFX_FCNTL_GENERATE:
761 		fc_conf->mode = RTE_FC_TX_PAUSE;
762 		break;
763 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
764 		fc_conf->mode = RTE_FC_FULL;
765 		break;
766 	default:
767 		sfc_err(sa, "%s: unexpected flow control value %#x",
768 			__func__, link_fc);
769 	}
770 
771 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
772 
773 	sfc_adapter_unlock(sa);
774 
775 	return 0;
776 }
777 
778 static int
779 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
780 {
781 	struct sfc_adapter *sa = dev->data->dev_private;
782 	struct sfc_port *port = &sa->port;
783 	unsigned int fcntl;
784 	int rc;
785 
786 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
787 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
788 	    fc_conf->mac_ctrl_frame_fwd != 0) {
789 		sfc_err(sa, "unsupported flow control settings specified");
790 		rc = EINVAL;
791 		goto fail_inval;
792 	}
793 
794 	switch (fc_conf->mode) {
795 	case RTE_FC_NONE:
796 		fcntl = 0;
797 		break;
798 	case RTE_FC_RX_PAUSE:
799 		fcntl = EFX_FCNTL_RESPOND;
800 		break;
801 	case RTE_FC_TX_PAUSE:
802 		fcntl = EFX_FCNTL_GENERATE;
803 		break;
804 	case RTE_FC_FULL:
805 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
806 		break;
807 	default:
808 		rc = EINVAL;
809 		goto fail_inval;
810 	}
811 
812 	sfc_adapter_lock(sa);
813 
814 	if (sa->state == SFC_ADAPTER_STARTED) {
815 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
816 		if (rc != 0)
817 			goto fail_mac_fcntl_set;
818 	}
819 
820 	port->flow_ctrl = fcntl;
821 	port->flow_ctrl_autoneg = fc_conf->autoneg;
822 
823 	sfc_adapter_unlock(sa);
824 
825 	return 0;
826 
827 fail_mac_fcntl_set:
828 	sfc_adapter_unlock(sa);
829 fail_inval:
830 	SFC_ASSERT(rc > 0);
831 	return -rc;
832 }
833 
834 static int
835 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
836 {
837 	struct sfc_adapter *sa = dev->data->dev_private;
838 	size_t pdu = EFX_MAC_PDU(mtu);
839 	size_t old_pdu;
840 	int rc;
841 
842 	sfc_log_init(sa, "mtu=%u", mtu);
843 
844 	rc = EINVAL;
845 	if (pdu < EFX_MAC_PDU_MIN) {
846 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
847 			(unsigned int)mtu, (unsigned int)pdu,
848 			EFX_MAC_PDU_MIN);
849 		goto fail_inval;
850 	}
851 	if (pdu > EFX_MAC_PDU_MAX) {
852 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
853 			(unsigned int)mtu, (unsigned int)pdu,
854 			EFX_MAC_PDU_MAX);
855 		goto fail_inval;
856 	}
857 
858 	sfc_adapter_lock(sa);
859 
860 	if (pdu != sa->port.pdu) {
861 		if (sa->state == SFC_ADAPTER_STARTED) {
862 			sfc_stop(sa);
863 
864 			old_pdu = sa->port.pdu;
865 			sa->port.pdu = pdu;
866 			rc = sfc_start(sa);
867 			if (rc != 0)
868 				goto fail_start;
869 		} else {
870 			sa->port.pdu = pdu;
871 		}
872 	}
873 
874 	/*
875 	 * The driver does not use it, but other PMDs update jumbo frame
876 	 * flag and max_rx_pkt_len when MTU is set.
877 	 */
878 	if (mtu > ETHER_MAX_LEN) {
879 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
880 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
881 	}
882 
883 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
884 
885 	sfc_adapter_unlock(sa);
886 
887 	sfc_log_init(sa, "done");
888 	return 0;
889 
890 fail_start:
891 	sa->port.pdu = old_pdu;
892 	if (sfc_start(sa) != 0)
893 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
894 			"PDU max size - port is stopped",
895 			(unsigned int)pdu, (unsigned int)old_pdu);
896 	sfc_adapter_unlock(sa);
897 
898 fail_inval:
899 	sfc_log_init(sa, "failed %d", rc);
900 	SFC_ASSERT(rc > 0);
901 	return -rc;
902 }
903 static int
904 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
905 {
906 	struct sfc_adapter *sa = dev->data->dev_private;
907 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
908 	struct sfc_port *port = &sa->port;
909 	struct ether_addr *old_addr = &dev->data->mac_addrs[0];
910 	int rc = 0;
911 
912 	sfc_adapter_lock(sa);
913 
914 	/*
915 	 * Copy the address to the device private data so that
916 	 * it could be recalled in the case of adapter restart.
917 	 */
918 	ether_addr_copy(mac_addr, &port->default_mac_addr);
919 
920 	/*
921 	 * Neither of the two following checks can return
922 	 * an error. The new MAC address is preserved in
923 	 * the device private data and can be activated
924 	 * on the next port start if the user prevents
925 	 * isolated mode from being enabled.
