xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision fdceb100c7ced8511ea0191cb084e38ee6c1162f)
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 
17 #include "efx.h"
18 
19 #include "sfc.h"
20 #include "sfc_debug.h"
21 #include "sfc_log.h"
22 #include "sfc_kvargs.h"
23 #include "sfc_ev.h"
24 #include "sfc_rx.h"
25 #include "sfc_tx.h"
26 #include "sfc_flow.h"
27 #include "sfc_dp.h"
28 #include "sfc_dp_rx.h"
29 
30 uint32_t sfc_logtype_driver;
31 
32 static struct sfc_dp_list sfc_dp_head =
33 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
34 
35 static int
36 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
37 {
38 	struct sfc_adapter *sa = dev->data->dev_private;
39 	efx_nic_fw_info_t enfi;
40 	int ret;
41 	int rc;
42 
43 	/*
44 	 * Return value of the callback is likely supposed to be
45 	 * equal to or greater than 0, nevertheless, if an error
46 	 * occurs, it will be desirable to pass it to the caller
47 	 */
48 	if ((fw_version == NULL) || (fw_size == 0))
49 		return -EINVAL;
50 
51 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
52 	if (rc != 0)
53 		return -rc;
54 
55 	ret = snprintf(fw_version, fw_size,
56 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
57 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
58 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
59 	if (ret < 0)
60 		return ret;
61 
62 	if (enfi.enfi_dpcpu_fw_ids_valid) {
63 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
64 		int ret_extra;
65 
66 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
67 				     fw_size - dpcpu_fw_ids_offset,
68 				     " rx%" PRIx16 " tx%" PRIx16,
69 				     enfi.enfi_rx_dpcpu_fw_id,
70 				     enfi.enfi_tx_dpcpu_fw_id);
71 		if (ret_extra < 0)
72 			return ret_extra;
73 
74 		ret += ret_extra;
75 	}
76 
77 	if (fw_size < (size_t)(++ret))
78 		return ret;
79 	else
80 		return 0;
81 }
82 
83 static void
84 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
85 {
86 	struct sfc_adapter *sa = dev->data->dev_private;
87 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
88 	uint64_t txq_offloads_def = 0;
89 
90 	sfc_log_init(sa, "entry");
91 
92 	dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
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_40000FDX)
102 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
103 
104 	dev_info->max_rx_queues = sa->rxq_max;
105 	dev_info->max_tx_queues = sa->txq_max;
106 
107 	/* By default packets are dropped if no descriptors are available */
108 	dev_info->default_rxconf.rx_drop_en = 1;
109 
110 	dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
111 
112 	/*
113 	 * rx_offload_capa includes both device and queue offloads since
114 	 * the latter may be requested on a per device basis which makes
115 	 * sense when some offloads are needed to be set on all queues.
116 	 */
117 	dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
118 				    dev_info->rx_queue_offload_capa;
119 
120 	dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
121 
122 	/*
123 	 * tx_offload_capa includes both device and queue offloads since
124 	 * the latter may be requested on a per device basis which makes
125 	 * sense when some offloads are needed to be set on all queues.
126 	 */
127 	dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
128 				    dev_info->tx_queue_offload_capa;
129 
130 	if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
131 		txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
132 
133 	dev_info->default_txconf.offloads |= txq_offloads_def;
134 
135 	dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
136 	if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
137 	    !encp->enc_hw_tx_insert_vlan_enabled)
138 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
139 
140 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
141 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
142 
143 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
144 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
145 
146 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
147 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
148 
149 #if EFSYS_OPT_RX_SCALE
150 	if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
151 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
152 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
153 		dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
154 	}
155 #endif
156 
157 	/* Initialize to hardware limits */
158 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
159 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
160 	/* The RXQ hardware requires that the descriptor count is a power
161 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
162 	 */
163 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
164 
165 	/* Initialize to hardware limits */
166 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
167 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
168 	/*
169 	 * The TXQ hardware requires that the descriptor count is a power
170 	 * of 2, but tx_desc_lim cannot properly describe that constraint
171 	 */
172 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
173 
174 	if (sa->dp_rx->get_dev_info != NULL)
175 		sa->dp_rx->get_dev_info(dev_info);
176 	if (sa->dp_tx->get_dev_info != NULL)
177 		sa->dp_tx->get_dev_info(dev_info);
178 }
179 
180 static const uint32_t *
181 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
182 {
183 	struct sfc_adapter *sa = dev->data->dev_private;
184 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
185 	uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
186 
187 	return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
188 }
189 
190 static int
191 sfc_dev_configure(struct rte_eth_dev *dev)
192 {
193 	struct rte_eth_dev_data *dev_data = dev->data;
194 	struct sfc_adapter *sa = dev_data->dev_private;
195 	int rc;
196 
197 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
198 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
199 
200 	sfc_adapter_lock(sa);
201 	switch (sa->state) {
202 	case SFC_ADAPTER_CONFIGURED:
203 		/* FALLTHROUGH */
204 	case SFC_ADAPTER_INITIALIZED:
205 		rc = sfc_configure(sa);
206 		break;
207 	default:
208 		sfc_err(sa, "unexpected adapter state %u to configure",
209 			sa->state);
210 		rc = EINVAL;
211 		break;
212 	}
213 	sfc_adapter_unlock(sa);
214 
215 	sfc_log_init(sa, "done %d", rc);
216 	SFC_ASSERT(rc >= 0);
217 	return -rc;
218 }
219 
220 static int
221 sfc_dev_start(struct rte_eth_dev *dev)
222 {
223 	struct sfc_adapter *sa = dev->data->dev_private;
224 	int rc;
225 
226 	sfc_log_init(sa, "entry");
227 
228 	sfc_adapter_lock(sa);
229 	rc = sfc_start(sa);
230 	sfc_adapter_unlock(sa);
231 
232 	sfc_log_init(sa, "done %d", rc);
233 	SFC_ASSERT(rc >= 0);
234 	return -rc;
235 }
236 
237 static int
238 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
239 {
240 	struct sfc_adapter *sa = dev->data->dev_private;
241 	struct rte_eth_link current_link;
242 	int ret;
243 
244 	sfc_log_init(sa, "entry");
