xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 9cd9d3e702fba4700539c1a2eddac13dd14ecf70)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
4  * All rights reserved.
5  *
6  * This software was jointly developed between OKTET Labs (under contract
7  * for Solarflare) and Solarflare Communications, Inc.
8  */
9 
10 #include <rte_dev.h>
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
16 #include <rte_string_fns.h>
17 #include <rte_ether.h>
18 
19 #include "efx.h"
20 
21 #include "sfc.h"
22 #include "sfc_debug.h"
23 #include "sfc_log.h"
24 #include "sfc_kvargs.h"
25 #include "sfc_ev.h"
26 #include "sfc_rx.h"
27 #include "sfc_tx.h"
28 #include "sfc_flow.h"
29 #include "sfc_dp.h"
30 #include "sfc_dp_rx.h"
31 
32 uint32_t sfc_logtype_driver;
33 
34 static struct sfc_dp_list sfc_dp_head =
35 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
36 
37 
38 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
39 
40 
41 static int
42 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
43 {
44 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
45 	efx_nic_fw_info_t enfi;
46 	int ret;
47 	int rc;
48 
49 	/*
50 	 * Return value of the callback is likely supposed to be
51 	 * equal to or greater than 0, nevertheless, if an error
52 	 * occurs, it will be desirable to pass it to the caller
53 	 */
54 	if ((fw_version == NULL) || (fw_size == 0))
55 		return -EINVAL;
56 
57 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
58 	if (rc != 0)
59 		return -rc;
60 
61 	ret = snprintf(fw_version, fw_size,
62 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
63 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
64 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
65 	if (ret < 0)
66 		return ret;
67 
68 	if (enfi.enfi_dpcpu_fw_ids_valid) {
69 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
70 		int ret_extra;
71 
72 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
73 				     fw_size - dpcpu_fw_ids_offset,
74 				     " rx%" PRIx16 " tx%" PRIx16,
75 				     enfi.enfi_rx_dpcpu_fw_id,
76 				     enfi.enfi_tx_dpcpu_fw_id);
77 		if (ret_extra < 0)
78 			return ret_extra;
79 
80 		ret += ret_extra;
81 	}
82 
83 	if (fw_size < (size_t)(++ret))
84 		return ret;
85 	else
86 		return 0;
87 }
88 
89 static int
90 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
91 {
92 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
93 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
94 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
95 	struct sfc_rss *rss = &sas->rss;
96 	uint64_t txq_offloads_def = 0;
97 
98 	sfc_log_init(sa, "entry");
99 
100 	dev_info->min_mtu = RTE_ETHER_MIN_MTU;
101 	dev_info->max_mtu = EFX_MAC_SDU_MAX;
102 
103 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
104 
105 	/* Autonegotiation may be disabled */
106 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
107 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
108 		dev_info->speed_capa |= ETH_LINK_SPEED_1G;
109 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
110 		dev_info->speed_capa |= ETH_LINK_SPEED_10G;
111 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
112 		dev_info->speed_capa |= ETH_LINK_SPEED_25G;
113 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
114 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
115 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
116 		dev_info->speed_capa |= ETH_LINK_SPEED_50G;
117 	if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
118 		dev_info->speed_capa |= ETH_LINK_SPEED_100G;
119 
120 	dev_info->max_rx_queues = sa->rxq_max;
121 	dev_info->max_tx_queues = sa->txq_max;
122 
123 	/* By default packets are dropped if no descriptors are available */
124 	dev_info->default_rxconf.rx_drop_en = 1;
125 
126 	dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
127 
128 	/*
129 	 * rx_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->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
134 				    dev_info->rx_queue_offload_capa;
135 
136 	dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
137 
138 	/*
139 	 * tx_offload_capa includes both device and queue offloads since
140 	 * the latter may be requested on a per device basis which makes
141 	 * sense when some offloads are needed to be set on all queues.
142 	 */
143 	dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
144 				    dev_info->tx_queue_offload_capa;
145 
146 	if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
147 		txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
148 
149 	dev_info->default_txconf.offloads |= txq_offloads_def;
150 
151 	if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
152 		uint64_t rte_hf = 0;
153 		unsigned int i;
154 
155 		for (i = 0; i < rss->hf_map_nb_entries; ++i)
156 			rte_hf |= rss->hf_map[i].rte;
157 
158 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
159 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
160 		dev_info->flow_type_rss_offloads = rte_hf;
161 	}
162 
163 	/* Initialize to hardware limits */
164 	dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
165 	dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
166 	/* The RXQ hardware requires that the descriptor count is a power
167 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
168 	 */
169 	dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
170 
171 	/* Initialize to hardware limits */
172 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
173 	dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
174 	/*
175 	 * The TXQ hardware requires that the descriptor count is a power
176 	 * of 2, but tx_desc_lim cannot properly describe that constraint
177 	 */
178 	dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
179 
180 	if (sap->dp_rx->get_dev_info != NULL)
181 		sap->dp_rx->get_dev_info(dev_info);
182 	if (sap->dp_tx->get_dev_info != NULL)
183 		sap->dp_tx->get_dev_info(dev_info);
184 
185 	dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
186 			     RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
187 
188 	return 0;
189 }
190 
191 static const uint32_t *
192 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
193 {
194 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
195 
196 	return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
197 }
198 
199 static int
200 sfc_dev_configure(struct rte_eth_dev *dev)
201 {
202 	struct rte_eth_dev_data *dev_data = dev->data;
203 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
204 	int rc;
205 
206 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
207 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
208 
209 	sfc_adapter_lock(sa);
210 	switch (sa->state) {
211 	case SFC_ADAPTER_CONFIGURED:
212 		/* FALLTHROUGH */
213 	case SFC_ADAPTER_INITIALIZED:
214 		rc = sfc_configure(sa);
215 		break;
216 	default:
217 		sfc_err(sa, "unexpected adapter state %u to configure",
218 			sa->state);
219 		rc = EINVAL;
220 		break;
221 	}
222 	sfc_adapter_unlock(sa);
223 
224 	sfc_log_init(sa, "done %d", rc);
225 	SFC_ASSERT(rc >= 0);
226 	return -rc;
227 }
228 
229 static int
230 sfc_dev_start(struct rte_eth_dev *dev)
231 {
232 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
233 	int rc;
234 
235 	sfc_log_init(sa, "entry");
236 
237 	sfc_adapter_lock(sa);
238 	rc = sfc_start(sa);
239 	sfc_adapter_unlock(sa);
240 
241 	sfc_log_init(sa, "done %d", rc);
242 	SFC_ASSERT(rc >= 0);
243 	return -rc;
244 }
245 
246 static int
247 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
248 {
249 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
250 	struct rte_eth_link current_link;
251 	int ret;
252 
253 	sfc_log_init(sa, "entry");
254 
255 	if (sa->state != SFC_ADAPTER_STARTED) {
256 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
257 	} else if (wait_to_complete) {
258 		efx_link_mode_t link_mode;
259 
260 		if (efx_port_poll(sa->nic, &link_mode) != 0)
261 			link_mode = EFX_LINK_UNKNOWN;
262 		sfc_port_link_mode_to_info(link_mode, &current_link);
263 
264 	} else {
265 		sfc_ev_mgmt_qpoll(sa);
266 		rte_eth_linkstatus_get(dev, &current_link);
267 	}
268 
269 	ret = rte_eth_linkstatus_set(dev, &current_link);
270 	if (ret == 0)
271 		sfc_notice(sa, "Link status is %s",
272 			   current_link.link_status ? "UP" : "DOWN");
273 
274 	return ret;
275 }
276 
277 static void
278 sfc_dev_stop(struct rte_eth_dev *dev)
279 {
280 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
281 
282 	sfc_log_init(sa, "entry");
283 
284 	sfc_adapter_lock(sa);
285 	sfc_stop(sa);
286 	sfc_adapter_unlock(sa);
287 
288 	sfc_log_init(sa, "done");
289 }
290 
291 static int
292 sfc_dev_set_link_up(struct rte_eth_dev *dev)
293 {
294 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
295 	int rc;
296 
297 	sfc_log_init(sa, "entry");
298 
299 	sfc_adapter_lock(sa);
300 	rc = sfc_start(sa);
301 	sfc_adapter_unlock(sa);
302 
303 	SFC_ASSERT(rc >= 0);
304 	return -rc;
305 }
306 
307 static int
308 sfc_dev_set_link_down(struct rte_eth_dev *dev)
309 {
310 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
311 
312 	sfc_log_init(sa, "entry");
313 
314 	sfc_adapter_lock(sa);
315 	sfc_stop(sa);
316 	sfc_adapter_unlock(sa);
317 
318 	return 0;
319 }
320 
321 static void
322 sfc_dev_close(struct rte_eth_dev *dev)
323 {
324 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
325 
326 	sfc_log_init(sa, "entry");
327 
328 	sfc_adapter_lock(sa);
329 	switch (sa->state) {
330 	case SFC_ADAPTER_STARTED:
331 		sfc_stop(sa);
332 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
333 		/* FALLTHROUGH */
334 	case SFC_ADAPTER_CONFIGURED:
335 		sfc_close(sa);
336 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
337 		/* FALLTHROUGH */
338 	case SFC_ADAPTER_INITIALIZED:
339 		break;
340 	default:
341 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
342 		break;
343 	}
344 
345 	/*
346 	 * Cleanup all resources in accordance with RTE_ETH_DEV_CLOSE_REMOVE.
