xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 24f8a95917b36b313a49bd604c3958f992893efd)
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_BAD_PACKETS];
541 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
542 	} else {
543 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
544 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
545 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
546 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
547 		/*
548 		 * Take into account stats which are whenever supported
549 		 * on EF10. If some stat is not supported by current
550 		 * firmware variant or HW revision, it is guaranteed
551 		 * to be zero in mac_stats.
552 		 */
553 		stats->imissed =
554 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
555 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
556 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
557 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
558 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
559 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
560 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
561 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
562 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
563 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
564 		stats->ierrors =
565 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
566 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
567 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
568 		/* no oerrors counters supported on EF10 */
569 	}
570 
571 unlock:
572 	rte_spinlock_unlock(&port->mac_stats_lock);
573 	SFC_ASSERT(ret >= 0);
574 	return -ret;
575 }
576 
577 static void
578 sfc_stats_reset(struct rte_eth_dev *dev)
579 {
580 	struct sfc_adapter *sa = dev->data->dev_private;
581 	struct sfc_port *port = &sa->port;
582 	int rc;
583 
584 	if (sa->state != SFC_ADAPTER_STARTED) {
585 		/*
586 		 * The operation cannot be done if port is not started; it
587 		 * will be scheduled to be done during the next port start
588 		 */
589 		port->mac_stats_reset_pending = B_TRUE;
590 		return;
591 	}
592 
593 	rc = sfc_port_reset_mac_stats(sa);
594 	if (rc != 0)
595 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
596 }
597 
598 static int
599 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
600 	       unsigned int xstats_count)
601 {
602 	struct sfc_adapter *sa = dev->data->dev_private;
603 	struct sfc_port *port = &sa->port;
604 	uint64_t *mac_stats;
605 	int rc;
606 	unsigned int i;
607 	int nstats = 0;
608 
609 	rte_spinlock_lock(&port->mac_stats_lock);
610 
611 	rc = sfc_port_update_mac_stats(sa);
612 	if (rc != 0) {
613 		SFC_ASSERT(rc > 0);
614 		nstats = -rc;
615 		goto unlock;
616 	}
617 
618 	mac_stats = port->mac_stats_buf;
619 
620 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
621 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
622 			if (xstats != NULL && nstats < (int)xstats_count) {
623 				xstats[nstats].id = nstats;
624 				xstats[nstats].value = mac_stats[i];
625 			}
626 			nstats++;
627 		}
628 	}
629 
630 unlock:
631 	rte_spinlock_unlock(&port->mac_stats_lock);
632 
633 	return nstats;
634 }
635 
636 static int
637 sfc_xstats_get_names(struct rte_eth_dev *dev,
638 		     struct rte_eth_xstat_name *xstats_names,
639 		     unsigned int xstats_count)
640 {
641 	struct sfc_adapter *sa = dev->data->dev_private;
642 	struct sfc_port *port = &sa->port;
643 	unsigned int i;
644 	unsigned int nstats = 0;
645 
646 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
647 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
648 			if (xstats_names != NULL && nstats < xstats_count)
649 				strlcpy(xstats_names[nstats].name,
650 					efx_mac_stat_name(sa->nic, i),
651 					sizeof(xstats_names[0].name));
652 			nstats++;
653 		}
654 	}
655 
656 	return nstats;
657 }
658 
659 static int
660 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
661 		     uint64_t *values, unsigned int n)
662 {
663 	struct sfc_adapter *sa = dev->data->dev_private;
664 	struct sfc_port *port = &sa->port;
665 	uint64_t *mac_stats;
666 	unsigned int nb_supported = 0;
667 	unsigned int nb_written = 0;
668 	unsigned int i;
669 	int ret;
670 	int rc;
671 
672 	if (unlikely(values == NULL) ||
673 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
674 		return port->mac_stats_nb_supported;
675 
676 	rte_spinlock_lock(&port->mac_stats_lock);
677 
678 	rc = sfc_port_update_mac_stats(sa);
679 	if (rc != 0) {
680 		SFC_ASSERT(rc > 0);
681 		ret = -rc;
682 		goto unlock;
683 	}
684 
685 	mac_stats = port->mac_stats_buf;
686 
687 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
688 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
689 			continue;
690 
691 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
692 			values[nb_written++] = mac_stats[i];
693 
694 		++nb_supported;
695 	}
696 
697 	ret = nb_written;
698 
699 unlock:
700 	rte_spinlock_unlock(&port->mac_stats_lock);
701 
702 	return ret;
703 }
704 
705 static int
706 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
707 			   struct rte_eth_xstat_name *xstats_names,
708 			   const uint64_t *ids, unsigned int size)
709 {
710 	struct sfc_adapter *sa = dev->data->dev_private;
711 	struct sfc_port *port = &sa->port;
712 	unsigned int nb_supported = 0;
713 	unsigned int nb_written = 0;
714 	unsigned int i;
715 
716 	if (unlikely(xstats_names == NULL) ||
717 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
