xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 0ecc27f28d202a3356a8601e6762b601ea822c4c)
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 	return sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
409 }
410 
411 static int
412 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
413 {
414 	return sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
415 }
416 
417 static int
418 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
419 {
420 	return sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
421 }
422 
423 static int
424 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
425 {
426 	return sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
427 }
428 
429 static int
430 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
431 		   uint16_t nb_rx_desc, unsigned int socket_id,
432 		   const struct rte_eth_rxconf *rx_conf,
433 		   struct rte_mempool *mb_pool)
434 {
435 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
436 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
437 	int rc;
438 
439 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
440 		     rx_queue_id, nb_rx_desc, socket_id);
441 
442 	sfc_adapter_lock(sa);
443 
444 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
445 			  rx_conf, mb_pool);
446 	if (rc != 0)
447 		goto fail_rx_qinit;
448 
449 	dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
450 
451 	sfc_adapter_unlock(sa);
452 
453 	return 0;
454 
455 fail_rx_qinit:
456 	sfc_adapter_unlock(sa);
457 	SFC_ASSERT(rc > 0);
458 	return -rc;
459 }
460 
461 static void
462 sfc_rx_queue_release(void *queue)
463 {
464 	struct sfc_dp_rxq *dp_rxq = queue;
465 	struct sfc_rxq *rxq;
466 	struct sfc_adapter *sa;
467 	unsigned int sw_index;
468 
469 	if (dp_rxq == NULL)
470 		return;
471 
472 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
473 	sa = rxq->evq->sa;
474 	sfc_adapter_lock(sa);
475 
476 	sw_index = dp_rxq->dpq.queue_id;
477 
478 	sfc_log_init(sa, "RxQ=%u", sw_index);
479 
480 	sfc_rx_qfini(sa, sw_index);
481 
482 	sfc_adapter_unlock(sa);
483 }
484 
485 static int
486 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
487 		   uint16_t nb_tx_desc, unsigned int socket_id,
488 		   const struct rte_eth_txconf *tx_conf)
489 {
490 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
491 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
492 	int rc;
493 
494 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
495 		     tx_queue_id, nb_tx_desc, socket_id);
496 
497 	sfc_adapter_lock(sa);
498 
499 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
500 	if (rc != 0)
501 		goto fail_tx_qinit;
502 
503 	dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
504 
505 	sfc_adapter_unlock(sa);
506 	return 0;
507 
508 fail_tx_qinit:
509 	sfc_adapter_unlock(sa);
510 	SFC_ASSERT(rc > 0);
511 	return -rc;
512 }
513 
514 static void
515 sfc_tx_queue_release(void *queue)
516 {
517 	struct sfc_dp_txq *dp_txq = queue;
518 	struct sfc_txq *txq;
519 	unsigned int sw_index;
520 	struct sfc_adapter *sa;
521 
522 	if (dp_txq == NULL)
523 		return;
524 
525 	txq = sfc_txq_by_dp_txq(dp_txq);
526 	sw_index = dp_txq->dpq.queue_id;
527 
528 	SFC_ASSERT(txq->evq != NULL);
529 	sa = txq->evq->sa;
530 
531 	sfc_log_init(sa, "TxQ = %u", sw_index);
532 
533 	sfc_adapter_lock(sa);
534 
535 	sfc_tx_qfini(sa, sw_index);
536 
537 	sfc_adapter_unlock(sa);
538 }
539 
540 /*
541  * Some statistics are computed as A - B where A and B each increase
542  * monotonically with some hardware counter(s) and the counters are read
543  * asynchronously.
544  *
545  * If packet X is counted in A, but not counted in B yet, computed value is
546  * greater than real.
547  *
548  * If packet X is not counted in A at the moment of reading the counter,
549  * but counted in B at the moment of reading the counter, computed value
550  * is less than real.
551  *
552  * However, counter which grows backward is worse evil than slightly wrong
553  * value. So, let's try to guarantee that it never happens except may be
554  * the case when the MAC stats are zeroed as a result of a NIC reset.
555  */
556 static void
557 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
558 {
559 	if ((int64_t)(newval - *stat) > 0 || newval == 0)
560 		*stat = newval;
561 }
562 
563 static int
564 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
565 {
566 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
567 	struct sfc_port *port = &sa->port;
568 	uint64_t *mac_stats;
569 	int ret;
570 
571 	rte_spinlock_lock(&port->mac_stats_lock);
572 
573 	ret = sfc_port_update_mac_stats(sa);
574 	if (ret != 0)
575 		goto unlock;
576 
577 	mac_stats = port->mac_stats_buf;
578 
579 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
580 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
581 		stats->ipackets =
582 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
583 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
584 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
585 		stats->opackets =
586 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
587 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
588 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
589 		stats->ibytes =
590 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
591 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
592 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
593 		stats->obytes =
594 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
595 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
596 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
597 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
598 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
599 	} else {
600 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
601 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
602 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
603 		/*
604 		 * Take into account stats which are whenever supported
605 		 * on EF10. If some stat is not supported by current
606 		 * firmware variant or HW revision, it is guaranteed
607 		 * to be zero in mac_stats.
