xref: /dpdk/drivers/net/sfc/sfc_ethdev.c (revision cd8c7c7ce241d2ea7c059a9df07caa9411ef19ed)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
4  * All rights reserved.
5  *
6  * This software was jointly developed between OKTET Labs (under contract
7  * for Solarflare) and Solarflare Communications, Inc.
8  */
9 
10 #include <rte_dev.h>
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
16 
17 #include "efx.h"
18 
19 #include "sfc.h"
20 #include "sfc_debug.h"
21 #include "sfc_log.h"
22 #include "sfc_kvargs.h"
23 #include "sfc_ev.h"
24 #include "sfc_rx.h"
25 #include "sfc_tx.h"
26 #include "sfc_flow.h"
27 #include "sfc_dp.h"
28 #include "sfc_dp_rx.h"
29 
30 uint32_t sfc_logtype_driver;
31 
32 static struct sfc_dp_list sfc_dp_head =
33 	TAILQ_HEAD_INITIALIZER(sfc_dp_head);
34 
35 static int
36 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
37 {
38 	struct sfc_adapter *sa = dev->data->dev_private;
39 	efx_nic_fw_info_t enfi;
40 	int ret;
41 	int rc;
42 
43 	/*
44 	 * Return value of the callback is likely supposed to be
45 	 * equal to or greater than 0, nevertheless, if an error
46 	 * occurs, it will be desirable to pass it to the caller
47 	 */
48 	if ((fw_version == NULL) || (fw_size == 0))
49 		return -EINVAL;
50 
51 	rc = efx_nic_get_fw_version(sa->nic, &enfi);
52 	if (rc != 0)
53 		return -rc;
54 
55 	ret = snprintf(fw_version, fw_size,
56 		       "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
57 		       enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
58 		       enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
59 	if (ret < 0)
60 		return ret;
61 
62 	if (enfi.enfi_dpcpu_fw_ids_valid) {
63 		size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
64 		int ret_extra;
65 
66 		ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
67 				     fw_size - dpcpu_fw_ids_offset,
68 				     " rx%" PRIx16 " tx%" PRIx16,
69 				     enfi.enfi_rx_dpcpu_fw_id,
70 				     enfi.enfi_tx_dpcpu_fw_id);
71 		if (ret_extra < 0)
72 			return ret_extra;
73 
74 		ret += ret_extra;
75 	}
76 
77 	if (fw_size < (size_t)(++ret))
78 		return ret;
79 	else
80 		return 0;
81 }
82 
83 static void
84 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
85 {
86 	struct sfc_adapter *sa = dev->data->dev_private;
87 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
88 	uint64_t txq_offloads_def = 0;
89 
90 	sfc_log_init(sa, "entry");
91 
92 	dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
93 
94 	/* Autonegotiation may be disabled */
95 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
96 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
97 		dev_info->speed_capa |= ETH_LINK_SPEED_1G;
98 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
99 		dev_info->speed_capa |= ETH_LINK_SPEED_10G;
100 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_25000FDX)
101 		dev_info->speed_capa |= ETH_LINK_SPEED_25G;
102 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
103 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
104 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_50000FDX)
105 		dev_info->speed_capa |= ETH_LINK_SPEED_50G;
106 	if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_100000FDX)
107 		dev_info->speed_capa |= ETH_LINK_SPEED_100G;
108 
109 	dev_info->max_rx_queues = sa->rxq_max;
110 	dev_info->max_tx_queues = sa->txq_max;
111 
112 	/* By default packets are dropped if no descriptors are available */
113 	dev_info->default_rxconf.rx_drop_en = 1;
114 
115 	dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
116 
117 	/*
118 	 * rx_offload_capa includes both device and queue offloads since
119 	 * the latter may be requested on a per device basis which makes
120 	 * sense when some offloads are needed to be set on all queues.
