xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision 005870b4d3a1f6ac2e0f492b57d502e09f1565f3)
1 /*-
2  *   BSD LICENSE
3  *
4  * Copyright (c) 2016-2017 Solarflare Communications Inc.
5  * All rights reserved.
6  *
7  * This software was jointly developed between OKTET Labs (under contract
8  * for Solarflare) and Solarflare Communications, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions are met:
12  *
13  * 1. Redistributions of source code must retain the above copyright notice,
14  *    this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright notice,
16  *    this list of conditions and the following disclaimer in the documentation
17  *    and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
21  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
23  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
24  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
25  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
26  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
27  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
28  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
29  * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <rte_dev.h>
33 #include <rte_ethdev.h>
34 #include <rte_pci.h>
35 #include <rte_errno.h>
36 
37 #include "efx.h"
38 
39 #include "sfc.h"
40 #include "sfc_debug.h"
41 #include "sfc_log.h"
42 #include "sfc_kvargs.h"
43 #include "sfc_ev.h"
44 #include "sfc_rx.h"
45 #include "sfc_tx.h"
46 #include "sfc_flow.h"
47 #include "sfc_dp.h"
48 #include "sfc_dp_rx.h"
49 
50 static struct sfc_dp_list sfc_dp_head =
51 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
52 
53 static int
54 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
55 {
56 	struct sfc_adapter *sa = dev->data->dev_private;
57 	efx_nic_fw_info_t enfi;
58 	int ret;
59 	int rc;
60 
61 	/*
62 	 * Return value of the callback is likely supposed to be
63 	 * equal to or greater than 0, nevertheless, if an error
64 	 * occurs, it will be desirable to pass it to the caller
65 	 */
66 	if ((fw_version == NULL) || (fw_size == 0))
67 		return -EINVAL;
68 
69 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
70 	if (rc != 0)
71 		return -rc;
72 
73 	ret = snprintf(fw_version, fw_size,
74 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
75 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
76 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
77 	if (ret < 0)
78 		return ret;
79 
80 	if (enfi.enfi_dpcpu_fw_ids_valid) {
81 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
82 		int ret_extra;
83 
84 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
85 				     fw_size - dpcpu_fw_ids_offset,
86 				     " rx%" PRIx16 " tx%" PRIx16,
87 				     enfi.enfi_rx_dpcpu_fw_id,
88 				     enfi.enfi_tx_dpcpu_fw_id);
89 		if (ret_extra < 0)
90 			return ret_extra;
91 
92 		ret += ret_extra;
93 	}
94 
95 	if (fw_size < (size_t)(++ret))
96 		return ret;
97 	else
98 		return 0;
99 }
100 
101 static void
102 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
103 {
104 	struct sfc_adapter *sa = dev->data->dev_private;
105 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
106 
107 	sfc_log_init(sa, "entry");
108 
109 	dev_info->pci_dev = RTE_DEV_TO_PCI(dev->device);
110 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
111 
112 	/* Autonegotiation may be disabled */
113 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
114 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
115 		dev_info->speed_capa |= ETH_LINK_SPEED_1G;
116 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
117 		dev_info->speed_capa |= ETH_LINK_SPEED_10G;
118 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
119 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
120 
121 	dev_info->max_rx_queues = sa->rxq_max;
122 	dev_info->max_tx_queues = sa->txq_max;
123 
124 	/* By default packets are dropped if no descriptors are available */
125 	dev_info->default_rxconf.rx_drop_en = 1;
126 
127 	dev_info->rx_offload_capa =
128 		DEV_RX_OFFLOAD_IPV4_CKSUM |
129 		DEV_RX_OFFLOAD_UDP_CKSUM |
130 		DEV_RX_OFFLOAD_TCP_CKSUM;
131 
132 	dev_info->tx_offload_capa =
133 		DEV_TX_OFFLOAD_IPV4_CKSUM |
134 		DEV_TX_OFFLOAD_UDP_CKSUM |
135 		DEV_TX_OFFLOAD_TCP_CKSUM;
136 
137 	dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
138 	if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
139 	    !encp->enc_hw_tx_insert_vlan_enabled)
140 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
141 	else
142 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
143 
144 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
145 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
146 
147 #if EFSYS_OPT_RX_SCALE
148 	if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
149 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
150 		dev_info->hash_key_size = SFC_RSS_KEY_SIZE;
151 		dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
152 	}
153 #endif
154 
155 	if (sa->tso)
156 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
157 
158 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
159 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
160 	/* The RXQ hardware requires that the descriptor count is a power
161 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
162 	 */
163 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
164 
165 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
166 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
167 	/*
168 	 * The TXQ hardware requires that the descriptor count is a power
169 	 * of 2, but tx_desc_lim cannot properly describe that constraint
170 	 */
171 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
172 }
173 
174 static const uint32_t *
175 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
176 {
