xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision db4e81351fb85ff623bd0438d1b5a8fb55fe9fee)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright(c) 2019-2020 Xilinx, Inc.
4  * Copyright(c) 2016-2019 Solarflare Communications Inc.
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, rc = %d",
397 				 ((enabled) ? "enable" : "disable"), desc, rc);
398 
399 			/*
400 			 * For promiscuous and all-multicast filters a
401 			 * permission failure should be reported as an
402 			 * unsupported filter.
403 			 */
404 			if (rc == EPERM)
405 				rc = ENOTSUP;
406 		}
407 	}
408 
409 	sfc_adapter_unlock(sa);
410 	return rc;
411 }
412 
413 static int
414 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
415 {
416 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
417 
418 	SFC_ASSERT(rc >= 0);
419 	return -rc;
420 }
421 
422 static int
423 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
424 {
425 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
426 
427 	SFC_ASSERT(rc >= 0);
428 	return -rc;
429 }
430 
431 static int
432 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
433 {
434 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
435 
436 	SFC_ASSERT(rc >= 0);
437 	return -rc;
438 }
439 
440 static int
441 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
442 {
443 	int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
444 
445 	SFC_ASSERT(rc >= 0);
446 	return -rc;
447 }
448 
449 static int
450 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
451 		   uint16_t nb_rx_desc, unsigned int socket_id,
452 		   const struct rte_eth_rxconf *rx_conf,
453 		   struct rte_mempool *mb_pool)
454 {
455 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
456 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
457 	int rc;
458 
459 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
460 		     rx_queue_id, nb_rx_desc, socket_id);
461 
462 	sfc_adapter_lock(sa);
463 
464 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
465 			  rx_conf, mb_pool);
466 	if (rc != 0)
467 		goto fail_rx_qinit;
468 
469 	dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
470 
471 	sfc_adapter_unlock(sa);
472 
473 	return 0;
474 
475 fail_rx_qinit:
476 	sfc_adapter_unlock(sa);
477 	SFC_ASSERT(rc > 0);
478 	return -rc;
479 }
480 
481 static void
482 sfc_rx_queue_release(void *queue)
483 {
484 	struct sfc_dp_rxq *dp_rxq = queue;
485 	struct sfc_rxq *rxq;
486 	struct sfc_adapter *sa;
487 	unsigned int sw_index;
488 
489 	if (dp_rxq == NULL)
490 		return;
491 
492 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
493 	sa = rxq->evq->sa;
494 	sfc_adapter_lock(sa);
495 
496 	sw_index = dp_rxq->dpq.queue_id;
497 
498 	sfc_log_init(sa, "RxQ=%u", sw_index);
499 
500 	sfc_rx_qfini(sa, sw_index);
501 
502 	sfc_adapter_unlock(sa);
503 }
504 
505 static int
506 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
507 		   uint16_t nb_tx_desc, unsigned int socket_id,
508 		   const struct rte_eth_txconf *tx_conf)
509 {
510 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
511 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
512 	int rc;
513 
514 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
515 		     tx_queue_id, nb_tx_desc, socket_id);
516 
517 	sfc_adapter_lock(sa);
518 
519 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
520 	if (rc != 0)
521 		goto fail_tx_qinit;
522 
523 	dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
524 
525 	sfc_adapter_unlock(sa);
526 	return 0;
527 
528 fail_tx_qinit:
529 	sfc_adapter_unlock(sa);
530 	SFC_ASSERT(rc > 0);
531 	return -rc;
532 }
533 
534 static void
535 sfc_tx_queue_release(void *queue)
536 {
537 	struct sfc_dp_txq *dp_txq = queue;
538 	struct sfc_txq *txq;
539 	unsigned int sw_index;
540 	struct sfc_adapter *sa;
541 
542 	if (dp_txq == NULL)
543 		return;
544 
545 	txq = sfc_txq_by_dp_txq(dp_txq);
546 	sw_index = dp_txq->dpq.queue_id;
547 
548 	SFC_ASSERT(txq->evq != NULL);
549 	sa = txq->evq->sa;
550 
551 	sfc_log_init(sa, "TxQ = %u", sw_index);
552 
553 	sfc_adapter_lock(sa);
554 
555 	sfc_tx_qfini(sa, sw_index);
556 
557 	sfc_adapter_unlock(sa);
558 }
559 
560 /*
561  * Some statistics are computed as A - B where A and B each increase
562  * monotonically with some hardware counter(s) and the counters are read
563  * asynchronously.
564  *
565  * If packet X is counted in A, but not counted in B yet, computed value is
566  * greater than real.
567  *
568  * If packet X is not counted in A at the moment of reading the counter,
569  * but counted in B at the moment of reading the counter, computed value
570  * is less than real.
571  *
572  * However, counter which grows backward is worse evil than slightly wrong
573  * value. So, let's try to guarantee that it never happens except may be
574  * the case when the MAC stats are zeroed as a result of a NIC reset.
575  */
576 static void
577 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
578 {
579 	if ((int64_t)(newval - *stat) > 0 || newval == 0)
580 		*stat = newval;
581 }
582 
583 static int
584 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
585 {
586 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
587 	struct sfc_port *port = &sa->port;
588 	uint64_t *mac_stats;
589 	int ret;
590 
591 	rte_spinlock_lock(&port->mac_stats_lock);
592 
593 	ret = sfc_port_update_mac_stats(sa);
594 	if (ret != 0)
595 		goto unlock;
596 
597 	mac_stats = port->mac_stats_buf;
598 
599 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
600 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
601 		stats->ipackets =
602 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
603 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
604 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
605 		stats->opackets =
606 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
607 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
608 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
609 		stats->ibytes =
610 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
611 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
612 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
613 		stats->obytes =
614 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
615 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
616 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
617 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
618 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
619 	} else {
620 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
621 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
622 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
623 		/*
624 		 * Take into account stats which are whenever supported
625 		 * on EF10. If some stat is not supported by current
626 		 * firmware variant or HW revision, it is guaranteed
627 		 * to be zero in mac_stats.