245 
246 	if (sa->state != SFC_ADAPTER_STARTED) {
247 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
248 	} else if (wait_to_complete) {
249 		efx_link_mode_t link_mode;
250 
251 		if (efx_port_poll(sa->nic, &link_mode) != 0)
252 			link_mode = EFX_LINK_UNKNOWN;
253 		sfc_port_link_mode_to_info(link_mode, &current_link);
254 
255 	} else {
256 		sfc_ev_mgmt_qpoll(sa);
257 		rte_eth_linkstatus_get(dev, &current_link);
258 	}
259 
260 	ret = rte_eth_linkstatus_set(dev, &current_link);
261 	if (ret == 0)
262 		sfc_info(sa, "Link status is %s",
263 			 current_link.link_status ? "UP" : "DOWN");
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 	sa->eth_dev->data->rx_queues[sw_index] = NULL;
448 
449 	sfc_rx_qfini(sa, sw_index);
450 
451 	sfc_adapter_unlock(sa);
452 }
453 
454 static int
455 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
456 		   uint16_t nb_tx_desc, unsigned int socket_id,
457 		   const struct rte_eth_txconf *tx_conf)
458 {
459 	struct sfc_adapter *sa = dev->data->dev_private;
460 	int rc;
461 
462 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
463 		     tx_queue_id, nb_tx_desc, socket_id);
464 
465 	sfc_adapter_lock(sa);
466 
467 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
468 	if (rc != 0)
469 		goto fail_tx_qinit;
470 
471 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
472 
473 	sfc_adapter_unlock(sa);
474 	return 0;
475 
476 fail_tx_qinit:
477 	sfc_adapter_unlock(sa);
478 	SFC_ASSERT(rc > 0);
479 	return -rc;
480 }
481 
482 static void
483 sfc_tx_queue_release(void *queue)
484 {
485 	struct sfc_dp_txq *dp_txq = queue;
486 	struct sfc_txq *txq;
487 	unsigned int sw_index;
488 	struct sfc_adapter *sa;
489 
490 	if (dp_txq == NULL)
491 		return;
492 
493 	txq = sfc_txq_by_dp_txq(dp_txq);
494 	sw_index = sfc_txq_sw_index(txq);
495 
496 	SFC_ASSERT(txq->evq != NULL);
497 	sa = txq->evq->sa;
498 
499 	sfc_log_init(sa, "TxQ = %u", sw_index);
500 
501 	sfc_adapter_lock(sa);
502 
503 	SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
504 	sa->eth_dev->data->tx_queues[sw_index] = NULL;
505 
506 	sfc_tx_qfini(sa, sw_index);
507 
508 	sfc_adapter_unlock(sa);
509 }
510 
511 static int
512 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
513 {
514 	struct sfc_adapter *sa = dev->data->dev_private;
515 	struct sfc_port *port = &sa->port;
516 	uint64_t *mac_stats;
517 	int ret;
518 
519 	rte_spinlock_lock(&port->mac_stats_lock);
520 
521 	ret = sfc_port_update_mac_stats(sa);
522 	if (ret != 0)
523 		goto unlock;
524 
525 	mac_stats = port->mac_stats_buf;
526 
527 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
528 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
529 		stats->ipackets =
530 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
531 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
532 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
533 		stats->opackets =
534 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
535 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
536 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
537 		stats->ibytes =
538 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
539 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
540 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
541 		stats->obytes =
542 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
543 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
544 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
545 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
546 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
547 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
548 	} else {
549 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
550 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
551 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
552 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
553 		/*
554 		 * Take into account stats which are whenever supported
555 		 * on EF10. If some stat is not supported by current
556 		 * firmware variant or HW revision, it is guaranteed
557 		 * to be zero in mac_stats.
558 		 */
559 		stats->imissed =
560 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
561 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
562 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
563 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
564 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
565 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
566 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
567 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
568 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
569 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
570 		stats->ierrors =
571 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
572 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
573 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
574 		/* no oerrors counters supported on EF10 */
575 	}
576 
577 unlock:
578 	rte_spinlock_unlock(&port->mac_stats_lock);
579 	SFC_ASSERT(ret >= 0);
580 	return -ret;
581 }
582 
583 static void
584 sfc_stats_reset(struct rte_eth_dev *dev)
585 {
586 	struct sfc_adapter *sa = dev->data->dev_private;
587 	struct sfc_port *port = &sa->port;
588 	int rc;
589 
590 	if (sa->state != SFC_ADAPTER_STARTED) {
591 		/*
592 		 * The operation cannot be done if port is not started; it
593 		 * will be scheduled to be done during the next port start
594 		 */
595 		port->mac_stats_reset_pending = B_TRUE;
596 		return;
597 	}
598 
599 	rc = sfc_port_reset_mac_stats(sa);
600 	if (rc != 0)
601 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
602 }
603 
604 static int
605 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
606 	       unsigned int xstats_count)
607 {
608 	struct sfc_adapter *sa = dev->data->dev_private;
609 	struct sfc_port *port = &sa->port;
610 	uint64_t *mac_stats;
611 	int rc;
612 	unsigned int i;
613 	int nstats = 0;
614 
615 	rte_spinlock_lock(&port->mac_stats_lock);
616 
617 	rc = sfc_port_update_mac_stats(sa);
618 	if (rc != 0) {
619 		SFC_ASSERT(rc > 0);
620 		nstats = -rc;
621 		goto unlock;
622 	}
623 
624 	mac_stats = port->mac_stats_buf;
625 
626 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
627 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
628 			if (xstats != NULL && nstats < (int)xstats_count) {
629 				xstats[nstats].id = nstats;
630 				xstats[nstats].value = mac_stats[i];
631 			}
632 			nstats++;
633 		}
634 	}
635 
636 unlock:
637 	rte_spinlock_unlock(&port->mac_stats_lock);
638 
639 	return nstats;
640 }
641 
642 static int
643 sfc_xstats_get_names(struct rte_eth_dev *dev,
644 		     struct rte_eth_xstat_name *xstats_names,
645 		     unsigned int xstats_count)
646 {
647 	struct sfc_adapter *sa = dev->data->dev_private;
648 	struct sfc_port *port = &sa->port;
649 	unsigned int i;
650 	unsigned int nstats = 0;
651 
652 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