177 	struct sfc_adapter *sa = dev->data->dev_private;
178 
179 	return sa->dp_rx->supported_ptypes_get();
180 }
181 
182 static int
183 sfc_dev_configure(struct rte_eth_dev *dev)
184 {
185 	struct rte_eth_dev_data *dev_data = dev->data;
186 	struct sfc_adapter *sa = dev_data->dev_private;
187 	int rc;
188 
189 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
190 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
191 
192 	sfc_adapter_lock(sa);
193 	switch (sa->state) {
194 	case SFC_ADAPTER_CONFIGURED:
195 		/* FALLTHROUGH */
196 	case SFC_ADAPTER_INITIALIZED:
197 		rc = sfc_configure(sa);
198 		break;
199 	default:
200 		sfc_err(sa, "unexpected adapter state %u to configure",
201 			sa->state);
202 		rc = EINVAL;
203 		break;
204 	}
205 	sfc_adapter_unlock(sa);
206 
207 	sfc_log_init(sa, "done %d", rc);
208 	SFC_ASSERT(rc >= 0);
209 	return -rc;
210 }
211 
212 static int
213 sfc_dev_start(struct rte_eth_dev *dev)
214 {
215 	struct sfc_adapter *sa = dev->data->dev_private;
216 	int rc;
217 
218 	sfc_log_init(sa, "entry");
219 
220 	sfc_adapter_lock(sa);
221 	rc = sfc_start(sa);
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_link_update(struct rte_eth_dev *dev, int wait_to_complete)
231 {
232 	struct sfc_adapter *sa = dev->data->dev_private;
233 	struct rte_eth_link *dev_link = &dev->data->dev_link;
234 	struct rte_eth_link old_link;
235 	struct rte_eth_link current_link;
236 
237 	sfc_log_init(sa, "entry");
238 
239 retry:
240 	EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
241 	*(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
242 
243 	if (sa->state != SFC_ADAPTER_STARTED) {
244 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
245 		if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
246 					 *(uint64_t *)&old_link,
247 					 *(uint64_t *)&current_link))
248 			goto retry;
249 	} else if (wait_to_complete) {
250 		efx_link_mode_t link_mode;
251 
252 		if (efx_port_poll(sa->nic, &link_mode) != 0)
253 			link_mode = EFX_LINK_UNKNOWN;
254 		sfc_port_link_mode_to_info(link_mode, &current_link);
255 
256 		if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
257 					 *(uint64_t *)&old_link,
258 					 *(uint64_t *)&current_link))
259 			goto retry;
260 	} else {
261 		sfc_ev_mgmt_qpoll(sa);
262 		*(int64_t *)&current_link =
263 			rte_atomic64_read((rte_atomic64_t *)dev_link);
264 	}
265 
266 	if (old_link.link_status != current_link.link_status)
267 		sfc_info(sa, "Link status is %s",
268 			 current_link.link_status ? "UP" : "DOWN");
269 
270 	return old_link.link_status == current_link.link_status ? 0 : -1;
271 }
272 
273 static void
274 sfc_dev_stop(struct rte_eth_dev *dev)
275 {
276 	struct sfc_adapter *sa = dev->data->dev_private;
277 
278 	sfc_log_init(sa, "entry");
279 
280 	sfc_adapter_lock(sa);
281 	sfc_stop(sa);
282 	sfc_adapter_unlock(sa);
283 
284 	sfc_log_init(sa, "done");
285 }
286 
287 static int
288 sfc_dev_set_link_up(struct rte_eth_dev *dev)
289 {
290 	struct sfc_adapter *sa = dev->data->dev_private;
291 	int rc;
292 
293 	sfc_log_init(sa, "entry");
294 
295 	sfc_adapter_lock(sa);
296 	rc = sfc_start(sa);
297 	sfc_adapter_unlock(sa);
298 
299 	SFC_ASSERT(rc >= 0);
300 	return -rc;
301 }
302 
303 static int
304 sfc_dev_set_link_down(struct rte_eth_dev *dev)
305 {
306 	struct sfc_adapter *sa = dev->data->dev_private;
307 
308 	sfc_log_init(sa, "entry");
309 
310 	sfc_adapter_lock(sa);
311 	sfc_stop(sa);
312 	sfc_adapter_unlock(sa);
313 
314 	return 0;
315 }
316 
317 static void
318 sfc_dev_close(struct rte_eth_dev *dev)
319 {
320 	struct sfc_adapter *sa = dev->data->dev_private;
321 
322 	sfc_log_init(sa, "entry");
323 
324 	sfc_adapter_lock(sa);
325 	switch (sa->state) {
326 	case SFC_ADAPTER_STARTED:
327 		sfc_stop(sa);
328 		SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
329 		/* FALLTHROUGH */
330 	case SFC_ADAPTER_CONFIGURED:
331 		sfc_close(sa);
332 		SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
333 		/* FALLTHROUGH */
334 	case SFC_ADAPTER_INITIALIZED:
335 		break;
336 	default:
337 		sfc_err(sa, "unexpected adapter state %u on close", sa->state);
338 		break;
339 	}
340 	sfc_adapter_unlock(sa);
341 
342 	sfc_log_init(sa, "done");
343 }
344 
345 static void
346 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
347 		   boolean_t enabled)
348 {
349 	struct sfc_port *port;
350 	boolean_t *toggle;
351 	struct sfc_adapter *sa = dev->data->dev_private;
352 	boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
353 	const char *desc = (allmulti) ? "all-multi" : "promiscuous";
354 
355 	sfc_adapter_lock(sa);
356 
357 	port = &sa->port;
358 	toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
359 
360 	if (*toggle != enabled) {
361 		*toggle = enabled;
362 
363 		if ((sa->state == SFC_ADAPTER_STARTED) &&
364 		    (sfc_set_rx_mode(sa) != 0)) {
365 			*toggle = !(enabled);
366 			sfc_warn(sa, "Failed to %s %s mode",
367 				 ((enabled) ? "enable" : "disable"), desc);
368 		}
369 	}
370 
371 	sfc_adapter_unlock(sa);
372 }
373 
374 static void
375 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
376 {
377 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
378 }
379 
380 static void
381 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
382 {
383 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
384 }
385 
386 static void
387 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
388 {
389 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
390 }
391 
392 static void
393 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
394 {
395 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
396 }
397 
398 static int
399 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
400 		   uint16_t nb_rx_desc, unsigned int socket_id,
401 		   const struct rte_eth_rxconf *rx_conf,
402 		   struct rte_mempool *mb_pool)
403 {