628 		 */
629 		stats->imissed =
630 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
631 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
632 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
633 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
634 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
635 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
636 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
637 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
638 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
639 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
640 		stats->ierrors =
641 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
642 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
643 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
644 		/* no oerrors counters supported on EF10 */
645 
646 		/* Exclude missed, errors and pauses from Rx packets */
647 		sfc_update_diff_stat(&port->ipackets,
648 			mac_stats[EFX_MAC_RX_PKTS] -
649 			mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
650 			stats->imissed - stats->ierrors);
651 		stats->ipackets = port->ipackets;
652 	}
653 
654 unlock:
655 	rte_spinlock_unlock(&port->mac_stats_lock);
656 	SFC_ASSERT(ret >= 0);
657 	return -ret;
658 }
659 
660 static int
661 sfc_stats_reset(struct rte_eth_dev *dev)
662 {
663 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
664 	struct sfc_port *port = &sa->port;
665 	int rc;
666 
667 	if (sa->state != SFC_ADAPTER_STARTED) {
668 		/*
669 		 * The operation cannot be done if port is not started; it
670 		 * will be scheduled to be done during the next port start
671 		 */
672 		port->mac_stats_reset_pending = B_TRUE;
673 		return 0;
674 	}
675 
676 	rc = sfc_port_reset_mac_stats(sa);
677 	if (rc != 0)
678 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
679 
680 	SFC_ASSERT(rc >= 0);
681 	return -rc;
682 }
683 
684 static int
685 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
686 	       unsigned int xstats_count)
687 {
688 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
689 	struct sfc_port *port = &sa->port;
690 	uint64_t *mac_stats;
691 	int rc;
692 	unsigned int i;
693 	int nstats = 0;
694 
695 	rte_spinlock_lock(&port->mac_stats_lock);
696 
697 	rc = sfc_port_update_mac_stats(sa);
698 	if (rc != 0) {
699 		SFC_ASSERT(rc > 0);
700 		nstats = -rc;
701 		goto unlock;
702 	}
703 
704 	mac_stats = port->mac_stats_buf;
705 
706 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
707 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
708 			if (xstats != NULL && nstats < (int)xstats_count) {
709 				xstats[nstats].id = nstats;
710 				xstats[nstats].value = mac_stats[i];
711 			}
712 			nstats++;
713 		}
714 	}
715 
716 unlock:
717 	rte_spinlock_unlock(&port->mac_stats_lock);
718 
719 	return nstats;
720 }
721 
722 static int
723 sfc_xstats_get_names(struct rte_eth_dev *dev,
724 		     struct rte_eth_xstat_name *xstats_names,
725 		     unsigned int xstats_count)
726 {
727 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
728 	struct sfc_port *port = &sa->port;
729 	unsigned int i;
730 	unsigned int nstats = 0;
731 
732 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
733 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
734 			if (xstats_names != NULL && nstats < xstats_count)
735 				strlcpy(xstats_names[nstats].name,
736 					efx_mac_stat_name(sa->nic, i),
737 					sizeof(xstats_names[0].name));
738 			nstats++;
739 		}
740 	}
741 
742 	return nstats;
743 }
744 
745 static int
746 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
747 		     uint64_t *values, unsigned int n)
748 {
749 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
750 	struct sfc_port *port = &sa->port;
751 	uint64_t *mac_stats;
752 	unsigned int nb_supported = 0;
753 	unsigned int nb_written = 0;
754 	unsigned int i;
755 	int ret;
756 	int rc;
757 
758 	if (unlikely(values == NULL) ||
759 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
760 		return port->mac_stats_nb_supported;
761 
762 	rte_spinlock_lock(&port->mac_stats_lock);
763 
764 	rc = sfc_port_update_mac_stats(sa);
765 	if (rc != 0) {
766 		SFC_ASSERT(rc > 0);
767 		ret = -rc;
768 		goto unlock;
769 	}
770 
771 	mac_stats = port->mac_stats_buf;
772 
773 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
774 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
775 			continue;
776 
777 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
778 			values[nb_written++] = mac_stats[i];
779 
780 		++nb_supported;
781 	}
782 
783 	ret = nb_written;
784 
785 unlock:
786 	rte_spinlock_unlock(&port->mac_stats_lock);
787 
788 	return ret;
789 }
790 
791 static int
792 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
793 			   struct rte_eth_xstat_name *xstats_names,
794 			   const uint64_t *ids, unsigned int size)
795 {
796 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
797 	struct sfc_port *port = &sa->port;
798 	unsigned int nb_supported = 0;
799 	unsigned int nb_written = 0;
800 	unsigned int i;
801 
802 	if (unlikely(xstats_names == NULL) ||
803 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
804 		return port->mac_stats_nb_supported;
805 
806 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
807 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
808 			continue;
809 
810 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
811 			char *name = xstats_names[nb_written++].name;
812 
813 			strlcpy(name, efx_mac_stat_name(sa->nic, i),
814 				sizeof(xstats_names[0].name));
815 		}
816 
817 		++nb_supported;
818 	}
819 
820 	return nb_written;
821 }
822 
823 static int
824 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
825 {
826 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
827 	unsigned int wanted_fc, link_fc;
828 
829 	memset(fc_conf, 0, sizeof(*fc_conf));
830 
831 	sfc_adapter_lock(sa);
832 
833 	if (sa->state == SFC_ADAPTER_STARTED)
834 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
835 	else
836 		link_fc = sa->port.flow_ctrl;
837 
838 	switch (link_fc) {
839 	case 0:
840 		fc_conf->mode = RTE_FC_NONE;
841 		break;
842 	case EFX_FCNTL_RESPOND:
843 		fc_conf->mode = RTE_FC_RX_PAUSE;
844 		break;
845 	case EFX_FCNTL_GENERATE:
846 		fc_conf->mode = RTE_FC_TX_PAUSE;
847 		break;
848 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
849 		fc_conf->mode = RTE_FC_FULL;
850 		break;
851 	default:
852 		sfc_err(sa, "%s: unexpected flow control value %#x",
853 			__func__, link_fc);
854 	}
855 
856 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
857 
858 	sfc_adapter_unlock(sa);
859 
860 	return 0;
861 }
862 
863 static int
864 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
865 {
866 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
867 	struct sfc_port *port = &sa->port;
868 	unsigned int fcntl;
869 	int rc;
870 
871 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
872 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
873 	    fc_conf->mac_ctrl_frame_fwd != 0) {
874 		sfc_err(sa, "unsupported flow control settings specified");
875 		rc = EINVAL;
876 		goto fail_inval;
877 	}
878 
879 	switch (fc_conf->mode) {
880 	case RTE_FC_NONE:
881 		fcntl = 0;
882 		break;
883 	case RTE_FC_RX_PAUSE:
884 		fcntl = EFX_FCNTL_RESPOND;
885 		break;
886 	case RTE_FC_TX_PAUSE:
887 		fcntl = EFX_FCNTL_GENERATE;
888 		break;
889 	case RTE_FC_FULL:
890 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
891 		break;
892 	default:
893 		rc = EINVAL;
894 		goto fail_inval;
895 	}
896 
897 	sfc_adapter_lock(sa);
898 
899 	if (sa->state == SFC_ADAPTER_STARTED) {
900 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
901 		if (rc != 0)
902 			goto fail_mac_fcntl_set;
903 	}
904 
905 	port->flow_ctrl = fcntl;
906 	port->flow_ctrl_autoneg = fc_conf->autoneg;
907 
908 	sfc_adapter_unlock(sa);
909 
910 	return 0;
911 
912 fail_mac_fcntl_set:
913 	sfc_adapter_unlock(sa);
914 fail_inval:
915 	SFC_ASSERT(rc > 0);
916 	return -rc;
917 }
918 
919 static int
920 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
921 {
922 	struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
923 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
924 	boolean_t scatter_enabled;
925 	const char *error;
926 	unsigned int i;
927 
928 	for (i = 0; i < sas->rxq_count; i++) {
929 		if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
930 			continue;
931 
932 		scatter_enabled = (sas->rxq_info[i].type_flags &
933 				   EFX_RXQ_FLAG_SCATTER);
934 
935 		if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
936 					  encp->enc_rx_prefix_size,
937 					  scatter_enabled, &error)) {
938 			sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
939 				error);
940 			return EINVAL;
941 		}
942 	}
943 
944 	return 0;
945 }
946 
947 static int
948 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
949 {
950 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
951 	size_t pdu = EFX_MAC_PDU(mtu);
952 	size_t old_pdu;
953 	int rc;
954 
955 	sfc_log_init(sa, "mtu=%u", mtu);
956 
957 	rc = EINVAL;
958 	if (pdu < EFX_MAC_PDU_MIN) {
959 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
960 			(unsigned int)mtu, (unsigned int)pdu,
961 			EFX_MAC_PDU_MIN);
962 		goto fail_inval;
963 	}
964 	if (pdu > EFX_MAC_PDU_MAX) {
965 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
966 			(unsigned int)mtu, (unsigned int)pdu,
967 			(unsigned int)EFX_MAC_PDU_MAX);
968 		goto fail_inval;
969 	}
970 
971 	sfc_adapter_lock(sa);
972 
973 	rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
974 	if (rc != 0)
975 		goto fail_check_scatter;
976 
977 	if (pdu != sa->port.pdu) {
978 		if (sa->state == SFC_ADAPTER_STARTED) {
979 			sfc_stop(sa);
980 
981 			old_pdu = sa->port.pdu;
982 			sa->port.pdu = pdu;
983 			rc = sfc_start(sa);
984 			if (rc != 0)
985 				goto fail_start;
986 		} else {
987 			sa->port.pdu = pdu;
988 		}
989 	}
990 
991 	/*
992 	 * The driver does not use it, but other PMDs update jumbo frame
993 	 * flag and max_rx_pkt_len when MTU is set.