347 	 * Rollback primary process sfc_eth_dev_init() below.
348 	 */
349 
350 	sfc_eth_dev_clear_ops(dev);
351 
352 	sfc_detach(sa);
353 	sfc_unprobe(sa);
354 
355 	sfc_kvargs_cleanup(sa);
356 
357 	sfc_adapter_unlock(sa);
358 	sfc_adapter_lock_fini(sa);
359 
360 	sfc_log_init(sa, "done");
361 
362 	/* Required for logging, so cleanup last */
363 	sa->eth_dev = NULL;
364 
365 	dev->process_private = NULL;
366 	free(sa);
367 }
368 
369 static int
370 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
371 		   boolean_t enabled)
372 {
373 	struct sfc_port *port;
374 	boolean_t *toggle;
375 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
376 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
377 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
378 	int rc = 0;
379 
380 	sfc_adapter_lock(sa);
381 
382 	port = &sa->port;
383 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
384 
385 	if (*toggle != enabled) {
386 		*toggle = enabled;
387 
388 		if (sfc_sa2shared(sa)->isolated) {
389 			sfc_warn(sa, "isolated mode is active on the port");
390 			sfc_warn(sa, "the change is to be applied on the next "
391 				     "start provided that isolated mode is "
392 				     "disabled prior the next start");
393 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
394 			   ((rc = sfc_set_rx_mode(sa)) != 0)) {
395 			*toggle = !(enabled);
396 			sfc_warn(sa, "Failed to %s %s mode",
397 				 ((enabled) ? "enable" : "disable"), desc);
398 		}
399 	}
400 
401 	sfc_adapter_unlock(sa);
402 	return rc;
403 }
404 
405 static int
406 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
407 {
408 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
409 
410 	SFC_ASSERT(rc >= 0);
411 	return -rc;
412 }
413 
414 static int
415 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
416 {
417 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
418 
419 	SFC_ASSERT(rc >= 0);
420 	return -rc;
421 }
422 
423 static int
424 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
425 {
426 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
427 
428 	SFC_ASSERT(rc >= 0);
429 	return -rc;
430 }
431 
432 static int
433 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
434 {
435 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
436 
437 	SFC_ASSERT(rc >= 0);
438 	return -rc;
439 }
440 
441 static int
442 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
443 		   uint16_t nb_rx_desc, unsigned int socket_id,
444 		   const struct rte_eth_rxconf *rx_conf,
445 		   struct rte_mempool *mb_pool)
446 {
447 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
448 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
449 	int rc;
450 
451 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
452 		     rx_queue_id, nb_rx_desc, socket_id);
453 
454 	sfc_adapter_lock(sa);
455 
456 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
457 			  rx_conf, mb_pool);
458 	if (rc != 0)
459 		goto fail_rx_qinit;
460 
461 	dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
462 
463 	sfc_adapter_unlock(sa);
464 
465 	return 0;
466 
467 fail_rx_qinit:
468 	sfc_adapter_unlock(sa);
469 	SFC_ASSERT(rc > 0);
470 	return -rc;
471 }
472 
473 static void
474 sfc_rx_queue_release(void *queue)
475 {
476 	struct sfc_dp_rxq *dp_rxq = queue;
477 	struct sfc_rxq *rxq;
478 	struct sfc_adapter *sa;
479 	unsigned int sw_index;
480 
481 	if (dp_rxq == NULL)
482 		return;
483 
484 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
485 	sa = rxq->evq->sa;
486 	sfc_adapter_lock(sa);
487 
488 	sw_index = dp_rxq->dpq.queue_id;
489 
490 	sfc_log_init(sa, "RxQ=%u", sw_index);
491 
492 	sfc_rx_qfini(sa, sw_index);
493 
494 	sfc_adapter_unlock(sa);
495 }
496 
497 static int
498 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
499 		   uint16_t nb_tx_desc, unsigned int socket_id,
500 		   const struct rte_eth_txconf *tx_conf)
501 {
502 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
503 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
504 	int rc;
505 
506 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
507 		     tx_queue_id, nb_tx_desc, socket_id);
508 
509 	sfc_adapter_lock(sa);
510 
511 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
512 	if (rc != 0)
513 		goto fail_tx_qinit;
514 
515 	dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
516 
517 	sfc_adapter_unlock(sa);
518 	return 0;
519 
520 fail_tx_qinit:
521 	sfc_adapter_unlock(sa);
522 	SFC_ASSERT(rc > 0);
523 	return -rc;
524 }
525 
526 static void
527 sfc_tx_queue_release(void *queue)
528 {
529 	struct sfc_dp_txq *dp_txq = queue;
530 	struct sfc_txq *txq;
531 	unsigned int sw_index;
532 	struct sfc_adapter *sa;
533 
534 	if (dp_txq == NULL)
535 		return;
536 
537 	txq = sfc_txq_by_dp_txq(dp_txq);
538 	sw_index = dp_txq->dpq.queue_id;
539 
540 	SFC_ASSERT(txq->evq != NULL);
541 	sa = txq->evq->sa;
542 
543 	sfc_log_init(sa, "TxQ = %u", sw_index);
544 
545 	sfc_adapter_lock(sa);
546 
547 	sfc_tx_qfini(sa, sw_index);
548 
549 	sfc_adapter_unlock(sa);
550 }
551 
552 /*
553  * Some statistics are computed as A - B where A and B each increase
554  * monotonically with some hardware counter(s) and the counters are read
555  * asynchronously.
556  *
557  * If packet X is counted in A, but not counted in B yet, computed value is
558  * greater than real.
559  *
560  * If packet X is not counted in A at the moment of reading the counter,
561  * but counted in B at the moment of reading the counter, computed value
562  * is less than real.
563  *
564  * However, counter which grows backward is worse evil than slightly wrong
565  * value. So, let's try to guarantee that it never happens except may be
566  * the case when the MAC stats are zeroed as a result of a NIC reset.
567  */
568 static void
569 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
570 {
571 	if ((int64_t)(newval - *stat) > 0 || newval == 0)
572 		*stat = newval;
573 }
574 
575 static int
576 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
577 {
578 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
579 	struct sfc_port *port = &sa->port;
580 	uint64_t *mac_stats;
581 	int ret;
582 
583 	rte_spinlock_lock(&port->mac_stats_lock);
584 
585 	ret = sfc_port_update_mac_stats(sa);
586 	if (ret != 0)
587 		goto unlock;
588 
589 	mac_stats = port->mac_stats_buf;
590 
591 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
592 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
593 		stats->ipackets =
594 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
595 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
596 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
597 		stats->opackets =
598 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
599 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
600 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
601 		stats->ibytes =
602 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
603 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
604 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
605 		stats->obytes =
606 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
607 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
608 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
609 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
610 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
611 	} else {
612 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
613 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
614 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
615 		/*
616 		 * Take into account stats which are whenever supported
617 		 * on EF10. If some stat is not supported by current
618 		 * firmware variant or HW revision, it is guaranteed
619 		 * to be zero in mac_stats.