121 	 */
122 	dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
123 				    dev_info->rx_queue_offload_capa;
124 
125 	dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
126 
127 	/*
128 	 * tx_offload_capa includes both device and queue offloads since
129 	 * the latter may be requested on a per device basis which makes
130 	 * sense when some offloads are needed to be set on all queues.
131 	 */
132 	dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
133 				    dev_info->tx_queue_offload_capa;
134 
135 	if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
136 		txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
137 
138 	dev_info->default_txconf.offloads |= txq_offloads_def;
139 
140 	dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
141 	if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
142 	    !encp->enc_hw_tx_insert_vlan_enabled)
143 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
144 
145 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
146 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
147 
148 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
149 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
150 
151 	if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
152 		dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
153 
154 #if EFSYS_OPT_RX_SCALE
155 	if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
156 		dev_info->reta_size = EFX_RSS_TBL_SIZE;
157 		dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
158 		dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
159 	}
160 #endif
161 
162 	/* Initialize to hardware limits */
163 	dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
164 	dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
165 	/* The RXQ hardware requires that the descriptor count is a power
166 	 * of 2, but rx_desc_lim cannot properly describe that constraint.
167 	 */
168 	dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
169 
170 	/* Initialize to hardware limits */
171 	dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
172 	dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
173 	/*
174 	 * The TXQ hardware requires that the descriptor count is a power
175 	 * of 2, but tx_desc_lim cannot properly describe that constraint
176 	 */
177 	dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
178 
179 	if (sa->dp_rx->get_dev_info != NULL)
180 		sa->dp_rx->get_dev_info(dev_info);
181 	if (sa->dp_tx->get_dev_info != NULL)
182 		sa->dp_tx->get_dev_info(dev_info);
183 }
184 
185 static const uint32_t *
186 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
187 {
188 	struct sfc_adapter *sa = dev->data->dev_private;
189 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
190 	uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
191 
192 	return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
193 }
194 
195 static int
196 sfc_dev_configure(struct rte_eth_dev *dev)
197 {
198 	struct rte_eth_dev_data *dev_data = dev->data;
199 	struct sfc_adapter *sa = dev_data->dev_private;
200 	int rc;
201 
202 	sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
203 		     dev_data->nb_rx_queues, dev_data->nb_tx_queues);
204 
205 	sfc_adapter_lock(sa);
206 	switch (sa->state) {
207 	case SFC_ADAPTER_CONFIGURED:
208 		/* FALLTHROUGH */
209 	case SFC_ADAPTER_INITIALIZED:
210 		rc = sfc_configure(sa);
211 		break;
212 	default:
213 		sfc_err(sa, "unexpected adapter state %u to configure",
214 			sa->state);
215 		rc = EINVAL;
216 		break;
217 	}
218 	sfc_adapter_unlock(sa);
219 
220 	sfc_log_init(sa, "done %d", rc);
221 	SFC_ASSERT(rc >= 0);
222 	return -rc;
223 }
224 
225 static int
226 sfc_dev_start(struct rte_eth_dev *dev)
227 {
228 	struct sfc_adapter *sa = dev->data->dev_private;
229 	int rc;
230 
231 	sfc_log_init(sa, "entry");
232 
233 	sfc_adapter_lock(sa);
234 	rc = sfc_start(sa);
235 	sfc_adapter_unlock(sa);
236 
237 	sfc_log_init(sa, "done %d", rc);
238 	SFC_ASSERT(rc >= 0);
239 	return -rc;
240 }
241 
242 static int