994 	 */
995 	if (mtu > RTE_ETHER_MAX_LEN) {
996 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
997 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
998 	}
999 
1000 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1001 
1002 	sfc_adapter_unlock(sa);
1003 
1004 	sfc_log_init(sa, "done");
1005 	return 0;
1006 
1007 fail_start:
1008 	sa->port.pdu = old_pdu;
1009 	if (sfc_start(sa) != 0)
1010 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1011 			"PDU max size - port is stopped",
1012 			(unsigned int)pdu, (unsigned int)old_pdu);
1013 
1014 fail_check_scatter:
1015 	sfc_adapter_unlock(sa);
1016 
1017 fail_inval:
1018 	sfc_log_init(sa, "failed %d", rc);
1019 	SFC_ASSERT(rc > 0);
1020 	return -rc;
1021 }
1022 static int
1023 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1024 {
1025 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1026 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1027 	struct sfc_port *port = &sa->port;
1028 	struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1029 	int rc = 0;
1030 
1031 	sfc_adapter_lock(sa);
1032 
1033 	if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1034 		goto unlock;
1035 
1036 	/*
1037 	 * Copy the address to the device private data so that
1038 	 * it could be recalled in the case of adapter restart.
1039 	 */
1040 	rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1041 
1042 	/*
1043 	 * Neither of the two following checks can return
1044 	 * an error. The new MAC address is preserved in
1045 	 * the device private data and can be activated
1046 	 * on the next port start if the user prevents
1047 	 * isolated mode from being enabled.
1048 	 */
1049 	if (sfc_sa2shared(sa)->isolated) {
1050 		sfc_warn(sa, "isolated mode is active on the port");
1051 		sfc_warn(sa, "will not set MAC address");
1052 		goto unlock;
1053 	}
1054 
1055 	if (sa->state != SFC_ADAPTER_STARTED) {
1056 		sfc_notice(sa, "the port is not started");
1057 		sfc_notice(sa, "the new MAC address will be set on port start");
1058 
1059 		goto unlock;
1060 	}
1061 
1062 	if (encp->enc_allow_set_mac_with_installed_filters) {
1063 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1064 		if (rc != 0) {
1065 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1066 			goto unlock;
1067 		}
1068 
1069 		/*
1070 		 * Changing the MAC address by means of MCDI request
1071 		 * has no effect on received traffic, therefore
1072 		 * we also need to update unicast filters
1073 		 */
1074 		rc = sfc_set_rx_mode_unchecked(sa);
1075 		if (rc != 0) {
1076 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
1077 			/* Rollback the old address */
1078 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1079 			(void)sfc_set_rx_mode_unchecked(sa);
1080 		}
1081 	} else {
1082 		sfc_warn(sa, "cannot set MAC address with filters installed");
1083 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1084 		sfc_warn(sa, "(some traffic may be dropped)");
1085 
1086 		/*
1087 		 * Since setting MAC address with filters installed is not
1088 		 * allowed on the adapter, the new MAC address will be set
1089 		 * by means of adapter restart. sfc_start() shall retrieve
1090 		 * the new address from the device private data and set it.
1091 		 */
1092 		sfc_stop(sa);
1093 		rc = sfc_start(sa);
1094 		if (rc != 0)
1095 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1096 	}
1097 
1098 unlock:
1099 	if (rc != 0)
1100 		rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1101 
1102 	sfc_adapter_unlock(sa);
1103 
1104 	SFC_ASSERT(rc >= 0);
1105 	return -rc;
1106 }
1107 
1108 
1109 static int
1110 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1111 		struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1112 {
1113 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1114 	struct sfc_port *port = &sa->port;
1115 	uint8_t *mc_addrs = port->mcast_addrs;
1116 	int rc;
1117 	unsigned int i;
1118 
1119 	if (sfc_sa2shared(sa)->isolated) {
1120 		sfc_err(sa, "isolated mode is active on the port");
1121 		sfc_err(sa, "will not set multicast address list");
1122 		return -ENOTSUP;
1123 	}
1124 
1125 	if (mc_addrs == NULL)
1126 		return -ENOBUFS;
1127 
1128 	if (nb_mc_addr > port->max_mcast_addrs) {
1129 		sfc_err(sa, "too many multicast addresses: %u > %u",
1130 			 nb_mc_addr, port->max_mcast_addrs);
1131 		return -EINVAL;
1132 	}
1133 
1134 	for (i = 0; i < nb_mc_addr; ++i) {
1135 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1136 				 EFX_MAC_ADDR_LEN);
1137 		mc_addrs += EFX_MAC_ADDR_LEN;
1138 	}
1139 
1140 	port->nb_mcast_addrs = nb_mc_addr;
1141 
1142 	if (sa->state != SFC_ADAPTER_STARTED)
1143 		return 0;
1144 
1145 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1146 					port->nb_mcast_addrs);
1147 	if (rc != 0)
1148 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1149 
1150 	SFC_ASSERT(rc >= 0);
1151 	return -rc;
1152 }
1153 
1154 /*
1155  * The function is used by the secondary process as well. It must not
1156  * use any process-local pointers from the adapter data.