608 		 */
609 		stats->imissed =
610 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
611 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
612 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
613 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
614 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
615 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
616 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
617 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
618 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
619 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
620 		stats->ierrors =
621 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
622 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
623 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
624 		/* no oerrors counters supported on EF10 */
625 
626 		/* Exclude missed, errors and pauses from Rx packets */
627 		sfc_update_diff_stat(&port->ipackets,
628 			mac_stats[EFX_MAC_RX_PKTS] -
629 			mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
630 			stats->imissed - stats->ierrors);
631 		stats->ipackets = port->ipackets;
632 	}
633 
634 unlock:
635 	rte_spinlock_unlock(&port->mac_stats_lock);
636 	SFC_ASSERT(ret >= 0);
637 	return -ret;
638 }
639 
640 static int
641 sfc_stats_reset(struct rte_eth_dev *dev)
642 {
643 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
644 	struct sfc_port *port = &sa->port;
645 	int rc;
646 
647 	if (sa->state != SFC_ADAPTER_STARTED) {
648 		/*
649 		 * The operation cannot be done if port is not started; it
650 		 * will be scheduled to be done during the next port start
651 		 */
652 		port->mac_stats_reset_pending = B_TRUE;
653 		return 0;
654 	}
655 
656 	rc = sfc_port_reset_mac_stats(sa);
657 	if (rc != 0)
658 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
659 
660 	SFC_ASSERT(rc >= 0);
661 	return -rc;
662 }
663 
664 static int
665 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
666 	       unsigned int xstats_count)
667 {
668 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
669 	struct sfc_port *port = &sa->port;
670 	uint64_t *mac_stats;
671 	int rc;
672 	unsigned int i;
673 	int nstats = 0;
674 
675 	rte_spinlock_lock(&port->mac_stats_lock);
676 
677 	rc = sfc_port_update_mac_stats(sa);
678 	if (rc != 0) {
679 		SFC_ASSERT(rc > 0);
680 		nstats = -rc;
681 		goto unlock;
682 	}
683 
684 	mac_stats = port->mac_stats_buf;
685 
686 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
687 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
688 			if (xstats != NULL && nstats < (int)xstats_count) {
689 				xstats[nstats].id = nstats;
690 				xstats[nstats].value = mac_stats[i];
691 			}
692 			nstats++;
693 		}
694 	}
695 
696 unlock:
697 	rte_spinlock_unlock(&port->mac_stats_lock);
698 
699 	return nstats;
700 }
701 
702 static int
703 sfc_xstats_get_names(struct rte_eth_dev *dev,
704 		     struct rte_eth_xstat_name *xstats_names,
705 		     unsigned int xstats_count)
706 {
707 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
708 	struct sfc_port *port = &sa->port;
709 	unsigned int i;
710 	unsigned int nstats = 0;
711 
712 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
713 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
714 			if (xstats_names != NULL && nstats < xstats_count)
715 				strlcpy(xstats_names[nstats].name,
716 					efx_mac_stat_name(sa->nic, i),
717 					sizeof(xstats_names[0].name));
718 			nstats++;
719 		}
720 	}
721 
722 	return nstats;
723 }
724 
725 static int
726 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
727 		     uint64_t *values, unsigned int n)
728 {
729 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
730 	struct sfc_port *port = &sa->port;
731 	uint64_t *mac_stats;
732 	unsigned int nb_supported = 0;
733 	unsigned int nb_written = 0;
734 	unsigned int i;
735 	int ret;
736 	int rc;
737 
738 	if (unlikely(values == NULL) ||
739 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
740 		return port->mac_stats_nb_supported;
741 
742 	rte_spinlock_lock(&port->mac_stats_lock);
743 
744 	rc = sfc_port_update_mac_stats(sa);
745 	if (rc != 0) {
746 		SFC_ASSERT(rc > 0);
747 		ret = -rc;
748 		goto unlock;
749 	}
750 
751 	mac_stats = port->mac_stats_buf;
752 
753 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
754 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
755 			continue;
756 
757 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
758 			values[nb_written++] = mac_stats[i];
759 
760 		++nb_supported;
761 	}
762 
763 	ret = nb_written;
764 
765 unlock:
766 	rte_spinlock_unlock(&port->mac_stats_lock);
767 
768 	return ret;
769 }
770 
771 static int
772 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
773 			   struct rte_eth_xstat_name *xstats_names,
774 			   const uint64_t *ids, unsigned int size)
775 {
776 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
777 	struct sfc_port *port = &sa->port;
778 	unsigned int nb_supported = 0;
779 	unsigned int nb_written = 0;
780 	unsigned int i;
781 
782 	if (unlikely(xstats_names == NULL) ||
783 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
784 		return port->mac_stats_nb_supported;
785 
786 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
787 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
788 			continue;
789 
790 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
791 			char *name = xstats_names[nb_written++].name;
792 
793 			strlcpy(name, efx_mac_stat_name(sa->nic, i),
794 				sizeof(xstats_names[0].name));
795 		}
796 
797 		++nb_supported;
798 	}
799 
800 	return nb_written;
801 }
802 
803 static int
804 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
805 {
806 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
807 	unsigned int wanted_fc, link_fc;
808 
809 	memset(fc_conf, 0, sizeof(*fc_conf));
810 
811 	sfc_adapter_lock(sa);
812 
813 	if (sa->state == SFC_ADAPTER_STARTED)
814 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
815 	else
816 		link_fc = sa->port.flow_ctrl;
817 
818 	switch (link_fc) {
819 	case 0:
820 		fc_conf->mode = RTE_FC_NONE;
821 		break;
822 	case EFX_FCNTL_RESPOND:
823 		fc_conf->mode = RTE_FC_RX_PAUSE;
824 		break;
825 	case EFX_FCNTL_GENERATE:
826 		fc_conf->mode = RTE_FC_TX_PAUSE;
827 		break;
828 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
829 		fc_conf->mode = RTE_FC_FULL;
830 		break;
831 	default:
832 		sfc_err(sa, "%s: unexpected flow control value %#x",
833 			__func__, link_fc);
834 	}
835 
836 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
837 
838 	sfc_adapter_unlock(sa);
839 
840 	return 0;
841 }
842 
843 static int
844 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
845 {
846 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
847 	struct sfc_port *port = &sa->port;
848 	unsigned int fcntl;
849 	int rc;
850 
851 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
852 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
853 	    fc_conf->mac_ctrl_frame_fwd != 0) {
854 		sfc_err(sa, "unsupported flow control settings specified");
855 		rc = EINVAL;
856 		goto fail_inval;
857 	}
858 
859 	switch (fc_conf->mode) {
860 	case RTE_FC_NONE:
861 		fcntl = 0;
862 		break;
863 	case RTE_FC_RX_PAUSE:
864 		fcntl = EFX_FCNTL_RESPOND;
865 		break;
866 	case RTE_FC_TX_PAUSE:
867 		fcntl = EFX_FCNTL_GENERATE;
868 		break;
869 	case RTE_FC_FULL:
870 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
871 		break;
872 	default:
873 		rc = EINVAL;
874 		goto fail_inval;
875 	}
876 
877 	sfc_adapter_lock(sa);
878 
879 	if (sa->state == SFC_ADAPTER_STARTED) {
880 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
881 		if (rc != 0)
882 			goto fail_mac_fcntl_set;
883 	}
884 
885 	port->flow_ctrl = fcntl;
886 	port->flow_ctrl_autoneg = fc_conf->autoneg;
887 
888 	sfc_adapter_unlock(sa);
889 
890 	return 0;
891 
892 fail_mac_fcntl_set:
893 	sfc_adapter_unlock(sa);
894 fail_inval:
895 	SFC_ASSERT(rc > 0);
896 	return -rc;
897 }
898 
899 static int
900 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
901 {
902 	struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
903 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
904 	boolean_t scatter_enabled;
905 	const char *error;
906 	unsigned int i;
907 
908 	for (i = 0; i < sas->rxq_count; i++) {
909 		if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
910 			continue;
911 
912 		scatter_enabled = (sas->rxq_info[i].type_flags &
913 				   EFX_RXQ_FLAG_SCATTER);
914 
915 		if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
916 					  encp->enc_rx_prefix_size,
917 					  scatter_enabled, &error)) {
918 			sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
919 				error);
920 			return EINVAL;
921 		}
922 	}
923 
924 	return 0;
925 }
926 
927 static int
928 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
929 {
930 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
931 	size_t pdu = EFX_MAC_PDU(mtu);
932 	size_t old_pdu;
933 	int rc;
934 
935 	sfc_log_init(sa, "mtu=%u", mtu);
936 
937 	rc = EINVAL;
938 	if (pdu < EFX_MAC_PDU_MIN) {
939 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
940 			(unsigned int)mtu, (unsigned int)pdu,
941 			EFX_MAC_PDU_MIN);
942 		goto fail_inval;
943 	}
944 	if (pdu > EFX_MAC_PDU_MAX) {
945 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
946 			(unsigned int)mtu, (unsigned int)pdu,
947 			(unsigned int)EFX_MAC_PDU_MAX);
948 		goto fail_inval;
949 	}
950 
951 	sfc_adapter_lock(sa);
952 
953 	rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
954 	if (rc != 0)
955 		goto fail_check_scatter;
956 
957 	if (pdu != sa->port.pdu) {
958 		if (sa->state == SFC_ADAPTER_STARTED) {
959 			sfc_stop(sa);
960 
961 			old_pdu = sa->port.pdu;
962 			sa->port.pdu = pdu;
963 			rc = sfc_start(sa);
964 			if (rc != 0)
965 				goto fail_start;
966 		} else {
967 			sa->port.pdu = pdu;
968 		}
969 	}
970 
971 	/*
972 	 * The driver does not use it, but other PMDs update jumbo frame
973 	 * flag and max_rx_pkt_len when MTU is set.
