xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision e11bdd37745229bf26b557305c07d118c3dbaad7)
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 	/*
1034 	 * Copy the address to the device private data so that
1035 	 * it could be recalled in the case of adapter restart.
1036 	 */
1037 	rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1038 
1039 	/*
1040 	 * Neither of the two following checks can return
1041 	 * an error. The new MAC address is preserved in
1042 	 * the device private data and can be activated
1043 	 * on the next port start if the user prevents
1044 	 * isolated mode from being enabled.
1045 	 */
1046 	if (sfc_sa2shared(sa)->isolated) {
1047 		sfc_warn(sa, "isolated mode is active on the port");
1048 		sfc_warn(sa, "will not set MAC address");
1049 		goto unlock;
1050 	}
1051 
1052 	if (sa->state != SFC_ADAPTER_STARTED) {
1053 		sfc_notice(sa, "the port is not started");
1054 		sfc_notice(sa, "the new MAC address will be set on port start");
1055 
1056 		goto unlock;
1057 	}
1058 
1059 	if (encp->enc_allow_set_mac_with_installed_filters) {
1060 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1061 		if (rc != 0) {
1062 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1063 			goto unlock;
1064 		}
1065 
1066 		/*
1067 		 * Changing the MAC address by means of MCDI request
1068 		 * has no effect on received traffic, therefore
1069 		 * we also need to update unicast filters
1070 		 */
1071 		rc = sfc_set_rx_mode_unchecked(sa);
1072 		if (rc != 0) {
1073 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
1074 			/* Rollback the old address */
1075 			(void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1076 			(void)sfc_set_rx_mode_unchecked(sa);
1077 		}
1078 	} else {
1079 		sfc_warn(sa, "cannot set MAC address with filters installed");
1080 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1081 		sfc_warn(sa, "(some traffic may be dropped)");
1082 
1083 		/*
1084 		 * Since setting MAC address with filters installed is not
1085 		 * allowed on the adapter, the new MAC address will be set
1086 		 * by means of adapter restart. sfc_start() shall retrieve
1087 		 * the new address from the device private data and set it.
1088 		 */
1089 		sfc_stop(sa);
1090 		rc = sfc_start(sa);
1091 		if (rc != 0)
1092 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1093 	}
1094 
1095 unlock:
1096 	if (rc != 0)
1097 		rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1098 
1099 	sfc_adapter_unlock(sa);
1100 
1101 	SFC_ASSERT(rc >= 0);
1102 	return -rc;
1103 }
1104 
1105 
1106 static int
1107 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1108 		struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1109 {
1110 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1111 	struct sfc_port *port = &sa->port;
1112 	uint8_t *mc_addrs = port->mcast_addrs;
1113 	int rc;
1114 	unsigned int i;
1115 
1116 	if (sfc_sa2shared(sa)->isolated) {
1117 		sfc_err(sa, "isolated mode is active on the port");
1118 		sfc_err(sa, "will not set multicast address list");
1119 		return -ENOTSUP;
1120 	}
1121 
1122 	if (mc_addrs == NULL)
1123 		return -ENOBUFS;
1124 
1125 	if (nb_mc_addr > port->max_mcast_addrs) {
1126 		sfc_err(sa, "too many multicast addresses: %u > %u",
1127 			 nb_mc_addr, port->max_mcast_addrs);
1128 		return -EINVAL;
1129 	}
1130 
1131 	for (i = 0; i < nb_mc_addr; ++i) {
1132 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1133 				 EFX_MAC_ADDR_LEN);
1134 		mc_addrs += EFX_MAC_ADDR_LEN;
1135 	}
1136 
1137 	port->nb_mcast_addrs = nb_mc_addr;
1138 
1139 	if (sa->state != SFC_ADAPTER_STARTED)
1140 		return 0;
1141 
1142 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1143 					port->nb_mcast_addrs);
1144 	if (rc != 0)
1145 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1146 
1147 	SFC_ASSERT(rc >= 0);
1148 	return -rc;
1149 }
1150 
1151 /*
1152  * The function is used by the secondary process as well. It must not
1153  * use any process-local pointers from the adapter data.
1154  */
1155 static void
1156 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1157 		      struct rte_eth_rxq_info *qinfo)
1158 {
1159 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1160 	struct sfc_rxq_info *rxq_info;
1161 
1162 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1163 
1164 	rxq_info = &sas->rxq_info[rx_queue_id];
1165 
1166 	qinfo->mp = rxq_info->refill_mb_pool;
1167 	qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1168 	qinfo->conf.rx_drop_en = 1;
1169 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1170 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1171 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1172 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1173 		qinfo->scattered_rx = 1;
1174 	}
1175 	qinfo->nb_desc = rxq_info->entries;
1176 }
1177 
1178 /*
1179  * The function is used by the secondary process as well. It must not
1180  * use any process-local pointers from the adapter data.