1157  */
1158 static void
1159 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1160 		      struct rte_eth_rxq_info *qinfo)
1161 {
1162 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1163 	struct sfc_rxq_info *rxq_info;
1164 
1165 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1166 
1167 	rxq_info = &sas->rxq_info[rx_queue_id];
1168 
1169 	qinfo->mp = rxq_info->refill_mb_pool;
1170 	qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1171 	qinfo->conf.rx_drop_en = 1;
1172 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1173 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1174 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1175 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1176 		qinfo->scattered_rx = 1;
1177 	}
1178 	qinfo->nb_desc = rxq_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 void
1186 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1187 		      struct rte_eth_txq_info *qinfo)
1188 {
1189 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1190 	struct sfc_txq_info *txq_info;
1191 
1192 	SFC_ASSERT(tx_queue_id < sas->txq_count);
1193 
1194 	txq_info = &sas->txq_info[tx_queue_id];
1195 
1196 	memset(qinfo, 0, sizeof(*qinfo));
1197 
1198 	qinfo->conf.offloads = txq_info->offloads;
1199 	qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1200 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1201 	qinfo->nb_desc = txq_info->entries;
1202 }
1203 
1204 /*
1205  * The function is used by the secondary process as well. It must not
1206  * use any process-local pointers from the adapter data.
1207  */
1208 static uint32_t
1209 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1210 {
1211 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1212 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1213 	struct sfc_rxq_info *rxq_info;
1214 
1215 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1216 	rxq_info = &sas->rxq_info[rx_queue_id];
1217 
1218 	if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1219 		return 0;
1220 
1221 	return sap->dp_rx->qdesc_npending(rxq_info->dp);
1222 }
1223 
1224 /*
1225  * The function is used by the secondary process as well. It must not
1226  * use any process-local pointers from the adapter data.
1227  */
1228 static int
1229 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1230 {
1231 	struct sfc_dp_rxq *dp_rxq = queue;
1232 	const struct sfc_dp_rx *dp_rx;
1233 
1234 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1235 
1236 	return offset < dp_rx->qdesc_npending(dp_rxq);
1237 }
1238 
1239 /*
1240  * The function is used by the secondary process as well. It must not
1241  * use any process-local pointers from the adapter data.
1242  */
1243 static int
1244 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1245 {
1246 	struct sfc_dp_rxq *dp_rxq = queue;
1247 	const struct sfc_dp_rx *dp_rx;
1248 
1249 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1250 
1251 	return dp_rx->qdesc_status(dp_rxq, offset);
1252 }
1253 
1254 /*
1255  * The function is used by the secondary process as well. It must not
1256  * use any process-local pointers from the adapter data.
1257  */
1258 static int
1259 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1260 {
1261 	struct sfc_dp_txq *dp_txq = queue;
1262 	const struct sfc_dp_tx *dp_tx;
1263 
1264 	dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1265 
1266 	return dp_tx->qdesc_status(dp_txq, offset);
1267 }
1268 
1269 static int
1270 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1271 {
1272 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1273 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1274 	int rc;
1275 
1276 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1277 
1278 	sfc_adapter_lock(sa);
1279 
1280 	rc = EINVAL;
1281 	if (sa->state != SFC_ADAPTER_STARTED)
1282 		goto fail_not_started;
1283 
1284 	if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1285 		goto fail_not_setup;
1286 
1287 	rc = sfc_rx_qstart(sa, rx_queue_id);
1288 	if (rc != 0)
1289 		goto fail_rx_qstart;
1290 
1291 	sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1292 
1293 	sfc_adapter_unlock(sa);
1294 
1295 	return 0;
1296 
1297 fail_rx_qstart:
1298 fail_not_setup:
1299 fail_not_started:
1300 	sfc_adapter_unlock(sa);
1301 	SFC_ASSERT(rc > 0);
1302 	return -rc;
1303 }
1304 
1305 static int
1306 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1307 {
1308 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1309 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1310 
1311 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1312 
1313 	sfc_adapter_lock(sa);
1314 	sfc_rx_qstop(sa, rx_queue_id);
1315 
1316 	sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1317 
1318 	sfc_adapter_unlock(sa);
1319 
1320 	return 0;
1321 }
1322 
1323 static int
1324 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1325 {
1326 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1327 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1328 	int rc;
1329 
1330 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1331 
1332 	sfc_adapter_lock(sa);
1333 
1334 	rc = EINVAL;
1335 	if (sa->state != SFC_ADAPTER_STARTED)
1336 		goto fail_not_started;
1337 
1338 	if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1339 		goto fail_not_setup;
1340 
1341 	rc = sfc_tx_qstart(sa, tx_queue_id);
1342 	if (rc != 0)
1343 		goto fail_tx_qstart;
1344 
1345 	sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1346 
1347 	sfc_adapter_unlock(sa);
1348 	return 0;
1349 
1350 fail_tx_qstart:
1351 
1352 fail_not_setup:
1353 fail_not_started:
1354 	sfc_adapter_unlock(sa);
1355 	SFC_ASSERT(rc > 0);
1356 	return -rc;
1357 }
1358 
1359 static int
1360 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1361 {
1362 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1363 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1364 
1365 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1366 
1367 	sfc_adapter_lock(sa);
1368 
1369 	sfc_tx_qstop(sa, tx_queue_id);
1370 
1371 	sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1372 
1373 	sfc_adapter_unlock(sa);
1374 	return 0;
1375 }
1376 
1377 static efx_tunnel_protocol_t
1378 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1379 {
1380 	switch (rte_type) {
1381 	case RTE_TUNNEL_TYPE_VXLAN:
1382 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1383 	case RTE_TUNNEL_TYPE_GENEVE:
1384 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1385 	default:
1386 		return EFX_TUNNEL_NPROTOS;
1387 	}
1388 }
1389 
1390 enum sfc_udp_tunnel_op_e {
1391 	SFC_UDP_TUNNEL_ADD_PORT,
1392 	SFC_UDP_TUNNEL_DEL_PORT,
1393 };
1394 
1395 static int
1396 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1397 		      struct rte_eth_udp_tunnel *tunnel_udp,
1398 		      enum sfc_udp_tunnel_op_e op)
1399 {
1400 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1401 	efx_tunnel_protocol_t tunnel_proto;
1402 	int rc;
1403 
1404 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1405 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1406 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1407 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1408 
1409 	tunnel_proto =
1410 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1411 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1412 		rc = ENOTSUP;
1413 		goto fail_bad_proto;
1414 	}
1415 
1416 	sfc_adapter_lock(sa);
1417 
1418 	switch (op) {
1419 	case SFC_UDP_TUNNEL_ADD_PORT:
1420 		rc = efx_tunnel_config_udp_add(sa->nic,
1421 					       tunnel_udp->udp_port,
1422 					       tunnel_proto);
1423 		break;
1424 	case SFC_UDP_TUNNEL_DEL_PORT:
1425 		rc = efx_tunnel_config_udp_remove(sa->nic,
1426 						  tunnel_udp->udp_port,
1427 						  tunnel_proto);
1428 		break;
1429 	default:
1430 		rc = EINVAL;
1431 		goto fail_bad_op;
1432 	}
1433 
1434 	if (rc != 0)
1435 		goto fail_op;
1436 
1437 	if (sa->state == SFC_ADAPTER_STARTED) {
1438 		rc = efx_tunnel_reconfigure(sa->nic);
1439 		if (rc == EAGAIN) {
1440 			/*
1441 			 * Configuration is accepted by FW and MC reboot
1442 			 * is initiated to apply the changes. MC reboot
1443 			 * will be handled in a usual way (MC reboot
1444 			 * event on management event queue and adapter
1445 			 * restart).