718 		return port->mac_stats_nb_supported;
719 
720 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
721 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
722 			continue;
723 
724 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
725 			char *name = xstats_names[nb_written++].name;
726 
727 			strlcpy(name, efx_mac_stat_name(sa->nic, i),
728 				sizeof(xstats_names[0].name));
729 		}
730 
731 		++nb_supported;
732 	}
733 
734 	return nb_written;
735 }
736 
737 static int
738 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
739 {
740 	struct sfc_adapter *sa = dev->data->dev_private;
741 	unsigned int wanted_fc, link_fc;
742 
743 	memset(fc_conf, 0, sizeof(*fc_conf));
744 
745 	sfc_adapter_lock(sa);
746 
747 	if (sa->state == SFC_ADAPTER_STARTED)
748 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
749 	else
750 		link_fc = sa->port.flow_ctrl;
751 
752 	switch (link_fc) {
753 	case 0:
754 		fc_conf->mode = RTE_FC_NONE;
755 		break;
756 	case EFX_FCNTL_RESPOND:
757 		fc_conf->mode = RTE_FC_RX_PAUSE;
758 		break;
759 	case EFX_FCNTL_GENERATE:
760 		fc_conf->mode = RTE_FC_TX_PAUSE;
761 		break;
762 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
763 		fc_conf->mode = RTE_FC_FULL;
764 		break;
765 	default:
766 		sfc_err(sa, "%s: unexpected flow control value %#x",
767 			__func__, link_fc);
768 	}
769 
770 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
771 
772 	sfc_adapter_unlock(sa);
773 
774 	return 0;
775 }
776 
777 static int
778 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
779 {
780 	struct sfc_adapter *sa = dev->data->dev_private;
781 	struct sfc_port *port = &sa->port;
782 	unsigned int fcntl;
783 	int rc;
784 
785 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
786 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
787 	    fc_conf->mac_ctrl_frame_fwd != 0) {
788 		sfc_err(sa, "unsupported flow control settings specified");
789 		rc = EINVAL;
790 		goto fail_inval;
791 	}
792 
793 	switch (fc_conf->mode) {
794 	case RTE_FC_NONE:
795 		fcntl = 0;
796 		break;
797 	case RTE_FC_RX_PAUSE:
798 		fcntl = EFX_FCNTL_RESPOND;
799 		break;
800 	case RTE_FC_TX_PAUSE:
801 		fcntl = EFX_FCNTL_GENERATE;
802 		break;
803 	case RTE_FC_FULL:
804 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
805 		break;
806 	default:
807 		rc = EINVAL;
808 		goto fail_inval;
809 	}
810 
811 	sfc_adapter_lock(sa);
812 
813 	if (sa->state == SFC_ADAPTER_STARTED) {
814 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
815 		if (rc != 0)
816 			goto fail_mac_fcntl_set;
817 	}
818 
819 	port->flow_ctrl = fcntl;
820 	port->flow_ctrl_autoneg = fc_conf->autoneg;
821 
822 	sfc_adapter_unlock(sa);
823 
824 	return 0;
825 
826 fail_mac_fcntl_set:
827 	sfc_adapter_unlock(sa);
828 fail_inval:
829 	SFC_ASSERT(rc > 0);
830 	return -rc;
831 }
832 
833 static int
834 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
835 {
836 	struct sfc_adapter *sa = dev->data->dev_private;
837 	size_t pdu = EFX_MAC_PDU(mtu);
838 	size_t old_pdu;
839 	int rc;
840 
841 	sfc_log_init(sa, "mtu=%u", mtu);
842 
843 	rc = EINVAL;
844 	if (pdu < EFX_MAC_PDU_MIN) {
845 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
846 			(unsigned int)mtu, (unsigned int)pdu,
847 			EFX_MAC_PDU_MIN);
848 		goto fail_inval;
849 	}
850 	if (pdu > EFX_MAC_PDU_MAX) {
851 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
852 			(unsigned int)mtu, (unsigned int)pdu,
853 			EFX_MAC_PDU_MAX);
854 		goto fail_inval;
855 	}
856 
857 	sfc_adapter_lock(sa);
858 
859 	if (pdu != sa->port.pdu) {
860 		if (sa->state == SFC_ADAPTER_STARTED) {
861 			sfc_stop(sa);
862 
863 			old_pdu = sa->port.pdu;
864 			sa->port.pdu = pdu;
865 			rc = sfc_start(sa);
866 			if (rc != 0)
867 				goto fail_start;
868 		} else {
869 			sa->port.pdu = pdu;
870 		}
871 	}
872 
873 	/*
874 	 * The driver does not use it, but other PMDs update jumbo frame
875 	 * flag and max_rx_pkt_len when MTU is set.
876 	 */
877 	if (mtu > ETHER_MAX_LEN) {
878 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
879 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
880 	}
881 
882 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
883 
884 	sfc_adapter_unlock(sa);
885 
886 	sfc_log_init(sa, "done");
887 	return 0;
888 
889 fail_start:
890 	sa->port.pdu = old_pdu;
891 	if (sfc_start(sa) != 0)
892 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
893 			"PDU max size - port is stopped",
894 			(unsigned int)pdu, (unsigned int)old_pdu);
895 	sfc_adapter_unlock(sa);
896 
897 fail_inval:
898 	sfc_log_init(sa, "failed %d", rc);
899 	SFC_ASSERT(rc > 0);
900 	return -rc;
901 }
902 static int
903 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
904 {
905 	struct sfc_adapter *sa = dev->data->dev_private;
906 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
907 	struct sfc_port *port = &sa->port;
908 	struct ether_addr *old_addr = &dev->data->mac_addrs[0];
909 	int rc = 0;
910 
911 	sfc_adapter_lock(sa);
912 
913 	/*
914 	 * Copy the address to the device private data so that
915 	 * it could be recalled in the case of adapter restart.
916 	 */
917 	ether_addr_copy(mac_addr, &port->default_mac_addr);
918 
919 	/*
920 	 * Neither of the two following checks can return
921 	 * an error. The new MAC address is preserved in
922 	 * the device private data and can be activated
923 	 * on the next port start if the user prevents
924 	 * isolated mode from being enabled.