620 		 */
621 		stats->imissed =
622 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
623 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
624 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
625 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
626 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
627 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
628 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
629 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
630 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
631 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
632 		stats->ierrors =
633 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
634 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
635 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
636 		/* no oerrors counters supported on EF10 */
637 
638 		/* Exclude missed, errors and pauses from Rx packets */
639 		sfc_update_diff_stat(&port->ipackets,
640 			mac_stats[EFX_MAC_RX_PKTS] -
641 			mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
642 			stats->imissed - stats->ierrors);
643 		stats->ipackets = port->ipackets;
644 	}
645 
646 unlock:
647 	rte_spinlock_unlock(&port->mac_stats_lock);
648 	SFC_ASSERT(ret >= 0);
649 	return -ret;
650 }
651 
652 static int
653 sfc_stats_reset(struct rte_eth_dev *dev)
654 {
655 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
656 	struct sfc_port *port = &sa->port;
657 	int rc;
658 
659 	if (sa->state != SFC_ADAPTER_STARTED) {
660 		/*
661 		 * The operation cannot be done if port is not started; it
662 		 * will be scheduled to be done during the next port start
663 		 */
664 		port->mac_stats_reset_pending = B_TRUE;
665 		return 0;
666 	}
667 
668 	rc = sfc_port_reset_mac_stats(sa);
669 	if (rc != 0)
670 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
671 
672 	SFC_ASSERT(rc >= 0);
673 	return -rc;
674 }
675 
676 static int
677 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
678 	       unsigned int xstats_count)
679 {
680 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
681 	struct sfc_port *port = &sa->port;
682 	uint64_t *mac_stats;
683 	int rc;
684 	unsigned int i;
685 	int nstats = 0;
686 
687 	rte_spinlock_lock(&port->mac_stats_lock);
688 
689 	rc = sfc_port_update_mac_stats(sa);
690 	if (rc != 0) {
691 		SFC_ASSERT(rc > 0);
692 		nstats = -rc;
693 		goto unlock;
694 	}
695 
696 	mac_stats = port->mac_stats_buf;
697 
698 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
699 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
700 			if (xstats != NULL && nstats < (int)xstats_count) {
701 				xstats[nstats].id = nstats;
702 				xstats[nstats].value = mac_stats[i];
703 			}
704 			nstats++;
705 		}
706 	}
707 
708 unlock:
709 	rte_spinlock_unlock(&port->mac_stats_lock);
710 
711 	return nstats;
712 }
713 
714 static int
715 sfc_xstats_get_names(struct rte_eth_dev *dev,
716 		     struct rte_eth_xstat_name *xstats_names,
717 		     unsigned int xstats_count)
718 {
719 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
720 	struct sfc_port *port = &sa->port;
721 	unsigned int i;
722 	unsigned int nstats = 0;
723 
724 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
725 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
726 			if (xstats_names != NULL && nstats < xstats_count)
727 				strlcpy(xstats_names[nstats].name,
728 					efx_mac_stat_name(sa->nic, i),
729 					sizeof(xstats_names[0].name));
730 			nstats++;
731 		}
732 	}
733 
734 	return nstats;
735 }
736 
737 static int
738 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
739 		     uint64_t *values, unsigned int n)
740 {
741 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
742 	struct sfc_port *port = &sa->port;
743 	uint64_t *mac_stats;
744 	unsigned int nb_supported = 0;
745 	unsigned int nb_written = 0;
746 	unsigned int i;
747 	int ret;
748 	int rc;
749 
750 	if (unlikely(values == NULL) ||
751 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
752 		return port->mac_stats_nb_supported;
753 
754 	rte_spinlock_lock(&port->mac_stats_lock);
755 
756 	rc = sfc_port_update_mac_stats(sa);
757 	if (rc != 0) {
758 		SFC_ASSERT(rc > 0);
759 		ret = -rc;
760 		goto unlock;
761 	}
762 
763 	mac_stats = port->mac_stats_buf;
764 
765 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
766 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
767 			continue;
768 
769 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
770 			values[nb_written++] = mac_stats[i];
771 
772 		++nb_supported;
773 	}
774 
775 	ret = nb_written;
776 
777 unlock:
778 	rte_spinlock_unlock(&port->mac_stats_lock);
779 
780 	return ret;
781 }
782 
783 static int
784 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
785 			   struct rte_eth_xstat_name *xstats_names,
786 			   const uint64_t *ids, unsigned int size)
787 {
788 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
789 	struct sfc_port *port = &sa->port;
790 	unsigned int nb_supported = 0;
791 	unsigned int nb_written = 0;
792 	unsigned int i;
793 
794 	if (unlikely(xstats_names == NULL) ||
795 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
796 		return port->mac_stats_nb_supported;
797 
798 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
799 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
800 			continue;
801 
802 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
803 			char *name = xstats_names[nb_written++].name;
804 
805 			strlcpy(name, efx_mac_stat_name(sa->nic, i),
806 				sizeof(xstats_names[0].name));
807 		}
808 
809 		++nb_supported;
810 	}
811 
812 	return nb_written;
813 }
814 
815 static int
816 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
817 {
818 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
819 	unsigned int wanted_fc, link_fc;
820 
821 	memset(fc_conf, 0, sizeof(*fc_conf));
822 
823 	sfc_adapter_lock(sa);
824 
825 	if (sa->state == SFC_ADAPTER_STARTED)
826 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
827 	else
828 		link_fc = sa->port.flow_ctrl;
829 
830 	switch (link_fc) {
831 	case 0:
832 		fc_conf->mode = RTE_FC_NONE;
833 		break;
834 	case EFX_FCNTL_RESPOND:
835 		fc_conf->mode = RTE_FC_RX_PAUSE;
836 		break;
837 	case EFX_FCNTL_GENERATE:
838 		fc_conf->mode = RTE_FC_TX_PAUSE;
839 		break;
840 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
841 		fc_conf->mode = RTE_FC_FULL;
842 		break;
843 	default:
844 		sfc_err(sa, "%s: unexpected flow control value %#x",
845 			__func__, link_fc);
846 	}
847 
848 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
849 
850 	sfc_adapter_unlock(sa);
851 
852 	return 0;
853 }
854 
855 static int
856 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
857 {
858 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
859 	struct sfc_port *port = &sa->port;
860 	unsigned int fcntl;
861 	int rc;
862 
863 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
864 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
865 	    fc_conf->mac_ctrl_frame_fwd != 0) {
866 		sfc_err(sa, "unsupported flow control settings specified");
867 		rc = EINVAL;
868 		goto fail_inval;
869 	}
870 
871 	switch (fc_conf->mode) {
872 	case RTE_FC_NONE:
873 		fcntl = 0;
874 		break;
875 	case RTE_FC_RX_PAUSE:
876 		fcntl = EFX_FCNTL_RESPOND;
877 		break;
878 	case RTE_FC_TX_PAUSE:
879 		fcntl = EFX_FCNTL_GENERATE;
880 		break;
881 	case RTE_FC_FULL:
882 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
883 		break;
884 	default:
885 		rc = EINVAL;
886 		goto fail_inval;
887 	}
888 
889 	sfc_adapter_lock(sa);
890 
891 	if (sa->state == SFC_ADAPTER_STARTED) {
892 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
893 		if (rc != 0)
894 			goto fail_mac_fcntl_set;
895 	}
896 
897 	port->flow_ctrl = fcntl;
898 	port->flow_ctrl_autoneg = fc_conf->autoneg;
899 
900 	sfc_adapter_unlock(sa);
901 
902 	return 0;
903 
904 fail_mac_fcntl_set:
905 	sfc_adapter_unlock(sa);
906 fail_inval:
907 	SFC_ASSERT(rc > 0);
908 	return -rc;
909 }
910 
911 static int
912 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
913 {
914 	struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
915 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
916 	boolean_t scatter_enabled;
917 	const char *error;
918 	unsigned int i;
919 
920 	for (i = 0; i < sas->rxq_count; i++) {
921 		if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
922 			continue;
923 
924 		scatter_enabled = (sas->rxq_info[i].type_flags &
925 				   EFX_RXQ_FLAG_SCATTER);
926 
927 		if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
928 					  encp->enc_rx_prefix_size,
929 					  scatter_enabled, &error)) {
930 			sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
931 				error);
932 			return EINVAL;
933 		}
934 	}
935 
936 	return 0;
937 }
938 
939 static int
940 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
941 {
942 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
943 	size_t pdu = EFX_MAC_PDU(mtu);
944 	size_t old_pdu;
945 	int rc;
946 
947 	sfc_log_init(sa, "mtu=%u", mtu);
948 
949 	rc = EINVAL;
950 	if (pdu < EFX_MAC_PDU_MIN) {
951 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
952 			(unsigned int)mtu, (unsigned int)pdu,
953 			EFX_MAC_PDU_MIN);
954 		goto fail_inval;
955 	}
956 	if (pdu > EFX_MAC_PDU_MAX) {
957 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
958 			(unsigned int)mtu, (unsigned int)pdu,
959 			(unsigned int)EFX_MAC_PDU_MAX);
960 		goto fail_inval;
961 	}
962 
963 	sfc_adapter_lock(sa);
964 
965 	rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
966 	if (rc != 0)
967 		goto fail_check_scatter;
968 
969 	if (pdu != sa->port.pdu) {
970 		if (sa->state == SFC_ADAPTER_STARTED) {
971 			sfc_stop(sa);
972 
973 			old_pdu = sa->port.pdu;
974 			sa->port.pdu = pdu;
975 			rc = sfc_start(sa);
976 			if (rc != 0)
977 				goto fail_start;
978 		} else {
979 			sa->port.pdu = pdu;
980 		}
981 	}
982 
983 	/*
984 	 * The driver does not use it, but other PMDs update jumbo frame
985 	 * flag and max_rx_pkt_len when MTU is set.