1446 			 */
1447 			rc = 0;
1448 		} else if (rc != 0) {
1449 			goto fail_reconfigure;
1450 		}
1451 	}
1452 
1453 	sfc_adapter_unlock(sa);
1454 	return 0;
1455 
1456 fail_reconfigure:
1457 	/* Remove/restore entry since the change makes the trouble */
1458 	switch (op) {
1459 	case SFC_UDP_TUNNEL_ADD_PORT:
1460 		(void)efx_tunnel_config_udp_remove(sa->nic,
1461 						   tunnel_udp->udp_port,
1462 						   tunnel_proto);
1463 		break;
1464 	case SFC_UDP_TUNNEL_DEL_PORT:
1465 		(void)efx_tunnel_config_udp_add(sa->nic,
1466 						tunnel_udp->udp_port,
1467 						tunnel_proto);
1468 		break;
1469 	}
1470 
1471 fail_op:
1472 fail_bad_op:
1473 	sfc_adapter_unlock(sa);
1474 
1475 fail_bad_proto:
1476 	SFC_ASSERT(rc > 0);
1477 	return -rc;
1478 }
1479 
1480 static int
1481 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1482 			    struct rte_eth_udp_tunnel *tunnel_udp)
1483 {
1484 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1485 }
1486 
1487 static int
1488 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1489 			    struct rte_eth_udp_tunnel *tunnel_udp)
1490 {
1491 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1492 }
1493 
1494 /*
1495  * The function is used by the secondary process as well. It must not
1496  * use any process-local pointers from the adapter data.
1497  */
1498 static int
1499 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1500 			  struct rte_eth_rss_conf *rss_conf)
1501 {
1502 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1503 	struct sfc_rss *rss = &sas->rss;
1504 
1505 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1506 		return -ENOTSUP;
1507 
1508 	/*
1509 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1510 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1511 	 * flags which corresponds to the active EFX configuration stored
1512 	 * locally in 'sfc_adapter' and kept up-to-date
1513 	 */
1514 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1515 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1516 	if (rss_conf->rss_key != NULL)
1517 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1518 
1519 	return 0;
1520 }
1521 
1522 static int
1523 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1524 			struct rte_eth_rss_conf *rss_conf)
1525 {
1526 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1527 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1528 	unsigned int efx_hash_types;
1529 	int rc = 0;
1530 
1531 	if (sfc_sa2shared(sa)->isolated)
1532 		return -ENOTSUP;
1533 
1534 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1535 		sfc_err(sa, "RSS is not available");
1536 		return -ENOTSUP;
1537 	}
1538 
1539 	if (rss->channels == 0) {
1540 		sfc_err(sa, "RSS is not configured");
1541 		return -EINVAL;
1542 	}
1543 
1544 	if ((rss_conf->rss_key != NULL) &&
1545 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1546 		sfc_err(sa, "RSS key size is wrong (should be %zu)",
1547 			sizeof(rss->key));
1548 		return -EINVAL;
1549 	}
1550 
1551 	sfc_adapter_lock(sa);
1552 
1553 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1554 	if (rc != 0)
1555 		goto fail_rx_hf_rte_to_efx;
1556 
1557 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1558 				   rss->hash_alg, efx_hash_types, B_TRUE);
1559 	if (rc != 0)
1560 		goto fail_scale_mode_set;
1561 
1562 	if (rss_conf->rss_key != NULL) {
1563 		if (sa->state == SFC_ADAPTER_STARTED) {
1564 			rc = efx_rx_scale_key_set(sa->nic,
1565 						  EFX_RSS_CONTEXT_DEFAULT,
1566 						  rss_conf->rss_key,
1567 						  sizeof(rss->key));
1568 			if (rc != 0)
1569 				goto fail_scale_key_set;
1570 		}
1571 
1572 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1573 	}
1574 
1575 	rss->hash_types = efx_hash_types;
1576 
1577 	sfc_adapter_unlock(sa);
1578 
1579 	return 0;
1580 
1581 fail_scale_key_set:
1582 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1583 				  EFX_RX_HASHALG_TOEPLITZ,
1584 				  rss->hash_types, B_TRUE) != 0)
1585 		sfc_err(sa, "failed to restore RSS mode");
1586 
1587 fail_scale_mode_set:
1588 fail_rx_hf_rte_to_efx:
1589 	sfc_adapter_unlock(sa);
1590 	return -rc;
1591 }
1592 
1593 /*
1594  * The function is used by the secondary process as well. It must not
1595  * use any process-local pointers from the adapter data.