404 	struct sfc_adapter *sa = dev->data->dev_private;
405 	int rc;
406 
407 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
408 		     rx_queue_id, nb_rx_desc, socket_id);
409 
410 	sfc_adapter_lock(sa);
411 
412 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
413 			  rx_conf, mb_pool);
414 	if (rc != 0)
415 		goto fail_rx_qinit;
416 
417 	dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
418 
419 	sfc_adapter_unlock(sa);
420 
421 	return 0;
422 
423 fail_rx_qinit:
424 	sfc_adapter_unlock(sa);
425 	SFC_ASSERT(rc > 0);
426 	return -rc;
427 }
428 
429 static void
430 sfc_rx_queue_release(void *queue)
431 {
432 	struct sfc_dp_rxq *dp_rxq = queue;
433 	struct sfc_rxq *rxq;
434 	struct sfc_adapter *sa;
435 	unsigned int sw_index;
436 
437 	if (dp_rxq == NULL)
438 		return;
439 
440 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
441 	sa = rxq->evq->sa;
442 	sfc_adapter_lock(sa);
443 
444 	sw_index = sfc_rxq_sw_index(rxq);
445 
446 	sfc_log_init(sa, "RxQ=%u", sw_index);
447 
448 	sa->eth_dev->data->rx_queues[sw_index] = NULL;
449 
450 	sfc_rx_qfini(sa, sw_index);
451 
452 	sfc_adapter_unlock(sa);
453 }
454 
455 static int
456 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
457 		   uint16_t nb_tx_desc, unsigned int socket_id,
458 		   const struct rte_eth_txconf *tx_conf)
459 {
460 	struct sfc_adapter *sa = dev->data->dev_private;
461 	int rc;
462 
463 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
464 		     tx_queue_id, nb_tx_desc, socket_id);
465 
466 	sfc_adapter_lock(sa);
467 
468 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
469 	if (rc != 0)
470 		goto fail_tx_qinit;
471 
472 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
473 
474 	sfc_adapter_unlock(sa);
475 	return 0;
476 
477 fail_tx_qinit:
478 	sfc_adapter_unlock(sa);
479 	SFC_ASSERT(rc > 0);
480 	return -rc;
481 }
482 
483 static void
484 sfc_tx_queue_release(void *queue)
485 {
486 	struct sfc_dp_txq *dp_txq = queue;
487 	struct sfc_txq *txq;
488 	unsigned int sw_index;
489 	struct sfc_adapter *sa;
490 
491 	if (dp_txq == NULL)
492 		return;
493 
494 	txq = sfc_txq_by_dp_txq(dp_txq);
495 	sw_index = sfc_txq_sw_index(txq);
496 
497 	SFC_ASSERT(txq->evq != NULL);
498 	sa = txq->evq->sa;
499 
500 	sfc_log_init(sa, "TxQ = %u", sw_index);
501 
502 	sfc_adapter_lock(sa);
503 
504 	SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
505 	sa->eth_dev->data->tx_queues[sw_index] = NULL;
506 
507 	sfc_tx_qfini(sa, sw_index);
508 
509 	sfc_adapter_unlock(sa);
510 }
511 
512 static void
513 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
514 {
515 	struct sfc_adapter *sa = dev->data->dev_private;
516 	struct sfc_port *port = &sa->port;
517 	uint64_t *mac_stats;
518 
519 	rte_spinlock_lock(&port->mac_stats_lock);
520 
521 	if (sfc_port_update_mac_stats(sa) != 0)
522 		goto unlock;
523 
524 	mac_stats = port->mac_stats_buf;
525 
526 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
527 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
528 		stats->ipackets =
529 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
530 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
531 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
532 		stats->opackets =
533 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
534 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
535 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
536 		stats->ibytes =
537 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
538 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
539 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
540 		stats->obytes =
541 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
542 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
543 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
544 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
545 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
546 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
547 	} else {
548 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
549 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
550 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
551 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
552 		/*
553 		 * Take into account stats which are whenever supported
554 		 * on EF10. If some stat is not supported by current
555 		 * firmware variant or HW revision, it is guaranteed
556 		 * to be zero in mac_stats.
557 		 */
558 		stats->imissed =
559 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
560 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
561 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
562 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
563 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
564 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
565 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
566 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
567 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
568 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
569 		stats->ierrors =
570 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
571 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
572 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
573 		/* no oerrors counters supported on EF10 */
574 	}
575 
576 unlock:
577 	rte_spinlock_unlock(&port->mac_stats_lock);
578 }
579 
580 static void
581 sfc_stats_reset(struct rte_eth_dev *dev)
582 {
583 	struct sfc_adapter *sa = dev->data->dev_private;
584 	struct sfc_port *port = &sa->port;
585 	int rc;
586 
587 	if (sa->state != SFC_ADAPTER_STARTED) {
588 		/*
589 		 * The operation cannot be done if port is not started; it
590 		 * will be scheduled to be done during the next port start
591 		 */
592 		port->mac_stats_reset_pending = B_TRUE;
593 		return;