986 	 */
987 	if (mtu > RTE_ETHER_MAX_LEN) {
988 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
989 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
990 	}
991 
992 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
993 
994 	sfc_adapter_unlock(sa);
995 
996 	sfc_log_init(sa, "done");
997 	return 0;
998 
999 fail_start:
1000 	sa->port.pdu = old_pdu;
1001 	if (sfc_start(sa) != 0)
1002 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1003 			"PDU max size - port is stopped",
1004 			(unsigned int)pdu, (unsigned int)old_pdu);
1005 
1006 fail_check_scatter:
1007 	sfc_adapter_unlock(sa);
1008 
1009 fail_inval:
1010 	sfc_log_init(sa, "failed %d", rc);
1011 	SFC_ASSERT(rc > 0);
1012 	return -rc;
1013 }
1014 static int
1015 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1016 {
1017 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1018 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1019 	struct sfc_port *port = &sa->port;
1020 	struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1021 	int rc = 0;
1022 
1023 	sfc_adapter_lock(sa);
1024 
1025 	/*
1026 	 * Copy the address to the device private data so that
1027 	 * it could be recalled in the case of adapter restart.
1028 	 */
1029 	rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1030 
1031 	/*
1032 	 * Neither of the two following checks can return
1033 	 * an error. The new MAC address is preserved in
1034 	 * the device private data and can be activated
1035 	 * on the next port start if the user prevents
1036 	 * isolated mode from being enabled.
1037 	 */
1038 	if (sfc_sa2shared(sa)->isolated) {
1039 		sfc_warn(sa, "isolated mode is active on the port");
1040 		sfc_warn(sa, "will not set MAC address");
1041 		goto unlock;
1042 	}
1043 
1044 	if (sa->state != SFC_ADAPTER_STARTED) {
1045 		sfc_notice(sa, "the port is not started");
1046 		sfc_notice(sa, "the new MAC address will be set on port start");
1047 
1048 		goto unlock;
1049 	}
1050 
1051 	if (encp->enc_allow_set_mac_with_installed_filters) {
1052 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1053 		if (rc != 0) {
1054 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1055 			goto unlock;
1056 		}
1057 
1058 		/*
1059 		 * Changing the MAC address by means of MCDI request
1060 		 * has no effect on received traffic, therefore
1061 		 * we also need to update unicast filters
1062 		 */
1063 		rc = sfc_set_rx_mode(sa);
1064 		if (rc != 0) {
1065 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
1066 			/* Rollback the old address */
1067 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1068 			(void)sfc_set_rx_mode(sa);
1069 		}
1070 	} else {
1071 		sfc_warn(sa, "cannot set MAC address with filters installed");
1072 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1073 		sfc_warn(sa, "(some traffic may be dropped)");
1074 
1075 		/*
1076 		 * Since setting MAC address with filters installed is not
1077 		 * allowed on the adapter, the new MAC address will be set
1078 		 * by means of adapter restart. sfc_start() shall retrieve
1079 		 * the new address from the device private data and set it.
1080 		 */
1081 		sfc_stop(sa);
1082 		rc = sfc_start(sa);
1083 		if (rc != 0)
1084 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1085 	}
1086 
1087 unlock:
1088 	if (rc != 0)
1089 		rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1090 
1091 	sfc_adapter_unlock(sa);
1092 
1093 	SFC_ASSERT(rc >= 0);
1094 	return -rc;
1095 }
1096 
1097 
1098 static int
1099 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1100 		struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1101 {
1102 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1103 	struct sfc_port *port = &sa->port;
1104 	uint8_t *mc_addrs = port->mcast_addrs;
1105 	int rc;
1106 	unsigned int i;
1107 
1108 	if (sfc_sa2shared(sa)->isolated) {
1109 		sfc_err(sa, "isolated mode is active on the port");
1110 		sfc_err(sa, "will not set multicast address list");
1111 		return -ENOTSUP;
1112 	}
1113 
1114 	if (mc_addrs == NULL)
1115 		return -ENOBUFS;
1116 
1117 	if (nb_mc_addr > port->max_mcast_addrs) {
1118 		sfc_err(sa, "too many multicast addresses: %u > %u",
1119 			 nb_mc_addr, port->max_mcast_addrs);
1120 		return -EINVAL;
1121 	}
1122 
1123 	for (i = 0; i < nb_mc_addr; ++i) {
1124 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1125 				 EFX_MAC_ADDR_LEN);
1126 		mc_addrs += EFX_MAC_ADDR_LEN;
1127 	}
1128 
1129 	port->nb_mcast_addrs = nb_mc_addr;
1130 
1131 	if (sa->state != SFC_ADAPTER_STARTED)
1132 		return 0;
1133 
1134 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1135 					port->nb_mcast_addrs);
1136 	if (rc != 0)
1137 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1138 
1139 	SFC_ASSERT(rc >= 0);
1140 	return -rc;
1141 }
1142 
1143 /*
1144  * The function is used by the secondary process as well. It must not
1145  * use any process-local pointers from the adapter data.
1146  */
1147 static void
1148 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1149 		      struct rte_eth_rxq_info *qinfo)
1150 {
1151 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1152 	struct sfc_rxq_info *rxq_info;
1153 
1154 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1155 
1156 	rxq_info = &sas->rxq_info[rx_queue_id];
1157 
1158 	qinfo->mp = rxq_info->refill_mb_pool;
1159 	qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1160 	qinfo->conf.rx_drop_en = 1;
1161 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1162 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1163 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1164 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1165 		qinfo->scattered_rx = 1;
1166 	}
1167 	qinfo->nb_desc = rxq_info->entries;
1168 }
1169 
1170 /*
1171  * The function is used by the secondary process as well. It must not
1172  * use any process-local pointers from the adapter data.
1173  */
1174 static void
1175 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1176 		      struct rte_eth_txq_info *qinfo)
1177 {
1178 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1179 	struct sfc_txq_info *txq_info;
1180 
1181 	SFC_ASSERT(tx_queue_id < sas->txq_count);
1182 
1183 	txq_info = &sas->txq_info[tx_queue_id];
1184 
1185 	memset(qinfo, 0, sizeof(*qinfo));
1186 
1187 	qinfo->conf.offloads = txq_info->offloads;
1188 	qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1189 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1190 	qinfo->nb_desc = txq_info->entries;
1191 }
1192 
1193 /*
1194  * The function is used by the secondary process as well. It must not
1195  * use any process-local pointers from the adapter data.
1196  */
1197 static uint32_t
1198 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1199 {
1200 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1201 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1202 	struct sfc_rxq_info *rxq_info;
1203 
1204 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1205 	rxq_info = &sas->rxq_info[rx_queue_id];
1206 
1207 	if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1208 		return 0;
1209 
1210 	return sap->dp_rx->qdesc_npending(rxq_info->dp);
1211 }
1212 
1213 /*
1214  * The function is used by the secondary process as well. It must not
1215  * use any process-local pointers from the adapter data.