1596  */
1597 static int
1598 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1599 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1600 		       uint16_t reta_size)
1601 {
1602 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1603 	struct sfc_rss *rss = &sas->rss;
1604 	int entry;
1605 
1606 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1607 		return -ENOTSUP;
1608 
1609 	if (rss->channels == 0)
1610 		return -EINVAL;
1611 
1612 	if (reta_size != EFX_RSS_TBL_SIZE)
1613 		return -EINVAL;
1614 
1615 	for (entry = 0; entry < reta_size; entry++) {
1616 		int grp = entry / RTE_RETA_GROUP_SIZE;
1617 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1618 
1619 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1620 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1621 	}
1622 
1623 	return 0;
1624 }
1625 
1626 static int
1627 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1628 			struct rte_eth_rss_reta_entry64 *reta_conf,
1629 			uint16_t reta_size)
1630 {
1631 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1632 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1633 	unsigned int *rss_tbl_new;
1634 	uint16_t entry;
1635 	int rc = 0;
1636 
1637 
1638 	if (sfc_sa2shared(sa)->isolated)
1639 		return -ENOTSUP;
1640 
1641 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1642 		sfc_err(sa, "RSS is not available");
1643 		return -ENOTSUP;
1644 	}
1645 
1646 	if (rss->channels == 0) {
1647 		sfc_err(sa, "RSS is not configured");
1648 		return -EINVAL;
1649 	}
1650 
1651 	if (reta_size != EFX_RSS_TBL_SIZE) {
1652 		sfc_err(sa, "RETA size is wrong (should be %u)",
1653 			EFX_RSS_TBL_SIZE);
1654 		return -EINVAL;
1655 	}
1656 
1657 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1658 	if (rss_tbl_new == NULL)
1659 		return -ENOMEM;
1660 
1661 	sfc_adapter_lock(sa);
1662 
1663 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1664 
1665 	for (entry = 0; entry < reta_size; entry++) {
1666 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1667 		struct rte_eth_rss_reta_entry64 *grp;
1668 
1669 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1670 
1671 		if (grp->mask & (1ull << grp_idx)) {
1672 			if (grp->reta[grp_idx] >= rss->channels) {
1673 				rc = EINVAL;
1674 				goto bad_reta_entry;
1675 			}
1676 			rss_tbl_new[entry] = grp->reta[grp_idx];
1677 		}
1678 	}
1679 
1680 	if (sa->state == SFC_ADAPTER_STARTED) {
1681 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1682 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1683 		if (rc != 0)
1684 			goto fail_scale_tbl_set;
1685 	}
1686 
1687 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1688 
1689 fail_scale_tbl_set:
1690 bad_reta_entry:
1691 	sfc_adapter_unlock(sa);
1692 
1693 	rte_free(rss_tbl_new);
1694 
1695 	SFC_ASSERT(rc >= 0);
1696 	return -rc;
1697 }
1698 
1699 static int
1700 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1701 		    enum rte_filter_op filter_op,
1702 		    void *arg)
1703 {
1704 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1705 	int rc = ENOTSUP;
1706 
1707 	sfc_log_init(sa, "entry");
1708 
1709 	switch (filter_type) {
1710 	case RTE_ETH_FILTER_NONE:
1711 		sfc_err(sa, "Global filters configuration not supported");
1712 		break;
1713 	case RTE_ETH_FILTER_MACVLAN:
1714 		sfc_err(sa, "MACVLAN filters not supported");
1715 		break;
1716 	case RTE_ETH_FILTER_ETHERTYPE:
1717 		sfc_err(sa, "EtherType filters not supported");
1718 		break;
1719 	case RTE_ETH_FILTER_FLEXIBLE:
1720 		sfc_err(sa, "Flexible filters not supported");
1721 		break;
1722 	case RTE_ETH_FILTER_SYN:
1723 		sfc_err(sa, "SYN filters not supported");
1724 		break;
1725 	case RTE_ETH_FILTER_NTUPLE:
1726 		sfc_err(sa, "NTUPLE filters not supported");
1727 		break;
1728 	case RTE_ETH_FILTER_TUNNEL:
1729 		sfc_err(sa, "Tunnel filters not supported");
1730 		break;
1731 	case RTE_ETH_FILTER_FDIR:
1732 		sfc_err(sa, "Flow Director filters not supported");
1733 		break;
1734 	case RTE_ETH_FILTER_HASH:
1735 		sfc_err(sa, "Hash filters not supported");
1736 		break;
1737 	case RTE_ETH_FILTER_GENERIC:
1738 		if (filter_op != RTE_ETH_FILTER_GET) {
1739 			rc = EINVAL;
1740 		} else {
1741 			*(const void **)arg = &sfc_flow_ops;
1742 			rc = 0;
1743 		}
1744 		break;
1745 	default:
1746 		sfc_err(sa, "Unknown filter type %u", filter_type);
1747 		break;
1748 	}
1749 
1750 	sfc_log_init(sa, "exit: %d", -rc);
1751 	SFC_ASSERT(rc >= 0);
1752 	return -rc;
1753 }
1754 
1755 static int
1756 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1757 {
1758 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1759 
1760 	/*
1761 	 * If Rx datapath does not provide callback to check mempool,
1762 	 * all pools are supported.
1763 	 */
1764 	if (sap->dp_rx->pool_ops_supported == NULL)
1765 		return 1;
1766 
1767 	return sap->dp_rx->pool_ops_supported(pool);
1768 }
1769 
1770 static int
1771 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1772 {
1773 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1774 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1775 	struct sfc_rxq_info *rxq_info;
1776 
1777 	SFC_ASSERT(queue_id < sas->rxq_count);
1778 	rxq_info = &sas->rxq_info[queue_id];
1779 
1780 	return sap->dp_rx->intr_enable(rxq_info->dp);
1781 }
1782 
1783 static int
1784 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1785 {
1786 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1787 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1788 	struct sfc_rxq_info *rxq_info;
1789 
1790 	SFC_ASSERT(queue_id < sas->rxq_count);
1791 	rxq_info = &sas->rxq_info[queue_id];
1792 
1793 	return sap->dp_rx->intr_disable(rxq_info->dp);
1794 }
1795 
1796 static const struct eth_dev_ops sfc_eth_dev_ops = {
1797 	.dev_configure			= sfc_dev_configure,
1798 	.dev_start			= sfc_dev_start,
1799 	.dev_stop			= sfc_dev_stop,
1800 	.dev_set_link_up		= sfc_dev_set_link_up,
1801 	.dev_set_link_down		= sfc_dev_set_link_down,
1802 	.dev_close			= sfc_dev_close,
1803 	.promiscuous_enable		= sfc_dev_promisc_enable,
1804 	.promiscuous_disable		= sfc_dev_promisc_disable,
1805 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1806 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1807 	.link_update			= sfc_dev_link_update,
1808 	.stats_get			= sfc_stats_get,
1809 	.stats_reset			= sfc_stats_reset,
1810 	.xstats_get			= sfc_xstats_get,
1811 	.xstats_reset			= sfc_stats_reset,
1812 	.xstats_get_names		= sfc_xstats_get_names,
1813 	.dev_infos_get			= sfc_dev_infos_get,
1814 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1815 	.mtu_set			= sfc_dev_set_mtu,
1816 	.rx_queue_start			= sfc_rx_queue_start,
1817 	.rx_queue_stop			= sfc_rx_queue_stop,
1818 	.tx_queue_start			= sfc_tx_queue_start,
1819 	.tx_queue_stop			= sfc_tx_queue_stop,
1820 	.rx_queue_setup			= sfc_rx_queue_setup,
1821 	.rx_queue_release		= sfc_rx_queue_release,
1822 	.rx_queue_intr_enable		= sfc_rx_queue_intr_enable,
1823 	.rx_queue_intr_disable		= sfc_rx_queue_intr_disable,
1824 	.tx_queue_setup			= sfc_tx_queue_setup,
1825 	.tx_queue_release		= sfc_tx_queue_release,
1826 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1827 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1828 	.mac_addr_set			= sfc_mac_addr_set,
1829 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1830 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1831 	.reta_update			= sfc_dev_rss_reta_update,
1832 	.reta_query			= sfc_dev_rss_reta_query,
1833 	.rss_hash_update		= sfc_dev_rss_hash_update,
1834 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1835 	.filter_ctrl			= sfc_dev_filter_ctrl,
1836 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1837 	.rxq_info_get			= sfc_rx_queue_info_get,
1838 	.txq_info_get			= sfc_tx_queue_info_get,
1839 	.fw_version_get			= sfc_fw_version_get,
1840 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1841 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1842 	.pool_ops_supported		= sfc_pool_ops_supported,
1843 };
1844 
1845 /**
1846  * Duplicate a string in potentially shared memory required for
1847  * multi-process support.