925 	 */
926 	if (port->isolated) {
927 		sfc_warn(sa, "isolated mode is active on the port");
928 		sfc_warn(sa, "will not set MAC address");
929 		goto unlock;
930 	}
931 
932 	if (sa->state != SFC_ADAPTER_STARTED) {
933 		sfc_notice(sa, "the port is not started");
934 		sfc_notice(sa, "the new MAC address will be set on port start");
935 
936 		goto unlock;
937 	}
938 
939 	if (encp->enc_allow_set_mac_with_installed_filters) {
940 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
941 		if (rc != 0) {
942 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
943 			goto unlock;
944 		}
945 
946 		/*
947 		 * Changing the MAC address by means of MCDI request
948 		 * has no effect on received traffic, therefore
949 		 * we also need to update unicast filters
950 		 */
951 		rc = sfc_set_rx_mode(sa);
952 		if (rc != 0) {
953 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
954 			/* Rollback the old address */
955 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
956 			(void)sfc_set_rx_mode(sa);
957 		}
958 	} else {
959 		sfc_warn(sa, "cannot set MAC address with filters installed");
960 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
961 		sfc_warn(sa, "(some traffic may be dropped)");
962 
963 		/*
964 		 * Since setting MAC address with filters installed is not
965 		 * allowed on the adapter, the new MAC address will be set
966 		 * by means of adapter restart. sfc_start() shall retrieve
967 		 * the new address from the device private data and set it.
968 		 */
969 		sfc_stop(sa);
970 		rc = sfc_start(sa);
971 		if (rc != 0)
972 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
973 	}
974 
975 unlock:
976 	if (rc != 0)
977 		ether_addr_copy(old_addr, &port->default_mac_addr);
978 
979 	sfc_adapter_unlock(sa);
980 
981 	SFC_ASSERT(rc >= 0);
982 	return -rc;
983 }
984 
985 
986 static int
987 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
988 		     uint32_t nb_mc_addr)
989 {
990 	struct sfc_adapter *sa = dev->data->dev_private;
991 	struct sfc_port *port = &sa->port;
992 	uint8_t *mc_addrs = port->mcast_addrs;
993 	int rc;
994 	unsigned int i;
995 
996 	if (port->isolated) {
997 		sfc_err(sa, "isolated mode is active on the port");
998 		sfc_err(sa, "will not set multicast address list");
999 		return -ENOTSUP;
1000 	}
1001 
1002 	if (mc_addrs == NULL)
1003 		return -ENOBUFS;
1004 
1005 	if (nb_mc_addr > port->max_mcast_addrs) {
1006 		sfc_err(sa, "too many multicast addresses: %u > %u",
1007 			 nb_mc_addr, port->max_mcast_addrs);
1008 		return -EINVAL;
1009 	}
1010 
1011 	for (i = 0; i < nb_mc_addr; ++i) {
1012 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1013 				 EFX_MAC_ADDR_LEN);
1014 		mc_addrs += EFX_MAC_ADDR_LEN;
1015 	}
1016 
1017 	port->nb_mcast_addrs = nb_mc_addr;
1018 
1019 	if (sa->state != SFC_ADAPTER_STARTED)
1020 		return 0;
1021 
1022 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1023 					port->nb_mcast_addrs);
1024 	if (rc != 0)
1025 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1026 
1027 	SFC_ASSERT(rc >= 0);
1028 	return -rc;
1029 }
1030 
1031 /*
1032  * The function is used by the secondary process as well. It must not
1033  * use any process-local pointers from the adapter data.
1034  */
1035 static void
1036 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1037 		      struct rte_eth_rxq_info *qinfo)
1038 {
1039 	struct sfc_adapter *sa = dev->data->dev_private;
1040 	struct sfc_rxq_info *rxq_info;
1041 	struct sfc_rxq *rxq;
1042 
1043 	sfc_adapter_lock(sa);
1044 
1045 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1046 
1047 	rxq_info = &sa->rxq_info[rx_queue_id];
1048 	rxq = rxq_info->rxq;
1049 	SFC_ASSERT(rxq != NULL);
1050 
1051 	qinfo->mp = rxq->refill_mb_pool;
1052 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1053 	qinfo->conf.rx_drop_en = 1;
1054 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1055 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
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.offloads = txq_info->txq->offloads;
1086 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1087 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1088 	qinfo->nb_desc = txq_info->entries;
1089 
1090 	sfc_adapter_unlock(sa);
1091 }
1092 
1093 static uint32_t
1094 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1095 {
1096 	struct sfc_adapter *sa = dev->data->dev_private;
1097 
1098 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1099 
1100 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
1101 }
1102 
1103 static int
1104 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1105 {
1106 	struct sfc_dp_rxq *dp_rxq = queue;
1107 
1108 	return sfc_rx_qdesc_done(dp_rxq, offset);
1109 }
1110 
1111 static int
1112 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1113 {
1114 	struct sfc_dp_rxq *dp_rxq = queue;
1115 	struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1116 