974 	 */
975 	if (mtu > RTE_ETHER_MAX_LEN) {
976 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
977 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
978 	}
979 
980 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
981 
982 	sfc_adapter_unlock(sa);
983 
984 	sfc_log_init(sa, "done");
985 	return 0;
986 
987 fail_start:
988 	sa->port.pdu = old_pdu;
989 	if (sfc_start(sa) != 0)
990 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
991 			"PDU max size - port is stopped",
992 			(unsigned int)pdu, (unsigned int)old_pdu);
993 
994 fail_check_scatter:
995 	sfc_adapter_unlock(sa);
996 
997 fail_inval:
998 	sfc_log_init(sa, "failed %d", rc);
999 	SFC_ASSERT(rc > 0);
1000 	return -rc;
1001 }
1002 static int
1003 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1004 {
1005 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1006 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1007 	struct sfc_port *port = &sa->port;
1008 	struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1009 	int rc = 0;
1010 
1011 	sfc_adapter_lock(sa);
1012 
1013 	/*
1014 	 * Copy the address to the device private data so that
1015 	 * it could be recalled in the case of adapter restart.
1016 	 */
1017 	rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1018 
1019 	/*
1020 	 * Neither of the two following checks can return
1021 	 * an error. The new MAC address is preserved in
1022 	 * the device private data and can be activated
1023 	 * on the next port start if the user prevents
1024 	 * isolated mode from being enabled.
1025 	 */
1026 	if (sfc_sa2shared(sa)->isolated) {
1027 		sfc_warn(sa, "isolated mode is active on the port");
1028 		sfc_warn(sa, "will not set MAC address");
1029 		goto unlock;
1030 	}
1031 
1032 	if (sa->state != SFC_ADAPTER_STARTED) {
1033 		sfc_notice(sa, "the port is not started");
1034 		sfc_notice(sa, "the new MAC address will be set on port start");
1035 
1036 		goto unlock;
1037 	}
1038 
1039 	if (encp->enc_allow_set_mac_with_installed_filters) {
1040 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1041 		if (rc != 0) {
1042 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1043 			goto unlock;
1044 		}
1045 
1046 		/*
1047 		 * Changing the MAC address by means of MCDI request
1048 		 * has no effect on received traffic, therefore
1049 		 * we also need to update unicast filters
1050 		 */
1051 		rc = sfc_set_rx_mode(sa);
1052 		if (rc != 0) {
1053 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
1054 			/* Rollback the old address */
1055 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1056 			(void)sfc_set_rx_mode(sa);
1057 		}
1058 	} else {
1059 		sfc_warn(sa, "cannot set MAC address with filters installed");
1060 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1061 		sfc_warn(sa, "(some traffic may be dropped)");
1062 
1063 		/*
1064 		 * Since setting MAC address with filters installed is not
1065 		 * allowed on the adapter, the new MAC address will be set
1066 		 * by means of adapter restart. sfc_start() shall retrieve
1067 		 * the new address from the device private data and set it.
1068 		 */
1069 		sfc_stop(sa);
1070 		rc = sfc_start(sa);
1071 		if (rc != 0)
1072 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1073 	}
1074 
1075 unlock:
1076 	if (rc != 0)
1077 		rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1078 
1079 	sfc_adapter_unlock(sa);
1080 
1081 	SFC_ASSERT(rc >= 0);
1082 	return -rc;
1083 }
1084 
1085 
1086 static int
1087 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1088 		struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1089 {
1090 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1091 	struct sfc_port *port = &sa->port;
1092 	uint8_t *mc_addrs = port->mcast_addrs;
1093 	int rc;
1094 	unsigned int i;
1095 
1096 	if (sfc_sa2shared(sa)->isolated) {
1097 		sfc_err(sa, "isolated mode is active on the port");
1098 		sfc_err(sa, "will not set multicast address list");
1099 		return -ENOTSUP;
1100 	}
1101 
1102 	if (mc_addrs == NULL)
1103 		return -ENOBUFS;
1104 
1105 	if (nb_mc_addr > port->max_mcast_addrs) {
1106 		sfc_err(sa, "too many multicast addresses: %u > %u",
1107 			 nb_mc_addr, port->max_mcast_addrs);
1108 		return -EINVAL;
1109 	}
1110 
1111 	for (i = 0; i < nb_mc_addr; ++i) {
1112 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1113 				 EFX_MAC_ADDR_LEN);
1114 		mc_addrs += EFX_MAC_ADDR_LEN;
1115 	}
1116 
1117 	port->nb_mcast_addrs = nb_mc_addr;
1118 
1119 	if (sa->state != SFC_ADAPTER_STARTED)
1120 		return 0;
1121 
1122 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1123 					port->nb_mcast_addrs);
1124 	if (rc != 0)
1125 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1126 
1127 	SFC_ASSERT(rc >= 0);
1128 	return -rc;
1129 }
1130 
1131 /*
1132  * The function is used by the secondary process as well. It must not
1133  * use any process-local pointers from the adapter data.
1134  */
1135 static void
1136 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1137 		      struct rte_eth_rxq_info *qinfo)
1138 {
1139 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1140 	struct sfc_rxq_info *rxq_info;
1141 
1142 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1143 
1144 	rxq_info = &sas->rxq_info[rx_queue_id];
1145 
1146 	qinfo->mp = rxq_info->refill_mb_pool;
1147 	qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1148 	qinfo->conf.rx_drop_en = 1;
1149 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1150 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1151 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1152 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1153 		qinfo->scattered_rx = 1;
1154 	}
1155 	qinfo->nb_desc = rxq_info->entries;
1156 }
1157 
1158 /*
1159  * The function is used by the secondary process as well. It must not
1160  * use any process-local pointers from the adapter data.