1848  *
1849  * strdup() allocates from process-local heap/memory.
1850  */
1851 static char *
1852 sfc_strdup(const char *str)
1853 {
1854 	size_t size;
1855 	char *copy;
1856 
1857 	if (str == NULL)
1858 		return NULL;
1859 
1860 	size = strlen(str) + 1;
1861 	copy = rte_malloc(__func__, size, 0);
1862 	if (copy != NULL)
1863 		rte_memcpy(copy, str, size);
1864 
1865 	return copy;
1866 }
1867 
1868 static int
1869 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1870 {
1871 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1872 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1873 	const struct sfc_dp_rx *dp_rx;
1874 	const struct sfc_dp_tx *dp_tx;
1875 	const efx_nic_cfg_t *encp;
1876 	unsigned int avail_caps = 0;
1877 	const char *rx_name = NULL;
1878 	const char *tx_name = NULL;
1879 	int rc;
1880 
1881 	switch (sa->family) {
1882 	case EFX_FAMILY_HUNTINGTON:
1883 	case EFX_FAMILY_MEDFORD:
1884 	case EFX_FAMILY_MEDFORD2:
1885 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1886 		break;
1887 	default:
1888 		break;
1889 	}
1890 
1891 	encp = efx_nic_cfg_get(sa->nic);
1892 	if (encp->enc_rx_es_super_buffer_supported)
1893 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1894 
1895 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1896 				sfc_kvarg_string_handler, &rx_name);
1897 	if (rc != 0)
1898 		goto fail_kvarg_rx_datapath;
1899 
1900 	if (rx_name != NULL) {
1901 		dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1902 		if (dp_rx == NULL) {
1903 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1904 			rc = ENOENT;
1905 			goto fail_dp_rx;
1906 		}
1907 		if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1908 			sfc_err(sa,
1909 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1910 				rx_name);
1911 			rc = EINVAL;
1912 			goto fail_dp_rx_caps;
1913 		}
1914 	} else {
1915 		dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1916 		if (dp_rx == NULL) {
1917 			sfc_err(sa, "Rx datapath by caps %#x not found",
1918 				avail_caps);
1919 			rc = ENOENT;
1920 			goto fail_dp_rx;
1921 		}
1922 	}
1923 
1924 	sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1925 	if (sas->dp_rx_name == NULL) {
1926 		rc = ENOMEM;
1927 		goto fail_dp_rx_name;
1928 	}
1929 
1930 	sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1931 
1932 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1933 				sfc_kvarg_string_handler, &tx_name);
1934 	if (rc != 0)
1935 		goto fail_kvarg_tx_datapath;
1936 
1937 	if (tx_name != NULL) {
1938 		dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1939 		if (dp_tx == NULL) {
1940 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1941 			rc = ENOENT;
1942 			goto fail_dp_tx;
1943 		}
1944 		if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1945 			sfc_err(sa,
1946 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1947 				tx_name);
1948 			rc = EINVAL;
1949 			goto fail_dp_tx_caps;
1950 		}
1951 	} else {
1952 		dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1953 		if (dp_tx == NULL) {
1954 			sfc_err(sa, "Tx datapath by caps %#x not found",
1955 				avail_caps);
1956 			rc = ENOENT;
1957 			goto fail_dp_tx;
1958 		}
1959 	}
1960 
1961 	sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1962 	if (sas->dp_tx_name == NULL) {
1963 		rc = ENOMEM;
1964 		goto fail_dp_tx_name;
1965 	}
1966 
1967 	sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1968 
1969 	sa->priv.dp_rx = dp_rx;
1970 	sa->priv.dp_tx = dp_tx;
1971 
1972 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1973 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1974 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1975 
1976 	dev->rx_queue_count = sfc_rx_queue_count;
1977 	dev->rx_descriptor_done = sfc_rx_descriptor_done;
1978 	dev->rx_descriptor_status = sfc_rx_descriptor_status;
1979 	dev->tx_descriptor_status = sfc_tx_descriptor_status;
1980 	dev->dev_ops = &sfc_eth_dev_ops;
1981 
1982 	return 0;
1983 
1984 fail_dp_tx_name:
1985 fail_dp_tx_caps:
1986 fail_dp_tx:
1987 fail_kvarg_tx_datapath:
1988 	rte_free(sas->dp_rx_name);
1989 	sas->dp_rx_name = NULL;
1990 
1991 fail_dp_rx_name:
1992 fail_dp_rx_caps:
1993 fail_dp_rx:
1994 fail_kvarg_rx_datapath:
1995 	return rc;
1996 }
1997 
1998 static void
1999 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2000 {
2001 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2002 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2003 
2004 	dev->dev_ops = NULL;
2005 	dev->tx_pkt_prepare = NULL;
2006 	dev->rx_pkt_burst = NULL;
2007 	dev->tx_pkt_burst = NULL;
2008 
2009 	rte_free(sas->dp_tx_name);
2010 	sas->dp_tx_name = NULL;
2011 	sa->priv.dp_tx = NULL;
2012 
2013 	rte_free(sas->dp_rx_name);
2014 	sas->dp_rx_name = NULL;
2015 	sa->priv.dp_rx = NULL;
2016 }
2017 
2018 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2019 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
2020 	.reta_query			= sfc_dev_rss_reta_query,
2021 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
2022 	.rxq_info_get			= sfc_rx_queue_info_get,
2023 	.txq_info_get			= sfc_tx_queue_info_get,
2024 };
2025 
2026 static int
2027 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2028 {
2029 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2030 	struct sfc_adapter_priv *sap;
2031 	const struct sfc_dp_rx *dp_rx;
2032 	const struct sfc_dp_tx *dp_tx;
2033 	int rc;
2034 
2035 	/*
2036 	 * Allocate process private data from heap, since it should not
2037 	 * be located in shared memory allocated using rte_malloc() API.