1181  */
1182 static void
1183 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1184 		      struct rte_eth_txq_info *qinfo)
1185 {
1186 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1187 	struct sfc_txq_info *txq_info;
1188 
1189 	SFC_ASSERT(tx_queue_id < sas->txq_count);
1190 
1191 	txq_info = &sas->txq_info[tx_queue_id];
1192 
1193 	memset(qinfo, 0, sizeof(*qinfo));
1194 
1195 	qinfo->conf.offloads = txq_info->offloads;
1196 	qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1197 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1198 	qinfo->nb_desc = txq_info->entries;
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 uint32_t
1206 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1207 {
1208 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1209 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1210 	struct sfc_rxq_info *rxq_info;
1211 
1212 	SFC_ASSERT(rx_queue_id < sas->rxq_count);
1213 	rxq_info = &sas->rxq_info[rx_queue_id];
1214 
1215 	if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1216 		return 0;
1217 
1218 	return sap->dp_rx->qdesc_npending(rxq_info->dp);
1219 }
1220 
1221 /*
1222  * The function is used by the secondary process as well. It must not
1223  * use any process-local pointers from the adapter data.
1224  */
1225 static int
1226 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1227 {
1228 	struct sfc_dp_rxq *dp_rxq = queue;
1229 	const struct sfc_dp_rx *dp_rx;
1230 
1231 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1232 
1233 	return offset < dp_rx->qdesc_npending(dp_rxq);
1234 }
1235 
1236 /*
1237  * The function is used by the secondary process as well. It must not
1238  * use any process-local pointers from the adapter data.
1239  */
1240 static int
1241 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1242 {
1243 	struct sfc_dp_rxq *dp_rxq = queue;
1244 	const struct sfc_dp_rx *dp_rx;
1245 
1246 	dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1247 
1248 	return dp_rx->qdesc_status(dp_rxq, offset);
1249 }
1250 
1251 /*
1252  * The function is used by the secondary process as well. It must not
1253  * use any process-local pointers from the adapter data.
1254  */
1255 static int
1256 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1257 {
1258 	struct sfc_dp_txq *dp_txq = queue;
1259 	const struct sfc_dp_tx *dp_tx;
1260 
1261 	dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1262 
1263 	return dp_tx->qdesc_status(dp_txq, offset);
1264 }
1265 
1266 static int
1267 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1268 {
1269 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1270 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1271 	int rc;
1272 
1273 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1274 
1275 	sfc_adapter_lock(sa);
1276 
1277 	rc = EINVAL;
1278 	if (sa->state != SFC_ADAPTER_STARTED)
1279 		goto fail_not_started;
1280 
1281 	if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1282 		goto fail_not_setup;
1283 
1284 	rc = sfc_rx_qstart(sa, rx_queue_id);
1285 	if (rc != 0)
1286 		goto fail_rx_qstart;
1287 
1288 	sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1289 
1290 	sfc_adapter_unlock(sa);
1291 
1292 	return 0;
1293 
1294 fail_rx_qstart:
1295 fail_not_setup:
1296 fail_not_started:
1297 	sfc_adapter_unlock(sa);
1298 	SFC_ASSERT(rc > 0);
1299 	return -rc;
1300 }
1301 
1302 static int
1303 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1304 {
1305 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1306 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1307 
1308 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1309 
1310 	sfc_adapter_lock(sa);
1311 	sfc_rx_qstop(sa, rx_queue_id);
1312 
1313 	sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1314 
1315 	sfc_adapter_unlock(sa);
1316 
1317 	return 0;
1318 }
1319 
1320 static int
1321 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1322 {
1323 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1324 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1325 	int rc;
1326 
1327 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1328 
1329 	sfc_adapter_lock(sa);
1330 
1331 	rc = EINVAL;
1332 	if (sa->state != SFC_ADAPTER_STARTED)
1333 		goto fail_not_started;
1334 
1335 	if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1336 		goto fail_not_setup;