594 	}
595 
596 	rc = sfc_port_reset_mac_stats(sa);
597 	if (rc != 0)
598 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
599 }
600 
601 static int
602 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
603 	       unsigned int xstats_count)
604 {
605 	struct sfc_adapter *sa = dev->data->dev_private;
606 	struct sfc_port *port = &sa->port;
607 	uint64_t *mac_stats;
608 	int rc;
609 	unsigned int i;
610 	int nstats = 0;
611 
612 	rte_spinlock_lock(&port->mac_stats_lock);
613 
614 	rc = sfc_port_update_mac_stats(sa);
615 	if (rc != 0) {
616 		SFC_ASSERT(rc > 0);
617 		nstats = -rc;
618 		goto unlock;
619 	}
620 
621 	mac_stats = port->mac_stats_buf;
622 
623 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
624 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
625 			if (xstats != NULL && nstats < (int)xstats_count) {
626 				xstats[nstats].id = nstats;
627 				xstats[nstats].value = mac_stats[i];
628 			}
629 			nstats++;
630 		}
631 	}
632 
633 unlock:
634 	rte_spinlock_unlock(&port->mac_stats_lock);
635 
636 	return nstats;
637 }
638 
639 static int
640 sfc_xstats_get_names(struct rte_eth_dev *dev,
641 		     struct rte_eth_xstat_name *xstats_names,
642 		     unsigned int xstats_count)
643 {
644 	struct sfc_adapter *sa = dev->data->dev_private;
645 	struct sfc_port *port = &sa->port;
646 	unsigned int i;
647 	unsigned int nstats = 0;
648 
649 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
650 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
651 			if (xstats_names != NULL && nstats < xstats_count)
652 				strncpy(xstats_names[nstats].name,
653 					efx_mac_stat_name(sa->nic, i),
654 					sizeof(xstats_names[0].name));
655 			nstats++;
656 		}
657 	}
658 
659 	return nstats;
660 }
661 
662 static int
663 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
664 {
665 	struct sfc_adapter *sa = dev->data->dev_private;
666 	unsigned int wanted_fc, link_fc;
667 
668 	memset(fc_conf, 0, sizeof(*fc_conf));
669 
670 	sfc_adapter_lock(sa);
671 
672 	if (sa->state == SFC_ADAPTER_STARTED)
673 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
674 	else
675 		link_fc = sa->port.flow_ctrl;
676 
677 	switch (link_fc) {
678 	case 0:
679 		fc_conf->mode = RTE_FC_NONE;
680 		break;
681 	case EFX_FCNTL_RESPOND:
682 		fc_conf->mode = RTE_FC_RX_PAUSE;
683 		break;
684 	case EFX_FCNTL_GENERATE:
685 		fc_conf->mode = RTE_FC_TX_PAUSE;
686 		break;
687 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
688 		fc_conf->mode = RTE_FC_FULL;
689 		break;
690 	default:
691 		sfc_err(sa, "%s: unexpected flow control value %#x",
692 			__func__, link_fc);
693 	}
694 
695 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
696 
697 	sfc_adapter_unlock(sa);
698 
699 	return 0;
700 }
701 
702 static int
703 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
704 {
705 	struct sfc_adapter *sa = dev->data->dev_private;
706 	struct sfc_port *port = &sa->port;
707 	unsigned int fcntl;
708 	int rc;
709 
710 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
711 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
712 	    fc_conf->mac_ctrl_frame_fwd != 0) {
713 		sfc_err(sa, "unsupported flow control settings specified");
714 		rc = EINVAL;
715 		goto fail_inval;
716 	}
717 
718 	switch (fc_conf->mode) {
719 	case RTE_FC_NONE:
720 		fcntl = 0;
721 		break;
722 	case RTE_FC_RX_PAUSE:
723 		fcntl = EFX_FCNTL_RESPOND;
724 		break;
725 	case RTE_FC_TX_PAUSE:
726 		fcntl = EFX_FCNTL_GENERATE;
727 		break;
728 	case RTE_FC_FULL:
729 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
730 		break;
731 	default:
732 		rc = EINVAL;
733 		goto fail_inval;
734 	}
735 
736 	sfc_adapter_lock(sa);
737 
738 	if (sa->state == SFC_ADAPTER_STARTED) {
739 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
740 		if (rc != 0)
741 			goto fail_mac_fcntl_set;
742 	}
743 
744 	port->flow_ctrl = fcntl;
745 	port->flow_ctrl_autoneg = fc_conf->autoneg;
746 
747 	sfc_adapter_unlock(sa);
748 
749 	return 0;
750 
751 fail_mac_fcntl_set:
752 	sfc_adapter_unlock(sa);
753 fail_inval:
754 	SFC_ASSERT(rc > 0);
755 	return -rc;
756 }
757 
758 static int
759 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
760 {
761 	struct sfc_adapter *sa = dev->data->dev_private;
762 	size_t pdu = EFX_MAC_PDU(mtu);
763 	size_t old_pdu;
764 	int rc;
765 
766 	sfc_log_init(sa, "mtu=%u", mtu);
767 
768 	rc = EINVAL;
769 	if (pdu < EFX_MAC_PDU_MIN) {
770 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
771 			(unsigned int)mtu, (unsigned int)pdu,
772 			EFX_MAC_PDU_MIN);
773 		goto fail_inval;
774 	}
775 	if (pdu > EFX_MAC_PDU_MAX) {
776 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
777 			(unsigned int)mtu, (unsigned int)pdu,
778 			EFX_MAC_PDU_MAX);
779 		goto fail_inval;
780 	}
781 
782 	sfc_adapter_lock(sa);
783 
784 	if (pdu != sa->port.pdu) {
785 		if (sa->state == SFC_ADAPTER_STARTED) {
786 			sfc_stop(sa);
787 
788 			old_pdu = sa->port.pdu;
789 			sa->port.pdu = pdu;
790 			rc = sfc_start(sa);
791 			if (rc != 0)
792 				goto fail_start;
793 		} else {
794 			sa->port.pdu = pdu;
795 		}
796 	}
797 
798 	/*
799 	 * The driver does not use it, but other PMDs update jumbo_frame
800 	 * flag and max_rx_pkt_len when MTU is set.
801 	 */
802 	dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
803 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
804 
805 	sfc_adapter_unlock(sa);
806 
807 	sfc_log_init(sa, "done");
808 	return 0;
809 
810 fail_start:
811 	sa->port.pdu = old_pdu;
812 	if (sfc_start(sa) != 0)
813 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
814 			"PDU max size - port is stopped",
815 			(unsigned int)pdu, (unsigned int)old_pdu);
816 	sfc_adapter_unlock(sa);
817 
818 fail_inval:
819 	sfc_log_init(sa, "failed %d", rc);
820 	SFC_ASSERT(rc > 0);
821 	return -rc;
822 }
823 static void