1117 	return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1118 }
1119 
1120 static int
1121 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1122 {
1123 	struct sfc_dp_txq *dp_txq = queue;
1124 	struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1125 
1126 	return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1127 }
1128 
1129 static int
1130 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1131 {
1132 	struct sfc_adapter *sa = dev->data->dev_private;
1133 	int rc;
1134 
1135 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1136 
1137 	sfc_adapter_lock(sa);
1138 
1139 	rc = EINVAL;
1140 	if (sa->state != SFC_ADAPTER_STARTED)
1141 		goto fail_not_started;
1142 
1143 	if (sa->rxq_info[rx_queue_id].rxq == NULL)
1144 		goto fail_not_setup;
1145 
1146 	rc = sfc_rx_qstart(sa, rx_queue_id);
1147 	if (rc != 0)
1148 		goto fail_rx_qstart;
1149 
1150 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1151 
1152 	sfc_adapter_unlock(sa);
1153 
1154 	return 0;
1155 
1156 fail_rx_qstart:
1157 fail_not_setup:
1158 fail_not_started:
1159 	sfc_adapter_unlock(sa);
1160 	SFC_ASSERT(rc > 0);
1161 	return -rc;
1162 }
1163 
1164 static int
1165 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1166 {
1167 	struct sfc_adapter *sa = dev->data->dev_private;
1168 
1169 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1170 
1171 	sfc_adapter_lock(sa);
1172 	sfc_rx_qstop(sa, rx_queue_id);
1173 
1174 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1175 
1176 	sfc_adapter_unlock(sa);
1177 
1178 	return 0;
1179 }
1180 
1181 static int
1182 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1183 {
1184 	struct sfc_adapter *sa = dev->data->dev_private;
1185 	int rc;
1186 
1187 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1188 
1189 	sfc_adapter_lock(sa);
1190 
1191 	rc = EINVAL;
1192 	if (sa->state != SFC_ADAPTER_STARTED)
1193 		goto fail_not_started;
1194 
1195 	if (sa->txq_info[tx_queue_id].txq == NULL)
1196 		goto fail_not_setup;
1197 
1198 	rc = sfc_tx_qstart(sa, tx_queue_id);
1199 	if (rc != 0)
1200 		goto fail_tx_qstart;
1201 
1202 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1203 
1204 	sfc_adapter_unlock(sa);
1205 	return 0;
1206 
1207 fail_tx_qstart:
1208 
1209 fail_not_setup:
1210 fail_not_started:
1211 	sfc_adapter_unlock(sa);
1212 	SFC_ASSERT(rc > 0);
1213 	return -rc;
1214 }
1215 
1216 static int
1217 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1218 {
1219 	struct sfc_adapter *sa = dev->data->dev_private;
1220 
1221 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1222 
1223 	sfc_adapter_lock(sa);
1224 
1225 	sfc_tx_qstop(sa, tx_queue_id);
1226 
1227 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1228 
1229 	sfc_adapter_unlock(sa);
1230 	return 0;
1231 }
1232 
1233 static efx_tunnel_protocol_t
1234 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1235 {
1236 	switch (rte_type) {
1237 	case RTE_TUNNEL_TYPE_VXLAN:
1238 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1239 	case RTE_TUNNEL_TYPE_GENEVE:
1240 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1241 	default:
1242 		return EFX_TUNNEL_NPROTOS;
1243 	}
1244 }
1245 
1246 enum sfc_udp_tunnel_op_e {
1247 	SFC_UDP_TUNNEL_ADD_PORT,
1248 	SFC_UDP_TUNNEL_DEL_PORT,
1249 };
1250 
1251 static int
1252 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1253 		      struct rte_eth_udp_tunnel *tunnel_udp,
1254 		      enum sfc_udp_tunnel_op_e op)
1255 {
1256 	struct sfc_adapter *sa = dev->data->dev_private;
1257 	efx_tunnel_protocol_t tunnel_proto;
1258 	int rc;
1259 
1260 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1261 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1262 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1263 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1264 
1265 	tunnel_proto =
1266 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1267 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1268 		rc = ENOTSUP;
1269 		goto fail_bad_proto;
1270 	}
1271 
1272 	sfc_adapter_lock(sa);
1273 
1274 	switch (op) {
1275 	case SFC_UDP_TUNNEL_ADD_PORT:
1276 		rc = efx_tunnel_config_udp_add(sa->nic,
1277 					       tunnel_udp->udp_port,
1278 					       tunnel_proto);
1279 		break;
1280 	case SFC_UDP_TUNNEL_DEL_PORT:
1281 		rc = efx_tunnel_config_udp_remove(sa->nic,
1282 						  tunnel_udp->udp_port,
1283 						  tunnel_proto);
1284 		break;
1285 	default:
1286 		rc = EINVAL;
1287 		goto fail_bad_op;
1288 	}
1289 
1290 	if (rc != 0)
1291 		goto fail_op;
1292 
1293 	if (sa->state == SFC_ADAPTER_STARTED) {
1294 		rc = efx_tunnel_reconfigure(sa->nic);
1295 		if (rc == EAGAIN) {
1296 			/*
1297 			 * Configuration is accepted by FW and MC reboot
1298 			 * is initiated to apply the changes. MC reboot
1299 			 * will be handled in a usual way (MC reboot
1300 			 * event on management event queue and adapter
1301 			 * restart).