1337 
1338 	rc = sfc_tx_qstart(sa, tx_queue_id);
1339 	if (rc != 0)
1340 		goto fail_tx_qstart;
1341 
1342 	sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1343 
1344 	sfc_adapter_unlock(sa);
1345 	return 0;
1346 
1347 fail_tx_qstart:
1348 
1349 fail_not_setup:
1350 fail_not_started:
1351 	sfc_adapter_unlock(sa);
1352 	SFC_ASSERT(rc > 0);
1353 	return -rc;
1354 }
1355 
1356 static int
1357 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1358 {
1359 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1360 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1361 
1362 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1363 
1364 	sfc_adapter_lock(sa);
1365 
1366 	sfc_tx_qstop(sa, tx_queue_id);
1367 
1368 	sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1369 
1370 	sfc_adapter_unlock(sa);
1371 	return 0;
1372 }
1373 
1374 static efx_tunnel_protocol_t
1375 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1376 {
1377 	switch (rte_type) {
1378 	case RTE_TUNNEL_TYPE_VXLAN:
1379 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1380 	case RTE_TUNNEL_TYPE_GENEVE:
1381 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1382 	default:
1383 		return EFX_TUNNEL_NPROTOS;
1384 	}
1385 }
1386 
1387 enum sfc_udp_tunnel_op_e {
1388 	SFC_UDP_TUNNEL_ADD_PORT,
1389 	SFC_UDP_TUNNEL_DEL_PORT,
1390 };
1391 
1392 static int
1393 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1394 		      struct rte_eth_udp_tunnel *tunnel_udp,
1395 		      enum sfc_udp_tunnel_op_e op)
1396 {
1397 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1398 	efx_tunnel_protocol_t tunnel_proto;
1399 	int rc;
1400 
1401 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1402 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1403 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1404 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1405 
1406 	tunnel_proto =
1407 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1408 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1409 		rc = ENOTSUP;
1410 		goto fail_bad_proto;
1411 	}
1412 
1413 	sfc_adapter_lock(sa);
1414 
1415 	switch (op) {
1416 	case SFC_UDP_TUNNEL_ADD_PORT:
1417 		rc = efx_tunnel_config_udp_add(sa->nic,
1418 					       tunnel_udp->udp_port,
1419 					       tunnel_proto);
1420 		break;
1421 	case SFC_UDP_TUNNEL_DEL_PORT:
1422 		rc = efx_tunnel_config_udp_remove(sa->nic,
1423 						  tunnel_udp->udp_port,
1424 						  tunnel_proto);
1425 		break;
1426 	default:
1427 		rc = EINVAL;
1428 		goto fail_bad_op;
1429 	}
1430 
1431 	if (rc != 0)
1432 		goto fail_op;
1433 
1434 	if (sa->state == SFC_ADAPTER_STARTED) {
1435 		rc = efx_tunnel_reconfigure(sa->nic);
1436 		if (rc == EAGAIN) {
1437 			/*
1438 			 * Configuration is accepted by FW and MC reboot
1439 			 * is initiated to apply the changes. MC reboot
1440 			 * will be handled in a usual way (MC reboot
1441 			 * event on management event queue and adapter
1442 			 * restart).
1443 			 */
1444 			rc = 0;
1445 		} else if (rc != 0) {
1446 			goto fail_reconfigure;
1447 		}
1448 	}
1449 
1450 	sfc_adapter_unlock(sa);
1451 	return 0;
1452 
1453 fail_reconfigure:
1454 	/* Remove/restore entry since the change makes the trouble */
1455 	switch (op) {
1456 	case SFC_UDP_TUNNEL_ADD_PORT:
1457 		(void)efx_tunnel_config_udp_remove(sa->nic,
1458 						   tunnel_udp->udp_port,
1459 						   tunnel_proto);
1460 		break;
1461 	case SFC_UDP_TUNNEL_DEL_PORT:
1462 		(void)efx_tunnel_config_udp_add(sa->nic,
1463 						tunnel_udp->udp_port,
1464 						tunnel_proto);
1465 		break;
1466 	}
1467 
1468 fail_op:
1469 fail_bad_op:
1470 	sfc_adapter_unlock(sa);
1471 
1472 fail_bad_proto:
1473 	SFC_ASSERT(rc > 0);
1474 	return -rc;
1475 }
1476 
1477 static int
1478 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1479 			    struct rte_eth_udp_tunnel *tunnel_udp)
1480 {
1481 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1482 }
1483 
1484 static int
1485 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1486 			    struct rte_eth_udp_tunnel *tunnel_udp)
1487 {
1488 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1489 }
1490 
1491 /*
1492  * The function is used by the secondary process as well. It must not
1493  * use any process-local pointers from the adapter data.