926 	 */
927 	if (port->isolated) {
928 		sfc_warn(sa, "isolated mode is active on the port");
929 		sfc_warn(sa, "will not set MAC address");
930 		goto unlock;
931 	}
932 
933 	if (sa->state != SFC_ADAPTER_STARTED) {
934 		sfc_notice(sa, "the port is not started");
935 		sfc_notice(sa, "the new MAC address will be set on port start");
936 
937 		goto unlock;
938 	}
939 
940 	if (encp->enc_allow_set_mac_with_installed_filters) {
941 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
942 		if (rc != 0) {
943 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
944 			goto unlock;
945 		}
946 
947 		/*
948 		 * Changing the MAC address by means of MCDI request
949 		 * has no effect on received traffic, therefore
950 		 * we also need to update unicast filters
951 		 */
952 		rc = sfc_set_rx_mode(sa);
953 		if (rc != 0) {
954 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
955 			/* Rollback the old address */
956 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
957 			(void)sfc_set_rx_mode(sa);
958 		}
959 	} else {
960 		sfc_warn(sa, "cannot set MAC address with filters installed");
961 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
962 		sfc_warn(sa, "(some traffic may be dropped)");
963 
964 		/*
965 		 * Since setting MAC address with filters installed is not
966 		 * allowed on the adapter, the new MAC address will be set
967 		 * by means of adapter restart. sfc_start() shall retrieve
968 		 * the new address from the device private data and set it.
969 		 */
970 		sfc_stop(sa);
971 		rc = sfc_start(sa);
972 		if (rc != 0)
973 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
974 	}
975 
976 unlock:
977 	if (rc != 0)
978 		ether_addr_copy(old_addr, &port->default_mac_addr);
979 
980 	sfc_adapter_unlock(sa);
981 
982 	SFC_ASSERT(rc >= 0);
983 	return -rc;
984 }
985 
986 
987 static int
988 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
989 		     uint32_t nb_mc_addr)
990 {
991 	struct sfc_adapter *sa = dev->data->dev_private;
992 	struct sfc_port *port = &sa->port;
993 	uint8_t *mc_addrs = port->mcast_addrs;
994 	int rc;
995 	unsigned int i;
996 
997 	if (port->isolated) {
998 		sfc_err(sa, "isolated mode is active on the port");
999 		sfc_err(sa, "will not set multicast address list");
1000 		return -ENOTSUP;
1001 	}
1002 
1003 	if (mc_addrs == NULL)
1004 		return -ENOBUFS;
1005 
1006 	if (nb_mc_addr > port->max_mcast_addrs) {
1007 		sfc_err(sa, "too many multicast addresses: %u > %u",
1008 			 nb_mc_addr, port->max_mcast_addrs);
1009 		return -EINVAL;
1010 	}
1011 
1012 	for (i = 0; i < nb_mc_addr; ++i) {
1013 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1014 				 EFX_MAC_ADDR_LEN);
1015 		mc_addrs += EFX_MAC_ADDR_LEN;
1016 	}
1017 
1018 	port->nb_mcast_addrs = nb_mc_addr;
1019 
1020 	if (sa->state != SFC_ADAPTER_STARTED)
1021 		return 0;
1022 
1023 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1024 					port->nb_mcast_addrs);
1025 	if (rc != 0)
1026 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1027 
1028 	SFC_ASSERT(rc >= 0);
1029 	return -rc;
1030 }
1031 
1032 /*
1033  * The function is used by the secondary process as well. It must not
1034  * use any process-local pointers from the adapter data.
1035  */
1036 static void
1037 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1038 		      struct rte_eth_rxq_info *qinfo)
1039 {
1040 	struct sfc_adapter *sa = dev->data->dev_private;
1041 	struct sfc_rxq_info *rxq_info;
1042 	struct sfc_rxq *rxq;
1043 
1044 	sfc_adapter_lock(sa);
1045 
1046 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1047 
1048 	rxq_info = &sa->rxq_info[rx_queue_id];
1049 	rxq = rxq_info->rxq;
1050 	SFC_ASSERT(rxq != NULL);
1051 
1052 	qinfo->mp = rxq->refill_mb_pool;
1053 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1054 	qinfo->conf.rx_drop_en = 1;
1055 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1056 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1057 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1058 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1059 		qinfo->scattered_rx = 1;
1060 	}
1061 	qinfo->nb_desc = rxq_info->entries;
1062 
1063 	sfc_adapter_unlock(sa);
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_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1072 		      struct rte_eth_txq_info *qinfo)
1073 {
1074 	struct sfc_adapter *sa = dev->data->dev_private;
1075 	struct sfc_txq_info *txq_info;
1076 
1077 	sfc_adapter_lock(sa);
1078 
1079 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1080 
1081 	txq_info = &sa->txq_info[tx_queue_id];
1082 	SFC_ASSERT(txq_info->txq != NULL);
1083 
1084 	memset(qinfo, 0, sizeof(*qinfo));
1085 
1086 	qinfo->conf.offloads = txq_info->txq->offloads;
1087 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1088 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1089 	qinfo->nb_desc = txq_info->entries;
1090 
1091 	sfc_adapter_unlock(sa);
1092 }
1093 
1094 static uint32_t
1095 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1096 {
1097 	struct sfc_adapter *sa = dev->data->dev_private;
1098 
1099 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1100 