653 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
654 			if (xstats_names != NULL && nstats < xstats_count)
655 				strncpy(xstats_names[nstats].name,
656 					efx_mac_stat_name(sa->nic, i),
657 					sizeof(xstats_names[0].name));
658 			nstats++;
659 		}
660 	}
661 
662 	return nstats;
663 }
664 
665 static int
666 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
667 		     uint64_t *values, unsigned int n)
668 {
669 	struct sfc_adapter *sa = dev->data->dev_private;
670 	struct sfc_port *port = &sa->port;
671 	uint64_t *mac_stats;
672 	unsigned int nb_supported = 0;
673 	unsigned int nb_written = 0;
674 	unsigned int i;
675 	int ret;
676 	int rc;
677 
678 	if (unlikely(values == NULL) ||
679 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
680 		return port->mac_stats_nb_supported;
681 
682 	rte_spinlock_lock(&port->mac_stats_lock);
683 
684 	rc = sfc_port_update_mac_stats(sa);
685 	if (rc != 0) {
686 		SFC_ASSERT(rc > 0);
687 		ret = -rc;
688 		goto unlock;
689 	}
690 
691 	mac_stats = port->mac_stats_buf;
692 
693 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
694 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
695 			continue;
696 
697 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
698 			values[nb_written++] = mac_stats[i];
699 
700 		++nb_supported;
701 	}
702 
703 	ret = nb_written;
704 
705 unlock:
706 	rte_spinlock_unlock(&port->mac_stats_lock);
707 
708 	return ret;
709 }
710 
711 static int
712 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
713 			   struct rte_eth_xstat_name *xstats_names,
714 			   const uint64_t *ids, unsigned int size)
715 {
716 	struct sfc_adapter *sa = dev->data->dev_private;
717 	struct sfc_port *port = &sa->port;
718 	unsigned int nb_supported = 0;
719 	unsigned int nb_written = 0;
720 	unsigned int i;
721 
722 	if (unlikely(xstats_names == NULL) ||
723 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
724 		return port->mac_stats_nb_supported;
725 
726 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
727 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
728 			continue;
729 
730 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
731 			char *name = xstats_names[nb_written++].name;
732 
733 			strncpy(name, efx_mac_stat_name(sa->nic, i),
734 				sizeof(xstats_names[0].name));
735 			name[sizeof(xstats_names[0].name) - 1] = '\0';
736 		}
737 
738 		++nb_supported;
739 	}
740 
741 	return nb_written;
742 }
743 
744 static int
745 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
746 {
747 	struct sfc_adapter *sa = dev->data->dev_private;
748 	unsigned int wanted_fc, link_fc;
749 
750 	memset(fc_conf, 0, sizeof(*fc_conf));
751 
752 	sfc_adapter_lock(sa);
753 
754 	if (sa->state == SFC_ADAPTER_STARTED)
755 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
756 	else
757 		link_fc = sa->port.flow_ctrl;
758 
759 	switch (link_fc) {
760 	case 0:
761 		fc_conf->mode = RTE_FC_NONE;
762 		break;
763 	case EFX_FCNTL_RESPOND:
764 		fc_conf->mode = RTE_FC_RX_PAUSE;
765 		break;
766 	case EFX_FCNTL_GENERATE:
767 		fc_conf->mode = RTE_FC_TX_PAUSE;
768 		break;
769 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
770 		fc_conf->mode = RTE_FC_FULL;
771 		break;
772 	default:
773 		sfc_err(sa, "%s: unexpected flow control value %#x",
774 			__func__, link_fc);
775 	}
776 
777 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
778 
779 	sfc_adapter_unlock(sa);
780 
781 	return 0;
782 }
783 
784 static int
785 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
786 {
787 	struct sfc_adapter *sa = dev->data->dev_private;
788 	struct sfc_port *port = &sa->port;
789 	unsigned int fcntl;
790 	int rc;
791 
792 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
793 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
794 	    fc_conf->mac_ctrl_frame_fwd != 0) {
795 		sfc_err(sa, "unsupported flow control settings specified");
796 		rc = EINVAL;
797 		goto fail_inval;
798 	}
799 
800 	switch (fc_conf->mode) {
801 	case RTE_FC_NONE:
802 		fcntl = 0;
803 		break;
804 	case RTE_FC_RX_PAUSE:
805 		fcntl = EFX_FCNTL_RESPOND;
806 		break;
807 	case RTE_FC_TX_PAUSE:
808 		fcntl = EFX_FCNTL_GENERATE;
809 		break;
810 	case RTE_FC_FULL:
811 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
812 		break;
813 	default:
814 		rc = EINVAL;
815 		goto fail_inval;
816 	}
817 
818 	sfc_adapter_lock(sa);
819 
820 	if (sa->state == SFC_ADAPTER_STARTED) {
821 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
822 		if (rc != 0)
823 			goto fail_mac_fcntl_set;
824 	}
825 
826 	port->flow_ctrl = fcntl;
827 	port->flow_ctrl_autoneg = fc_conf->autoneg;
828 
829 	sfc_adapter_unlock(sa);
830 
831 	return 0;
832 
833 fail_mac_fcntl_set:
834 	sfc_adapter_unlock(sa);
835 fail_inval:
836 	SFC_ASSERT(rc > 0);
837 	return -rc;
838 }
839 
840 static int
841 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
842 {
843 	struct sfc_adapter *sa = dev->data->dev_private;
844 	size_t pdu = EFX_MAC_PDU(mtu);
845 	size_t old_pdu;
846 	int rc;
847 
848 	sfc_log_init(sa, "mtu=%u", mtu);
849 
850 	rc = EINVAL;
851 	if (pdu < EFX_MAC_PDU_MIN) {
852 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
853 			(unsigned int)mtu, (unsigned int)pdu,
854 			EFX_MAC_PDU_MIN);
855 		goto fail_inval;
856 	}
857 	if (pdu > EFX_MAC_PDU_MAX) {
858 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
859 			(unsigned int)mtu, (unsigned int)pdu,
860 			EFX_MAC_PDU_MAX);
861 		goto fail_inval;
862 	}
863 
864 	sfc_adapter_lock(sa);
865 
866 	if (pdu != sa->port.pdu) {
867 		if (sa->state == SFC_ADAPTER_STARTED) {
868 			sfc_stop(sa);
869 
870 			old_pdu = sa->port.pdu;
871 			sa->port.pdu = pdu;
872 			rc = sfc_start(sa);
873 			if (rc != 0)
874 				goto fail_start;
875 		} else {
876 			sa->port.pdu = pdu;
877 		}
878 	}
879 
880 	/*
881 	 * The driver does not use it, but other PMDs update jumbo_frame
882 	 * flag and max_rx_pkt_len when MTU is set.
883 	 */
884 	if (mtu > ETHER_MAX_LEN) {
885 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
886 
887 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
888 		rxmode->jumbo_frame = 1;
889 	}
890 
891 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
892 
893 	sfc_adapter_unlock(sa);
894 
895 	sfc_log_init(sa, "done");
896 	return 0;
897 
898 fail_start:
899 	sa->port.pdu = old_pdu;
900 	if (sfc_start(sa) != 0)
901 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
902 			"PDU max size - port is stopped",
903 			(unsigned int)pdu, (unsigned int)old_pdu);
904 	sfc_adapter_unlock(sa);
905 
906 fail_inval:
907 	sfc_log_init(sa, "failed %d", rc);
908 	SFC_ASSERT(rc > 0);
909 	return -rc;
910 }
911 static void
912 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
913 {
914 	struct sfc_adapter *sa = dev->data->dev_private;
915 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
916 	struct sfc_port *port = &sa->port;
917 	int rc;
918 
919 	sfc_adapter_lock(sa);
920 
921 	/*
922 	 * Copy the address to the device private data so that
923 	 * it could be recalled in the case of adapter restart.