1494  */
1495 static int
1496 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1497 			  struct rte_eth_rss_conf *rss_conf)
1498 {
1499 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1500 	struct sfc_rss *rss = &sas->rss;
1501 
1502 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1503 		return -ENOTSUP;
1504 
1505 	/*
1506 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1507 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1508 	 * flags which corresponds to the active EFX configuration stored
1509 	 * locally in 'sfc_adapter' and kept up-to-date
1510 	 */
1511 	rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1512 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1513 	if (rss_conf->rss_key != NULL)
1514 		rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1515 
1516 	return 0;
1517 }
1518 
1519 static int
1520 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1521 			struct rte_eth_rss_conf *rss_conf)
1522 {
1523 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1524 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1525 	unsigned int efx_hash_types;
1526 	int rc = 0;
1527 
1528 	if (sfc_sa2shared(sa)->isolated)
1529 		return -ENOTSUP;
1530 
1531 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1532 		sfc_err(sa, "RSS is not available");
1533 		return -ENOTSUP;
1534 	}
1535 
1536 	if (rss->channels == 0) {
1537 		sfc_err(sa, "RSS is not configured");
1538 		return -EINVAL;
1539 	}
1540 
1541 	if ((rss_conf->rss_key != NULL) &&
1542 	    (rss_conf->rss_key_len != sizeof(rss->key))) {
1543 		sfc_err(sa, "RSS key size is wrong (should be %zu)",
1544 			sizeof(rss->key));
1545 		return -EINVAL;
1546 	}
1547 
1548 	sfc_adapter_lock(sa);
1549 
1550 	rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1551 	if (rc != 0)
1552 		goto fail_rx_hf_rte_to_efx;
1553 
1554 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1555 				   rss->hash_alg, efx_hash_types, B_TRUE);
1556 	if (rc != 0)
1557 		goto fail_scale_mode_set;
1558 
1559 	if (rss_conf->rss_key != NULL) {
1560 		if (sa->state == SFC_ADAPTER_STARTED) {
1561 			rc = efx_rx_scale_key_set(sa->nic,
1562 						  EFX_RSS_CONTEXT_DEFAULT,
1563 						  rss_conf->rss_key,
1564 						  sizeof(rss->key));
1565 			if (rc != 0)
1566 				goto fail_scale_key_set;
1567 		}
1568 
1569 		rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1570 	}
1571 
1572 	rss->hash_types = efx_hash_types;
1573 
1574 	sfc_adapter_unlock(sa);
1575 
1576 	return 0;
1577 
1578 fail_scale_key_set:
1579 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1580 				  EFX_RX_HASHALG_TOEPLITZ,
1581 				  rss->hash_types, B_TRUE) != 0)
1582 		sfc_err(sa, "failed to restore RSS mode");
1583 
1584 fail_scale_mode_set:
1585 fail_rx_hf_rte_to_efx:
1586 	sfc_adapter_unlock(sa);
1587 	return -rc;
1588 }
1589 
1590 /*
1591  * The function is used by the secondary process as well. It must not
1592  * use any process-local pointers from the adapter data.
1593  */
1594 static int
1595 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1596 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1597 		       uint16_t reta_size)
1598 {
1599 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1600 	struct sfc_rss *rss = &sas->rss;
1601 	int entry;
1602 
1603 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1604 		return -ENOTSUP;
1605 
1606 	if (rss->channels == 0)
1607 		return -EINVAL;
1608 
1609 	if (reta_size != EFX_RSS_TBL_SIZE)
1610 		return -EINVAL;
1611 
1612 	for (entry = 0; entry < reta_size; entry++) {
1613 		int grp = entry / RTE_RETA_GROUP_SIZE;
1614 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1615 
1616 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1617 			reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1618 	}
1619 
1620 	return 0;
1621 }
1622 
1623 static int
1624 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1625 			struct rte_eth_rss_reta_entry64 *reta_conf,
1626 			uint16_t reta_size)
1627 {
1628 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1629 	struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1630 	unsigned int *rss_tbl_new;
1631 	uint16_t entry;
1632 	int rc = 0;
1633 
1634 
1635 	if (sfc_sa2shared(sa)->isolated)
1636 		return -ENOTSUP;
1637 
1638 	if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1639 		sfc_err(sa, "RSS is not available");
1640 		return -ENOTSUP;
1641 	}
1642 
1643 	if (rss->channels == 0) {
1644 		sfc_err(sa, "RSS is not configured");
1645 		return -EINVAL;
1646 	}
1647 
1648 	if (reta_size != EFX_RSS_TBL_SIZE) {
1649 		sfc_err(sa, "RETA size is wrong (should be %u)",
1650 			EFX_RSS_TBL_SIZE);
1651 		return -EINVAL;
1652 	}
1653 
1654 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1655 	if (rss_tbl_new == NULL)
1656 		return -ENOMEM;
1657 
1658 	sfc_adapter_lock(sa);
1659 
1660 	rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1661 
1662 	for (entry = 0; entry < reta_size; entry++) {
1663 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1664 		struct rte_eth_rss_reta_entry64 *grp;
1665 
1666 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1667 
1668 		if (grp->mask & (1ull << grp_idx)) {
1669 			if (grp->reta[grp_idx] >= rss->channels) {
1670 				rc = EINVAL;
1671 				goto bad_reta_entry;
1672 			}
1673 			rss_tbl_new[entry] = grp->reta[grp_idx];
1674 		}
1675 	}
1676 
1677 	if (sa->state == SFC_ADAPTER_STARTED) {
1678 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1679 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1680 		if (rc != 0)
1681 			goto fail_scale_tbl_set;
1682 	}
1683 
1684 	rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1685 