243 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
244 {
245 	struct sfc_adapter *sa = dev->data->dev_private;
246 	struct rte_eth_link current_link;
247 	int ret;
248 
249 	sfc_log_init(sa, "entry");
250 
251 	if (sa->state != SFC_ADAPTER_STARTED) {
252 		sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
253 	} else if (wait_to_complete) {
254 		efx_link_mode_t link_mode;
255 
256 		if (efx_port_poll(sa->nic, &link_mode) != 0)
257 			link_mode = EFX_LINK_UNKNOWN;
258 		sfc_port_link_mode_to_info(link_mode, &current_link);
259 
260 	} else {
261 		sfc_ev_mgmt_qpoll(sa);
262 		rte_eth_linkstatus_get(dev, &current_link);
263 	}
264 
265 	ret = rte_eth_linkstatus_set(dev, &current_link);
266 	if (ret == 0)
267 		sfc_notice(sa, "Link status is %s",
268 			   current_link.link_status ? "UP" : "DOWN");
269 
270 	return ret;
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 (port->isolated) {
364 			sfc_warn(sa, "isolated mode is active on the port");
365 			sfc_warn(sa, "the change is to be applied on the next "
366 				     "start provided that isolated mode is "
367 				     "disabled prior the next start");
368 		} else if ((sa->state == SFC_ADAPTER_STARTED) &&
369 			   (sfc_set_rx_mode(sa) != 0)) {
370 			*toggle = !(enabled);
371 			sfc_warn(sa, "Failed to %s %s mode",
372 				 ((enabled) ? "enable" : "disable"), desc);
373 		}
374 	}
375 
376 	sfc_adapter_unlock(sa);
377 }
378 
379 static void
380 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
381 {
382 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
383 }
384 
385 static void
386 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
387 {
388 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
389 }
390 
391 static void
392 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
393 {
394 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
395 }
396 
397 static void
398 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
399 {
400 	sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
401 }
402 
403 static int
404 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
405 		   uint16_t nb_rx_desc, unsigned int socket_id,
406 		   const struct rte_eth_rxconf *rx_conf,
407 		   struct rte_mempool *mb_pool)
408 {
409 	struct sfc_adapter *sa = dev->data->dev_private;
410 	int rc;
411 
412 	sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
413 		     rx_queue_id, nb_rx_desc, socket_id);
414 
415 	sfc_adapter_lock(sa);
416 
417 	rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
418 			  rx_conf, mb_pool);
419 	if (rc != 0)
420 		goto fail_rx_qinit;
421 
422 	dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
423 
424 	sfc_adapter_unlock(sa);
425 
426 	return 0;
427 
428 fail_rx_qinit:
429 	sfc_adapter_unlock(sa);
430 	SFC_ASSERT(rc > 0);
431 	return -rc;
432 }
433 
434 static void
435 sfc_rx_queue_release(void *queue)
436 {
437 	struct sfc_dp_rxq *dp_rxq = queue;
438 	struct sfc_rxq *rxq;
439 	struct sfc_adapter *sa;
440 	unsigned int sw_index;
441 
442 	if (dp_rxq == NULL)
443 		return;
444 
445 	rxq = sfc_rxq_by_dp_rxq(dp_rxq);
446 	sa = rxq->evq->sa;
447 	sfc_adapter_lock(sa);
448 
449 	sw_index = sfc_rxq_sw_index(rxq);
450 
451 	sfc_log_init(sa, "RxQ=%u", sw_index);
452 
453 	sa->eth_dev->data->rx_queues[sw_index] = NULL;
454 
455 	sfc_rx_qfini(sa, sw_index);
456 
457 	sfc_adapter_unlock(sa);
458 }
459 
460 static int
461 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
462 		   uint16_t nb_tx_desc, unsigned int socket_id,
463 		   const struct rte_eth_txconf *tx_conf)
464 {
465 	struct sfc_adapter *sa = dev->data->dev_private;
466 	int rc;
467 
468 	sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
469 		     tx_queue_id, nb_tx_desc, socket_id);
470 
471 	sfc_adapter_lock(sa);
472 
473 	rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
474 	if (rc != 0)
475 		goto fail_tx_qinit;
476 
477 	dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
478 
479 	sfc_adapter_unlock(sa);
480 	return 0;
481 
482 fail_tx_qinit:
483 	sfc_adapter_unlock(sa);
484 	SFC_ASSERT(rc > 0);
485 	return -rc;
486 }
487 
488 static void
489 sfc_tx_queue_release(void *queue)
490 {
491 	struct sfc_dp_txq *dp_txq = queue;
492 	struct sfc_txq *txq;
493 	unsigned int sw_index;
494 	struct sfc_adapter *sa;
495 
496 	if (dp_txq == NULL)
497 		return;
498 
499 	txq = sfc_txq_by_dp_txq(dp_txq);
500 	sw_index = sfc_txq_sw_index(txq);
501 
502 	SFC_ASSERT(txq->evq != NULL);
503 	sa = txq->evq->sa;
504 
505 	sfc_log_init(sa, "TxQ = %u", sw_index);
506 
507 	sfc_adapter_lock(sa);
508 
509 	SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
510 	sa->eth_dev->data->tx_queues[sw_index] = NULL;
511 
512 	sfc_tx_qfini(sa, sw_index);
513 
514 	sfc_adapter_unlock(sa);
515 }
516 
517 static int
518 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
519 {
520 	struct sfc_adapter *sa = dev->data->dev_private;
521 	struct sfc_port *port = &sa->port;
522 	uint64_t *mac_stats;
523 	int ret;
524 
525 	rte_spinlock_lock(&port->mac_stats_lock);
526 
527 	ret = sfc_port_update_mac_stats(sa);
528 	if (ret != 0)
529 		goto unlock;
530 
531 	mac_stats = port->mac_stats_buf;
532 
533 	if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
534 				   EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
535 		stats->ipackets =
536 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
537 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
538 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
539 		stats->opackets =
540 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
541 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
542 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
543 		stats->ibytes =
544 			mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
545 			mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
546 			mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
547 		stats->obytes =
548 			mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
549 			mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
550 			mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
551 		stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
552 		stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
553 		stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
554 	} else {
555 		stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
556 		stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
557 		stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
558 		stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
559 		/*
560 		 * Take into account stats which are whenever supported
561 		 * on EF10. If some stat is not supported by current
562 		 * firmware variant or HW revision, it is guaranteed
563 		 * to be zero in mac_stats.
564 		 */
565 		stats->imissed =
566 			mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
567 			mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
568 			mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
569 			mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
570 			mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
571 			mac_stats[EFX_MAC_PM_TRUNC_QBB] +
572 			mac_stats[EFX_MAC_PM_DISCARD_QBB] +
573 			mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
574 			mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
575 			mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
576 		stats->ierrors =
577 			mac_stats[EFX_MAC_RX_FCS_ERRORS] +
578 			mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
579 			mac_stats[EFX_MAC_RX_JABBER_PKTS];
580 		/* no oerrors counters supported on EF10 */
581 	}
582 
583 unlock:
584 	rte_spinlock_unlock(&port->mac_stats_lock);
585 	SFC_ASSERT(ret >= 0);
586 	return -ret;
587 }
588 
589 static void
590 sfc_stats_reset(struct rte_eth_dev *dev)
591 {
592 	struct sfc_adapter *sa = dev->data->dev_private;
593 	struct sfc_port *port = &sa->port;
594 	int rc;
595 
596 	if (sa->state != SFC_ADAPTER_STARTED) {
597 		/*
598 		 * The operation cannot be done if port is not started; it