924 	 */
925 	ether_addr_copy(mac_addr, &port->default_mac_addr);
926 
927 	if (port->isolated) {
928 		sfc_err(sa, "isolated mode is active on the port");
929 		sfc_err(sa, "will not set MAC address");
930 		goto unlock;
931 	}
932 
933 	if (sa->state != SFC_ADAPTER_STARTED) {
934 		sfc_info(sa, "the port is not started");
935 		sfc_info(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 	} else {
956 		sfc_warn(sa, "cannot set MAC address with filters installed");
957 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
958 		sfc_warn(sa, "(some traffic may be dropped)");
959 
960 		/*
961 		 * Since setting MAC address with filters installed is not
962 		 * allowed on the adapter, the new MAC address will be set
963 		 * by means of adapter restart. sfc_start() shall retrieve
964 		 * the new address from the device private data and set it.
965 		 */
966 		sfc_stop(sa);
967 		rc = sfc_start(sa);
968 		if (rc != 0)
969 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
970 	}
971 
972 unlock:
973 	/*
974 	 * In the case of failure sa->port->default_mac_addr does not
975 	 * need rollback since no error code is returned, and the upper
976 	 * API will anyway update the external MAC address storage.
977 	 * To be consistent with that new value it is better to keep
978 	 * the device private value the same.
979 	 */
980 	sfc_adapter_unlock(sa);
981 }
982 
983 
984 static int
985 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
986 		     uint32_t nb_mc_addr)
987 {
988 	struct sfc_adapter *sa = dev->data->dev_private;
989 	struct sfc_port *port = &sa->port;
990 	uint8_t *mc_addrs = port->mcast_addrs;
991 	int rc;
992 	unsigned int i;
993 
994 	if (port->isolated) {
995 		sfc_err(sa, "isolated mode is active on the port");
996 		sfc_err(sa, "will not set multicast address list");
997 		return -ENOTSUP;
998 	}
999 
1000 	if (mc_addrs == NULL)
1001 		return -ENOBUFS;
1002 
1003 	if (nb_mc_addr > port->max_mcast_addrs) {
1004 		sfc_err(sa, "too many multicast addresses: %u > %u",
1005 			 nb_mc_addr, port->max_mcast_addrs);
1006 		return -EINVAL;
1007 	}
1008 
1009 	for (i = 0; i < nb_mc_addr; ++i) {
1010 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1011 				 EFX_MAC_ADDR_LEN);
1012 		mc_addrs += EFX_MAC_ADDR_LEN;
1013 	}
1014 
1015 	port->nb_mcast_addrs = nb_mc_addr;
1016 
1017 	if (sa->state != SFC_ADAPTER_STARTED)
1018 		return 0;
1019 
1020 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1021 					port->nb_mcast_addrs);
1022 	if (rc != 0)
1023 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1024 
1025 	SFC_ASSERT(rc > 0);
1026 	return -rc;
1027 }
1028 
1029 /*
1030  * The function is used by the secondary process as well. It must not
1031  * use any process-local pointers from the adapter data.
1032  */
1033 static void
1034 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1035 		      struct rte_eth_rxq_info *qinfo)
1036 {
1037 	struct sfc_adapter *sa = dev->data->dev_private;
1038 	struct sfc_rxq_info *rxq_info;
1039 	struct sfc_rxq *rxq;
1040 
1041 	sfc_adapter_lock(sa);
1042 
1043 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1044 
1045 	rxq_info = &sa->rxq_info[rx_queue_id];
1046 	rxq = rxq_info->rxq;
1047 	SFC_ASSERT(rxq != NULL);
1048 
1049 	qinfo->mp = rxq->refill_mb_pool;
1050 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1051 	qinfo->conf.rx_drop_en = 1;
1052 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1053 	qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1054 			       DEV_RX_OFFLOAD_UDP_CKSUM |
1055 			       DEV_RX_OFFLOAD_TCP_CKSUM;
1056 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1057 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1058 		qinfo->scattered_rx = 1;
1059 	}
1060 	qinfo->nb_desc = rxq_info->entries;
1061 
1062 	sfc_adapter_unlock(sa);
1063 }
1064 
1065 /*
1066  * The function is used by the secondary process as well. It must not
1067  * use any process-local pointers from the adapter data.
1068  */
1069 static void
1070 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1071 		      struct rte_eth_txq_info *qinfo)
1072 {
1073 	struct sfc_adapter *sa = dev->data->dev_private;
1074 	struct sfc_txq_info *txq_info;
1075 
1076 	sfc_adapter_lock(sa);
1077 
1078 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1079 
1080 	txq_info = &sa->txq_info[tx_queue_id];
1081 	SFC_ASSERT(txq_info->txq != NULL);
1082 
1083 	memset(qinfo, 0, sizeof(*qinfo));
1084 
1085 	qinfo->conf.txq_flags = txq_info->txq->flags;
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 	rc = sfc_rx_qstart(sa, rx_queue_id);
1145 	if (rc != 0)
1146 		goto fail_rx_qstart;
1147 
1148 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1149 
1150 	sfc_adapter_unlock(sa);
1151 
1152 	return 0;
1153 
1154 fail_rx_qstart:
1155 fail_not_started:
1156 	sfc_adapter_unlock(sa);
1157 	SFC_ASSERT(rc > 0);
1158 	return -rc;
1159 }
1160 
1161 static int
1162 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1163 {
1164 	struct sfc_adapter *sa = dev->data->dev_private;
1165 
1166 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1167 
1168 	sfc_adapter_lock(sa);
1169 	sfc_rx_qstop(sa, rx_queue_id);
1170 
1171 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1172 
1173 	sfc_adapter_unlock(sa);
1174 
1175 	return 0;
1176 }
1177 
1178 static int
1179 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1180 {
1181 	struct sfc_adapter *sa = dev->data->dev_private;
1182 	int rc;
1183 
1184 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1185 
1186 	sfc_adapter_lock(sa);
1187 
1188 	rc = EINVAL;
1189 	if (sa->state != SFC_ADAPTER_STARTED)
1190 		goto fail_not_started;
1191 
1192 	rc = sfc_tx_qstart(sa, tx_queue_id);
1193 	if (rc != 0)
1194 		goto fail_tx_qstart;
1195 
1196 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1197 
1198 	sfc_adapter_unlock(sa);
1199 	return 0;
1200 
1201 fail_tx_qstart:
1202 
1203 fail_not_started:
1204 	sfc_adapter_unlock(sa);
1205 	SFC_ASSERT(rc > 0);
1206 	return -rc;
1207 }
1208 
1209 static int
1210 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1211 {
1212 	struct sfc_adapter *sa = dev->data->dev_private;
1213 
1214 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1215 
1216 	sfc_adapter_lock(sa);
1217 
1218 	sfc_tx_qstop(sa, tx_queue_id);
1219 
1220 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1221 
1222 	sfc_adapter_unlock(sa);
1223 	return 0;
1224 }
1225 
1226 static efx_tunnel_protocol_t
1227 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1228 {
1229 	switch (rte_type) {
1230 	case RTE_TUNNEL_TYPE_VXLAN:
1231 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1232 	case RTE_TUNNEL_TYPE_GENEVE:
1233 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1234 	default:
1235 		return EFX_TUNNEL_NPROTOS;
1236 	}
1237 }
1238 
1239 enum sfc_udp_tunnel_op_e {
1240 	SFC_UDP_TUNNEL_ADD_PORT,
1241 	SFC_UDP_TUNNEL_DEL_PORT,
1242 };
1243 
1244 static int
1245 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1246 		      struct rte_eth_udp_tunnel *tunnel_udp,
1247 		      enum sfc_udp_tunnel_op_e op)
1248 {
1249 	struct sfc_adapter *sa = dev->data->dev_private;
1250 	efx_tunnel_protocol_t tunnel_proto;
1251 	int rc;
1252 
1253 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1254 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1255 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1256 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1257 
1258 	tunnel_proto =
1259 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1260 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1261 		rc = ENOTSUP;
1262 		goto fail_bad_proto;
1263 	}
1264 
1265 	sfc_adapter_lock(sa);
1266 
1267 	switch (op) {
1268 	case SFC_UDP_TUNNEL_ADD_PORT:
1269 		rc = efx_tunnel_config_udp_add(sa->nic,
1270 					       tunnel_udp->udp_port,
1271 					       tunnel_proto);
1272 		break;
1273 	case SFC_UDP_TUNNEL_DEL_PORT:
1274 		rc = efx_tunnel_config_udp_remove(sa->nic,
1275 						  tunnel_udp->udp_port,
1276 						  tunnel_proto);
1277 		break;
1278 	default:
1279 		rc = EINVAL;
1280 		goto fail_bad_op;
1281 	}
1282 
1283 	if (rc != 0)
1284 		goto fail_op;
1285 
1286 	if (sa->state == SFC_ADAPTER_STARTED) {
1287 		rc = efx_tunnel_reconfigure(sa->nic);
1288 		if (rc == EAGAIN) {
1289 			/*
1290 			 * Configuration is accepted by FW and MC reboot
1291 			 * is initiated to apply the changes. MC reboot
1292 			 * will be handled in a usual way (MC reboot
1293 			 * event on management event queue and adapter
1294 			 * restart).
1295 			 */
1296 			rc = 0;
1297 		} else if (rc != 0) {
1298 			goto fail_reconfigure;
1299 		}
1300 	}
1301 
1302 	sfc_adapter_unlock(sa);
1303 	return 0;
1304 
1305 fail_reconfigure:
1306 	/* Remove/restore entry since the change makes the trouble */
1307 	switch (op) {
1308 	case SFC_UDP_TUNNEL_ADD_PORT:
1309 		(void)efx_tunnel_config_udp_remove(sa->nic,
1310 						   tunnel_udp->udp_port,
1311 						   tunnel_proto);
1312 		break;
1313 	case SFC_UDP_TUNNEL_DEL_PORT:
1314 		(void)efx_tunnel_config_udp_add(sa->nic,
1315 						tunnel_udp->udp_port,
1316 						tunnel_proto);
1317 		break;
1318 	}
1319 
1320 fail_op:
1321 fail_bad_op:
1322 	sfc_adapter_unlock(sa);
1323 
1324 fail_bad_proto:
1325 	SFC_ASSERT(rc > 0);
1326 	return -rc;
1327 }
1328 
1329 static int
1330 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1331 			    struct rte_eth_udp_tunnel *tunnel_udp)
1332 {
1333 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1334 }
1335 
1336 static int
1337 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1338 			    struct rte_eth_udp_tunnel *tunnel_udp)
1339 {
1340 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1341 }
1342 
1343 #if EFSYS_OPT_RX_SCALE
1344 static int
1345 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1346 			  struct rte_eth_rss_conf *rss_conf)
1347 {
1348 	struct sfc_adapter *sa = dev->data->dev_private;
1349 	struct sfc_port *port = &sa->port;
1350 
1351 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1352 		return -ENOTSUP;
1353 
1354 	if (sa->rss_channels == 0)
1355 		return -EINVAL;
1356 
1357 	sfc_adapter_lock(sa);
1358 
1359 	/*
1360 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1361 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1362 	 * flags which corresponds to the active EFX configuration stored
1363 	 * locally in 'sfc_adapter' and kept up-to-date
1364 	 */
1365 	rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1366 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1367 	if (rss_conf->rss_key != NULL)
1368 		rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1369 
1370 	sfc_adapter_unlock(sa);
1371 
1372 	return 0;
1373 }
1374 
1375 static int
1376 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1377 			struct rte_eth_rss_conf *rss_conf)
1378 {
1379 	struct sfc_adapter *sa = dev->data->dev_private;
1380 	struct sfc_port *port = &sa->port;
1381 	unsigned int efx_hash_types;
1382 	int rc = 0;
1383 
1384 	if (port->isolated)
1385 		return -ENOTSUP;
1386 
1387 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1388 		sfc_err(sa, "RSS is not available");
1389 		return -ENOTSUP;
1390 	}
1391 
1392 	if (sa->rss_channels == 0) {
1393 		sfc_err(sa, "RSS is not configured");
1394 		return -EINVAL;
1395 	}
1396 
1397 	if ((rss_conf->rss_key != NULL) &&
1398 	    (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1399 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1400 			sizeof(sa->rss_key));
1401 		return -EINVAL;
1402 	}
1403 
1404 	if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1405 		sfc_err(sa, "unsupported hash functions requested");
1406 		return -EINVAL;
1407 	}
1408 
1409 	sfc_adapter_lock(sa);
1410 
1411 	efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1412 
1413 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1414 				   EFX_RX_HASHALG_TOEPLITZ,
1415 				   efx_hash_types, B_TRUE);
1416 	if (rc != 0)
1417 		goto fail_scale_mode_set;
1418 
1419 	if (rss_conf->rss_key != NULL) {
1420 		if (sa->state == SFC_ADAPTER_STARTED) {
1421 			rc = efx_rx_scale_key_set(sa->nic,
1422 						  EFX_RSS_CONTEXT_DEFAULT,
1423 						  rss_conf->rss_key,
1424 						  sizeof(sa->rss_key));
1425 			if (rc != 0)
1426 				goto fail_scale_key_set;
1427 		}
1428 
1429 		rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1430 	}
1431 
1432 	sa->rss_hash_types = efx_hash_types;
1433 
1434 	sfc_adapter_unlock(sa);
1435 
1436 	return 0;
1437 
1438 fail_scale_key_set:
1439 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1440 				  EFX_RX_HASHALG_TOEPLITZ,
1441 				  sa->rss_hash_types, B_TRUE) != 0)
1442 		sfc_err(sa, "failed to restore RSS mode");
1443 
1444 fail_scale_mode_set:
1445 	sfc_adapter_unlock(sa);
1446 	return -rc;
1447 }
1448 
1449 static int
1450 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1451 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1452 		       uint16_t reta_size)
1453 {
1454 	struct sfc_adapter *sa = dev->data->dev_private;
1455 	struct sfc_port *port = &sa->port;
1456 	int entry;
1457 
1458 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1459 		return -ENOTSUP;
1460 
1461 	if (sa->rss_channels == 0)
1462 		return -EINVAL;