1101 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
1102 }
1103 
1104 static int
1105 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1106 {
1107 	struct sfc_dp_rxq *dp_rxq = queue;
1108 
1109 	return sfc_rx_qdesc_done(dp_rxq, offset);
1110 }
1111 
1112 static int
1113 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1114 {
1115 	struct sfc_dp_rxq *dp_rxq = queue;
1116 	struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1117 
1118 	return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1119 }
1120 
1121 static int
1122 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1123 {
1124 	struct sfc_dp_txq *dp_txq = queue;
1125 	struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1126 
1127 	return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1128 }
1129 
1130 static int
1131 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1132 {
1133 	struct sfc_adapter *sa = dev->data->dev_private;
1134 	int rc;
1135 
1136 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1137 
1138 	sfc_adapter_lock(sa);
1139 
1140 	rc = EINVAL;
1141 	if (sa->state != SFC_ADAPTER_STARTED)
1142 		goto fail_not_started;
1143 
1144 	if (sa->rxq_info[rx_queue_id].rxq == NULL)
1145 		goto fail_not_setup;
1146 
1147 	rc = sfc_rx_qstart(sa, rx_queue_id);
1148 	if (rc != 0)
1149 		goto fail_rx_qstart;
1150 
1151 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1152 
1153 	sfc_adapter_unlock(sa);
1154 
1155 	return 0;
1156 
1157 fail_rx_qstart:
1158 fail_not_setup:
1159 fail_not_started:
1160 	sfc_adapter_unlock(sa);
1161 	SFC_ASSERT(rc > 0);
1162 	return -rc;
1163 }
1164 
1165 static int
1166 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1167 {
1168 	struct sfc_adapter *sa = dev->data->dev_private;
1169 
1170 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1171 
1172 	sfc_adapter_lock(sa);
1173 	sfc_rx_qstop(sa, rx_queue_id);
1174 
1175 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1176 
1177 	sfc_adapter_unlock(sa);
1178 
1179 	return 0;
1180 }
1181 
1182 static int
1183 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1184 {
1185 	struct sfc_adapter *sa = dev->data->dev_private;
1186 	int rc;
1187 
1188 	sfc_log_init(sa, "TxQ = %u", tx_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 (sa->txq_info[tx_queue_id].txq == NULL)
1197 		goto fail_not_setup;
1198 
1199 	rc = sfc_tx_qstart(sa, tx_queue_id);
1200 	if (rc != 0)
1201 		goto fail_tx_qstart;
1202 
1203 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1204 
1205 	sfc_adapter_unlock(sa);
1206 	return 0;
1207 
1208 fail_tx_qstart:
1209 
1210 fail_not_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_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1219 {
1220 	struct sfc_adapter *sa = dev->data->dev_private;
1221 
1222 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1223 
1224 	sfc_adapter_lock(sa);
1225 
1226 	sfc_tx_qstop(sa, tx_queue_id);
1227 
1228 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1229 
1230 	sfc_adapter_unlock(sa);
1231 	return 0;
1232 }
1233 
1234 static efx_tunnel_protocol_t
1235 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1236 {
1237 	switch (rte_type) {
1238 	case RTE_TUNNEL_TYPE_VXLAN:
1239 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1240 	case RTE_TUNNEL_TYPE_GENEVE:
1241 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1242 	default:
1243 		return EFX_TUNNEL_NPROTOS;
1244 	}
1245 }
1246 
1247 enum sfc_udp_tunnel_op_e {
1248 	SFC_UDP_TUNNEL_ADD_PORT,
1249 	SFC_UDP_TUNNEL_DEL_PORT,
1250 };
1251 
1252 static int
1253 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1254 		      struct rte_eth_udp_tunnel *tunnel_udp,
1255 		      enum sfc_udp_tunnel_op_e op)
1256 {
1257 	struct sfc_adapter *sa = dev->data->dev_private;
1258 	efx_tunnel_protocol_t tunnel_proto;
1259 	int rc;
1260 
1261 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1262 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1263 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1264 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1265 
1266 	tunnel_proto =
1267 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1268 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1269 		rc = ENOTSUP;
1270 		goto fail_bad_proto;
1271 	}
1272 
1273 	sfc_adapter_lock(sa);
1274 
1275 	switch (op) {
1276 	case SFC_UDP_TUNNEL_ADD_PORT:
1277 		rc = efx_tunnel_config_udp_add(sa->nic,
1278 					       tunnel_udp->udp_port,
1279 					       tunnel_proto);
1280 		break;
1281 	case SFC_UDP_TUNNEL_DEL_PORT:
1282 		rc = efx_tunnel_config_udp_remove(sa->nic,
1283 						  tunnel_udp->udp_port,
1284 						  tunnel_proto);
1285 		break;
1286 	default:
1287 		rc = EINVAL;
1288 		goto fail_bad_op;
1289 	}
1290 
1291 	if (rc != 0)
1292 		goto fail_op;
1293 
1294 	if (sa->state == SFC_ADAPTER_STARTED) {
1295 		rc = efx_tunnel_reconfigure(sa->nic);
1296 		if (rc == EAGAIN) {
1297 			/*
1298 			 * Configuration is accepted by FW and MC reboot
1299 			 * is initiated to apply the changes. MC reboot
1300 			 * will be handled in a usual way (MC reboot
1301 			 * event on management event queue and adapter
1302 			 * restart).