1686 fail_scale_tbl_set:
1687 bad_reta_entry:
1688 	sfc_adapter_unlock(sa);
1689 
1690 	rte_free(rss_tbl_new);
1691 
1692 	SFC_ASSERT(rc >= 0);
1693 	return -rc;
1694 }
1695 
1696 static int
1697 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1698 		    enum rte_filter_op filter_op,
1699 		    void *arg)
1700 {
1701 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1702 	int rc = ENOTSUP;
1703 
1704 	sfc_log_init(sa, "entry");
1705 
1706 	switch (filter_type) {
1707 	case RTE_ETH_FILTER_NONE:
1708 		sfc_err(sa, "Global filters configuration not supported");
1709 		break;
1710 	case RTE_ETH_FILTER_MACVLAN:
1711 		sfc_err(sa, "MACVLAN filters not supported");
1712 		break;
1713 	case RTE_ETH_FILTER_ETHERTYPE:
1714 		sfc_err(sa, "EtherType filters not supported");
1715 		break;
1716 	case RTE_ETH_FILTER_FLEXIBLE:
1717 		sfc_err(sa, "Flexible filters not supported");
1718 		break;
1719 	case RTE_ETH_FILTER_SYN:
1720 		sfc_err(sa, "SYN filters not supported");
1721 		break;
1722 	case RTE_ETH_FILTER_NTUPLE:
1723 		sfc_err(sa, "NTUPLE filters not supported");
1724 		break;
1725 	case RTE_ETH_FILTER_TUNNEL:
1726 		sfc_err(sa, "Tunnel filters not supported");
1727 		break;
1728 	case RTE_ETH_FILTER_FDIR:
1729 		sfc_err(sa, "Flow Director filters not supported");
1730 		break;
1731 	case RTE_ETH_FILTER_HASH:
1732 		sfc_err(sa, "Hash filters not supported");
1733 		break;
1734 	case RTE_ETH_FILTER_GENERIC:
1735 		if (filter_op != RTE_ETH_FILTER_GET) {
1736 			rc = EINVAL;
1737 		} else {
1738 			*(const void **)arg = &sfc_flow_ops;
1739 			rc = 0;
1740 		}
1741 		break;
1742 	default:
1743 		sfc_err(sa, "Unknown filter type %u", filter_type);
1744 		break;
1745 	}
1746 
1747 	sfc_log_init(sa, "exit: %d", -rc);
1748 	SFC_ASSERT(rc >= 0);
1749 	return -rc;
1750 }
1751 
1752 static int
1753 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1754 {
1755 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1756 
1757 	/*
1758 	 * If Rx datapath does not provide callback to check mempool,
1759 	 * all pools are supported.
1760 	 */
1761 	if (sap->dp_rx->pool_ops_supported == NULL)
1762 		return 1;
1763 
1764 	return sap->dp_rx->pool_ops_supported(pool);
1765 }
1766 
1767 static int
1768 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1769 {
1770 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1771 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1772 	struct sfc_rxq_info *rxq_info;
1773 
1774 	SFC_ASSERT(queue_id < sas->rxq_count);
1775 	rxq_info = &sas->rxq_info[queue_id];
1776 
1777 	return sap->dp_rx->intr_enable(rxq_info->dp);
1778 }
1779 
1780 static int
1781 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1782 {
1783 	const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1784 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1785 	struct sfc_rxq_info *rxq_info;
1786 
1787 	SFC_ASSERT(queue_id < sas->rxq_count);
1788 	rxq_info = &sas->rxq_info[queue_id];
1789 
1790 	return sap->dp_rx->intr_disable(rxq_info->dp);
1791 }
1792 
1793 static const struct eth_dev_ops sfc_eth_dev_ops = {
1794 	.dev_configure			= sfc_dev_configure,
1795 	.dev_start			= sfc_dev_start,
1796 	.dev_stop			= sfc_dev_stop,
1797 	.dev_set_link_up		= sfc_dev_set_link_up,
1798 	.dev_set_link_down		= sfc_dev_set_link_down,
1799 	.dev_close			= sfc_dev_close,
1800 	.promiscuous_enable		= sfc_dev_promisc_enable,
1801 	.promiscuous_disable		= sfc_dev_promisc_disable,
1802 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1803 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1804 	.link_update			= sfc_dev_link_update,
1805 	.stats_get			= sfc_stats_get,
1806 	.stats_reset			= sfc_stats_reset,
1807 	.xstats_get			= sfc_xstats_get,
1808 	.xstats_reset			= sfc_stats_reset,
1809 	.xstats_get_names		= sfc_xstats_get_names,
1810 	.dev_infos_get			= sfc_dev_infos_get,
1811 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1812 	.mtu_set			= sfc_dev_set_mtu,
1813 	.rx_queue_start			= sfc_rx_queue_start,
1814 	.rx_queue_stop			= sfc_rx_queue_stop,
1815 	.tx_queue_start			= sfc_tx_queue_start,
1816 	.tx_queue_stop			= sfc_tx_queue_stop,
1817 	.rx_queue_setup			= sfc_rx_queue_setup,
1818 	.rx_queue_release		= sfc_rx_queue_release,
1819 	.rx_queue_count			= sfc_rx_queue_count,
1820 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1821 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1822 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1823 	.rx_queue_intr_enable		= sfc_rx_queue_intr_enable,
1824 	.rx_queue_intr_disable		= sfc_rx_queue_intr_disable,
1825 	.tx_queue_setup			= sfc_tx_queue_setup,
1826 	.tx_queue_release		= sfc_tx_queue_release,
1827 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1828 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1829 	.mac_addr_set			= sfc_mac_addr_set,
1830 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1831 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1832 	.reta_update			= sfc_dev_rss_reta_update,
1833 	.reta_query			= sfc_dev_rss_reta_query,
1834 	.rss_hash_update		= sfc_dev_rss_hash_update,
1835 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1836 	.filter_ctrl			= sfc_dev_filter_ctrl,
1837 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1838 	.rxq_info_get			= sfc_rx_queue_info_get,
1839 	.txq_info_get			= sfc_tx_queue_info_get,
1840 	.fw_version_get			= sfc_fw_version_get,
1841 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1842 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1843 	.pool_ops_supported		= sfc_pool_ops_supported,
1844 };
1845 
1846 /**
1847  * Duplicate a string in potentially shared memory required for
1848  * multi-process support.