1463 
1464 	if (reta_size != EFX_RSS_TBL_SIZE)
1465 		return -EINVAL;
1466 
1467 	sfc_adapter_lock(sa);
1468 
1469 	for (entry = 0; entry < reta_size; entry++) {
1470 		int grp = entry / RTE_RETA_GROUP_SIZE;
1471 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1472 
1473 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1474 			reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1475 	}
1476 
1477 	sfc_adapter_unlock(sa);
1478 
1479 	return 0;
1480 }
1481 
1482 static int
1483 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1484 			struct rte_eth_rss_reta_entry64 *reta_conf,
1485 			uint16_t reta_size)
1486 {
1487 	struct sfc_adapter *sa = dev->data->dev_private;
1488 	struct sfc_port *port = &sa->port;
1489 	unsigned int *rss_tbl_new;
1490 	uint16_t entry;
1491 	int rc = 0;
1492 
1493 
1494 	if (port->isolated)
1495 		return -ENOTSUP;
1496 
1497 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1498 		sfc_err(sa, "RSS is not available");
1499 		return -ENOTSUP;
1500 	}
1501 
1502 	if (sa->rss_channels == 0) {
1503 		sfc_err(sa, "RSS is not configured");
1504 		return -EINVAL;
1505 	}
1506 
1507 	if (reta_size != EFX_RSS_TBL_SIZE) {
1508 		sfc_err(sa, "RETA size is wrong (should be %u)",
1509 			EFX_RSS_TBL_SIZE);
1510 		return -EINVAL;
1511 	}
1512 
1513 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1514 	if (rss_tbl_new == NULL)
1515 		return -ENOMEM;
1516 
1517 	sfc_adapter_lock(sa);
1518 
1519 	rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1520 
1521 	for (entry = 0; entry < reta_size; entry++) {
1522 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1523 		struct rte_eth_rss_reta_entry64 *grp;
1524 
1525 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1526 
1527 		if (grp->mask & (1ull << grp_idx)) {
1528 			if (grp->reta[grp_idx] >= sa->rss_channels) {
1529 				rc = EINVAL;
1530 				goto bad_reta_entry;
1531 			}
1532 			rss_tbl_new[entry] = grp->reta[grp_idx];
1533 		}
1534 	}
1535 
1536 	if (sa->state == SFC_ADAPTER_STARTED) {
1537 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1538 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1539 		if (rc != 0)
1540 			goto fail_scale_tbl_set;
1541 	}
1542 
1543 	rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1544 
1545 fail_scale_tbl_set:
1546 bad_reta_entry:
1547 	sfc_adapter_unlock(sa);
1548 
1549 	rte_free(rss_tbl_new);
1550 
1551 	SFC_ASSERT(rc >= 0);
1552 	return -rc;
1553 }
1554 #endif
1555 
1556 static int
1557 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1558 		    enum rte_filter_op filter_op,
1559 		    void *arg)
1560 {
1561 	struct sfc_adapter *sa = dev->data->dev_private;
1562 	int rc = ENOTSUP;
1563 
1564 	sfc_log_init(sa, "entry");
1565 
1566 	switch (filter_type) {
1567 	case RTE_ETH_FILTER_NONE:
1568 		sfc_err(sa, "Global filters configuration not supported");
1569 		break;
1570 	case RTE_ETH_FILTER_MACVLAN:
1571 		sfc_err(sa, "MACVLAN filters not supported");
1572 		break;
1573 	case RTE_ETH_FILTER_ETHERTYPE:
1574 		sfc_err(sa, "EtherType filters not supported");
1575 		break;
1576 	case RTE_ETH_FILTER_FLEXIBLE:
1577 		sfc_err(sa, "Flexible filters not supported");
1578 		break;
1579 	case RTE_ETH_FILTER_SYN:
1580 		sfc_err(sa, "SYN filters not supported");
1581 		break;
1582 	case RTE_ETH_FILTER_NTUPLE:
1583 		sfc_err(sa, "NTUPLE filters not supported");
1584 		break;
1585 	case RTE_ETH_FILTER_TUNNEL:
1586 		sfc_err(sa, "Tunnel filters not supported");
1587 		break;
1588 	case RTE_ETH_FILTER_FDIR:
1589 		sfc_err(sa, "Flow Director filters not supported");
1590 		break;
1591 	case RTE_ETH_FILTER_HASH:
1592 		sfc_err(sa, "Hash filters not supported");
1593 		break;
1594 	case RTE_ETH_FILTER_GENERIC:
1595 		if (filter_op != RTE_ETH_FILTER_GET) {
1596 			rc = EINVAL;
1597 		} else {
1598 			*(const void **)arg = &sfc_flow_ops;
1599 			rc = 0;
1600 		}
1601 		break;
1602 	default:
1603 		sfc_err(sa, "Unknown filter type %u", filter_type);
1604 		break;
1605 	}
1606 
1607 	sfc_log_init(sa, "exit: %d", -rc);
1608 	SFC_ASSERT(rc >= 0);
1609 	return -rc;
1610 }
1611 
1612 static const struct eth_dev_ops sfc_eth_dev_ops = {
1613 	.dev_configure			= sfc_dev_configure,
1614 	.dev_start			= sfc_dev_start,
1615 	.dev_stop			= sfc_dev_stop,
1616 	.dev_set_link_up		= sfc_dev_set_link_up,
1617 	.dev_set_link_down		= sfc_dev_set_link_down,
1618 	.dev_close			= sfc_dev_close,
1619 	.promiscuous_enable		= sfc_dev_promisc_enable,
1620 	.promiscuous_disable		= sfc_dev_promisc_disable,
1621 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1622 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1623 	.link_update			= sfc_dev_link_update,
1624 	.stats_get			= sfc_stats_get,
1625 	.stats_reset			= sfc_stats_reset,
1626 	.xstats_get			= sfc_xstats_get,
1627 	.xstats_reset			= sfc_stats_reset,
1628 	.xstats_get_names		= sfc_xstats_get_names,
1629 	.dev_infos_get			= sfc_dev_infos_get,
1630 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1631 	.mtu_set			= sfc_dev_set_mtu,
1632 	.rx_queue_start			= sfc_rx_queue_start,
1633 	.rx_queue_stop			= sfc_rx_queue_stop,
1634 	.tx_queue_start			= sfc_tx_queue_start,
1635 	.tx_queue_stop			= sfc_tx_queue_stop,
1636 	.rx_queue_setup			= sfc_rx_queue_setup,
1637 	.rx_queue_release		= sfc_rx_queue_release,
1638 	.rx_queue_count			= sfc_rx_queue_count,
1639 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1640 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1641 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1642 	.tx_queue_setup			= sfc_tx_queue_setup,
1643 	.tx_queue_release		= sfc_tx_queue_release,
1644 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1645 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1646 	.mac_addr_set			= sfc_mac_addr_set,
1647 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1648 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1649 #if EFSYS_OPT_RX_SCALE
1650 	.reta_update			= sfc_dev_rss_reta_update,
1651 	.reta_query			= sfc_dev_rss_reta_query,
1652 	.rss_hash_update		= sfc_dev_rss_hash_update,
1653 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1654 #endif
1655 	.filter_ctrl			= sfc_dev_filter_ctrl,
1656 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1657 	.rxq_info_get			= sfc_rx_queue_info_get,
1658 	.txq_info_get			= sfc_tx_queue_info_get,
1659 	.fw_version_get			= sfc_fw_version_get,
1660 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1661 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1662 };
1663 
1664 /**
1665  * Duplicate a string in potentially shared memory required for
1666  * multi-process support.