1161  */
1162 static void
1163 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1164 		      struct rte_eth_txq_info *qinfo)
1165 {
1166 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1167 	struct sfc_txq_info *txq_info;
1168 
1169 	SFC_ASSERT(tx_queue_id < sas->txq_count);
1170 
1171 	txq_info = &sas->txq_info[tx_queue_id];
1172 
1173 	memset(qinfo, 0, sizeof(*qinfo));
1174 
1175 	qinfo->conf.offloads = txq_info->offloads;
1176 	qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1177 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1178 	qinfo->nb_desc = txq_info->entries;
1179 }
1180 
1181 /*
1182  * The function is used by the secondary process as well. It must not
1183  * use any process-local pointers from the adapter data.
1184  */
1185 static uint32_t
1186 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1187 {
1188 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1189 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1190 	struct sfc_rxq_info *rxq_info;
1191 
1192 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1193 	rxq_info = &sas->rxq_info[rx_queue_id];
1194 
1195 	if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1196 		return 0;
1197 
1198 	return sap->dp_rx->qdesc_npending(rxq_info->dp);
1199 }
1200 
1201 /*
1202  * The function is used by the secondary process as well. It must not
1203  * use any process-local pointers from the adapter data.
1204  */
1205 static int
1206 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1207 {
1208 	struct sfc_dp_rxq *dp_rxq = queue;
1209 	const struct sfc_dp_rx *dp_rx;
1210 
1211 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1212 
1213 	return offset < dp_rx->qdesc_npending(dp_rxq);
1214 }
1215 
1216 /*
1217  * The function is used by the secondary process as well. It must not
1218  * use any process-local pointers from the adapter data.
1219  */
1220 static int
1221 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1222 {
1223 	struct sfc_dp_rxq *dp_rxq = queue;
1224 	const struct sfc_dp_rx *dp_rx;
1225 
1226 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1227 
1228 	return dp_rx->qdesc_status(dp_rxq, offset);
1229 }
1230 
1231 /*
1232  * The function is used by the secondary process as well. It must not
1233  * use any process-local pointers from the adapter data.
1234  */
1235 static int
1236 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1237 {
1238 	struct sfc_dp_txq *dp_txq = queue;
1239 	const struct sfc_dp_tx *dp_tx;
1240 
1241 	dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1242 
1243 	return dp_tx->qdesc_status(dp_txq, offset);
1244 }
1245 
1246 static int
1247 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1248 {
1249 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1250 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1251 	int rc;
1252 
1253 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1254 
1255 	sfc_adapter_lock(sa);
1256 
1257 	rc = EINVAL;
1258 	if (sa->state != SFC_ADAPTER_STARTED)
1259 		goto fail_not_started;
1260 
1261 	if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1262 		goto fail_not_setup;
1263 
1264 	rc = sfc_rx_qstart(sa, rx_queue_id);
1265 	if (rc != 0)
1266 		goto fail_rx_qstart;
1267 
1268 	sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1269 
1270 	sfc_adapter_unlock(sa);
1271 
1272 	return 0;
1273 
1274 fail_rx_qstart:
1275 fail_not_setup:
1276 fail_not_started:
1277 	sfc_adapter_unlock(sa);
1278 	SFC_ASSERT(rc > 0);
1279 	return -rc;
1280 }
1281 
1282 static int
1283 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1284 {
1285 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1286 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1287 
1288 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1289 
1290 	sfc_adapter_lock(sa);
1291 	sfc_rx_qstop(sa, rx_queue_id);
1292 
1293 	sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1294 
1295 	sfc_adapter_unlock(sa);
1296 
1297 	return 0;
1298 }
1299 
1300 static int
1301 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1302 {
1303 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1304 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1305 	int rc;
1306 
1307 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1308 
1309 	sfc_adapter_lock(sa);
1310 
1311 	rc = EINVAL;
1312 	if (sa->state != SFC_ADAPTER_STARTED)
1313 		goto fail_not_started;
1314 
1315 	if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1316 		goto fail_not_setup;
1317 
1318 	rc = sfc_tx_qstart(sa, tx_queue_id);
1319 	if (rc != 0)
1320 		goto fail_tx_qstart;
1321 
1322 	sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1323 
1324 	sfc_adapter_unlock(sa);
1325 	return 0;
1326 
1327 fail_tx_qstart:
1328 
1329 fail_not_setup:
1330 fail_not_started:
1331 	sfc_adapter_unlock(sa);
1332 	SFC_ASSERT(rc > 0);
1333 	return -rc;
1334 }
1335 
1336 static int
1337 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1338 {
1339 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1340 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1341 
1342 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1343 
1344 	sfc_adapter_lock(sa);
1345 
1346 	sfc_tx_qstop(sa, tx_queue_id);
1347 
1348 	sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1349 
1350 	sfc_adapter_unlock(sa);
1351 	return 0;
1352 }
1353 
1354 static efx_tunnel_protocol_t
1355 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1356 {
1357 	switch (rte_type) {
1358 	case RTE_TUNNEL_TYPE_VXLAN:
1359 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1360 	case RTE_TUNNEL_TYPE_GENEVE:
1361 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1362 	default:
1363 		return EFX_TUNNEL_NPROTOS;
1364 	}
1365 }
1366 
1367 enum sfc_udp_tunnel_op_e {
1368 	SFC_UDP_TUNNEL_ADD_PORT,
1369 	SFC_UDP_TUNNEL_DEL_PORT,
1370 };
1371 
1372 static int
1373 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1374 		      struct rte_eth_udp_tunnel *tunnel_udp,
1375 		      enum sfc_udp_tunnel_op_e op)
1376 {
1377 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1378 	efx_tunnel_protocol_t tunnel_proto;
1379 	int rc;
1380 
1381 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1382 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1383 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1384 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1385 
1386 	tunnel_proto =
1387 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1388 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1389 		rc = ENOTSUP;
1390 		goto fail_bad_proto;
1391 	}
1392 
1393 	sfc_adapter_lock(sa);
1394 
1395 	switch (op) {
1396 	case SFC_UDP_TUNNEL_ADD_PORT:
1397 		rc = efx_tunnel_config_udp_add(sa->nic,
1398 					       tunnel_udp->udp_port,
1399 					       tunnel_proto);
1400 		break;
1401 	case SFC_UDP_TUNNEL_DEL_PORT:
1402 		rc = efx_tunnel_config_udp_remove(sa->nic,
1403 						  tunnel_udp->udp_port,
1404 						  tunnel_proto);
1405 		break;
1406 	default:
1407 		rc = EINVAL;
1408 		goto fail_bad_op;
1409 	}
1410 
1411 	if (rc != 0)
1412 		goto fail_op;
1413 
1414 	if (sa->state == SFC_ADAPTER_STARTED) {
1415 		rc = efx_tunnel_reconfigure(sa->nic);
1416 		if (rc == EAGAIN) {
1417 			/*
1418 			 * Configuration is accepted by FW and MC reboot
1419 			 * is initiated to apply the changes. MC reboot
1420 			 * will be handled in a usual way (MC reboot
1421 			 * event on management event queue and adapter
1422 			 * restart).