1302 			 */
1303 			rc = 0;
1304 		} else if (rc != 0) {
1305 			goto fail_reconfigure;
1306 		}
1307 	}
1308 
1309 	sfc_adapter_unlock(sa);
1310 	return 0;
1311 
1312 fail_reconfigure:
1313 	/* Remove/restore entry since the change makes the trouble */
1314 	switch (op) {
1315 	case SFC_UDP_TUNNEL_ADD_PORT:
1316 		(void)efx_tunnel_config_udp_remove(sa->nic,
1317 						   tunnel_udp->udp_port,
1318 						   tunnel_proto);
1319 		break;
1320 	case SFC_UDP_TUNNEL_DEL_PORT:
1321 		(void)efx_tunnel_config_udp_add(sa->nic,
1322 						tunnel_udp->udp_port,
1323 						tunnel_proto);
1324 		break;
1325 	}
1326 
1327 fail_op:
1328 fail_bad_op:
1329 	sfc_adapter_unlock(sa);
1330 
1331 fail_bad_proto:
1332 	SFC_ASSERT(rc > 0);
1333 	return -rc;
1334 }
1335 
1336 static int
1337 sfc_dev_udp_tunnel_port_add(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_ADD_PORT);
1341 }
1342 
1343 static int
1344 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1345 			    struct rte_eth_udp_tunnel *tunnel_udp)
1346 {
1347 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1348 }
1349 
1350 static int
1351 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1352 			  struct rte_eth_rss_conf *rss_conf)
1353 {
1354 	struct sfc_adapter *sa = dev->data->dev_private;
1355 	struct sfc_rss *rss = &sa->rss;
1356 
1357 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1358 		return -ENOTSUP;
1359 
1360 	sfc_adapter_lock(sa);
1361 
1362 	/*
1363 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1364 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1365 	 * flags which corresponds to the active EFX configuration stored
1366 	 * locally in 'sfc_adapter' and kept up-to-date
1367 	 */
1368 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1369 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1370 	if (rss_conf->rss_key != NULL)
1371 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1372 
1373 	sfc_adapter_unlock(sa);
1374 
1375 	return 0;
1376 }
1377 
1378 static int
1379 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1380 			struct rte_eth_rss_conf *rss_conf)
1381 {
1382 	struct sfc_adapter *sa = dev->data->dev_private;
1383 	struct sfc_rss *rss = &sa->rss;
1384 	struct sfc_port *port = &sa->port;
1385 	unsigned int efx_hash_types;
1386 	int rc = 0;
1387 
1388 	if (port->isolated)
1389 		return -ENOTSUP;
1390 
1391 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1392 		sfc_err(sa, "RSS is not available");
1393 		return -ENOTSUP;
1394 	}
1395 
1396 	if (rss->channels == 0) {
1397 		sfc_err(sa, "RSS is not configured");
1398 		return -EINVAL;
1399 	}
1400 
1401 	if ((rss_conf->rss_key != NULL) &&
1402 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1403 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1404 			sizeof(rss->key));
1405 		return -EINVAL;
1406 	}
1407 
1408 	sfc_adapter_lock(sa);
1409 
1410 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1411 	if (rc != 0)
1412 		goto fail_rx_hf_rte_to_efx;
1413 
1414 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1415 				   rss->hash_alg, 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(rss->key));
1425 			if (rc != 0)
1426 				goto fail_scale_key_set;
1427 		}
1428 
1429 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1430 	}
1431 
1432 	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 				  rss->hash_types, B_TRUE) != 0)
1442 		sfc_err(sa, "failed to restore RSS mode");
1443 
1444 fail_scale_mode_set:
1445 fail_rx_hf_rte_to_efx:
1446 	sfc_adapter_unlock(sa);
1447 	return -rc;
1448 }
1449 
1450 static int
1451 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1452 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1453 		       uint16_t reta_size)
1454 {
1455 	struct sfc_adapter *sa = dev->data->dev_private;
1456 	struct sfc_rss *rss = &sa->rss;
1457 	struct sfc_port *port = &sa->port;
1458 	int entry;
1459 
1460 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1461 		return -ENOTSUP;
1462 
1463 	if (rss->channels == 0)
1464 		return -EINVAL;
1465 
1466 	if (reta_size != EFX_RSS_TBL_SIZE)
1467 		return -EINVAL;
1468 
1469 	sfc_adapter_lock(sa);
1470 
1471 	for (entry = 0; entry < reta_size; entry++) {
1472 		int grp = entry / RTE_RETA_GROUP_SIZE;
1473 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1474 
1475 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1476 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1477 	}
1478 
1479 	sfc_adapter_unlock(sa);
1480 
1481 	return 0;
1482 }
1483 
1484 static int
1485 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1486 			struct rte_eth_rss_reta_entry64 *reta_conf,
1487 			uint16_t reta_size)
1488 {
1489 	struct sfc_adapter *sa = dev->data->dev_private;
1490 	struct sfc_rss *rss = &sa->rss;
1491 	struct sfc_port *port = &sa->port;
1492 	unsigned int *rss_tbl_new;
1493 	uint16_t entry;
1494 	int rc = 0;
1495 
1496 
1497 	if (port->isolated)
1498 		return -ENOTSUP;
1499 
1500 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1501 		sfc_err(sa, "RSS is not available");
1502 		return -ENOTSUP;
1503 	}
1504 
1505 	if (rss->channels == 0) {
1506 		sfc_err(sa, "RSS is not configured");
1507 		return -EINVAL;
1508 	}
1509 
1510 	if (reta_size != EFX_RSS_TBL_SIZE) {
1511 		sfc_err(sa, "RETA size is wrong (should be %u)",
1512 			EFX_RSS_TBL_SIZE);
1513 		return -EINVAL;
1514 	}
1515 
1516 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1517 	if (rss_tbl_new == NULL)
1518 		return -ENOMEM;
1519 
1520 	sfc_adapter_lock(sa);
1521 
1522 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1523 