1667  *
1668  * strdup() allocates from process-local heap/memory.
1669  */
1670 static char *
1671 sfc_strdup(const char *str)
1672 {
1673 	size_t size;
1674 	char *copy;
1675 
1676 	if (str == NULL)
1677 		return NULL;
1678 
1679 	size = strlen(str) + 1;
1680 	copy = rte_malloc(__func__, size, 0);
1681 	if (copy != NULL)
1682 		rte_memcpy(copy, str, size);
1683 
1684 	return copy;
1685 }
1686 
1687 static int
1688 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1689 {
1690 	struct sfc_adapter *sa = dev->data->dev_private;
1691 	unsigned int avail_caps = 0;
1692 	const char *rx_name = NULL;
1693 	const char *tx_name = NULL;
1694 	int rc;
1695 
1696 	switch (sa->family) {
1697 	case EFX_FAMILY_HUNTINGTON:
1698 	case EFX_FAMILY_MEDFORD:
1699 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1700 		break;
1701 	default:
1702 		break;
1703 	}
1704 
1705 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1706 				sfc_kvarg_string_handler, &rx_name);
1707 	if (rc != 0)
1708 		goto fail_kvarg_rx_datapath;
1709 
1710 	if (rx_name != NULL) {
1711 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1712 		if (sa->dp_rx == NULL) {
1713 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1714 			rc = ENOENT;
1715 			goto fail_dp_rx;
1716 		}
1717 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1718 			sfc_err(sa,
1719 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1720 				rx_name);
1721 			rc = EINVAL;
1722 			goto fail_dp_rx_caps;
1723 		}
1724 	} else {
1725 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1726 		if (sa->dp_rx == NULL) {
1727 			sfc_err(sa, "Rx datapath by caps %#x not found",
1728 				avail_caps);
1729 			rc = ENOENT;
1730 			goto fail_dp_rx;
1731 		}
1732 	}
1733 
1734 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1735 	if (sa->dp_rx_name == NULL) {
1736 		rc = ENOMEM;
1737 		goto fail_dp_rx_name;
1738 	}
1739 
1740 	sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1741 
1742 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1743 
1744 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1745 				sfc_kvarg_string_handler, &tx_name);
1746 	if (rc != 0)
1747 		goto fail_kvarg_tx_datapath;
1748 
1749 	if (tx_name != NULL) {
1750 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1751 		if (sa->dp_tx == NULL) {
1752 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1753 			rc = ENOENT;
1754 			goto fail_dp_tx;
1755 		}
1756 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1757 			sfc_err(sa,
1758 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1759 				tx_name);
1760 			rc = EINVAL;
1761 			goto fail_dp_tx_caps;
1762 		}
1763 	} else {
1764 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1765 		if (sa->dp_tx == NULL) {
1766 			sfc_err(sa, "Tx datapath by caps %#x not found",
1767 				avail_caps);
1768 			rc = ENOENT;
1769 			goto fail_dp_tx;
1770 		}
1771 	}
1772 
1773 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1774 	if (sa->dp_tx_name == NULL) {
1775 		rc = ENOMEM;
1776 		goto fail_dp_tx_name;
1777 	}
1778 
1779 	sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1780 
1781 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1782 
1783 	dev->dev_ops = &sfc_eth_dev_ops;
1784 
1785 	return 0;
1786 
1787 fail_dp_tx_name:
1788 fail_dp_tx_caps:
1789 	sa->dp_tx = NULL;
1790 
1791 fail_dp_tx:
1792 fail_kvarg_tx_datapath:
1793 	rte_free(sa->dp_rx_name);
1794 	sa->dp_rx_name = NULL;
1795 
1796 fail_dp_rx_name:
1797 fail_dp_rx_caps:
1798 	sa->dp_rx = NULL;
1799 
1800 fail_dp_rx:
1801 fail_kvarg_rx_datapath:
1802 	return rc;
1803 }
1804 
1805 static void
1806 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1807 {
1808 	struct sfc_adapter *sa = dev->data->dev_private;
1809 
1810 	dev->dev_ops = NULL;
1811 	dev->rx_pkt_burst = NULL;
1812 	dev->tx_pkt_burst = NULL;
1813 
1814 	rte_free(sa->dp_tx_name);
1815 	sa->dp_tx_name = NULL;
1816 	sa->dp_tx = NULL;
1817 
1818 	rte_free(sa->dp_rx_name);
1819 	sa->dp_rx_name = NULL;
1820 	sa->dp_rx = NULL;
1821 }
1822 
1823 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1824 	.rxq_info_get			= sfc_rx_queue_info_get,
1825 	.txq_info_get			= sfc_tx_queue_info_get,
1826 };
1827 
1828 static int
1829 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1830 {
1831 	/*
1832 	 * Device private data has really many process-local pointers.
1833 	 * Below code should be extremely careful to use data located
1834 	 * in shared memory only.