1216  */
1217 static int
1218 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1219 {
1220 	struct sfc_dp_rxq *dp_rxq = queue;
1221 	const struct sfc_dp_rx *dp_rx;
1222 
1223 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1224 
1225 	return offset < dp_rx->qdesc_npending(dp_rxq);
1226 }
1227 
1228 /*
1229  * The function is used by the secondary process as well. It must not
1230  * use any process-local pointers from the adapter data.
1231  */
1232 static int
1233 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1234 {
1235 	struct sfc_dp_rxq *dp_rxq = queue;
1236 	const struct sfc_dp_rx *dp_rx;
1237 
1238 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1239 
1240 	return dp_rx->qdesc_status(dp_rxq, offset);
1241 }
1242 
1243 /*
1244  * The function is used by the secondary process as well. It must not
1245  * use any process-local pointers from the adapter data.
1246  */
1247 static int
1248 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1249 {
1250 	struct sfc_dp_txq *dp_txq = queue;
1251 	const struct sfc_dp_tx *dp_tx;
1252 
1253 	dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1254 
1255 	return dp_tx->qdesc_status(dp_txq, offset);
1256 }
1257 
1258 static int
1259 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1260 {
1261 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1262 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1263 	int rc;
1264 
1265 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1266 
1267 	sfc_adapter_lock(sa);
1268 
1269 	rc = EINVAL;
1270 	if (sa->state != SFC_ADAPTER_STARTED)
1271 		goto fail_not_started;
1272 
1273 	if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1274 		goto fail_not_setup;
1275 
1276 	rc = sfc_rx_qstart(sa, rx_queue_id);
1277 	if (rc != 0)
1278 		goto fail_rx_qstart;
1279 
1280 	sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1281 
1282 	sfc_adapter_unlock(sa);
1283 
1284 	return 0;
1285 
1286 fail_rx_qstart:
1287 fail_not_setup:
1288 fail_not_started:
1289 	sfc_adapter_unlock(sa);
1290 	SFC_ASSERT(rc > 0);
1291 	return -rc;
1292 }
1293 
1294 static int
1295 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1296 {
1297 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1298 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1299 
1300 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1301 
1302 	sfc_adapter_lock(sa);
1303 	sfc_rx_qstop(sa, rx_queue_id);
1304 
1305 	sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1306 
1307 	sfc_adapter_unlock(sa);
1308 
1309 	return 0;
1310 }
1311 
1312 static int
1313 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1314 {
1315 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1316 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1317 	int rc;
1318 
1319 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1320 
1321 	sfc_adapter_lock(sa);
1322 
1323 	rc = EINVAL;
1324 	if (sa->state != SFC_ADAPTER_STARTED)
1325 		goto fail_not_started;
1326 
1327 	if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1328 		goto fail_not_setup;
1329 
1330 	rc = sfc_tx_qstart(sa, tx_queue_id);
1331 	if (rc != 0)
1332 		goto fail_tx_qstart;
1333 
1334 	sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1335 
1336 	sfc_adapter_unlock(sa);
1337 	return 0;
1338 
1339 fail_tx_qstart:
1340 
1341 fail_not_setup:
1342 fail_not_started:
1343 	sfc_adapter_unlock(sa);
1344 	SFC_ASSERT(rc > 0);
1345 	return -rc;
1346 }
1347 
1348 static int
1349 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1350 {
1351 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1352 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1353 
1354 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1355 
1356 	sfc_adapter_lock(sa);
1357 
1358 	sfc_tx_qstop(sa, tx_queue_id);
1359 
1360 	sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1361 
1362 	sfc_adapter_unlock(sa);
1363 	return 0;
1364 }
1365 
1366 static efx_tunnel_protocol_t
1367 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1368 {
1369 	switch (rte_type) {
1370 	case RTE_TUNNEL_TYPE_VXLAN:
1371 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1372 	case RTE_TUNNEL_TYPE_GENEVE:
1373 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1374 	default:
1375 		return EFX_TUNNEL_NPROTOS;
1376 	}
1377 }
1378 
1379 enum sfc_udp_tunnel_op_e {
1380 	SFC_UDP_TUNNEL_ADD_PORT,
1381 	SFC_UDP_TUNNEL_DEL_PORT,
1382 };
1383 
1384 static int
1385 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1386 		      struct rte_eth_udp_tunnel *tunnel_udp,
1387 		      enum sfc_udp_tunnel_op_e op)
1388 {
1389 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1390 	efx_tunnel_protocol_t tunnel_proto;
1391 	int rc;
1392 
1393 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1394 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1395 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1396 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1397 
1398 	tunnel_proto =
1399 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1400 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1401 		rc = ENOTSUP;
1402 		goto fail_bad_proto;
1403 	}
1404 
1405 	sfc_adapter_lock(sa);
1406 
1407 	switch (op) {
1408 	case SFC_UDP_TUNNEL_ADD_PORT:
1409 		rc = efx_tunnel_config_udp_add(sa->nic,
1410 					       tunnel_udp->udp_port,
1411 					       tunnel_proto);
1412 		break;
1413 	case SFC_UDP_TUNNEL_DEL_PORT:
1414 		rc = efx_tunnel_config_udp_remove(sa->nic,
1415 						  tunnel_udp->udp_port,
1416 						  tunnel_proto);
1417 		break;
1418 	default:
1419 		rc = EINVAL;
1420 		goto fail_bad_op;
1421 	}
1422 
1423 	if (rc != 0)
1424 		goto fail_op;
1425 
1426 	if (sa->state == SFC_ADAPTER_STARTED) {
1427 		rc = efx_tunnel_reconfigure(sa->nic);
1428 		if (rc == EAGAIN) {
1429 			/*
1430 			 * Configuration is accepted by FW and MC reboot
1431 			 * is initiated to apply the changes. MC reboot
1432 			 * will be handled in a usual way (MC reboot
1433 			 * event on management event queue and adapter
1434 			 * restart).
1435 			 */
1436 			rc = 0;
1437 		} else if (rc != 0) {
1438 			goto fail_reconfigure;
1439 		}
1440 	}
1441 
1442 	sfc_adapter_unlock(sa);
1443 	return 0;
1444 
1445 fail_reconfigure:
1446 	/* Remove/restore entry since the change makes the trouble */
1447 	switch (op) {
1448 	case SFC_UDP_TUNNEL_ADD_PORT:
1449 		(void)efx_tunnel_config_udp_remove(sa->nic,
1450 						   tunnel_udp->udp_port,
1451 						   tunnel_proto);
1452 		break;
1453 	case SFC_UDP_TUNNEL_DEL_PORT:
1454 		(void)efx_tunnel_config_udp_add(sa->nic,
1455 						tunnel_udp->udp_port,
1456 						tunnel_proto);
1457 		break;
1458 	}
1459 
1460 fail_op:
1461 fail_bad_op:
1462 	sfc_adapter_unlock(sa);
1463 
1464 fail_bad_proto:
1465 	SFC_ASSERT(rc > 0);
1466 	return -rc;
1467 }
1468 
1469 static int
1470 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1471 			    struct rte_eth_udp_tunnel *tunnel_udp)
1472 {
1473 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1474 }
1475 
1476 static int
1477 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1478 			    struct rte_eth_udp_tunnel *tunnel_udp)
1479 {
1480 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1481 }
1482 
1483 /*
1484  * The function is used by the secondary process as well. It must not
1485  * use any process-local pointers from the adapter data.