1524 	for (entry = 0; entry < reta_size; entry++) {
1525 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1526 		struct rte_eth_rss_reta_entry64 *grp;
1527 
1528 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1529 
1530 		if (grp->mask & (1ull << grp_idx)) {
1531 			if (grp->reta[grp_idx] >= rss->channels) {
1532 				rc = EINVAL;
1533 				goto bad_reta_entry;
1534 			}
1535 			rss_tbl_new[entry] = grp->reta[grp_idx];
1536 		}
1537 	}
1538 
1539 	if (sa->state == SFC_ADAPTER_STARTED) {
1540 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1541 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1542 		if (rc != 0)
1543 			goto fail_scale_tbl_set;
1544 	}
1545 
1546 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1547 
1548 fail_scale_tbl_set:
1549 bad_reta_entry:
1550 	sfc_adapter_unlock(sa);
1551 
1552 	rte_free(rss_tbl_new);
1553 
1554 	SFC_ASSERT(rc >= 0);
1555 	return -rc;
1556 }
1557 
1558 static int
1559 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1560 		    enum rte_filter_op filter_op,
1561 		    void *arg)
1562 {
1563 	struct sfc_adapter *sa = dev->data->dev_private;
1564 	int rc = ENOTSUP;
1565 
1566 	sfc_log_init(sa, "entry");
1567 
1568 	switch (filter_type) {
1569 	case RTE_ETH_FILTER_NONE:
1570 		sfc_err(sa, "Global filters configuration not supported");
1571 		break;
1572 	case RTE_ETH_FILTER_MACVLAN:
1573 		sfc_err(sa, "MACVLAN filters not supported");
1574 		break;
1575 	case RTE_ETH_FILTER_ETHERTYPE:
1576 		sfc_err(sa, "EtherType filters not supported");
1577 		break;
1578 	case RTE_ETH_FILTER_FLEXIBLE:
1579 		sfc_err(sa, "Flexible filters not supported");
1580 		break;
1581 	case RTE_ETH_FILTER_SYN:
1582 		sfc_err(sa, "SYN filters not supported");
1583 		break;
1584 	case RTE_ETH_FILTER_NTUPLE:
1585 		sfc_err(sa, "NTUPLE filters not supported");
1586 		break;
1587 	case RTE_ETH_FILTER_TUNNEL:
1588 		sfc_err(sa, "Tunnel filters not supported");
1589 		break;
1590 	case RTE_ETH_FILTER_FDIR:
1591 		sfc_err(sa, "Flow Director filters not supported");
1592 		break;
1593 	case RTE_ETH_FILTER_HASH:
1594 		sfc_err(sa, "Hash filters not supported");
1595 		break;
1596 	case RTE_ETH_FILTER_GENERIC:
1597 		if (filter_op != RTE_ETH_FILTER_GET) {
1598 			rc = EINVAL;
1599 		} else {
1600 			*(const void **)arg = &sfc_flow_ops;
1601 			rc = 0;
1602 		}
1603 		break;
1604 	default:
1605 		sfc_err(sa, "Unknown filter type %u", filter_type);
1606 		break;
1607 	}
1608 
1609 	sfc_log_init(sa, "exit: %d", -rc);
1610 	SFC_ASSERT(rc >= 0);
1611 	return -rc;
1612 }
1613 
1614 static int
1615 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1616 {
1617 	struct sfc_adapter *sa = dev->data->dev_private;
1618 
1619 	/*
1620 	 * If Rx datapath does not provide callback to check mempool,
1621 	 * all pools are supported.
1622 	 */
1623 	if (sa->dp_rx->pool_ops_supported == NULL)
1624 		return 1;
1625 
1626 	return sa->dp_rx->pool_ops_supported(pool);
1627 }
1628 
1629 static const struct eth_dev_ops sfc_eth_dev_ops = {
1630 	.dev_configure			= sfc_dev_configure,
1631 	.dev_start			= sfc_dev_start,
1632 	.dev_stop			= sfc_dev_stop,
1633 	.dev_set_link_up		= sfc_dev_set_link_up,
1634 	.dev_set_link_down		= sfc_dev_set_link_down,
1635 	.dev_close			= sfc_dev_close,
1636 	.promiscuous_enable		= sfc_dev_promisc_enable,
1637 	.promiscuous_disable		= sfc_dev_promisc_disable,
1638 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1639 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1640 	.link_update			= sfc_dev_link_update,
1641 	.stats_get			= sfc_stats_get,
1642 	.stats_reset			= sfc_stats_reset,
1643 	.xstats_get			= sfc_xstats_get,
1644 	.xstats_reset			= sfc_stats_reset,
1645 	.xstats_get_names		= sfc_xstats_get_names,
1646 	.dev_infos_get			= sfc_dev_infos_get,
1647 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1648 	.mtu_set			= sfc_dev_set_mtu,
1649 	.rx_queue_start			= sfc_rx_queue_start,
1650 	.rx_queue_stop			= sfc_rx_queue_stop,
1651 	.tx_queue_start			= sfc_tx_queue_start,
1652 	.tx_queue_stop			= sfc_tx_queue_stop,
1653 	.rx_queue_setup			= sfc_rx_queue_setup,
1654 	.rx_queue_release		= sfc_rx_queue_release,
1655 	.rx_queue_count			= sfc_rx_queue_count,
1656 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1657 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1658 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1659 	.tx_queue_setup			= sfc_tx_queue_setup,
1660 	.tx_queue_release		= sfc_tx_queue_release,
1661 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1662 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1663 	.mac_addr_set			= sfc_mac_addr_set,
1664 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1665 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1666 	.reta_update			= sfc_dev_rss_reta_update,
1667 	.reta_query			= sfc_dev_rss_reta_query,
1668 	.rss_hash_update		= sfc_dev_rss_hash_update,
1669 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1670 	.filter_ctrl			= sfc_dev_filter_ctrl,
1671 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1672 	.rxq_info_get			= sfc_rx_queue_info_get,
1673 	.txq_info_get			= sfc_tx_queue_info_get,
1674 	.fw_version_get			= sfc_fw_version_get,
1675 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1676 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1677 	.pool_ops_supported		= sfc_pool_ops_supported,
1678 };
1679 
1680 /**
1681  * Duplicate a string in potentially shared memory required for
1682  * multi-process support.
1683  *
1684  * strdup() allocates from process-local heap/memory.