824 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
825 {
826 	struct sfc_adapter *sa = dev->data->dev_private;
827 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
828 	int rc;
829 
830 	sfc_adapter_lock(sa);
831 
832 	if (sa->state != SFC_ADAPTER_STARTED) {
833 		sfc_info(sa, "the port is not started");
834 		sfc_info(sa, "the new MAC address will be set on port start");
835 
836 		goto unlock;
837 	}
838 
839 	if (encp->enc_allow_set_mac_with_installed_filters) {
840 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
841 		if (rc != 0) {
842 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
843 			goto unlock;
844 		}
845 
846 		/*
847 		 * Changing the MAC address by means of MCDI request
848 		 * has no effect on received traffic, therefore
849 		 * we also need to update unicast filters
850 		 */
851 		rc = sfc_set_rx_mode(sa);
852 		if (rc != 0)
853 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
854 	} else {
855 		sfc_warn(sa, "cannot set MAC address with filters installed");
856 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
857 		sfc_warn(sa, "(some traffic may be dropped)");
858 
859 		/*
860 		 * Since setting MAC address with filters installed is not
861 		 * allowed on the adapter, one needs to simply restart adapter
862 		 * so that the new MAC address will be taken from an outer
863 		 * storage and set flawlessly by means of sfc_start() call
864 		 */
865 		sfc_stop(sa);
866 		rc = sfc_start(sa);
867 		if (rc != 0)
868 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
869 	}
870 
871 unlock:
872 	sfc_adapter_unlock(sa);
873 }
874 
875 
876 static int
877 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
878 		     uint32_t nb_mc_addr)
879 {
880 	struct sfc_adapter *sa = dev->data->dev_private;
881 	struct sfc_port *port = &sa->port;
882 	uint8_t *mc_addrs = port->mcast_addrs;
883 	int rc;
884 	unsigned int i;
885 
886 	if (mc_addrs == NULL)
887 		return -ENOBUFS;
888 
889 	if (nb_mc_addr > port->max_mcast_addrs) {
890 		sfc_err(sa, "too many multicast addresses: %u > %u",
891 			 nb_mc_addr, port->max_mcast_addrs);
892 		return -EINVAL;
893 	}
894 
895 	for (i = 0; i < nb_mc_addr; ++i) {
896 		(void)rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
897 				 EFX_MAC_ADDR_LEN);
898 		mc_addrs += EFX_MAC_ADDR_LEN;
899 	}
900 
901 	port->nb_mcast_addrs = nb_mc_addr;
902 
903 	if (sa->state != SFC_ADAPTER_STARTED)
904 		return 0;
905 
906 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
907 					port->nb_mcast_addrs);
908 	if (rc != 0)
909 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
910 
911 	SFC_ASSERT(rc > 0);
912 	return -rc;
913 }
914 
915 static void
916 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
917 		      struct rte_eth_rxq_info *qinfo)
918 {
919 	struct sfc_adapter *sa = dev->data->dev_private;
920 	struct sfc_rxq_info *rxq_info;
921 	struct sfc_rxq *rxq;
922 
923 	sfc_adapter_lock(sa);
924 
925 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
926 
927 	rxq_info = &sa->rxq_info[rx_queue_id];
928 	rxq = rxq_info->rxq;
929 	SFC_ASSERT(rxq != NULL);
930 
931 	qinfo->mp = rxq->refill_mb_pool;
932 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
933 	qinfo->conf.rx_drop_en = 1;
934 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
935 	qinfo->scattered_rx = (rxq_info->type == EFX_RXQ_TYPE_SCATTER);
936 	qinfo->nb_desc = rxq_info->entries;
937 
938 	sfc_adapter_unlock(sa);
939 }
940 
941 static void
942 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
943 		      struct rte_eth_txq_info *qinfo)
944 {
945 	struct sfc_adapter *sa = dev->data->dev_private;
946 	struct sfc_txq_info *txq_info;
947 
948 	sfc_adapter_lock(sa);
949 
950 	SFC_ASSERT(tx_queue_id < sa->txq_count);
951 
952 	txq_info = &sa->txq_info[tx_queue_id];
953 	SFC_ASSERT(txq_info->txq != NULL);
954 
955 	memset(qinfo, 0, sizeof(*qinfo));
956 
957 	qinfo->conf.txq_flags = txq_info->txq->flags;
958 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
959 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
960 	qinfo->nb_desc = txq_info->entries;
961 
962 	sfc_adapter_unlock(sa);
963 }
964 
965 static uint32_t
966 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
967 {
968 	struct sfc_adapter *sa = dev->data->dev_private;
969 
970 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
971 
972 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
973 }
974 
975 static int
976 sfc_rx_descriptor_done(void *queue, uint16_t offset)
977 {
978 	struct sfc_dp_rxq *dp_rxq = queue;
979 
980 	return sfc_rx_qdesc_done(dp_rxq, offset);
981 }
982 
983 static int
984 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
985 {
986 	struct sfc_adapter *sa = dev->data->dev_private;
987 	int rc;
988 
989 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
990 
991 	sfc_adapter_lock(sa);
992 
993 	rc = EINVAL;
994 	if (sa->state != SFC_ADAPTER_STARTED)
995 		goto fail_not_started;
996 
997 	rc = sfc_rx_qstart(sa, rx_queue_id);
998 	if (rc != 0)
999 		goto fail_rx_qstart;
1000 
1001 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1002 
1003 	sfc_adapter_unlock(sa);
1004 
1005 	return 0;
1006 
1007 fail_rx_qstart:
1008 fail_not_started:
1009 	sfc_adapter_unlock(sa);
1010 	SFC_ASSERT(rc > 0);
1011 	return -rc;
1012 }
1013 
1014 static int
1015 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1016 {
1017 	struct sfc_adapter *sa = dev->data->dev_private;
1018 
1019 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1020 
1021 	sfc_adapter_lock(sa);
1022 	sfc_rx_qstop(sa, rx_queue_id);
1023 
1024 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1025 
1026 	sfc_adapter_unlock(sa);
1027 
1028 	return 0;
1029 }
1030 
1031 static int
1032 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1033 {
1034 	struct sfc_adapter *sa = dev->data->dev_private;