1486  */
1487 static int
1488 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1489 			  struct rte_eth_rss_conf *rss_conf)
1490 {
1491 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1492 	struct sfc_rss *rss = &sas->rss;
1493 
1494 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1495 		return -ENOTSUP;
1496 
1497 	/*
1498 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1499 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1500 	 * flags which corresponds to the active EFX configuration stored
1501 	 * locally in 'sfc_adapter' and kept up-to-date
1502 	 */
1503 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1504 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1505 	if (rss_conf->rss_key != NULL)
1506 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1507 
1508 	return 0;
1509 }
1510 
1511 static int
1512 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1513 			struct rte_eth_rss_conf *rss_conf)
1514 {
1515 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1516 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1517 	unsigned int efx_hash_types;
1518 	int rc = 0;
1519 
1520 	if (sfc_sa2shared(sa)->isolated)
1521 		return -ENOTSUP;
1522 
1523 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1524 		sfc_err(sa, "RSS is not available");
1525 		return -ENOTSUP;
1526 	}
1527 
1528 	if (rss->channels == 0) {
1529 		sfc_err(sa, "RSS is not configured");
1530 		return -EINVAL;
1531 	}
1532 
1533 	if ((rss_conf->rss_key != NULL) &&
1534 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1535 		sfc_err(sa, "RSS key size is wrong (should be %zu)",
1536 			sizeof(rss->key));
1537 		return -EINVAL;
1538 	}
1539 
1540 	sfc_adapter_lock(sa);
1541 
1542 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1543 	if (rc != 0)
1544 		goto fail_rx_hf_rte_to_efx;
1545 
1546 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1547 				   rss->hash_alg, efx_hash_types, B_TRUE);
1548 	if (rc != 0)
1549 		goto fail_scale_mode_set;
1550 
1551 	if (rss_conf->rss_key != NULL) {
1552 		if (sa->state == SFC_ADAPTER_STARTED) {
1553 			rc = efx_rx_scale_key_set(sa->nic,
1554 						  EFX_RSS_CONTEXT_DEFAULT,
1555 						  rss_conf->rss_key,
1556 						  sizeof(rss->key));
1557 			if (rc != 0)
1558 				goto fail_scale_key_set;
1559 		}
1560 
1561 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1562 	}
1563 
1564 	rss->hash_types = efx_hash_types;
1565 
1566 	sfc_adapter_unlock(sa);
1567 
1568 	return 0;
1569 
1570 fail_scale_key_set:
1571 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1572 				  EFX_RX_HASHALG_TOEPLITZ,
1573 				  rss->hash_types, B_TRUE) != 0)
1574 		sfc_err(sa, "failed to restore RSS mode");
1575 
1576 fail_scale_mode_set:
1577 fail_rx_hf_rte_to_efx:
1578 	sfc_adapter_unlock(sa);
1579 	return -rc;
1580 }
1581 
1582 /*
1583  * The function is used by the secondary process as well. It must not
1584  * use any process-local pointers from the adapter data.
1585  */
1586 static int
1587 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1588 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1589 		       uint16_t reta_size)
1590 {
1591 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1592 	struct sfc_rss *rss = &sas->rss;
1593 	int entry;
1594 
1595 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1596 		return -ENOTSUP;
1597 
1598 	if (rss->channels == 0)
1599 		return -EINVAL;
1600 
1601 	if (reta_size != EFX_RSS_TBL_SIZE)
1602 		return -EINVAL;
1603 
1604 	for (entry = 0; entry < reta_size; entry++) {
1605 		int grp = entry / RTE_RETA_GROUP_SIZE;
1606 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1607 
1608 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1609 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1610 	}
1611 
1612 	return 0;
1613 }
1614 
1615 static int
1616 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1617 			struct rte_eth_rss_reta_entry64 *reta_conf,
1618 			uint16_t reta_size)
1619 {
1620 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1621 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1622 	unsigned int *rss_tbl_new;
1623 	uint16_t entry;
1624 	int rc = 0;
1625 
1626 
1627 	if (sfc_sa2shared(sa)->isolated)
1628 		return -ENOTSUP;
1629 
1630 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1631 		sfc_err(sa, "RSS is not available");
1632 		return -ENOTSUP;
1633 	}
1634 
1635 	if (rss->channels == 0) {
1636 		sfc_err(sa, "RSS is not configured");
1637 		return -EINVAL;
1638 	}
1639 
1640 	if (reta_size != EFX_RSS_TBL_SIZE) {
1641 		sfc_err(sa, "RETA size is wrong (should be %u)",
1642 			EFX_RSS_TBL_SIZE);
1643 		return -EINVAL;
1644 	}
1645 
1646 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1647 	if (rss_tbl_new == NULL)
1648 		return -ENOMEM;
1649 
1650 	sfc_adapter_lock(sa);
1651 
1652 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1653 
1654 	for (entry = 0; entry < reta_size; entry++) {
1655 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1656 		struct rte_eth_rss_reta_entry64 *grp;
1657 
1658 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1659 
1660 		if (grp->mask & (1ull << grp_idx)) {
1661 			if (grp->reta[grp_idx] >= rss->channels) {
1662 				rc = EINVAL;
1663 				goto bad_reta_entry;
1664 			}
1665 			rss_tbl_new[entry] = grp->reta[grp_idx];
1666 		}
1667 	}
1668 
1669 	if (sa->state == SFC_ADAPTER_STARTED) {
1670 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1671 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1672 		if (rc != 0)
1673 			goto fail_scale_tbl_set;
1674 	}
1675 
1676 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1677 
1678 fail_scale_tbl_set:
1679 bad_reta_entry:
1680 	sfc_adapter_unlock(sa);
1681 
1682 	rte_free(rss_tbl_new);
1683 
1684 	SFC_ASSERT(rc >= 0);
1685 	return -rc;
1686 }
1687 
1688 static int
1689 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1690 		    enum rte_filter_op filter_op,
1691 		    void *arg)
1692 {
1693 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1694 	int rc = ENOTSUP;
1695 
1696 	sfc_log_init(sa, "entry");
1697 
1698 	switch (filter_type) {
1699 	case RTE_ETH_FILTER_NONE:
1700 		sfc_err(sa, "Global filters configuration not supported");
1701 		break;
1702 	case RTE_ETH_FILTER_MACVLAN:
1703 		sfc_err(sa, "MACVLAN filters not supported");
1704 		break;
1705 	case RTE_ETH_FILTER_ETHERTYPE:
1706 		sfc_err(sa, "EtherType filters not supported");
1707 		break;
1708 	case RTE_ETH_FILTER_FLEXIBLE:
1709 		sfc_err(sa, "Flexible filters not supported");
1710 		break;
1711 	case RTE_ETH_FILTER_SYN:
1712 		sfc_err(sa, "SYN filters not supported");
1713 		break;
1714 	case RTE_ETH_FILTER_NTUPLE:
1715 		sfc_err(sa, "NTUPLE filters not supported");
1716 		break;
1717 	case RTE_ETH_FILTER_TUNNEL:
1718 		sfc_err(sa, "Tunnel filters not supported");
1719 		break;
1720 	case RTE_ETH_FILTER_FDIR:
1721 		sfc_err(sa, "Flow Director filters not supported");
1722 		break;
1723 	case RTE_ETH_FILTER_HASH:
1724 		sfc_err(sa, "Hash filters not supported");
1725 		break;
1726 	case RTE_ETH_FILTER_GENERIC:
1727 		if (filter_op != RTE_ETH_FILTER_GET) {
1728 			rc = EINVAL;
1729 		} else {
1730 			*(const void **)arg = &sfc_flow_ops;
1731 			rc = 0;
1732 		}
1733 		break;
1734 	default:
1735 		sfc_err(sa, "Unknown filter type %u", filter_type);
1736 		break;
1737 	}
1738 
1739 	sfc_log_init(sa, "exit: %d", -rc);
1740 	SFC_ASSERT(rc >= 0);
1741 	return -rc;
1742 }
1743 
1744 static int
1745 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1746 {
1747 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1748 
1749 	/*
1750 	 * If Rx datapath does not provide callback to check mempool,
1751 	 * all pools are supported.