1685  */
1686 static char *
1687 sfc_strdup(const char *str)
1688 {
1689 	size_t size;
1690 	char *copy;
1691 
1692 	if (str == NULL)
1693 		return NULL;
1694 
1695 	size = strlen(str) + 1;
1696 	copy = rte_malloc(__func__, size, 0);
1697 	if (copy != NULL)
1698 		rte_memcpy(copy, str, size);
1699 
1700 	return copy;
1701 }
1702 
1703 static int
1704 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1705 {
1706 	struct sfc_adapter *sa = dev->data->dev_private;
1707 	const efx_nic_cfg_t *encp;
1708 	unsigned int avail_caps = 0;
1709 	const char *rx_name = NULL;
1710 	const char *tx_name = NULL;
1711 	int rc;
1712 
1713 	switch (sa->family) {
1714 	case EFX_FAMILY_HUNTINGTON:
1715 	case EFX_FAMILY_MEDFORD:
1716 	case EFX_FAMILY_MEDFORD2:
1717 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1718 		break;
1719 	default:
1720 		break;
1721 	}
1722 
1723 	encp = efx_nic_cfg_get(sa->nic);
1724 	if (encp->enc_rx_es_super_buffer_supported)
1725 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1726 
1727 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1728 				sfc_kvarg_string_handler, &rx_name);
1729 	if (rc != 0)
1730 		goto fail_kvarg_rx_datapath;
1731 
1732 	if (rx_name != NULL) {
1733 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1734 		if (sa->dp_rx == NULL) {
1735 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1736 			rc = ENOENT;
1737 			goto fail_dp_rx;
1738 		}
1739 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1740 			sfc_err(sa,
1741 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1742 				rx_name);
1743 			rc = EINVAL;
1744 			goto fail_dp_rx_caps;
1745 		}
1746 	} else {
1747 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1748 		if (sa->dp_rx == NULL) {
1749 			sfc_err(sa, "Rx datapath by caps %#x not found",
1750 				avail_caps);
1751 			rc = ENOENT;
1752 			goto fail_dp_rx;
1753 		}
1754 	}
1755 
1756 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1757 	if (sa->dp_rx_name == NULL) {
1758 		rc = ENOMEM;
1759 		goto fail_dp_rx_name;
1760 	}
1761 
1762 	sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1763 
1764 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1765 
1766 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1767 				sfc_kvarg_string_handler, &tx_name);
1768 	if (rc != 0)
1769 		goto fail_kvarg_tx_datapath;
1770 
1771 	if (tx_name != NULL) {
1772 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1773 		if (sa->dp_tx == NULL) {
1774 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1775 			rc = ENOENT;
1776 			goto fail_dp_tx;
1777 		}
1778 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1779 			sfc_err(sa,
1780 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1781 				tx_name);
1782 			rc = EINVAL;
1783 			goto fail_dp_tx_caps;
1784 		}
1785 	} else {
1786 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1787 		if (sa->dp_tx == NULL) {
1788 			sfc_err(sa, "Tx datapath by caps %#x not found",
1789 				avail_caps);
1790 			rc = ENOENT;
1791 			goto fail_dp_tx;
1792 		}
1793 	}
1794 
1795 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1796 	if (sa->dp_tx_name == NULL) {
1797 		rc = ENOMEM;
1798 		goto fail_dp_tx_name;
1799 	}
1800 
1801 	sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1802 
1803 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1804 
1805 	dev->dev_ops = &sfc_eth_dev_ops;
1806 
1807 	return 0;
1808 
1809 fail_dp_tx_name:
1810 fail_dp_tx_caps:
1811 	sa->dp_tx = NULL;
1812 
1813 fail_dp_tx:
1814 fail_kvarg_tx_datapath:
1815 	rte_free(sa->dp_rx_name);
1816 	sa->dp_rx_name = NULL;
1817 
1818 fail_dp_rx_name:
1819 fail_dp_rx_caps:
1820 	sa->dp_rx = NULL;
1821 
1822 fail_dp_rx:
1823 fail_kvarg_rx_datapath:
1824 	return rc;
1825 }
1826 
1827 static void
1828 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1829 {
1830 	struct sfc_adapter *sa = dev->data->dev_private;
1831 
1832 	dev->dev_ops = NULL;
1833 	dev->rx_pkt_burst = NULL;
1834 	dev->tx_pkt_burst = NULL;
1835 
1836 	rte_free(sa->dp_tx_name);
1837 	sa->dp_tx_name = NULL;
1838 	sa->dp_tx = NULL;
1839 
1840 	rte_free(sa->dp_rx_name);
1841 	sa->dp_rx_name = NULL;
1842 	sa->dp_rx = NULL;
1843 }
1844 
1845 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1846 	.rxq_info_get			= sfc_rx_queue_info_get,
1847 	.txq_info_get			= sfc_tx_queue_info_get,
1848 };
1849 
1850 static int
1851 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1852 {
1853 	/*
1854 	 * Device private data has really many process-local pointers.
1855 	 * Below code should be extremely careful to use data located
1856 	 * in shared memory only.