1035 	int rc;
1036 
1037 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1038 
1039 	sfc_adapter_lock(sa);
1040 
1041 	rc = EINVAL;
1042 	if (sa->state != SFC_ADAPTER_STARTED)
1043 		goto fail_not_started;
1044 
1045 	rc = sfc_tx_qstart(sa, tx_queue_id);
1046 	if (rc != 0)
1047 		goto fail_tx_qstart;
1048 
1049 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1050 
1051 	sfc_adapter_unlock(sa);
1052 	return 0;
1053 
1054 fail_tx_qstart:
1055 
1056 fail_not_started:
1057 	sfc_adapter_unlock(sa);
1058 	SFC_ASSERT(rc > 0);
1059 	return -rc;
1060 }
1061 
1062 static int
1063 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1064 {
1065 	struct sfc_adapter *sa = dev->data->dev_private;
1066 
1067 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1068 
1069 	sfc_adapter_lock(sa);
1070 
1071 	sfc_tx_qstop(sa, tx_queue_id);
1072 
1073 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1074 
1075 	sfc_adapter_unlock(sa);
1076 	return 0;
1077 }
1078 
1079 #if EFSYS_OPT_RX_SCALE
1080 static int
1081 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1082 			  struct rte_eth_rss_conf *rss_conf)
1083 {
1084 	struct sfc_adapter *sa = dev->data->dev_private;
1085 
1086 	if ((sa->rss_channels == 1) ||
1087 	    (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE))
1088 		return -ENOTSUP;
1089 
1090 	sfc_adapter_lock(sa);
1091 
1092 	/*
1093 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1094 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1095 	 * flags which corresponds to the active EFX configuration stored
1096 	 * locally in 'sfc_adapter' and kept up-to-date
1097 	 */
1098 	rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1099 	rss_conf->rss_key_len = SFC_RSS_KEY_SIZE;
1100 	if (rss_conf->rss_key != NULL)
1101 		rte_memcpy(rss_conf->rss_key, sa->rss_key, SFC_RSS_KEY_SIZE);
1102 
1103 	sfc_adapter_unlock(sa);
1104 
1105 	return 0;
1106 }
1107 
1108 static int
1109 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1110 			struct rte_eth_rss_conf *rss_conf)
1111 {
1112 	struct sfc_adapter *sa = dev->data->dev_private;
1113 	unsigned int efx_hash_types;
1114 	int rc = 0;
1115 
1116 	if ((sa->rss_channels == 1) ||
1117 	    (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE)) {
1118 		sfc_err(sa, "RSS is not available");
1119 		return -ENOTSUP;
1120 	}
1121 
1122 	if ((rss_conf->rss_key != NULL) &&
1123 	    (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1124 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1125 			sizeof(sa->rss_key));
1126 		return -EINVAL;
1127 	}
1128 
1129 	if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1130 		sfc_err(sa, "unsupported hash functions requested");
1131 		return -EINVAL;
1132 	}
1133 
1134 	sfc_adapter_lock(sa);
1135 
1136 	efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1137 
1138 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RX_HASHALG_TOEPLITZ,
1139 				   efx_hash_types, B_TRUE);
1140 	if (rc != 0)
1141 		goto fail_scale_mode_set;
1142 
1143 	if (rss_conf->rss_key != NULL) {
1144 		if (sa->state == SFC_ADAPTER_STARTED) {
1145 			rc = efx_rx_scale_key_set(sa->nic, rss_conf->rss_key,
1146 						  sizeof(sa->rss_key));
1147 			if (rc != 0)
1148 				goto fail_scale_key_set;
1149 		}
1150 
1151 		rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1152 	}
1153 
1154 	sa->rss_hash_types = efx_hash_types;
1155 
1156 	sfc_adapter_unlock(sa);
1157 
1158 	return 0;
1159 
1160 fail_scale_key_set:
1161 	if (efx_rx_scale_mode_set(sa->nic, EFX_RX_HASHALG_TOEPLITZ,
1162 				  sa->rss_hash_types, B_TRUE) != 0)
1163 		sfc_err(sa, "failed to restore RSS mode");
1164 
1165 fail_scale_mode_set:
1166 	sfc_adapter_unlock(sa);
1167 	return -rc;
1168 }
1169 
1170 static int
1171 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1172 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1173 		       uint16_t reta_size)
1174 {
1175 	struct sfc_adapter *sa = dev->data->dev_private;
1176 	int entry;
1177 
1178 	if ((sa->rss_channels == 1) ||
1179 	    (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE))
1180 		return -ENOTSUP;
1181 
1182 	if (reta_size != EFX_RSS_TBL_SIZE)
1183 		return -EINVAL;
1184 
1185 	sfc_adapter_lock(sa);
1186 
1187 	for (entry = 0; entry < reta_size; entry++) {
1188 		int grp = entry / RTE_RETA_GROUP_SIZE;
1189 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1190 
1191 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1192 			reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1193 	}
1194 
1195 	sfc_adapter_unlock(sa);
1196 
1197 	return 0;
1198 }
1199 
1200 static int
1201 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1202 			struct rte_eth_rss_reta_entry64 *reta_conf,
1203 			uint16_t reta_size)
1204 {
1205 	struct sfc_adapter *sa = dev->data->dev_private;
1206 	unsigned int *rss_tbl_new;
1207 	uint16_t entry;
1208 	int rc;
1209 
1210 
1211 	if ((sa->rss_channels == 1) ||
1212 	    (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE)) {
1213 		sfc_err(sa, "RSS is not available");
1214 		return -ENOTSUP;
1215 	}
1216 
1217 	if (reta_size != EFX_RSS_TBL_SIZE) {
1218 		sfc_err(sa, "RETA size is wrong (should be %u)",
1219 			EFX_RSS_TBL_SIZE);
1220 		return -EINVAL;
1221 	}
1222 
1223 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1224 	if (rss_tbl_new == NULL)
1225 		return -ENOMEM;
1226 
1227 	sfc_adapter_lock(sa);
1228 
1229 	rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1230 
1231 	for (entry = 0; entry < reta_size; entry++) {
1232 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1233 		struct rte_eth_rss_reta_entry64 *grp;
1234 
1235 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1236 
1237 		if (grp->mask & (1ull << grp_idx)) {
1238 			if (grp->reta[grp_idx] >= sa->rss_channels) {
1239 				rc = EINVAL;