1752 	 */
1753 	if (sap->dp_rx->pool_ops_supported == NULL)
1754 		return 1;
1755 
1756 	return sap->dp_rx->pool_ops_supported(pool);
1757 }
1758 
1759 static int
1760 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1761 {
1762 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1763 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1764 	struct sfc_rxq_info *rxq_info;
1765 
1766 	SFC_ASSERT(queue_id < sas->rxq_count);
1767 	rxq_info = &sas->rxq_info[queue_id];
1768 
1769 	return sap->dp_rx->intr_enable(rxq_info->dp);
1770 }
1771 
1772 static int
1773 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1774 {
1775 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1776 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1777 	struct sfc_rxq_info *rxq_info;
1778 
1779 	SFC_ASSERT(queue_id < sas->rxq_count);
1780 	rxq_info = &sas->rxq_info[queue_id];
1781 
1782 	return sap->dp_rx->intr_disable(rxq_info->dp);
1783 }
1784 
1785 static const struct eth_dev_ops sfc_eth_dev_ops = {
1786 	.dev_configure			= sfc_dev_configure,
1787 	.dev_start			= sfc_dev_start,
1788 	.dev_stop			= sfc_dev_stop,
1789 	.dev_set_link_up		= sfc_dev_set_link_up,
1790 	.dev_set_link_down		= sfc_dev_set_link_down,
1791 	.dev_close			= sfc_dev_close,
1792 	.promiscuous_enable		= sfc_dev_promisc_enable,
1793 	.promiscuous_disable		= sfc_dev_promisc_disable,
1794 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1795 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1796 	.link_update			= sfc_dev_link_update,
1797 	.stats_get			= sfc_stats_get,
1798 	.stats_reset			= sfc_stats_reset,
1799 	.xstats_get			= sfc_xstats_get,
1800 	.xstats_reset			= sfc_stats_reset,
1801 	.xstats_get_names		= sfc_xstats_get_names,
1802 	.dev_infos_get			= sfc_dev_infos_get,
1803 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1804 	.mtu_set			= sfc_dev_set_mtu,
1805 	.rx_queue_start			= sfc_rx_queue_start,
1806 	.rx_queue_stop			= sfc_rx_queue_stop,
1807 	.tx_queue_start			= sfc_tx_queue_start,
1808 	.tx_queue_stop			= sfc_tx_queue_stop,
1809 	.rx_queue_setup			= sfc_rx_queue_setup,
1810 	.rx_queue_release		= sfc_rx_queue_release,
1811 	.rx_queue_count			= sfc_rx_queue_count,
1812 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1813 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1814 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1815 	.rx_queue_intr_enable		= sfc_rx_queue_intr_enable,
1816 	.rx_queue_intr_disable		= sfc_rx_queue_intr_disable,
1817 	.tx_queue_setup			= sfc_tx_queue_setup,
1818 	.tx_queue_release		= sfc_tx_queue_release,
1819 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1820 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1821 	.mac_addr_set			= sfc_mac_addr_set,
1822 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1823 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1824 	.reta_update			= sfc_dev_rss_reta_update,
1825 	.reta_query			= sfc_dev_rss_reta_query,
1826 	.rss_hash_update		= sfc_dev_rss_hash_update,
1827 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1828 	.filter_ctrl			= sfc_dev_filter_ctrl,
1829 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1830 	.rxq_info_get			= sfc_rx_queue_info_get,
1831 	.txq_info_get			= sfc_tx_queue_info_get,
1832 	.fw_version_get			= sfc_fw_version_get,
1833 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1834 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1835 	.pool_ops_supported		= sfc_pool_ops_supported,
1836 };
1837 
1838 /**
1839  * Duplicate a string in potentially shared memory required for
1840  * multi-process support.
1841  *
1842  * strdup() allocates from process-local heap/memory.
1843  */
1844 static char *
1845 sfc_strdup(const char *str)
1846 {
1847 	size_t size;
1848 	char *copy;
1849 
1850 	if (str == NULL)
1851 		return NULL;
1852 
1853 	size = strlen(str) + 1;
1854 	copy = rte_malloc(__func__, size, 0);
1855 	if (copy != NULL)
1856 		rte_memcpy(copy, str, size);
1857 
1858 	return copy;
1859 }
1860 
1861 static int
1862 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1863 {
1864 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1865 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1866 	const struct sfc_dp_rx *dp_rx;
1867 	const struct sfc_dp_tx *dp_tx;
1868 	const efx_nic_cfg_t *encp;
1869 	unsigned int avail_caps = 0;
1870 	const char *rx_name = NULL;
1871 	const char *tx_name = NULL;
1872 	int rc;
1873 
1874 	switch (sa->family) {
1875 	case EFX_FAMILY_HUNTINGTON:
1876 	case EFX_FAMILY_MEDFORD:
1877 	case EFX_FAMILY_MEDFORD2:
1878 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1879 		break;
1880 	default:
1881 		break;
1882 	}
1883 
1884 	encp = efx_nic_cfg_get(sa->nic);
1885 	if (encp->enc_rx_es_super_buffer_supported)
1886 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1887 
1888 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1889 				sfc_kvarg_string_handler, &rx_name);
1890 	if (rc != 0)
1891 		goto fail_kvarg_rx_datapath;
1892 
1893 	if (rx_name != NULL) {
1894 		dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1895 		if (dp_rx == NULL) {
1896 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1897 			rc = ENOENT;
1898 			goto fail_dp_rx;
1899 		}
1900 		if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1901 			sfc_err(sa,
1902 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1903 				rx_name);
1904 			rc = EINVAL;
1905 			goto fail_dp_rx_caps;
1906 		}
1907 	} else {
1908 		dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1909 		if (dp_rx == NULL) {
1910 			sfc_err(sa, "Rx datapath by caps %#x not found",
1911 				avail_caps);
1912 			rc = ENOENT;
1913 			goto fail_dp_rx;
1914 		}
1915 	}
1916 
1917 	sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1918 	if (sas->dp_rx_name == NULL) {
1919 		rc = ENOMEM;
1920 		goto fail_dp_rx_name;
1921 	}
1922 
1923 	sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1924 
1925 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1926 				sfc_kvarg_string_handler, &tx_name);
1927 	if (rc != 0)
1928 		goto fail_kvarg_tx_datapath;
1929 
1930 	if (tx_name != NULL) {
1931 		dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1932 		if (dp_tx == NULL) {
1933 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1934 			rc = ENOENT;
1935 			goto fail_dp_tx;
1936 		}
1937 		if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1938 			sfc_err(sa,
1939 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1940 				tx_name);
1941 			rc = EINVAL;
1942 			goto fail_dp_tx_caps;
1943 		}
1944 	} else {
1945 		dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1946 		if (dp_tx == NULL) {
1947 			sfc_err(sa, "Tx datapath by caps %#x not found",
1948 				avail_caps);
1949 			rc = ENOENT;
1950 			goto fail_dp_tx;
1951 		}
1952 	}
1953 
1954 	sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1955 	if (sas->dp_tx_name == NULL) {
1956 		rc = ENOMEM;
1957 		goto fail_dp_tx_name;
1958 	}
1959 
1960 	sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1961 
1962 	sa->priv.dp_rx = dp_rx;
1963 	sa->priv.dp_tx = dp_tx;
1964 
1965 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1966 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1967 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1968 
1969 	dev->dev_ops = &sfc_eth_dev_ops;
1970 
1971 	return 0;
1972 
1973 fail_dp_tx_name:
1974 fail_dp_tx_caps:
1975 fail_dp_tx:
1976 fail_kvarg_tx_datapath:
1977 	rte_free(sas->dp_rx_name);
1978 	sas->dp_rx_name = NULL;
1979 
1980 fail_dp_rx_name:
1981 fail_dp_rx_caps:
1982 fail_dp_rx:
1983 fail_kvarg_rx_datapath:
1984 	return rc;
1985 }
1986 
1987 static void
1988 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1989 {
1990 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1991 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1992 
1993 	dev->dev_ops = NULL;
1994 	dev->tx_pkt_prepare = NULL;
1995 	dev->rx_pkt_burst = NULL;
1996 	dev->tx_pkt_burst = NULL;
1997 
1998 	rte_free(sas->dp_tx_name);
1999 	sas->dp_tx_name = NULL;
2000 	sa->priv.dp_tx = NULL;
2001 
2002 	rte_free(sas->dp_rx_name);
2003 	sas->dp_rx_name = NULL;
2004 	sa->priv.dp_rx = NULL;
2005 }
2006 
2007 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2008 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
2009 	.rx_queue_count			= sfc_rx_queue_count,
2010 	.rx_descriptor_done		= sfc_rx_descriptor_done,
2011 	.rx_descriptor_status		= sfc_rx_descriptor_status,
2012 	.tx_descriptor_status		= sfc_tx_descriptor_status,
2013 	.reta_query			= sfc_dev_rss_reta_query,
2014 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
2015 	.rxq_info_get			= sfc_rx_queue_info_get,
2016 	.txq_info_get			= sfc_tx_queue_info_get,
2017 };
2018 
2019 static int
2020 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2021 {
2022 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2023 	struct sfc_adapter_priv *sap;
2024 	const struct sfc_dp_rx *dp_rx;
2025 	const struct sfc_dp_tx *dp_tx;
2026 	int rc;
2027 
2028 	/*
2029 	 * Allocate process private data from heap, since it should not
2030 	 * be located in shared memory allocated using rte_malloc() API.