1849  *
1850  * strdup() allocates from process-local heap/memory.
1851  */
1852 static char *
1853 sfc_strdup(const char *str)
1854 {
1855 	size_t size;
1856 	char *copy;
1857 
1858 	if (str == NULL)
1859 		return NULL;
1860 
1861 	size = strlen(str) + 1;
1862 	copy = rte_malloc(__func__, size, 0);
1863 	if (copy != NULL)
1864 		rte_memcpy(copy, str, size);
1865 
1866 	return copy;
1867 }
1868 
1869 static int
1870 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1871 {
1872 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1873 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1874 	const struct sfc_dp_rx *dp_rx;
1875 	const struct sfc_dp_tx *dp_tx;
1876 	const efx_nic_cfg_t *encp;
1877 	unsigned int avail_caps = 0;
1878 	const char *rx_name = NULL;
1879 	const char *tx_name = NULL;
1880 	int rc;
1881 
1882 	switch (sa->family) {
1883 	case EFX_FAMILY_HUNTINGTON:
1884 	case EFX_FAMILY_MEDFORD:
1885 	case EFX_FAMILY_MEDFORD2:
1886 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1887 		break;
1888 	default:
1889 		break;
1890 	}
1891 
1892 	encp = efx_nic_cfg_get(sa->nic);
1893 	if (encp->enc_rx_es_super_buffer_supported)
1894 		avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1895 
1896 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1897 				sfc_kvarg_string_handler, &rx_name);
1898 	if (rc != 0)
1899 		goto fail_kvarg_rx_datapath;
1900 
1901 	if (rx_name != NULL) {
1902 		dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1903 		if (dp_rx == NULL) {
1904 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1905 			rc = ENOENT;
1906 			goto fail_dp_rx;
1907 		}
1908 		if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1909 			sfc_err(sa,
1910 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1911 				rx_name);
1912 			rc = EINVAL;
1913 			goto fail_dp_rx_caps;
1914 		}
1915 	} else {
1916 		dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1917 		if (dp_rx == NULL) {
1918 			sfc_err(sa, "Rx datapath by caps %#x not found",
1919 				avail_caps);
1920 			rc = ENOENT;
1921 			goto fail_dp_rx;
1922 		}
1923 	}
1924 
1925 	sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1926 	if (sas->dp_rx_name == NULL) {
1927 		rc = ENOMEM;
1928 		goto fail_dp_rx_name;
1929 	}
1930 
1931 	sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1932 
1933 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1934 				sfc_kvarg_string_handler, &tx_name);
1935 	if (rc != 0)
1936 		goto fail_kvarg_tx_datapath;
1937 
1938 	if (tx_name != NULL) {
1939 		dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1940 		if (dp_tx == NULL) {
1941 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1942 			rc = ENOENT;
1943 			goto fail_dp_tx;
1944 		}
1945 		if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1946 			sfc_err(sa,
1947 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1948 				tx_name);
1949 			rc = EINVAL;
1950 			goto fail_dp_tx_caps;
1951 		}
1952 	} else {
1953 		dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1954 		if (dp_tx == NULL) {
1955 			sfc_err(sa, "Tx datapath by caps %#x not found",
1956 				avail_caps);
1957 			rc = ENOENT;
1958 			goto fail_dp_tx;
1959 		}
1960 	}
1961 
1962 	sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1963 	if (sas->dp_tx_name == NULL) {
1964 		rc = ENOMEM;
1965 		goto fail_dp_tx_name;
1966 	}
1967 
1968 	sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1969 
1970 	sa->priv.dp_rx = dp_rx;
1971 	sa->priv.dp_tx = dp_tx;
1972 
1973 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1974 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1975 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1976 
1977 	dev->dev_ops = &sfc_eth_dev_ops;
1978 
1979 	return 0;
1980 
1981 fail_dp_tx_name:
1982 fail_dp_tx_caps:
1983 fail_dp_tx:
1984 fail_kvarg_tx_datapath:
1985 	rte_free(sas->dp_rx_name);
1986 	sas->dp_rx_name = NULL;
1987 
1988 fail_dp_rx_name:
1989 fail_dp_rx_caps:
1990 fail_dp_rx:
1991 fail_kvarg_rx_datapath:
1992 	return rc;
1993 }
1994 
1995 static void
1996 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1997 {
1998 	struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1999 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2000 
2001 	dev->dev_ops = NULL;
2002 	dev->tx_pkt_prepare = NULL;
2003 	dev->rx_pkt_burst = NULL;
2004 	dev->tx_pkt_burst = NULL;
2005 
2006 	rte_free(sas->dp_tx_name);
2007 	sas->dp_tx_name = NULL;
2008 	sa->priv.dp_tx = NULL;
2009 
2010 	rte_free(sas->dp_rx_name);