1303 			 */
1304 			rc = 0;
1305 		} else if (rc != 0) {
1306 			goto fail_reconfigure;
1307 		}
1308 	}
1309 
1310 	sfc_adapter_unlock(sa);
1311 	return 0;
1312 
1313 fail_reconfigure:
1314 	/* Remove/restore entry since the change makes the trouble */
1315 	switch (op) {
1316 	case SFC_UDP_TUNNEL_ADD_PORT:
1317 		(void)efx_tunnel_config_udp_remove(sa->nic,
1318 						   tunnel_udp->udp_port,
1319 						   tunnel_proto);
1320 		break;
1321 	case SFC_UDP_TUNNEL_DEL_PORT:
1322 		(void)efx_tunnel_config_udp_add(sa->nic,
1323 						tunnel_udp->udp_port,
1324 						tunnel_proto);
1325 		break;
1326 	}
1327 
1328 fail_op:
1329 fail_bad_op:
1330 	sfc_adapter_unlock(sa);
1331 
1332 fail_bad_proto:
1333 	SFC_ASSERT(rc > 0);
1334 	return -rc;
1335 }
1336 
1337 static int
1338 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1339 			    struct rte_eth_udp_tunnel *tunnel_udp)
1340 {
1341 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1342 }
1343 
1344 static int
1345 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1346 			    struct rte_eth_udp_tunnel *tunnel_udp)
1347 {
1348 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1349 }
1350 
1351 static int
1352 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1353 			  struct rte_eth_rss_conf *rss_conf)
1354 {
1355 	struct sfc_adapter *sa = dev->data->dev_private;
1356 	struct sfc_rss *rss = &sa->rss;
1357 
1358 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1359 		return -ENOTSUP;
1360 
1361 	sfc_adapter_lock(sa);
1362 
1363 	/*
1364 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1365 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1366 	 * flags which corresponds to the active EFX configuration stored
1367 	 * locally in 'sfc_adapter' and kept up-to-date
1368 	 */
1369 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1370 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1371 	if (rss_conf->rss_key != NULL)
1372 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1373 
1374 	sfc_adapter_unlock(sa);
1375 
1376 	return 0;
1377 }
1378 
1379 static int
1380 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1381 			struct rte_eth_rss_conf *rss_conf)
1382 {
1383 	struct sfc_adapter *sa = dev->data->dev_private;
1384 	struct sfc_rss *rss = &sa->rss;
1385 	struct sfc_port *port = &sa->port;
1386 	unsigned int efx_hash_types;
1387 	int rc = 0;
1388 
1389 	if (port->isolated)
1390 		return -ENOTSUP;
1391 
1392 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1393 		sfc_err(sa, "RSS is not available");
1394 		return -ENOTSUP;
1395 	}
1396 
1397 	if (rss->channels == 0) {
1398 		sfc_err(sa, "RSS is not configured");
1399 		return -EINVAL;
1400 	}
1401 
1402 	if ((rss_conf->rss_key != NULL) &&
1403 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1404 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1405 			sizeof(rss->key));
1406 		return -EINVAL;
1407 	}
1408 
1409 	sfc_adapter_lock(sa);
1410 
1411 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1412 	if (rc != 0)
1413 		goto fail_rx_hf_rte_to_efx;
1414 
1415 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1416 				   rss->hash_alg, efx_hash_types, B_TRUE);
1417 	if (rc != 0)
1418 		goto fail_scale_mode_set;
1419 
1420 	if (rss_conf->rss_key != NULL) {
1421 		if (sa->state == SFC_ADAPTER_STARTED) {
1422 			rc = efx_rx_scale_key_set(sa->nic,
1423 						  EFX_RSS_CONTEXT_DEFAULT,
1424 						  rss_conf->rss_key,
1425 						  sizeof(rss->key));
1426 			if (rc != 0)
1427 				goto fail_scale_key_set;
1428 		}
1429 
1430 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1431 	}
1432 
1433 	rss->hash_types = efx_hash_types;
1434 
1435 	sfc_adapter_unlock(sa);
1436 
1437 	return 0;
1438 
1439 fail_scale_key_set:
1440 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1441 				  EFX_RX_HASHALG_TOEPLITZ,
1442 				  rss->hash_types, B_TRUE) != 0)
1443 		sfc_err(sa, "failed to restore RSS mode");
1444 
1445 fail_scale_mode_set:
1446 fail_rx_hf_rte_to_efx:
1447 	sfc_adapter_unlock(sa);
1448 	return -rc;
1449 }
1450 
1451 static int
1452 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1453 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1454 		       uint16_t reta_size)
1455 {
1456 	struct sfc_adapter *sa = dev->data->dev_private;
1457 	struct sfc_rss *rss = &sa->rss;
1458 	struct sfc_port *port = &sa->port;
1459 	int entry;
1460 
1461 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1462 		return -ENOTSUP;