1423 			 */
1424 			rc = 0;
1425 		} else if (rc != 0) {
1426 			goto fail_reconfigure;
1427 		}
1428 	}
1429 
1430 	sfc_adapter_unlock(sa);
1431 	return 0;
1432 
1433 fail_reconfigure:
1434 	/* Remove/restore entry since the change makes the trouble */
1435 	switch (op) {
1436 	case SFC_UDP_TUNNEL_ADD_PORT:
1437 		(void)efx_tunnel_config_udp_remove(sa->nic,
1438 						   tunnel_udp->udp_port,
1439 						   tunnel_proto);
1440 		break;
1441 	case SFC_UDP_TUNNEL_DEL_PORT:
1442 		(void)efx_tunnel_config_udp_add(sa->nic,
1443 						tunnel_udp->udp_port,
1444 						tunnel_proto);
1445 		break;
1446 	}
1447 
1448 fail_op:
1449 fail_bad_op:
1450 	sfc_adapter_unlock(sa);
1451 
1452 fail_bad_proto:
1453 	SFC_ASSERT(rc > 0);
1454 	return -rc;
1455 }
1456 
1457 static int
1458 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1459 			    struct rte_eth_udp_tunnel *tunnel_udp)
1460 {
1461 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1462 }
1463 
1464 static int
1465 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1466 			    struct rte_eth_udp_tunnel *tunnel_udp)
1467 {
1468 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1469 }
1470 
1471 /*
1472  * The function is used by the secondary process as well. It must not
1473  * use any process-local pointers from the adapter data.
1474  */
1475 static int
1476 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1477 			  struct rte_eth_rss_conf *rss_conf)
1478 {
1479 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1480 	struct sfc_rss *rss = &sas->rss;
1481 
1482 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1483 		return -ENOTSUP;
1484 
1485 	/*
1486 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1487 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1488 	 * flags which corresponds to the active EFX configuration stored
1489 	 * locally in 'sfc_adapter' and kept up-to-date
1490 	 */
1491 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1492 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1493 	if (rss_conf->rss_key != NULL)
1494 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1495 
1496 	return 0;
1497 }
1498 
1499 static int
1500 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1501 			struct rte_eth_rss_conf *rss_conf)
1502 {
1503 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1504 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1505 	unsigned int efx_hash_types;
1506 	int rc = 0;
1507 
1508 	if (sfc_sa2shared(sa)->isolated)
1509 		return -ENOTSUP;
1510 
1511 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1512 		sfc_err(sa, "RSS is not available");
1513 		return -ENOTSUP;
1514 	}
1515 
1516 	if (rss->channels == 0) {
1517 		sfc_err(sa, "RSS is not configured");
1518 		return -EINVAL;
1519 	}
1520 
1521 	if ((rss_conf->rss_key != NULL) &&
1522 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1523 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1524 			sizeof(rss->key));
1525 		return -EINVAL;
1526 	}
1527 
1528 	sfc_adapter_lock(sa);
1529 
1530 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1531 	if (rc != 0)
1532 		goto fail_rx_hf_rte_to_efx;
1533 
1534 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1535 				   rss->hash_alg, efx_hash_types, B_TRUE);
1536 	if (rc != 0)
1537 		goto fail_scale_mode_set;
1538 
1539 	if (rss_conf->rss_key != NULL) {
1540 		if (sa->state == SFC_ADAPTER_STARTED) {
1541 			rc = efx_rx_scale_key_set(sa->nic,
1542 						  EFX_RSS_CONTEXT_DEFAULT,
1543 						  rss_conf->rss_key,
1544 						  sizeof(rss->key));
1545 			if (rc != 0)
1546 				goto fail_scale_key_set;
1547 		}
1548 
1549 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1550 	}
1551 
1552 	rss->hash_types = efx_hash_types;
1553 
1554 	sfc_adapter_unlock(sa);
1555 
1556 	return 0;
1557 
1558 fail_scale_key_set:
1559 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1560 				  EFX_RX_HASHALG_TOEPLITZ,
1561 				  rss->hash_types, B_TRUE) != 0)
1562 		sfc_err(sa, "failed to restore RSS mode");
1563 
1564 fail_scale_mode_set:
1565 fail_rx_hf_rte_to_efx:
1566 	sfc_adapter_unlock(sa);
1567 	return -rc;
1568 }
1569 
1570 /*
1571  * The function is used by the secondary process as well. It must not
1572  * use any process-local pointers from the adapter data.