2031 	 */
2032 	sap = calloc(1, sizeof(*sap));
2033 	if (sap == NULL) {
2034 		rc = ENOMEM;
2035 		goto fail_alloc_priv;
2036 	}
2037 
2038 	sap->logtype_main = logtype_main;
2039 
2040 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2041 	if (dp_rx == NULL) {
2042 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2043 			"cannot find %s Rx datapath", sas->dp_rx_name);
2044 		rc = ENOENT;
2045 		goto fail_dp_rx;
2046 	}
2047 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2048 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2049 			"%s Rx datapath does not support multi-process",
2050 			sas->dp_rx_name);
2051 		rc = EINVAL;
2052 		goto fail_dp_rx_multi_process;
2053 	}
2054 
2055 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2056 	if (dp_tx == NULL) {
2057 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2058 			"cannot find %s Tx datapath", sas->dp_tx_name);
2059 		rc = ENOENT;
2060 		goto fail_dp_tx;
2061 	}
2062 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2063 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2064 			"%s Tx datapath does not support multi-process",
2065 			sas->dp_tx_name);
2066 		rc = EINVAL;
2067 		goto fail_dp_tx_multi_process;
2068 	}
2069 
2070 	sap->dp_rx = dp_rx;
2071 	sap->dp_tx = dp_tx;
2072 
2073 	dev->process_private = sap;
2074 	dev->rx_pkt_burst = dp_rx->pkt_burst;
2075 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2076 	dev->tx_pkt_burst = dp_tx->pkt_burst;
2077 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
2078 
2079 	return 0;
2080 
2081 fail_dp_tx_multi_process:
2082 fail_dp_tx:
2083 fail_dp_rx_multi_process:
2084 fail_dp_rx:
2085 	free(sap);
2086 
2087 fail_alloc_priv:
2088 	return rc;
2089 }
2090 
2091 static void
2092 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
2093 {
2094 	free(dev->process_private);
2095 	dev->process_private = NULL;
2096 	dev->dev_ops = NULL;
2097 	dev->tx_pkt_prepare = NULL;
2098 	dev->tx_pkt_burst = NULL;
2099 	dev->rx_pkt_burst = NULL;
2100 }
2101 
2102 static void
2103 sfc_register_dp(void)
2104 {
2105 	/* Register once */
2106 	if (TAILQ_EMPTY(&sfc_dp_head)) {
2107 		/* Prefer EF10 datapath */
2108 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2109 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2110 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2111 
2112 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2113 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2114 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2115 	}
2116 }
2117 
2118 static int
2119 sfc_eth_dev_init(struct rte_eth_dev *dev)
2120 {
2121 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2122 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2123 	uint32_t logtype_main;
2124 	struct sfc_adapter *sa;
2125 	int rc;
2126 	const efx_nic_cfg_t *encp;
2127 	const struct rte_ether_addr *from;
2128 
2129 	sfc_register_dp();
2130 
2131 	logtype_main = sfc_register_logtype(&pci_dev->addr,
2132 					    SFC_LOGTYPE_MAIN_STR,
2133 					    RTE_LOG_NOTICE);
2134 
2135 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2136 		return -sfc_eth_dev_secondary_init(dev, logtype_main);
2137 
2138 	/* Required for logging */
2139 	sas->pci_addr = pci_dev->addr;
2140 	sas->port_id = dev->data->port_id;
2141 
2142 	/*
2143 	 * Allocate process private data from heap, since it should not
2144 	 * be located in shared memory allocated using rte_malloc() API.
2145 	 */
2146 	sa = calloc(1, sizeof(*sa));
2147 	if (sa == NULL) {
2148 		rc = ENOMEM;
2149 		goto fail_alloc_sa;
2150 	}
2151 
2152 	dev->process_private = sa;
2153 
2154 	/* Required for logging */
2155 	sa->priv.shared = sas;
2156 	sa->priv.logtype_main = logtype_main;
2157 
2158 	sa->eth_dev = dev;
2159 
2160 	/* Copy PCI device info to the dev->data */
2161 	rte_eth_copy_pci_info(dev, pci_dev);
2162 
2163 	rc = sfc_kvargs_parse(sa);
2164 	if (rc != 0)
2165 		goto fail_kvargs_parse;
2166 
2167 	sfc_log_init(sa, "entry");
2168 
2169 	dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
2170 
2171 	dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2172 	if (dev->data->mac_addrs == NULL) {
2173 		rc = ENOMEM;
2174 		goto fail_mac_addrs;
2175 	}
2176 
2177 	sfc_adapter_lock_init(sa);
2178 	sfc_adapter_lock(sa);
2179 
2180 	sfc_log_init(sa, "probing");
2181 	rc = sfc_probe(sa);
2182 	if (rc != 0)
2183 		goto fail_probe;
2184 
2185 	sfc_log_init(sa, "set device ops");
2186 	rc = sfc_eth_dev_set_ops(dev);
2187 	if (rc != 0)
2188 		goto fail_set_ops;
2189 
2190 	sfc_log_init(sa, "attaching");
2191 	rc = sfc_attach(sa);
2192 	if (rc != 0)
2193 		goto fail_attach;
2194 
2195 	encp = efx_nic_cfg_get(sa->nic);
2196 
2197 	/*
2198 	 * The arguments are really reverse order in comparison to
2199 	 * Linux kernel. Copy from NIC config to Ethernet device data.
2200 	 */
2201 	from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2202 	rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2203 
2204 	sfc_adapter_unlock(sa);
2205 
2206 	sfc_log_init(sa, "done");
2207 	return 0;
2208 
2209 fail_attach:
2210 	sfc_eth_dev_clear_ops(dev);
2211 
2212 fail_set_ops:
2213 	sfc_unprobe(sa);
2214 
2215 fail_probe:
2216 	sfc_adapter_unlock(sa);
2217 	sfc_adapter_lock_fini(sa);
2218 	rte_free(dev->data->mac_addrs);
2219 	dev->data->mac_addrs = NULL;
2220 
2221 fail_mac_addrs:
2222 	sfc_kvargs_cleanup(sa);
2223 
2224 fail_kvargs_parse:
2225 	sfc_log_init(sa, "failed %d", rc);
2226 	dev->process_private = NULL;
2227 	free(sa);
2228 
2229 fail_alloc_sa:
2230 	SFC_ASSERT(rc > 0);
2231 	return -rc;
2232 }
2233 
2234 static int
2235 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2236 {
2237 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2238 		sfc_eth_dev_secondary_clear_ops(dev);
2239 		return 0;
2240 	}
2241 
2242 	sfc_dev_close(dev);
2243 
2244 	return 0;
2245 }
2246 
2247 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2248 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2249 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2250 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2251 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2252 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2253 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2254 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2255 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2256 	{ .vendor_id = 0 /* sentinel */ }
2257 };
2258 
2259 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2260 	struct rte_pci_device *pci_dev)
2261 {
2262 	return rte_eth_dev_pci_generic_probe(pci_dev,
2263 		sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2264 }
2265 
2266 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2267 {
2268 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2269 }
2270 
2271 static struct rte_pci_driver sfc_efx_pmd = {
2272 	.id_table = pci_id_sfc_efx_map,
2273 	.drv_flags =
2274 		RTE_PCI_DRV_INTR_LSC |
2275 		RTE_PCI_DRV_NEED_MAPPING,
2276 	.probe = sfc_eth_dev_pci_probe,
2277 	.remove = sfc_eth_dev_pci_remove,
2278 };
2279 
2280 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2281 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2282 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2283 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2284 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2285 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2286 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2287 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2288 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2289 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2290 
2291 RTE_INIT(sfc_driver_register_logtype)
2292 {
2293 	int ret;
2294 
2295 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2296 						   RTE_LOG_NOTICE);
2297 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2298 }
2299