1463 
1464 	if (rss->channels == 0)
1465 		return -EINVAL;
1466 
1467 	if (reta_size != EFX_RSS_TBL_SIZE)
1468 		return -EINVAL;
1469 
1470 	sfc_adapter_lock(sa);
1471 
1472 	for (entry = 0; entry < reta_size; entry++) {
1473 		int grp = entry / RTE_RETA_GROUP_SIZE;
1474 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1475 
1476 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1477 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1478 	}
1479 
1480 	sfc_adapter_unlock(sa);
1481 
1482 	return 0;
1483 }
1484 
1485 static int
1486 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1487 			struct rte_eth_rss_reta_entry64 *reta_conf,
1488 			uint16_t reta_size)
1489 {
1490 	struct sfc_adapter *sa = dev->data->dev_private;
1491 	struct sfc_rss *rss = &sa->rss;
1492 	struct sfc_port *port = &sa->port;
1493 	unsigned int *rss_tbl_new;
1494 	uint16_t entry;
1495 	int rc = 0;
1496 
1497 
1498 	if (port->isolated)
1499 		return -ENOTSUP;
1500 
1501 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1502 		sfc_err(sa, "RSS is not available");
1503 		return -ENOTSUP;
1504 	}
1505 
1506 	if (rss->channels == 0) {
1507 		sfc_err(sa, "RSS is not configured");
1508 		return -EINVAL;
1509 	}
1510 
1511 	if (reta_size != EFX_RSS_TBL_SIZE) {
1512 		sfc_err(sa, "RETA size is wrong (should be %u)",
1513 			EFX_RSS_TBL_SIZE);
1514 		return -EINVAL;
1515 	}
1516 
1517 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1518 	if (rss_tbl_new == NULL)
1519 		return -ENOMEM;
1520 
1521 	sfc_adapter_lock(sa);
1522 
1523 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1524 
1525 	for (entry = 0; entry < reta_size; entry++) {
1526 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1527 		struct rte_eth_rss_reta_entry64 *grp;
1528 
1529 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1530 
1531 		if (grp->mask & (1ull << grp_idx)) {
1532 			if (grp->reta[grp_idx] >= rss->channels) {
1533 				rc = EINVAL;
1534 				goto bad_reta_entry;
1535 			}
1536 			rss_tbl_new[entry] = grp->reta[grp_idx];
1537 		}
1538 	}
1539 
1540 	if (sa->state == SFC_ADAPTER_STARTED) {
1541 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1542 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1543 		if (rc != 0)
1544 			goto fail_scale_tbl_set;
1545 	}
1546 
1547 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1548 
1549 fail_scale_tbl_set:
1550 bad_reta_entry:
1551 	sfc_adapter_unlock(sa);
1552 
1553 	rte_free(rss_tbl_new);
1554 
1555 	SFC_ASSERT(rc >= 0);
1556 	return -rc;
1557 }
1558 
1559 static int
1560 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1561 		    enum rte_filter_op filter_op,
1562 		    void *arg)
1563 {
1564 	struct sfc_adapter *sa = dev->data->dev_private;
1565 	int rc = ENOTSUP;
1566 
1567 	sfc_log_init(sa, "entry");
1568 
1569 	switch (filter_type) {
1570 	case RTE_ETH_FILTER_NONE:
1571 		sfc_err(sa, "Global filters configuration not supported");
1572 		break;
1573 	case RTE_ETH_FILTER_MACVLAN:
1574 		sfc_err(sa, "MACVLAN filters not supported");
1575 		break;
1576 	case RTE_ETH_FILTER_ETHERTYPE:
1577 		sfc_err(sa, "EtherType filters not supported");
1578 		break;
1579 	case RTE_ETH_FILTER_FLEXIBLE:
1580 		sfc_err(sa, "Flexible filters not supported");
1581 		break;
1582 	case RTE_ETH_FILTER_SYN:
1583 		sfc_err(sa, "SYN filters not supported");
1584 		break;
1585 	case RTE_ETH_FILTER_NTUPLE:
1586 		sfc_err(sa, "NTUPLE filters not supported");
1587 		break;
1588 	case RTE_ETH_FILTER_TUNNEL:
1589 		sfc_err(sa, "Tunnel filters not supported");
1590 		break;
1591 	case RTE_ETH_FILTER_FDIR:
1592 		sfc_err(sa, "Flow Director filters not supported");
1593 		break;
1594 	case RTE_ETH_FILTER_HASH:
1595 		sfc_err(sa, "Hash filters not supported");
1596 		break;
1597 	case RTE_ETH_FILTER_GENERIC:
1598 		if (filter_op != RTE_ETH_FILTER_GET) {
1599 			rc = EINVAL;
1600 		} else {
1601 			*(const void **)arg = &sfc_flow_ops;
1602 			rc = 0;
1603 		}
1604 		break;
1605 	default:
1606 		sfc_err(sa, "Unknown filter type %u", filter_type);
1607 		break;
1608 	}
1609 
1610 	sfc_log_init(sa, "exit: %d", -rc);
1611 	SFC_ASSERT(rc >= 0);
1612 	return -rc;
1613 }
1614 
1615 static int
1616 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1617 {
1618 	struct sfc_adapter *sa = dev->data->dev_private;
1619 
1620 	/*
1621 	 * If Rx datapath does not provide callback to check mempool,
1622 	 * all pools are supported.