2011 	sas->dp_rx_name = NULL;
2012 	sa->priv.dp_rx = NULL;
2013 }
2014 
2015 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2016 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
2017 	.rx_queue_count			= sfc_rx_queue_count,
2018 	.rx_descriptor_done		= sfc_rx_descriptor_done,
2019 	.rx_descriptor_status		= sfc_rx_descriptor_status,
2020 	.tx_descriptor_status		= sfc_tx_descriptor_status,
2021 	.reta_query			= sfc_dev_rss_reta_query,
2022 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
2023 	.rxq_info_get			= sfc_rx_queue_info_get,
2024 	.txq_info_get			= sfc_tx_queue_info_get,
2025 };
2026 
2027 static int
2028 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2029 {
2030 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2031 	struct sfc_adapter_priv *sap;
2032 	const struct sfc_dp_rx *dp_rx;
2033 	const struct sfc_dp_tx *dp_tx;
2034 	int rc;
2035 
2036 	/*
2037 	 * Allocate process private data from heap, since it should not
2038 	 * be located in shared memory allocated using rte_malloc() API.
2039 	 */
2040 	sap = calloc(1, sizeof(*sap));
2041 	if (sap == NULL) {
2042 		rc = ENOMEM;
2043 		goto fail_alloc_priv;
2044 	}
2045 
2046 	sap->logtype_main = logtype_main;
2047 
2048 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2049 	if (dp_rx == NULL) {
2050 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2051 			"cannot find %s Rx datapath", sas->dp_rx_name);
2052 		rc = ENOENT;
2053 		goto fail_dp_rx;
2054 	}
2055 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2056 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2057 			"%s Rx datapath does not support multi-process",
2058 			sas->dp_rx_name);
2059 		rc = EINVAL;
2060 		goto fail_dp_rx_multi_process;
2061 	}
2062 
2063 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2064 	if (dp_tx == NULL) {
2065 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2066 			"cannot find %s Tx datapath", sas->dp_tx_name);
2067 		rc = ENOENT;
2068 		goto fail_dp_tx;
2069 	}
2070 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2071 		SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2072 			"%s Tx datapath does not support multi-process",
2073 			sas->dp_tx_name);
2074 		rc = EINVAL;
2075 		goto fail_dp_tx_multi_process;
2076 	}
2077 
2078 	sap->dp_rx = dp_rx;
2079 	sap->dp_tx = dp_tx;
2080 
2081 	dev->process_private = sap;
2082 	dev->rx_pkt_burst = dp_rx->pkt_burst;
2083 	dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2084 	dev->tx_pkt_burst = dp_tx->pkt_burst;
2085 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
2086 
2087 	return 0;
2088 
2089 fail_dp_tx_multi_process:
2090 fail_dp_tx:
2091 fail_dp_rx_multi_process:
2092 fail_dp_rx:
2093 	free(sap);
2094 
2095 fail_alloc_priv:
2096 	return rc;
2097 }
2098 
2099 static void
2100 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
2101 {
2102 	free(dev->process_private);
2103 	dev->process_private = NULL;
2104 	dev->dev_ops = NULL;
2105 	dev->tx_pkt_prepare = NULL;
2106 	dev->tx_pkt_burst = NULL;
2107 	dev->rx_pkt_burst = NULL;
2108 }
2109 
2110 static void
2111 sfc_register_dp(void)
2112 {
2113 	/* Register once */
2114 	if (TAILQ_EMPTY(&sfc_dp_head)) {
2115 		/* Prefer EF10 datapath */
2116 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2117 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2118 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2119 
2120 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2121 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2122 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2123 	}
2124 }
2125 
2126 static int
2127 sfc_eth_dev_init(struct rte_eth_dev *dev)
2128 {
2129 	struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2130 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2131 	uint32_t logtype_main;
2132 	struct sfc_adapter *sa;
2133 	int rc;
2134 	const efx_nic_cfg_t *encp;
2135 	const struct rte_ether_addr *from;
2136 
2137 	sfc_register_dp();
2138 
2139 	logtype_main = sfc_register_logtype(&pci_dev->addr,
2140 					    SFC_LOGTYPE_MAIN_STR,
2141 					    RTE_LOG_NOTICE);
2142 
2143 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2144 		return -sfc_eth_dev_secondary_init(dev, logtype_main);
2145 
2146 	/* Required for logging */
2147 	sas->pci_addr = pci_dev->addr;
2148 	sas->port_id = dev->data->port_id;
2149 
2150 	/*
2151 	 * Allocate process private data from heap, since it should not
2152 	 * be located in shared memory allocated using rte_malloc() API.