2038 	 */
2039 	sap = calloc(1, sizeof(*sap));
2040 	if (sap == NULL) {
2041 		rc = ENOMEM;
2042 		goto fail_alloc_priv;
2043 	}
2044 
2045 	sap->logtype_main = logtype_main;
2046 
2047 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2048 	if (dp_rx == NULL) {
2049 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2050 			"cannot find %s Rx datapath", sas->dp_rx_name);
2051 		rc = ENOENT;
2052 		goto fail_dp_rx;
2053 	}
2054 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2055 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2056 			"%s Rx datapath does not support multi-process",
2057 			sas->dp_rx_name);
2058 		rc = EINVAL;
2059 		goto fail_dp_rx_multi_process;
2060 	}
2061 
2062 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2063 	if (dp_tx == NULL) {
2064 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2065 			"cannot find %s Tx datapath", sas->dp_tx_name);
2066 		rc = ENOENT;
2067 		goto fail_dp_tx;
2068 	}
2069 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2070 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2071 			"%s Tx datapath does not support multi-process",
2072 			sas->dp_tx_name);
2073 		rc = EINVAL;
2074 		goto fail_dp_tx_multi_process;
2075 	}
2076 
2077 	sap->dp_rx = dp_rx;
2078 	sap->dp_tx = dp_tx;
2079 
2080 	dev->process_private = sap;
2081 	dev->rx_pkt_burst = dp_rx->pkt_burst;
2082 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2083 	dev->tx_pkt_burst = dp_tx->pkt_burst;
2084 	dev->rx_queue_count = sfc_rx_queue_count;
2085 	dev->rx_descriptor_done = sfc_rx_descriptor_done;
2086 	dev->rx_descriptor_status = sfc_rx_descriptor_status;
2087 	dev->tx_descriptor_status = sfc_tx_descriptor_status;
2088 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
2089 
2090 	return 0;
2091 
2092 fail_dp_tx_multi_process:
2093 fail_dp_tx:
2094 fail_dp_rx_multi_process:
2095 fail_dp_rx:
2096 	free(sap);
2097 
2098 fail_alloc_priv:
2099 	return rc;
2100 }
2101 
2102 static void
2103 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
2104 {
2105 	free(dev->process_private);
2106 	dev->process_private = NULL;
2107 	dev->dev_ops = NULL;
2108 	dev->tx_pkt_prepare = NULL;
2109 	dev->tx_pkt_burst = NULL;
2110 	dev->rx_pkt_burst = NULL;
2111 }
2112 
2113 static void
2114 sfc_register_dp(void)
2115 {
2116 	/* Register once */
2117 	if (TAILQ_EMPTY(&sfc_dp_head)) {
2118 		/* Prefer EF10 datapath */
2119 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2120 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2121 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2122 
2123 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2124 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2125 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2126 	}
2127 }
2128 
2129 static int
2130 sfc_eth_dev_init(struct rte_eth_dev *dev)
2131 {
2132 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2133 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2134 	uint32_t logtype_main;
2135 	struct sfc_adapter *sa;
2136 	int rc;
2137 	const efx_nic_cfg_t *encp;
2138 	const struct rte_ether_addr *from;
2139 
2140 	sfc_register_dp();
2141 
2142 	logtype_main = sfc_register_logtype(&pci_dev->addr,
2143 					    SFC_LOGTYPE_MAIN_STR,
2144 					    RTE_LOG_NOTICE);
2145 
2146 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2147 		return -sfc_eth_dev_secondary_init(dev, logtype_main);
2148 
2149 	/* Required for logging */
2150 	sas->pci_addr = pci_dev->addr;
2151 	sas->port_id = dev->data->port_id;
2152 
2153 	/*
2154 	 * Allocate process private data from heap, since it should not
2155 	 * be located in shared memory allocated using rte_malloc() API.
2156 	 */
2157 	sa = calloc(1, sizeof(*sa));
2158 	if (sa == NULL) {
2159 		rc = ENOMEM;
2160 		goto fail_alloc_sa;
2161 	}
2162 
2163 	dev->process_private = sa;
2164 
2165 	/* Required for logging */
2166 	sa->priv.shared = sas;
2167 	sa->priv.logtype_main = logtype_main;
2168 
2169 	sa->eth_dev = dev;
2170 
2171 	/* Copy PCI device info to the dev->data */
2172 	rte_eth_copy_pci_info(dev, pci_dev);
2173 
2174 	rc = sfc_kvargs_parse(sa);
2175 	if (rc != 0)
2176 		goto fail_kvargs_parse;
2177 
2178 	sfc_log_init(sa, "entry");
2179 
2180 	dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
2181 
2182 	dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2183 	if (dev->data->mac_addrs == NULL) {
2184 		rc = ENOMEM;
2185 		goto fail_mac_addrs;
2186 	}
2187 
2188 	sfc_adapter_lock_init(sa);
2189 	sfc_adapter_lock(sa);
2190 
2191 	sfc_log_init(sa, "probing");
2192 	rc = sfc_probe(sa);
2193 	if (rc != 0)
2194 		goto fail_probe;
2195 
2196 	sfc_log_init(sa, "set device ops");
2197 	rc = sfc_eth_dev_set_ops(dev);
2198 	if (rc != 0)
2199 		goto fail_set_ops;
2200 
2201 	sfc_log_init(sa, "attaching");
2202 	rc = sfc_attach(sa);
2203 	if (rc != 0)
2204 		goto fail_attach;
2205 
2206 	encp = efx_nic_cfg_get(sa->nic);
2207 
2208 	/*
2209 	 * The arguments are really reverse order in comparison to
2210 	 * Linux kernel. Copy from NIC config to Ethernet device data.
2211 	 */
2212 	from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2213 	rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2214 
2215 	sfc_adapter_unlock(sa);
2216 
2217 	sfc_log_init(sa, "done");
2218 	return 0;
2219 
2220 fail_attach:
2221 	sfc_eth_dev_clear_ops(dev);
2222 
2223 fail_set_ops:
2224 	sfc_unprobe(sa);
2225 
2226 fail_probe:
2227 	sfc_adapter_unlock(sa);
2228 	sfc_adapter_lock_fini(sa);
2229 	rte_free(dev->data->mac_addrs);
2230 	dev->data->mac_addrs = NULL;
2231 
2232 fail_mac_addrs:
2233 	sfc_kvargs_cleanup(sa);
2234 
2235 fail_kvargs_parse:
2236 	sfc_log_init(sa, "failed %d", rc);
2237 	dev->process_private = NULL;
2238 	free(sa);
2239 
2240 fail_alloc_sa:
2241 	SFC_ASSERT(rc > 0);
2242 	return -rc;
2243 }
2244 
2245 static int
2246 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2247 {
2248 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2249 		sfc_eth_dev_secondary_clear_ops(dev);
2250 		return 0;
2251 	}
2252 
2253 	sfc_dev_close(dev);
2254 
2255 	return 0;
2256 }
2257 
2258 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2259 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2260 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2261 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2262 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2263 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2264 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2265 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2266 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2267 	{ .vendor_id = 0 /* sentinel */ }
2268 };
2269 
2270 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2271 	struct rte_pci_device *pci_dev)
2272 {
2273 	return rte_eth_dev_pci_generic_probe(pci_dev,
2274 		sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2275 }
2276 
2277 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2278 {
2279 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2280 }
2281 
2282 static struct rte_pci_driver sfc_efx_pmd = {
2283 	.id_table = pci_id_sfc_efx_map,
2284 	.drv_flags =
2285 		RTE_PCI_DRV_INTR_LSC |
2286 		RTE_PCI_DRV_NEED_MAPPING,
2287 	.probe = sfc_eth_dev_pci_probe,
2288 	.remove = sfc_eth_dev_pci_remove,
2289 };
2290 
2291 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2292 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2293 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2294 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2295 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2296 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2297 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2298 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2299 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2300 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2301 
2302 RTE_INIT(sfc_driver_register_logtype)
2303 {
2304 	int ret;
2305 
2306 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2307 						   RTE_LOG_NOTICE);
2308 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2309 }
2310