1835 	 */
1836 	struct sfc_adapter *sa = dev->data->dev_private;
1837 	const struct sfc_dp_rx *dp_rx;
1838 	const struct sfc_dp_tx *dp_tx;
1839 	int rc;
1840 
1841 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1842 	if (dp_rx == NULL) {
1843 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1844 		rc = ENOENT;
1845 		goto fail_dp_rx;
1846 	}
1847 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1848 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1849 			sa->dp_tx_name);
1850 		rc = EINVAL;
1851 		goto fail_dp_rx_multi_process;
1852 	}
1853 
1854 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1855 	if (dp_tx == NULL) {
1856 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1857 		rc = ENOENT;
1858 		goto fail_dp_tx;
1859 	}
1860 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1861 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1862 			sa->dp_tx_name);
1863 		rc = EINVAL;
1864 		goto fail_dp_tx_multi_process;
1865 	}
1866 
1867 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1868 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1869 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1870 
1871 	return 0;
1872 
1873 fail_dp_tx_multi_process:
1874 fail_dp_tx:
1875 fail_dp_rx_multi_process:
1876 fail_dp_rx:
1877 	return rc;
1878 }
1879 
1880 static void
1881 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1882 {
1883 	dev->dev_ops = NULL;
1884 	dev->tx_pkt_burst = NULL;
1885 	dev->rx_pkt_burst = NULL;
1886 }
1887 
1888 static void
1889 sfc_register_dp(void)
1890 {
1891 	/* Register once */
1892 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1893 		/* Prefer EF10 datapath */
1894 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1895 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1896 
1897 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1898 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1899 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1900 	}
1901 }
1902 
1903 static int
1904 sfc_eth_dev_init(struct rte_eth_dev *dev)
1905 {
1906 	struct sfc_adapter *sa = dev->data->dev_private;
1907 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1908 	int rc;
1909 	const efx_nic_cfg_t *encp;
1910 	const struct ether_addr *from;
1911 
1912 	sfc_register_dp();
1913 
1914 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1915 		return -sfc_eth_dev_secondary_set_ops(dev);
1916 
1917 	/* Required for logging */
1918 	sa->pci_addr = pci_dev->addr;
1919 	sa->port_id = dev->data->port_id;
1920 
1921 	sa->eth_dev = dev;
1922 
1923 	/* Copy PCI device info to the dev->data */
1924 	rte_eth_copy_pci_info(dev, pci_dev);
1925 
1926 	rc = sfc_kvargs_parse(sa);
1927 	if (rc != 0)
1928 		goto fail_kvargs_parse;
1929 
1930 	rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1931 				sfc_kvarg_bool_handler, &sa->debug_init);
1932 	if (rc != 0)
1933 		goto fail_kvarg_debug_init;
1934 
1935 	sfc_log_init(sa, "entry");
1936 
1937 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1938 	if (dev->data->mac_addrs == NULL) {
1939 		rc = ENOMEM;
1940 		goto fail_mac_addrs;
1941 	}
1942 
1943 	sfc_adapter_lock_init(sa);
1944 	sfc_adapter_lock(sa);
1945 
1946 	sfc_log_init(sa, "probing");
1947 	rc = sfc_probe(sa);
1948 	if (rc != 0)
1949 		goto fail_probe;
1950 
1951 	sfc_log_init(sa, "set device ops");
1952 	rc = sfc_eth_dev_set_ops(dev);
1953 	if (rc != 0)
1954 		goto fail_set_ops;
1955 
1956 	sfc_log_init(sa, "attaching");
1957 	rc = sfc_attach(sa);
1958 	if (rc != 0)
1959 		goto fail_attach;
1960 
1961 	encp = efx_nic_cfg_get(sa->nic);
1962 
1963 	/*
1964 	 * The arguments are really reverse order in comparison to
1965 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1966 	 */
1967 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1968 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1969 
1970 	sfc_adapter_unlock(sa);
1971 
1972 	sfc_log_init(sa, "done");
1973 	return 0;
1974 
1975 fail_attach:
1976 	sfc_eth_dev_clear_ops(dev);
1977 
1978 fail_set_ops:
1979 	sfc_unprobe(sa);
1980 
1981 fail_probe:
1982 	sfc_adapter_unlock(sa);
1983 	sfc_adapter_lock_fini(sa);
1984 	rte_free(dev->data->mac_addrs);
1985 	dev->data->mac_addrs = NULL;
1986 
1987 fail_mac_addrs:
1988 fail_kvarg_debug_init:
1989 	sfc_kvargs_cleanup(sa);
1990 
1991 fail_kvargs_parse:
1992 	sfc_log_init(sa, "failed %d", rc);
1993 	SFC_ASSERT(rc > 0);
1994 	return -rc;
1995 }
1996 
1997 static int
1998 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1999 {
2000 	struct sfc_adapter *sa;
2001 
2002 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2003 		sfc_eth_dev_secondary_clear_ops(dev);
2004 		return 0;
2005 	}
2006 
2007 	sa = dev->data->dev_private;
2008 	sfc_log_init(sa, "entry");
2009 
2010 	sfc_adapter_lock(sa);
2011 
2012 	sfc_eth_dev_clear_ops(dev);
2013 
2014 	sfc_detach(sa);
2015 	sfc_unprobe(sa);
2016 
2017 	rte_free(dev->data->mac_addrs);
2018 	dev->data->mac_addrs = NULL;
2019 
2020 	sfc_kvargs_cleanup(sa);
2021 
2022 	sfc_adapter_unlock(sa);
2023 	sfc_adapter_lock_fini(sa);
2024 
2025 	sfc_log_init(sa, "done");
2026 
2027 	/* Required for logging, so cleanup last */
2028 	sa->eth_dev = NULL;
2029 	return 0;
2030 }
2031 
2032 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2033 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2034 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2035 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2036 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2037 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2038 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2039 	{ .vendor_id = 0 /* sentinel */ }
2040 };
2041 
2042 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2043 	struct rte_pci_device *pci_dev)
2044 {
2045 	return rte_eth_dev_pci_generic_probe(pci_dev,
2046 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2047 }
2048 
2049 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2050 {
2051 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2052 }
2053 
2054 static struct rte_pci_driver sfc_efx_pmd = {
2055 	.id_table = pci_id_sfc_efx_map,
2056 	.drv_flags =
2057 		RTE_PCI_DRV_INTR_LSC |
2058 		RTE_PCI_DRV_NEED_MAPPING,
2059 	.probe = sfc_eth_dev_pci_probe,
2060 	.remove = sfc_eth_dev_pci_remove,
2061 };
2062 
2063 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2064 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2065 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2066 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2067 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2068 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2069 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2070 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
2071 	SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
2072 	SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);
2073 
2074 RTE_INIT(sfc_driver_register_logtype);
2075 static void
2076 sfc_driver_register_logtype(void)
2077 {
2078 	int ret;
2079 
2080 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2081 						   RTE_LOG_NOTICE);
2082 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2083 }
2084