1573  */
1574 static int
1575 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1576 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1577 		       uint16_t reta_size)
1578 {
1579 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1580 	struct sfc_rss *rss = &sas->rss;
1581 	int entry;
1582 
1583 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1584 		return -ENOTSUP;
1585 
1586 	if (rss->channels == 0)
1587 		return -EINVAL;
1588 
1589 	if (reta_size != EFX_RSS_TBL_SIZE)
1590 		return -EINVAL;
1591 
1592 	for (entry = 0; entry < reta_size; entry++) {
1593 		int grp = entry / RTE_RETA_GROUP_SIZE;
1594 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1595 
1596 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1597 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1598 	}
1599 
1600 	return 0;
1601 }
1602 
1603 static int
1604 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1605 			struct rte_eth_rss_reta_entry64 *reta_conf,
1606 			uint16_t reta_size)
1607 {
1608 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1609 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1610 	unsigned int *rss_tbl_new;
1611 	uint16_t entry;
1612 	int rc = 0;
1613 
1614 
1615 	if (sfc_sa2shared(sa)->isolated)
1616 		return -ENOTSUP;
1617 
1618 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1619 		sfc_err(sa, "RSS is not available");
1620 		return -ENOTSUP;
1621 	}
1622 
1623 	if (rss->channels == 0) {
1624 		sfc_err(sa, "RSS is not configured");
1625 		return -EINVAL;
1626 	}
1627 
1628 	if (reta_size != EFX_RSS_TBL_SIZE) {
1629 		sfc_err(sa, "RETA size is wrong (should be %u)",
1630 			EFX_RSS_TBL_SIZE);
1631 		return -EINVAL;
1632 	}
1633 
1634 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1635 	if (rss_tbl_new == NULL)
1636 		return -ENOMEM;
1637 
1638 	sfc_adapter_lock(sa);
1639 
1640 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1641 
1642 	for (entry = 0; entry < reta_size; entry++) {
1643 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1644 		struct rte_eth_rss_reta_entry64 *grp;
1645 
1646 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1647 
1648 		if (grp->mask & (1ull << grp_idx)) {
1649 			if (grp->reta[grp_idx] >= rss->channels) {
1650 				rc = EINVAL;
1651 				goto bad_reta_entry;
1652 			}
1653 			rss_tbl_new[entry] = grp->reta[grp_idx];
1654 		}
1655 	}
1656 
1657 	if (sa->state == SFC_ADAPTER_STARTED) {
1658 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1659 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1660 		if (rc != 0)
1661 			goto fail_scale_tbl_set;
1662 	}
1663 
1664 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1665 
1666 fail_scale_tbl_set:
1667 bad_reta_entry:
1668 	sfc_adapter_unlock(sa);
1669 
1670 	rte_free(rss_tbl_new);
1671 
1672 	SFC_ASSERT(rc >= 0);
1673 	return -rc;
1674 }
1675 
1676 static int
1677 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1678 		    enum rte_filter_op filter_op,
1679 		    void *arg)
1680 {
1681 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1682 	int rc = ENOTSUP;
1683 
1684 	sfc_log_init(sa, "entry");
1685 
1686 	switch (filter_type) {
1687 	case RTE_ETH_FILTER_NONE:
1688 		sfc_err(sa, "Global filters configuration not supported");
1689 		break;
1690 	case RTE_ETH_FILTER_MACVLAN:
1691 		sfc_err(sa, "MACVLAN filters not supported");
1692 		break;
1693 	case RTE_ETH_FILTER_ETHERTYPE:
1694 		sfc_err(sa, "EtherType filters not supported");
1695 		break;
1696 	case RTE_ETH_FILTER_FLEXIBLE:
1697 		sfc_err(sa, "Flexible filters not supported");
1698 		break;
1699 	case RTE_ETH_FILTER_SYN:
1700 		sfc_err(sa, "SYN filters not supported");
1701 		break;
1702 	case RTE_ETH_FILTER_NTUPLE:
1703 		sfc_err(sa, "NTUPLE filters not supported");
1704 		break;
1705 	case RTE_ETH_FILTER_TUNNEL:
1706 		sfc_err(sa, "Tunnel filters not supported");
1707 		break;
1708 	case RTE_ETH_FILTER_FDIR:
1709 		sfc_err(sa, "Flow Director filters not supported");
1710 		break;
1711 	case RTE_ETH_FILTER_HASH:
1712 		sfc_err(sa, "Hash filters not supported");
1713 		break;
1714 	case RTE_ETH_FILTER_GENERIC:
1715 		if (filter_op != RTE_ETH_FILTER_GET) {
1716 			rc = EINVAL;
1717 		} else {
1718 			*(const void **)arg = &sfc_flow_ops;
1719 			rc = 0;
1720 		}
1721 		break;
1722 	default:
1723 		sfc_err(sa, "Unknown filter type %u", filter_type);
1724 		break;
1725 	}
1726 
1727 	sfc_log_init(sa, "exit: %d", -rc);
1728 	SFC_ASSERT(rc >= 0);
1729 	return -rc;
1730 }
1731 
1732 static int
1733 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1734 {
1735 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1736 
1737 	/*
1738 	 * If Rx datapath does not provide callback to check mempool,
1739 	 * all pools are supported.
1740 	 */
1741 	if (sap->dp_rx->pool_ops_supported == NULL)
1742 		return 1;
1743 
1744 	return sap->dp_rx->pool_ops_supported(pool);
1745 }
1746 
1747 static int
1748 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1749 {
1750 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1751 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1752 	struct sfc_rxq_info *rxq_info;
1753 
1754 	SFC_ASSERT(queue_id < sas->rxq_count);
1755 	rxq_info = &sas->rxq_info[queue_id];
1756 
1757 	return sap->dp_rx->intr_enable(rxq_info->dp);
1758 }
1759 
1760 static int
1761 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1762 {
1763 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1764 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1765 	struct sfc_rxq_info *rxq_info;
1766 
1767 	SFC_ASSERT(queue_id < sas->rxq_count);
1768 	rxq_info = &sas->rxq_info[queue_id];
1769 
1770 	return sap->dp_rx->intr_disable(rxq_info->dp);
1771 }
1772 
1773 static const struct eth_dev_ops sfc_eth_dev_ops = {
1774 	.dev_configure			= sfc_dev_configure,
1775 	.dev_start			= sfc_dev_start,
1776 	.dev_stop			= sfc_dev_stop,
1777 	.dev_set_link_up		= sfc_dev_set_link_up,
1778 	.dev_set_link_down		= sfc_dev_set_link_down,
1779 	.dev_close			= sfc_dev_close,
1780 	.promiscuous_enable		= sfc_dev_promisc_enable,
1781 	.promiscuous_disable		= sfc_dev_promisc_disable,
1782 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1783 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1784 	.link_update			= sfc_dev_link_update,
1785 	.stats_get			= sfc_stats_get,
1786 	.stats_reset			= sfc_stats_reset,
1787 	.xstats_get			= sfc_xstats_get,
1788 	.xstats_reset			= sfc_stats_reset,
1789 	.xstats_get_names		= sfc_xstats_get_names,
1790 	.dev_infos_get			= sfc_dev_infos_get,
1791 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1792 	.mtu_set			= sfc_dev_set_mtu,
1793 	.rx_queue_start			= sfc_rx_queue_start,
1794 	.rx_queue_stop			= sfc_rx_queue_stop,
1795 	.tx_queue_start			= sfc_tx_queue_start,
1796 	.tx_queue_stop			= sfc_tx_queue_stop,
1797 	.rx_queue_setup			= sfc_rx_queue_setup,
1798 	.rx_queue_release		= sfc_rx_queue_release,
1799 	.rx_queue_count			= sfc_rx_queue_count,
1800 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1801 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1802 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1803 	.rx_queue_intr_enable		= sfc_rx_queue_intr_enable,
1804 	.rx_queue_intr_disable		= sfc_rx_queue_intr_disable,
1805 	.tx_queue_setup			= sfc_tx_queue_setup,
1806 	.tx_queue_release		= sfc_tx_queue_release,
1807 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1808 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1809 	.mac_addr_set			= sfc_mac_addr_set,
1810 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1811 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1812 	.reta_update			= sfc_dev_rss_reta_update,
1813 	.reta_query			= sfc_dev_rss_reta_query,
1814 	.rss_hash_update		= sfc_dev_rss_hash_update,
1815 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1816 	.filter_ctrl			= sfc_dev_filter_ctrl,
1817 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1818 	.rxq_info_get			= sfc_rx_queue_info_get,
1819 	.txq_info_get			= sfc_tx_queue_info_get,
1820 	.fw_version_get			= sfc_fw_version_get,
1821 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1822 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1823 	.pool_ops_supported		= sfc_pool_ops_supported,
1824 };
1825 
1826 /**
1827  * Duplicate a string in potentially shared memory required for
1828  * multi-process support.