1240 				goto bad_reta_entry;
1241 			}
1242 			rss_tbl_new[entry] = grp->reta[grp_idx];
1243 		}
1244 	}
1245 
1246 	rc = efx_rx_scale_tbl_set(sa->nic, rss_tbl_new, EFX_RSS_TBL_SIZE);
1247 	if (rc == 0)
1248 		rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1249 
1250 bad_reta_entry:
1251 	sfc_adapter_unlock(sa);
1252 
1253 	rte_free(rss_tbl_new);
1254 
1255 	SFC_ASSERT(rc >= 0);
1256 	return -rc;
1257 }
1258 #endif
1259 
1260 static int
1261 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1262 		    enum rte_filter_op filter_op,
1263 		    void *arg)
1264 {
1265 	struct sfc_adapter *sa = dev->data->dev_private;
1266 	int rc = ENOTSUP;
1267 
1268 	sfc_log_init(sa, "entry");
1269 
1270 	switch (filter_type) {
1271 	case RTE_ETH_FILTER_NONE:
1272 		sfc_err(sa, "Global filters configuration not supported");
1273 		break;
1274 	case RTE_ETH_FILTER_MACVLAN:
1275 		sfc_err(sa, "MACVLAN filters not supported");
1276 		break;
1277 	case RTE_ETH_FILTER_ETHERTYPE:
1278 		sfc_err(sa, "EtherType filters not supported");
1279 		break;
1280 	case RTE_ETH_FILTER_FLEXIBLE:
1281 		sfc_err(sa, "Flexible filters not supported");
1282 		break;
1283 	case RTE_ETH_FILTER_SYN:
1284 		sfc_err(sa, "SYN filters not supported");
1285 		break;
1286 	case RTE_ETH_FILTER_NTUPLE:
1287 		sfc_err(sa, "NTUPLE filters not supported");
1288 		break;
1289 	case RTE_ETH_FILTER_TUNNEL:
1290 		sfc_err(sa, "Tunnel filters not supported");
1291 		break;
1292 	case RTE_ETH_FILTER_FDIR:
1293 		sfc_err(sa, "Flow Director filters not supported");
1294 		break;
1295 	case RTE_ETH_FILTER_HASH:
1296 		sfc_err(sa, "Hash filters not supported");
1297 		break;
1298 	case RTE_ETH_FILTER_GENERIC:
1299 		if (filter_op != RTE_ETH_FILTER_GET) {
1300 			rc = EINVAL;
1301 		} else {
1302 			*(const void **)arg = &sfc_flow_ops;
1303 			rc = 0;
1304 		}
1305 		break;
1306 	default:
1307 		sfc_err(sa, "Unknown filter type %u", filter_type);
1308 		break;
1309 	}
1310 
1311 	sfc_log_init(sa, "exit: %d", -rc);
1312 	SFC_ASSERT(rc >= 0);
1313 	return -rc;
1314 }
1315 
1316 static const struct eth_dev_ops sfc_eth_dev_ops = {
1317 	.dev_configure			= sfc_dev_configure,
1318 	.dev_start			= sfc_dev_start,
1319 	.dev_stop			= sfc_dev_stop,
1320 	.dev_set_link_up		= sfc_dev_set_link_up,
1321 	.dev_set_link_down		= sfc_dev_set_link_down,
1322 	.dev_close			= sfc_dev_close,
1323 	.promiscuous_enable		= sfc_dev_promisc_enable,
1324 	.promiscuous_disable		= sfc_dev_promisc_disable,
1325 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1326 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1327 	.link_update			= sfc_dev_link_update,
1328 	.stats_get			= sfc_stats_get,
1329 	.stats_reset			= sfc_stats_reset,
1330 	.xstats_get			= sfc_xstats_get,
1331 	.xstats_reset			= sfc_stats_reset,
1332 	.xstats_get_names		= sfc_xstats_get_names,
1333 	.dev_infos_get			= sfc_dev_infos_get,
1334 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1335 	.mtu_set			= sfc_dev_set_mtu,
1336 	.rx_queue_start			= sfc_rx_queue_start,
1337 	.rx_queue_stop			= sfc_rx_queue_stop,
1338 	.tx_queue_start			= sfc_tx_queue_start,
1339 	.tx_queue_stop			= sfc_tx_queue_stop,
1340 	.rx_queue_setup			= sfc_rx_queue_setup,
1341 	.rx_queue_release		= sfc_rx_queue_release,
1342 	.rx_queue_count			= sfc_rx_queue_count,
1343 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1344 	.tx_queue_setup			= sfc_tx_queue_setup,
1345 	.tx_queue_release		= sfc_tx_queue_release,
1346 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1347 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1348 	.mac_addr_set			= sfc_mac_addr_set,
1349 #if EFSYS_OPT_RX_SCALE
1350 	.reta_update			= sfc_dev_rss_reta_update,
1351 	.reta_query			= sfc_dev_rss_reta_query,
1352 	.rss_hash_update		= sfc_dev_rss_hash_update,
1353 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1354 #endif
1355 	.filter_ctrl			= sfc_dev_filter_ctrl,
1356 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1357 	.rxq_info_get			= sfc_rx_queue_info_get,
1358 	.txq_info_get			= sfc_tx_queue_info_get,
1359 	.fw_version_get			= sfc_fw_version_get,
1360 };
1361 
1362 static int
1363 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1364 {
1365 	struct sfc_adapter *sa = dev->data->dev_private;
1366 	unsigned int avail_caps = 0;
1367 	const char *rx_name = NULL;
1368 	const char *tx_name = NULL;
1369 	int rc;
1370 
1371 	switch (sa->family) {
1372 	case EFX_FAMILY_HUNTINGTON:
1373 	case EFX_FAMILY_MEDFORD:
1374 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1375 		break;
1376 	default:
1377 		break;
1378 	}
1379 
1380 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1381 				sfc_kvarg_string_handler, &rx_name);
1382 	if (rc != 0)
1383 		goto fail_kvarg_rx_datapath;
1384 
1385 	if (rx_name != NULL) {
1386 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1387 		if (sa->dp_rx == NULL) {
1388 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1389 			rc = ENOENT;
1390 			goto fail_dp_rx;
1391 		}
1392 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1393 			sfc_err(sa,
1394 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1395 				rx_name);
1396 			rc = EINVAL;
1397 			goto fail_dp_rx;
1398 		}
1399 	} else {
1400 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1401 		if (sa->dp_rx == NULL) {
1402 			sfc_err(sa, "Rx datapath by caps %#x not found",
1403 				avail_caps);
1404 			rc = ENOENT;
1405 			goto fail_dp_rx;
1406 		}
1407 	}
1408 
1409 	sfc_info(sa, "use %s Rx datapath", sa->dp_rx->dp.name);
1410 
1411 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1412 
1413 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1414 				sfc_kvarg_string_handler, &tx_name);
1415 	if (rc != 0)
1416 		goto fail_kvarg_tx_datapath;
1417 
1418 	if (tx_name != NULL) {
1419 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1420 		if (sa->dp_tx == NULL) {