1857 	 */
1858 	struct sfc_adapter *sa = dev->data->dev_private;
1859 	const struct sfc_dp_rx *dp_rx;
1860 	const struct sfc_dp_tx *dp_tx;
1861 	int rc;
1862 
1863 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1864 	if (dp_rx == NULL) {
1865 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1866 		rc = ENOENT;
1867 		goto fail_dp_rx;
1868 	}
1869 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1870 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1871 			sa->dp_tx_name);
1872 		rc = EINVAL;
1873 		goto fail_dp_rx_multi_process;
1874 	}
1875 
1876 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1877 	if (dp_tx == NULL) {
1878 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1879 		rc = ENOENT;
1880 		goto fail_dp_tx;
1881 	}
1882 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1883 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1884 			sa->dp_tx_name);
1885 		rc = EINVAL;
1886 		goto fail_dp_tx_multi_process;
1887 	}
1888 
1889 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1890 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1891 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1892 
1893 	return 0;
1894 
1895 fail_dp_tx_multi_process:
1896 fail_dp_tx:
1897 fail_dp_rx_multi_process:
1898 fail_dp_rx:
1899 	return rc;
1900 }
1901 
1902 static void
1903 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1904 {
1905 	dev->dev_ops = NULL;
1906 	dev->tx_pkt_burst = NULL;
1907 	dev->rx_pkt_burst = NULL;
1908 }
1909 
1910 static void
1911 sfc_register_dp(void)
1912 {
1913 	/* Register once */
1914 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1915 		/* Prefer EF10 datapath */
1916 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1917 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1918 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1919 
1920 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1921 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1922 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1923 	}
1924 }
1925 
1926 static int
1927 sfc_eth_dev_init(struct rte_eth_dev *dev)
1928 {
1929 	struct sfc_adapter *sa = dev->data->dev_private;
1930 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1931 	int rc;
1932 	const efx_nic_cfg_t *encp;
1933 	const struct ether_addr *from;
1934 
1935 	sfc_register_dp();
1936 
1937 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1938 		return -sfc_eth_dev_secondary_set_ops(dev);
1939 
1940 	/* Required for logging */
1941 	sa->pci_addr = pci_dev->addr;
1942 	sa->port_id = dev->data->port_id;
1943 
1944 	sa->eth_dev = dev;
1945 
1946 	/* Copy PCI device info to the dev->data */
1947 	rte_eth_copy_pci_info(dev, pci_dev);
1948 
1949 	sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1950 						RTE_LOG_NOTICE);
1951 
1952 	rc = sfc_kvargs_parse(sa);
1953 	if (rc != 0)
1954 		goto fail_kvargs_parse;
1955 
1956 	sfc_log_init(sa, "entry");
1957 
1958 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1959 	if (dev->data->mac_addrs == NULL) {
1960 		rc = ENOMEM;
1961 		goto fail_mac_addrs;
1962 	}
1963 
1964 	sfc_adapter_lock_init(sa);
1965 	sfc_adapter_lock(sa);
1966 
1967 	sfc_log_init(sa, "probing");
1968 	rc = sfc_probe(sa);
1969 	if (rc != 0)
1970 		goto fail_probe;
1971 
1972 	sfc_log_init(sa, "set device ops");
1973 	rc = sfc_eth_dev_set_ops(dev);
1974 	if (rc != 0)
1975 		goto fail_set_ops;
1976 
1977 	sfc_log_init(sa, "attaching");
1978 	rc = sfc_attach(sa);
1979 	if (rc != 0)
1980 		goto fail_attach;
1981 
1982 	encp = efx_nic_cfg_get(sa->nic);
1983 
1984 	/*
1985 	 * The arguments are really reverse order in comparison to
1986 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1987 	 */
1988 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1989 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1990 
1991 	sfc_adapter_unlock(sa);
1992 
1993 	sfc_log_init(sa, "done");
1994 	return 0;
1995 
1996 fail_attach:
1997 	sfc_eth_dev_clear_ops(dev);
1998 
1999 fail_set_ops:
2000 	sfc_unprobe(sa);
2001 
2002 fail_probe:
2003 	sfc_adapter_unlock(sa);
2004 	sfc_adapter_lock_fini(sa);
2005 	rte_free(dev->data->mac_addrs);
2006 	dev->data->mac_addrs = NULL;
2007 
2008 fail_mac_addrs:
2009 	sfc_kvargs_cleanup(sa);
2010 
2011 fail_kvargs_parse:
2012 	sfc_log_init(sa, "failed %d", rc);
2013 	SFC_ASSERT(rc > 0);
2014 	return -rc;
2015 }
2016 
2017 static int
2018 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2019 {
2020 	struct sfc_adapter *sa;
2021 
2022 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2023 		sfc_eth_dev_secondary_clear_ops(dev);
2024 		return 0;
2025 	}
2026 
2027 	sa = dev->data->dev_private;
2028 	sfc_log_init(sa, "entry");
2029 
2030 	sfc_adapter_lock(sa);
2031 
2032 	sfc_eth_dev_clear_ops(dev);
2033 
2034 	sfc_detach(sa);
2035 	sfc_unprobe(sa);
2036 
2037 	sfc_kvargs_cleanup(sa);
2038 
2039 	sfc_adapter_unlock(sa);
2040 	sfc_adapter_lock_fini(sa);
2041 
2042 	sfc_log_init(sa, "done");
2043 
2044 	/* Required for logging, so cleanup last */
2045 	sa->eth_dev = NULL;
2046 	return 0;
2047 }
2048 
2049 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2050 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2051 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2052 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2053 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2054 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2055 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2056 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2057 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2058 	{ .vendor_id = 0 /* sentinel */ }
2059 };
2060 
2061 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2062 	struct rte_pci_device *pci_dev)
2063 {
2064 	return rte_eth_dev_pci_generic_probe(pci_dev,
2065 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2066 }
2067 
2068 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2069 {
2070 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2071 }
2072 
2073 static struct rte_pci_driver sfc_efx_pmd = {
2074 	.id_table = pci_id_sfc_efx_map,
2075 	.drv_flags =
2076 		RTE_PCI_DRV_INTR_LSC |
2077 		RTE_PCI_DRV_NEED_MAPPING,
2078 	.probe = sfc_eth_dev_pci_probe,
2079 	.remove = sfc_eth_dev_pci_remove,
2080 };
2081 
2082 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2083 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2084 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2085 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2086 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2087 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2088 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2089 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2090 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2091 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2092 
2093 RTE_INIT(sfc_driver_register_logtype)
2094 {
2095 	int ret;
2096 
2097 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2098 						   RTE_LOG_NOTICE);
2099 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2100 }
2101