1623 	 */
1624 	if (sa->dp_rx->pool_ops_supported == NULL)
1625 		return 1;
1626 
1627 	return sa->dp_rx->pool_ops_supported(pool);
1628 }
1629 
1630 static const struct eth_dev_ops sfc_eth_dev_ops = {
1631 	.dev_configure			= sfc_dev_configure,
1632 	.dev_start			= sfc_dev_start,
1633 	.dev_stop			= sfc_dev_stop,
1634 	.dev_set_link_up		= sfc_dev_set_link_up,
1635 	.dev_set_link_down		= sfc_dev_set_link_down,
1636 	.dev_close			= sfc_dev_close,
1637 	.promiscuous_enable		= sfc_dev_promisc_enable,
1638 	.promiscuous_disable		= sfc_dev_promisc_disable,
1639 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1640 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1641 	.link_update			= sfc_dev_link_update,
1642 	.stats_get			= sfc_stats_get,
1643 	.stats_reset			= sfc_stats_reset,
1644 	.xstats_get			= sfc_xstats_get,
1645 	.xstats_reset			= sfc_stats_reset,
1646 	.xstats_get_names		= sfc_xstats_get_names,
1647 	.dev_infos_get			= sfc_dev_infos_get,
1648 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1649 	.mtu_set			= sfc_dev_set_mtu,
1650 	.rx_queue_start			= sfc_rx_queue_start,
1651 	.rx_queue_stop			= sfc_rx_queue_stop,
1652 	.tx_queue_start			= sfc_tx_queue_start,
1653 	.tx_queue_stop			= sfc_tx_queue_stop,
1654 	.rx_queue_setup			= sfc_rx_queue_setup,
1655 	.rx_queue_release		= sfc_rx_queue_release,
1656 	.rx_queue_count			= sfc_rx_queue_count,
1657 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1658 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1659 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1660 	.tx_queue_setup			= sfc_tx_queue_setup,
1661 	.tx_queue_release		= sfc_tx_queue_release,
1662 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1663 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1664 	.mac_addr_set			= sfc_mac_addr_set,
1665 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1666 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1667 	.reta_update			= sfc_dev_rss_reta_update,
1668 	.reta_query			= sfc_dev_rss_reta_query,
1669 	.rss_hash_update		= sfc_dev_rss_hash_update,
1670 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1671 	.filter_ctrl			= sfc_dev_filter_ctrl,
1672 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1673 	.rxq_info_get			= sfc_rx_queue_info_get,
1674 	.txq_info_get			= sfc_tx_queue_info_get,
1675 	.fw_version_get			= sfc_fw_version_get,
1676 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1677 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1678 	.pool_ops_supported		= sfc_pool_ops_supported,
1679 };
1680 
1681 /**
1682  * Duplicate a string in potentially shared memory required for
1683  * multi-process support.
1684  *
1685  * strdup() allocates from process-local heap/memory.
1686  */
1687 static char *
1688 sfc_strdup(const char *str)
1689 {
1690 	size_t size;
1691 	char *copy;
1692 
1693 	if (str == NULL)
1694 		return NULL;
1695 
1696 	size = strlen(str) + 1;
1697 	copy = rte_malloc(__func__, size, 0);
1698 	if (copy != NULL)
1699 		rte_memcpy(copy, str, size);
1700 
1701 	return copy;
1702 }
1703 
1704 static int
1705 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1706 {
1707 	struct sfc_adapter *sa = dev->data->dev_private;
1708 	const efx_nic_cfg_t *encp;
1709 	unsigned int avail_caps = 0;
1710 	const char *rx_name = NULL;
1711 	const char *tx_name = NULL;
1712 	int rc;
1713 
1714 	switch (sa->family) {
1715 	case EFX_FAMILY_HUNTINGTON:
1716 	case EFX_FAMILY_MEDFORD:
1717 	case EFX_FAMILY_MEDFORD2:
1718 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1719 		break;
1720 	default:
1721 		break;
1722 	}
1723 
1724 	encp = efx_nic_cfg_get(sa->nic);
1725 	if (encp->enc_rx_es_super_buffer_supported)
1726 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1727 
1728 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1729 				sfc_kvarg_string_handler, &rx_name);
1730 	if (rc != 0)
1731 		goto fail_kvarg_rx_datapath;
1732 
1733 	if (rx_name != NULL) {
1734 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1735 		if (sa->dp_rx == NULL) {
1736 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1737 			rc = ENOENT;
1738 			goto fail_dp_rx;
1739 		}
1740 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1741 			sfc_err(sa,
1742 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1743 				rx_name);
1744 			rc = EINVAL;
1745 			goto fail_dp_rx_caps;
1746 		}
1747 	} else {
1748 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1749 		if (sa->dp_rx == NULL) {
1750 			sfc_err(sa, "Rx datapath by caps %#x not found",
1751 				avail_caps);
1752 			rc = ENOENT;
1753 			goto fail_dp_rx;
1754 		}
1755 	}
1756 
1757 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1758 	if (sa->dp_rx_name == NULL) {
1759 		rc = ENOMEM;
1760 		goto fail_dp_rx_name;
1761 	}
1762 
1763 	sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1764 