2153 	 */
2154 	sa = calloc(1, sizeof(*sa));
2155 	if (sa == NULL) {
2156 		rc = ENOMEM;
2157 		goto fail_alloc_sa;
2158 	}
2159 
2160 	dev->process_private = sa;
2161 
2162 	/* Required for logging */
2163 	sa->priv.shared = sas;
2164 	sa->priv.logtype_main = logtype_main;
2165 
2166 	sa->eth_dev = dev;
2167 
2168 	/* Copy PCI device info to the dev->data */
2169 	rte_eth_copy_pci_info(dev, pci_dev);
2170 
2171 	rc = sfc_kvargs_parse(sa);
2172 	if (rc != 0)
2173 		goto fail_kvargs_parse;
2174 
2175 	sfc_log_init(sa, "entry");
2176 
2177 	dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
2178 
2179 	dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2180 	if (dev->data->mac_addrs == NULL) {
2181 		rc = ENOMEM;
2182 		goto fail_mac_addrs;
2183 	}
2184 
2185 	sfc_adapter_lock_init(sa);
2186 	sfc_adapter_lock(sa);
2187 
2188 	sfc_log_init(sa, "probing");
2189 	rc = sfc_probe(sa);
2190 	if (rc != 0)
2191 		goto fail_probe;
2192 
2193 	sfc_log_init(sa, "set device ops");
2194 	rc = sfc_eth_dev_set_ops(dev);
2195 	if (rc != 0)
2196 		goto fail_set_ops;
2197 
2198 	sfc_log_init(sa, "attaching");
2199 	rc = sfc_attach(sa);
2200 	if (rc != 0)
2201 		goto fail_attach;
2202 
2203 	encp = efx_nic_cfg_get(sa->nic);
2204 
2205 	/*
2206 	 * The arguments are really reverse order in comparison to
2207 	 * Linux kernel. Copy from NIC config to Ethernet device data.
2208 	 */
2209 	from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2210 	rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2211 
2212 	sfc_adapter_unlock(sa);
2213 
2214 	sfc_log_init(sa, "done");
2215 	return 0;
2216 
2217 fail_attach:
2218 	sfc_eth_dev_clear_ops(dev);
2219 
2220 fail_set_ops:
2221 	sfc_unprobe(sa);
2222 
2223 fail_probe:
2224 	sfc_adapter_unlock(sa);
2225 	sfc_adapter_lock_fini(sa);
2226 	rte_free(dev->data->mac_addrs);
2227 	dev->data->mac_addrs = NULL;
2228 
2229 fail_mac_addrs:
2230 	sfc_kvargs_cleanup(sa);
2231 
2232 fail_kvargs_parse:
2233 	sfc_log_init(sa, "failed %d", rc);
2234 	dev->process_private = NULL;
2235 	free(sa);
2236 
2237 fail_alloc_sa:
2238 	SFC_ASSERT(rc > 0);
2239 	return -rc;
2240 }
2241 
2242 static int
2243 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2244 {
2245 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2246 		sfc_eth_dev_secondary_clear_ops(dev);
2247 		return 0;
2248 	}
2249 
2250 	sfc_dev_close(dev);
2251 
2252 	return 0;
2253 }
2254 
2255 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2256 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2257 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2258 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2259 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2260 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2261 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2262 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2263 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2264 	{ .vendor_id = 0 /* sentinel */ }
2265 };
2266 
2267 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2268 	struct rte_pci_device *pci_dev)
2269 {
2270 	return rte_eth_dev_pci_generic_probe(pci_dev,
2271 		sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2272 }
2273 
2274 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2275 {
2276 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2277 }
2278 
2279 static struct rte_pci_driver sfc_efx_pmd = {
2280 	.id_table = pci_id_sfc_efx_map,
2281 	.drv_flags =
2282 		RTE_PCI_DRV_INTR_LSC |
2283 		RTE_PCI_DRV_NEED_MAPPING,
2284 	.probe = sfc_eth_dev_pci_probe,
2285 	.remove = sfc_eth_dev_pci_remove,
2286 };
2287 
2288 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2289 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2290 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2291 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2292 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2293 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2294 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2295 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2296 	SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2297 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2298 
2299 RTE_INIT(sfc_driver_register_logtype)
2300 {
2301 	int ret;
2302 
2303 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2304 						   RTE_LOG_NOTICE);
2305 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2306 }
2307