1829  *
1830  * strdup() allocates from process-local heap/memory.
1831  */
1832 static char *
1833 sfc_strdup(const char *str)
1834 {
1835 	size_t size;
1836 	char *copy;
1837 
1838 	if (str == NULL)
1839 		return NULL;
1840 
1841 	size = strlen(str) + 1;
1842 	copy = rte_malloc(__func__, size, 0);
1843 	if (copy != NULL)
1844 		rte_memcpy(copy, str, size);
1845 
1846 	return copy;
1847 }
1848 
1849 static int
1850 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1851 {
1852 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1853 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1854 	const struct sfc_dp_rx *dp_rx;
1855 	const struct sfc_dp_tx *dp_tx;
1856 	const efx_nic_cfg_t *encp;
1857 	unsigned int avail_caps = 0;
1858 	const char *rx_name = NULL;
1859 	const char *tx_name = NULL;
1860 	int rc;
1861 
1862 	switch (sa->family) {
1863 	case EFX_FAMILY_HUNTINGTON:
1864 	case EFX_FAMILY_MEDFORD:
1865 	case EFX_FAMILY_MEDFORD2:
1866 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1867 		break;
1868 	default:
1869 		break;
1870 	}
1871 
1872 	encp = efx_nic_cfg_get(sa->nic);
1873 	if (encp->enc_rx_es_super_buffer_supported)
1874 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1875 
1876 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1877 				sfc_kvarg_string_handler, &rx_name);
1878 	if (rc != 0)
1879 		goto fail_kvarg_rx_datapath;
1880 
1881 	if (rx_name != NULL) {
1882 		dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1883 		if (dp_rx == NULL) {
1884 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1885 			rc = ENOENT;
1886 			goto fail_dp_rx;
1887 		}
1888 		if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1889 			sfc_err(sa,
1890 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1891 				rx_name);
1892 			rc = EINVAL;
1893 			goto fail_dp_rx_caps;
1894 		}
1895 	} else {
1896 		dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1897 		if (dp_rx == NULL) {
1898 			sfc_err(sa, "Rx datapath by caps %#x not found",
1899 				avail_caps);
1900 			rc = ENOENT;
1901 			goto fail_dp_rx;
1902 		}
1903 	}
1904 
1905 	sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1906 	if (sas->dp_rx_name == NULL) {
1907 		rc = ENOMEM;
1908 		goto fail_dp_rx_name;
1909 	}
1910 
1911 	sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1912 
1913 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1914 				sfc_kvarg_string_handler, &tx_name);
1915 	if (rc != 0)
1916 		goto fail_kvarg_tx_datapath;
1917 
1918 	if (tx_name != NULL) {
1919 		dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1920 		if (dp_tx == NULL) {
1921 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1922 			rc = ENOENT;
1923 			goto fail_dp_tx;
1924 		}
1925 		if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1926 			sfc_err(sa,
1927 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1928 				tx_name);
1929 			rc = EINVAL;
1930 			goto fail_dp_tx_caps;
1931 		}
1932 	} else {
1933 		dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1934 		if (dp_tx == NULL) {
1935 			sfc_err(sa, "Tx datapath by caps %#x not found",
1936 				avail_caps);
1937 			rc = ENOENT;
1938 			goto fail_dp_tx;
1939 		}
1940 	}
1941 
1942 	sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1943 	if (sas->dp_tx_name == NULL) {
1944 		rc = ENOMEM;
1945 		goto fail_dp_tx_name;
1946 	}
1947 
1948 	sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1949 
1950 	sa->priv.dp_rx = dp_rx;
1951 	sa->priv.dp_tx = dp_tx;
1952 
1953 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1954 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1955 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1956 
1957 	dev->dev_ops = &sfc_eth_dev_ops;
1958 
1959 	return 0;
1960 
1961 fail_dp_tx_name:
1962 fail_dp_tx_caps:
1963 fail_dp_tx:
1964 fail_kvarg_tx_datapath:
1965 	rte_free(sas->dp_rx_name);
1966 	sas->dp_rx_name = NULL;
1967 
1968 fail_dp_rx_name:
1969 fail_dp_rx_caps:
1970 fail_dp_rx:
1971 fail_kvarg_rx_datapath:
1972 	return rc;
1973 }
1974 
1975 static void
1976 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1977 {
1978 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1979 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1980 
1981 	dev->dev_ops = NULL;
1982 	dev->tx_pkt_prepare = NULL;
1983 	dev->rx_pkt_burst = NULL;
1984 	dev->tx_pkt_burst = NULL;
1985 
1986 	rte_free(sas->dp_tx_name);
1987 	sas->dp_tx_name = NULL;
1988 	sa->priv.dp_tx = NULL;
1989 
1990 	rte_free(sas->dp_rx_name);
1991 	sas->dp_rx_name = NULL;
1992 	sa->priv.dp_rx = NULL;
1993 }
1994 
1995 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1996 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1997 	.rx_queue_count			= sfc_rx_queue_count,
1998 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1999 	.rx_descriptor_status		= sfc_rx_descriptor_status,
2000 	.tx_descriptor_status		= sfc_tx_descriptor_status,
2001 	.reta_query			= sfc_dev_rss_reta_query,
2002 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
2003 	.rxq_info_get			= sfc_rx_queue_info_get,
2004 	.txq_info_get			= sfc_tx_queue_info_get,
2005 };
2006 
2007 static int
2008 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2009 {
2010 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2011 	struct sfc_adapter_priv *sap;
2012 	const struct sfc_dp_rx *dp_rx;
2013 	const struct sfc_dp_tx *dp_tx;
2014 	int rc;
2015 
2016 	/*
2017 	 * Allocate process private data from heap, since it should not
2018 	 * be located in shared memory allocated using rte_malloc() API.