1765 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1766 
1767 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1768 				sfc_kvarg_string_handler, &tx_name);
1769 	if (rc != 0)
1770 		goto fail_kvarg_tx_datapath;
1771 
1772 	if (tx_name != NULL) {
1773 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1774 		if (sa->dp_tx == NULL) {
1775 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1776 			rc = ENOENT;
1777 			goto fail_dp_tx;
1778 		}
1779 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1780 			sfc_err(sa,
1781 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1782 				tx_name);
1783 			rc = EINVAL;
1784 			goto fail_dp_tx_caps;
1785 		}
1786 	} else {
1787 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1788 		if (sa->dp_tx == NULL) {
1789 			sfc_err(sa, "Tx datapath by caps %#x not found",
1790 				avail_caps);
1791 			rc = ENOENT;
1792 			goto fail_dp_tx;
1793 		}
1794 	}
1795 
1796 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1797 	if (sa->dp_tx_name == NULL) {
1798 		rc = ENOMEM;
1799 		goto fail_dp_tx_name;
1800 	}
1801 
1802 	sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1803 
1804 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1805 
1806 	dev->dev_ops = &sfc_eth_dev_ops;
1807 
1808 	return 0;
1809 
1810 fail_dp_tx_name:
1811 fail_dp_tx_caps:
1812 	sa->dp_tx = NULL;
1813 
1814 fail_dp_tx:
1815 fail_kvarg_tx_datapath:
1816 	rte_free(sa->dp_rx_name);
1817 	sa->dp_rx_name = NULL;
1818 
1819 fail_dp_rx_name:
1820 fail_dp_rx_caps:
1821 	sa->dp_rx = NULL;
1822 
1823 fail_dp_rx:
1824 fail_kvarg_rx_datapath:
1825 	return rc;
1826 }
1827 
1828 static void
1829 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1830 {
1831 	struct sfc_adapter *sa = dev->data->dev_private;
1832 
1833 	dev->dev_ops = NULL;
1834 	dev->rx_pkt_burst = NULL;
1835 	dev->tx_pkt_burst = NULL;
1836 
1837 	rte_free(sa->dp_tx_name);
1838 	sa->dp_tx_name = NULL;
1839 	sa->dp_tx = NULL;
1840 
1841 	rte_free(sa->dp_rx_name);
1842 	sa->dp_rx_name = NULL;
1843 	sa->dp_rx = NULL;
1844 }
1845 
1846 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1847 	.rxq_info_get			= sfc_rx_queue_info_get,
1848 	.txq_info_get			= sfc_tx_queue_info_get,
1849 };
1850 
1851 static int
1852 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1853 {
1854 	/*
1855 	 * Device private data has really many process-local pointers.
1856 	 * Below code should be extremely careful to use data located
1857 	 * in shared memory only.
1858 	 */
1859 	struct sfc_adapter *sa = dev->data->dev_private;
1860 	const struct sfc_dp_rx *dp_rx;
1861 	const struct sfc_dp_tx *dp_tx;
1862 	int rc;
1863 
1864 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1865 	if (dp_rx == NULL) {
1866 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1867 		rc = ENOENT;
1868 		goto fail_dp_rx;
1869 	}
1870 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1871 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1872 			sa->dp_tx_name);
1873 		rc = EINVAL;
1874 		goto fail_dp_rx_multi_process;
1875 	}
1876 
1877 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1878 	if (dp_tx == NULL) {
1879 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1880 		rc = ENOENT;
1881 		goto fail_dp_tx;
1882 	}
1883 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1884 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1885 			sa->dp_tx_name);
1886 		rc = EINVAL;
1887 		goto fail_dp_tx_multi_process;
1888 	}
1889 
1890 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1891 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1892 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1893 
1894 	return 0;
1895 
1896 fail_dp_tx_multi_process:
1897 fail_dp_tx:
1898 fail_dp_rx_multi_process:
1899 fail_dp_rx:
1900 	return rc;
1901 }
1902 
1903 static void
1904 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1905 {
1906 	dev->dev_ops = NULL;
1907 	dev->tx_pkt_burst = NULL;
1908 	dev->rx_pkt_burst = NULL;
1909 }
1910 
1911 static void
1912 sfc_register_dp(void)
1913 {
1914 	/* Register once */
1915 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1916 		/* Prefer EF10 datapath */
1917 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1918 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1919 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1920 
1921 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1922 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1923 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1924 	}
1925 }
1926 
1927 static int
1928 sfc_eth_dev_init(struct rte_eth_dev *dev)
1929 {
1930 	struct sfc_adapter *sa = dev->data->dev_private;
1931 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1932 	int rc;
1933 	const efx_nic_cfg_t *encp;
1934 	const struct ether_addr *from;
1935 
1936 	sfc_register_dp();