1421 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1422 			rc = ENOENT;
1423 			goto fail_dp_tx;
1424 		}
1425 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1426 			sfc_err(sa,
1427 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1428 				tx_name);
1429 			rc = EINVAL;
1430 			goto fail_dp_tx;
1431 		}
1432 	} else {
1433 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1434 		if (sa->dp_tx == NULL) {
1435 			sfc_err(sa, "Tx datapath by caps %#x not found",
1436 				avail_caps);
1437 			rc = ENOENT;
1438 			goto fail_dp_tx;
1439 		}
1440 	}
1441 
1442 	sfc_info(sa, "use %s Tx datapath", sa->dp_tx->dp.name);
1443 
1444 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1445 
1446 	dev->dev_ops = &sfc_eth_dev_ops;
1447 
1448 	return 0;
1449 
1450 fail_dp_tx:
1451 fail_kvarg_tx_datapath:
1452 fail_dp_rx:
1453 fail_kvarg_rx_datapath:
1454 	return rc;
1455 }
1456 
1457 static void
1458 sfc_register_dp(void)
1459 {
1460 	/* Register once */
1461 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1462 		/* Prefer EF10 datapath */
1463 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1464 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1465 
1466 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1467 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1468 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1469 	}
1470 }
1471 
1472 static int
1473 sfc_eth_dev_init(struct rte_eth_dev *dev)
1474 {
1475 	struct sfc_adapter *sa = dev->data->dev_private;
1476 	struct rte_pci_device *pci_dev = SFC_DEV_TO_PCI(dev);
1477 	int rc;
1478 	const efx_nic_cfg_t *encp;
1479 	const struct ether_addr *from;
1480 
1481 	sfc_register_dp();
1482 
1483 	/* Required for logging */
1484 	sa->eth_dev = dev;
1485 
1486 	/* Copy PCI device info to the dev->data */
1487 	rte_eth_copy_pci_info(dev, pci_dev);
1488 
1489 	rc = sfc_kvargs_parse(sa);
1490 	if (rc != 0)
1491 		goto fail_kvargs_parse;
1492 
1493 	rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1494 				sfc_kvarg_bool_handler, &sa->debug_init);
1495 	if (rc != 0)
1496 		goto fail_kvarg_debug_init;
1497 
1498 	sfc_log_init(sa, "entry");
1499 
1500 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1501 	if (dev->data->mac_addrs == NULL) {
1502 		rc = ENOMEM;
1503 		goto fail_mac_addrs;
1504 	}
1505 
1506 	sfc_adapter_lock_init(sa);
1507 	sfc_adapter_lock(sa);
1508 
1509 	sfc_log_init(sa, "probing");
1510 	rc = sfc_probe(sa);
1511 	if (rc != 0)
1512 		goto fail_probe;
1513 
1514 	sfc_log_init(sa, "set device ops");
1515 	rc = sfc_eth_dev_set_ops(dev);
1516 	if (rc != 0)
1517 		goto fail_set_ops;
1518 
1519 	sfc_log_init(sa, "attaching");
1520 	rc = sfc_attach(sa);
1521 	if (rc != 0)
1522 		goto fail_attach;
1523 
1524 	encp = efx_nic_cfg_get(sa->nic);
1525 
1526 	/*
1527 	 * The arguments are really reverse order in comparison to
1528 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1529 	 */
1530 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1531 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1532 
1533 	sfc_adapter_unlock(sa);
1534 
1535 	sfc_log_init(sa, "done");
1536 	return 0;
1537 
1538 fail_attach:
1539 fail_set_ops:
1540 	sfc_unprobe(sa);
1541 
1542 fail_probe:
1543 	sfc_adapter_unlock(sa);
1544 	sfc_adapter_lock_fini(sa);
1545 	rte_free(dev->data->mac_addrs);
1546 	dev->data->mac_addrs = NULL;
1547 
1548 fail_mac_addrs:
1549 fail_kvarg_debug_init:
1550 	sfc_kvargs_cleanup(sa);
1551 
1552 fail_kvargs_parse:
1553 	sfc_log_init(sa, "failed %d", rc);
1554 	SFC_ASSERT(rc > 0);
1555 	return -rc;
1556 }
1557 
1558 static int
1559 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1560 {
1561 	struct sfc_adapter *sa = dev->data->dev_private;
1562 
1563 	sfc_log_init(sa, "entry");
1564 
1565 	sfc_adapter_lock(sa);
1566 
1567 	sfc_detach(sa);
1568 	sfc_unprobe(sa);
1569 
1570 	rte_free(dev->data->mac_addrs);
1571 	dev->data->mac_addrs = NULL;
1572 
1573 	dev->dev_ops = NULL;
1574 	dev->rx_pkt_burst = NULL;
1575 	dev->tx_pkt_burst = NULL;
1576 
1577 	sfc_kvargs_cleanup(sa);
1578 
1579 	sfc_adapter_unlock(sa);
1580 	sfc_adapter_lock_fini(sa);
1581 
1582 	sfc_log_init(sa, "done");
1583 
1584 	/* Required for logging, so cleanup last */
1585 	sa->eth_dev = NULL;
1586 	return 0;
1587 }
1588 
1589 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
1590 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
1591 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
1592 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
1593 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
1594 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
1595 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
1596 	{ .vendor_id = 0 /* sentinel */ }
1597 };
1598 
1599 static struct eth_driver sfc_efx_pmd = {
1600 	.pci_drv = {
1601 		.id_table = pci_id_sfc_efx_map,
1602 		.drv_flags =
1603 			RTE_PCI_DRV_INTR_LSC |
1604 			RTE_PCI_DRV_NEED_MAPPING,
1605 		.probe = rte_eth_dev_pci_probe,
1606 		.remove = rte_eth_dev_pci_remove,
1607 	},
1608 	.eth_dev_init = sfc_eth_dev_init,
1609 	.eth_dev_uninit = sfc_eth_dev_uninit,
1610 	.dev_private_size = sizeof(struct sfc_adapter),
1611 };
1612 
1613 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd.pci_drv);
1614 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
1615 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio");
1616 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
1617 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
1618 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
1619 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
1620 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
1621 	SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
1622 	SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);
1623