2019 	 */
2020 	sap = calloc(1, sizeof(*sap));
2021 	if (sap == NULL) {
2022 		rc = ENOMEM;
2023 		goto fail_alloc_priv;
2024 	}
2025 
2026 	sap->logtype_main = logtype_main;
2027 
2028 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2029 	if (dp_rx == NULL) {
2030 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2031 			"cannot find %s Rx datapath", sas->dp_rx_name);
2032 		rc = ENOENT;
2033 		goto fail_dp_rx;
2034 	}
2035 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2036 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2037 			"%s Rx datapath does not support multi-process",
2038 			sas->dp_rx_name);
2039 		rc = EINVAL;
2040 		goto fail_dp_rx_multi_process;
2041 	}
2042 
2043 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2044 	if (dp_tx == NULL) {
2045 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2046 			"cannot find %s Tx datapath", sas->dp_tx_name);
2047 		rc = ENOENT;
2048 		goto fail_dp_tx;
2049 	}
2050 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2051 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2052 			"%s Tx datapath does not support multi-process",
2053 			sas->dp_tx_name);
2054 		rc = EINVAL;
2055 		goto fail_dp_tx_multi_process;
2056 	}
2057 
2058 	sap->dp_rx = dp_rx;
2059 	sap->dp_tx = dp_tx;
2060 
2061 	dev->process_private = sap;
2062 	dev->rx_pkt_burst = dp_rx->pkt_burst;
2063 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2064 	dev->tx_pkt_burst = dp_tx->pkt_burst;
2065 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
2066 
2067 	return 0;
2068 
2069 fail_dp_tx_multi_process:
2070 fail_dp_tx:
2071 fail_dp_rx_multi_process:
2072 fail_dp_rx:
2073 	free(sap);
2074 
2075 fail_alloc_priv:
2076 	return rc;
2077 }
2078 
2079 static void
2080 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
2081 {
2082 	free(dev->process_private);
2083 	dev->process_private = NULL;
2084 	dev->dev_ops = NULL;
2085 	dev->tx_pkt_prepare = NULL;
2086 	dev->tx_pkt_burst = NULL;
2087 	dev->rx_pkt_burst = NULL;
2088 }
2089 
2090 static void
2091 sfc_register_dp(void)
2092 {
2093 	/* Register once */
2094 	if (TAILQ_EMPTY(&sfc_dp_head)) {
2095 		/* Prefer EF10 datapath */
2096 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2097 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2098 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2099 
2100 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2101 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2102 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2103 	}
2104 }
2105 
2106 static int
2107 sfc_eth_dev_init(struct rte_eth_dev *dev)
2108 {
2109 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2110 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2111 	uint32_t logtype_main;
2112 	struct sfc_adapter *sa;
2113 	int rc;
2114 	const efx_nic_cfg_t *encp;
2115 	const struct rte_ether_addr *from;
2116 
2117 	sfc_register_dp();
2118 
2119 	logtype_main = sfc_register_logtype(&pci_dev->addr,
2120 					    SFC_LOGTYPE_MAIN_STR,
2121 					    RTE_LOG_NOTICE);
2122 
2123 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2124 		return -sfc_eth_dev_secondary_init(dev, logtype_main);
2125 
2126 	/* Required for logging */
2127 	sas->pci_addr = pci_dev->addr;
2128 	sas->port_id = dev->data->port_id;
2129 
2130 	/*
2131 	 * Allocate process private data from heap, since it should not
2132 	 * be located in shared memory allocated using rte_malloc() API.
2133 	 */
2134 	sa = calloc(1, sizeof(*sa));
2135 	if (sa == NULL) {
2136 		rc = ENOMEM;
2137 		goto fail_alloc_sa;
2138 	}
2139 
2140 	dev->process_private = sa;
2141 
2142 	/* Required for logging */
2143 	sa->priv.shared = sas;
2144 	sa->priv.logtype_main = logtype_main;
2145 
2146 	sa->eth_dev = dev;
2147 
2148 	/* Copy PCI device info to the dev->data */
2149 	rte_eth_copy_pci_info(dev, pci_dev);
2150 
2151 	rc = sfc_kvargs_parse(sa);
2152 	if (rc != 0)
2153 		goto fail_kvargs_parse;
2154 
2155 	sfc_log_init(sa, "entry");
2156 
2157 	dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
2158 
2159 	dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2160 	if (dev->data->mac_addrs == NULL) {
2161 		rc = ENOMEM;
2162 		goto fail_mac_addrs;
2163 	}
2164 
2165 	sfc_adapter_lock_init(sa);
2166 	sfc_adapter_lock(sa);
2167 
2168 	sfc_log_init(sa, "probing");
2169 	rc = sfc_probe(sa);
2170 	if (rc != 0)
2171 		goto fail_probe;
2172 
2173 	sfc_log_init(sa, "set device ops");
2174 	rc = sfc_eth_dev_set_ops(dev);
2175 	if (rc != 0)
2176 		goto fail_set_ops;
2177 
2178 	sfc_log_init(sa, "attaching");
2179 	rc = sfc_attach(sa);
2180 	if (rc != 0)
2181 		goto fail_attach;
2182 
2183 	encp = efx_nic_cfg_get(sa->nic);
2184 
2185 	/*
2186 	 * The arguments are really reverse order in comparison to
2187 	 * Linux kernel. Copy from NIC config to Ethernet device data.
2188 	 */
2189 	from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2190 	rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2191 
2192 	sfc_adapter_unlock(sa);
2193 
2194 	sfc_log_init(sa, "done");
2195 	return 0;
2196 
2197 fail_attach:
2198 	sfc_eth_dev_clear_ops(dev);
2199 
2200 fail_set_ops:
2201 	sfc_unprobe(sa);
2202 
2203 fail_probe:
2204 	sfc_adapter_unlock(sa);
2205 	sfc_adapter_lock_fini(sa);
2206 	rte_free(dev->data->mac_addrs);
2207 	dev->data->mac_addrs = NULL;
2208 
2209 fail_mac_addrs:
2210 	sfc_kvargs_cleanup(sa);
2211 
2212 fail_kvargs_parse:
2213 	sfc_log_init(sa, "failed %d", rc);
2214 	dev->process_private = NULL;
2215 	free(sa);
2216 
2217 fail_alloc_sa:
2218 	SFC_ASSERT(rc > 0);
2219 	return -rc;
2220 }
2221 
2222 static int
2223 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2224 {
2225 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2226 		sfc_eth_dev_secondary_clear_ops(dev);
2227 		return 0;
2228 	}
2229 
2230 	sfc_dev_close(dev);
2231 
2232 	return 0;
2233 }
2234 
2235 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2236 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2237 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2238 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2239 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2240 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2241 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2242 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2243 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2244 	{ .vendor_id = 0 /* sentinel */ }
2245 };
2246 
2247 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2248 	struct rte_pci_device *pci_dev)
2249 {
2250 	return rte_eth_dev_pci_generic_probe(pci_dev,
2251 		sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2252 }
2253 
2254 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2255 {
2256 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2257 }
2258 
2259 static struct rte_pci_driver sfc_efx_pmd = {
2260 	.id_table = pci_id_sfc_efx_map,
2261 	.drv_flags =
2262 		RTE_PCI_DRV_INTR_LSC |
2263 		RTE_PCI_DRV_NEED_MAPPING,
2264 	.probe = sfc_eth_dev_pci_probe,
2265 	.remove = sfc_eth_dev_pci_remove,
2266 };
2267 
2268 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2269 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2270 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2271 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2272 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2273 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2274 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2275 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2276 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2277 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2278 
2279 RTE_INIT(sfc_driver_register_logtype)
2280 {
2281 	int ret;
2282 
2283 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2284 						   RTE_LOG_NOTICE);
2285 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2286 }
2287