1937 
1938 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1939 		return -sfc_eth_dev_secondary_set_ops(dev);
1940 
1941 	/* Required for logging */
1942 	sa->pci_addr = pci_dev->addr;
1943 	sa->port_id = dev->data->port_id;
1944 
1945 	sa->eth_dev = dev;
1946 
1947 	/* Copy PCI device info to the dev->data */
1948 	rte_eth_copy_pci_info(dev, pci_dev);
1949 
1950 	sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1951 						RTE_LOG_NOTICE);
1952 
1953 	rc = sfc_kvargs_parse(sa);
1954 	if (rc != 0)
1955 		goto fail_kvargs_parse;
1956 
1957 	sfc_log_init(sa, "entry");
1958 
1959 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1960 	if (dev->data->mac_addrs == NULL) {
1961 		rc = ENOMEM;
1962 		goto fail_mac_addrs;
1963 	}
1964 
1965 	sfc_adapter_lock_init(sa);
1966 	sfc_adapter_lock(sa);
1967 
1968 	sfc_log_init(sa, "probing");
1969 	rc = sfc_probe(sa);
1970 	if (rc != 0)
1971 		goto fail_probe;
1972 
1973 	sfc_log_init(sa, "set device ops");
1974 	rc = sfc_eth_dev_set_ops(dev);
1975 	if (rc != 0)
1976 		goto fail_set_ops;
1977 
1978 	sfc_log_init(sa, "attaching");
1979 	rc = sfc_attach(sa);
1980 	if (rc != 0)
1981 		goto fail_attach;
1982 
1983 	encp = efx_nic_cfg_get(sa->nic);
1984 
1985 	/*
1986 	 * The arguments are really reverse order in comparison to
1987 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1988 	 */
1989 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1990 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1991 
1992 	sfc_adapter_unlock(sa);
1993 
1994 	sfc_log_init(sa, "done");
1995 	return 0;
1996 
1997 fail_attach:
1998 	sfc_eth_dev_clear_ops(dev);
1999 
2000 fail_set_ops:
2001 	sfc_unprobe(sa);
2002 
2003 fail_probe:
2004 	sfc_adapter_unlock(sa);
2005 	sfc_adapter_lock_fini(sa);
2006 	rte_free(dev->data->mac_addrs);
2007 	dev->data->mac_addrs = NULL;
2008 
2009 fail_mac_addrs:
2010 	sfc_kvargs_cleanup(sa);
2011 
2012 fail_kvargs_parse:
2013 	sfc_log_init(sa, "failed %d", rc);
2014 	SFC_ASSERT(rc > 0);
2015 	return -rc;
2016 }
2017 
2018 static int
2019 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2020 {
2021 	struct sfc_adapter *sa;
2022 
2023 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2024 		sfc_eth_dev_secondary_clear_ops(dev);
2025 		return 0;
2026 	}
2027 
2028 	sa = dev->data->dev_private;
2029 	sfc_log_init(sa, "entry");
2030 
2031 	sfc_adapter_lock(sa);
2032 
2033 	sfc_eth_dev_clear_ops(dev);
2034 
2035 	sfc_detach(sa);
2036 	sfc_unprobe(sa);
2037 
2038 	rte_free(dev->data->mac_addrs);
2039 	dev->data->mac_addrs = NULL;
2040 
2041 	sfc_kvargs_cleanup(sa);
2042 
2043 	sfc_adapter_unlock(sa);
2044 	sfc_adapter_lock_fini(sa);
2045 
2046 	sfc_log_init(sa, "done");
2047 
2048 	/* Required for logging, so cleanup last */
2049 	sa->eth_dev = NULL;
2050 	return 0;
2051 }
2052 
2053 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2054 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2055 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2056 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2057 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2058 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2059 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2060 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2061 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2062 	{ .vendor_id = 0 /* sentinel */ }
2063 };
2064 
2065 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2066 	struct rte_pci_device *pci_dev)
2067 {
2068 	return rte_eth_dev_pci_generic_probe(pci_dev,
2069 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2070 }
2071 
2072 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2073 {
2074 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2075 }
2076 
2077 static struct rte_pci_driver sfc_efx_pmd = {
2078 	.id_table = pci_id_sfc_efx_map,
2079 	.drv_flags =
2080 		RTE_PCI_DRV_INTR_LSC |
2081 		RTE_PCI_DRV_NEED_MAPPING,
2082 	.probe = sfc_eth_dev_pci_probe,
2083 	.remove = sfc_eth_dev_pci_remove,
2084 };
2085 
2086 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2087 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2088 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2089 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2090 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2091 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2092 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2093 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2094 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2095 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2096 
2097 RTE_INIT(sfc_driver_register_logtype)
2098 {
2099 	int ret;
2100 
2101 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2102 						   RTE_LOG_NOTICE);
2103 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2104 }
2105