599 		 * will be scheduled to be done during the next port start
600 		 */
601 		port->mac_stats_reset_pending = B_TRUE;
602 		return;
603 	}
604 
605 	rc = sfc_port_reset_mac_stats(sa);
606 	if (rc != 0)
607 		sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
608 }
609 
610 static int
611 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
612 	       unsigned int xstats_count)
613 {
614 	struct sfc_adapter *sa = dev->data->dev_private;
615 	struct sfc_port *port = &sa->port;
616 	uint64_t *mac_stats;
617 	int rc;
618 	unsigned int i;
619 	int nstats = 0;
620 
621 	rte_spinlock_lock(&port->mac_stats_lock);
622 
623 	rc = sfc_port_update_mac_stats(sa);
624 	if (rc != 0) {
625 		SFC_ASSERT(rc > 0);
626 		nstats = -rc;
627 		goto unlock;
628 	}
629 
630 	mac_stats = port->mac_stats_buf;
631 
632 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
633 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
634 			if (xstats != NULL && nstats < (int)xstats_count) {
635 				xstats[nstats].id = nstats;
636 				xstats[nstats].value = mac_stats[i];
637 			}
638 			nstats++;
639 		}
640 	}
641 
642 unlock:
643 	rte_spinlock_unlock(&port->mac_stats_lock);
644 
645 	return nstats;
646 }
647 
648 static int
649 sfc_xstats_get_names(struct rte_eth_dev *dev,
650 		     struct rte_eth_xstat_name *xstats_names,
651 		     unsigned int xstats_count)
652 {
653 	struct sfc_adapter *sa = dev->data->dev_private;
654 	struct sfc_port *port = &sa->port;
655 	unsigned int i;
656 	unsigned int nstats = 0;
657 
658 	for (i = 0; i < EFX_MAC_NSTATS; ++i) {
659 		if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
660 			if (xstats_names != NULL && nstats < xstats_count)
661 				strncpy(xstats_names[nstats].name,
662 					efx_mac_stat_name(sa->nic, i),
663 					sizeof(xstats_names[0].name));
664 			nstats++;
665 		}
666 	}
667 
668 	return nstats;
669 }
670 
671 static int
672 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
673 		     uint64_t *values, unsigned int n)
674 {
675 	struct sfc_adapter *sa = dev->data->dev_private;
676 	struct sfc_port *port = &sa->port;
677 	uint64_t *mac_stats;
678 	unsigned int nb_supported = 0;
679 	unsigned int nb_written = 0;
680 	unsigned int i;
681 	int ret;
682 	int rc;
683 
684 	if (unlikely(values == NULL) ||
685 	    unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
686 		return port->mac_stats_nb_supported;
687 
688 	rte_spinlock_lock(&port->mac_stats_lock);
689 
690 	rc = sfc_port_update_mac_stats(sa);
691 	if (rc != 0) {
692 		SFC_ASSERT(rc > 0);
693 		ret = -rc;
694 		goto unlock;
695 	}
696 
697 	mac_stats = port->mac_stats_buf;
698 
699 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
700 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
701 			continue;
702 
703 		if ((ids == NULL) || (ids[nb_written] == nb_supported))
704 			values[nb_written++] = mac_stats[i];
705 
706 		++nb_supported;
707 	}
708 
709 	ret = nb_written;
710 
711 unlock:
712 	rte_spinlock_unlock(&port->mac_stats_lock);
713 
714 	return ret;
715 }
716 
717 static int
718 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
719 			   struct rte_eth_xstat_name *xstats_names,
720 			   const uint64_t *ids, unsigned int size)
721 {
722 	struct sfc_adapter *sa = dev->data->dev_private;
723 	struct sfc_port *port = &sa->port;
724 	unsigned int nb_supported = 0;
725 	unsigned int nb_written = 0;
726 	unsigned int i;
727 
728 	if (unlikely(xstats_names == NULL) ||
729 	    unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
730 		return port->mac_stats_nb_supported;
731 
732 	for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
733 		if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
734 			continue;
735 
736 		if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
737 			char *name = xstats_names[nb_written++].name;
738 
739 			strncpy(name, efx_mac_stat_name(sa->nic, i),
740 				sizeof(xstats_names[0].name));
741 			name[sizeof(xstats_names[0].name) - 1] = '\0';
742 		}
743 
744 		++nb_supported;
745 	}
746 
747 	return nb_written;
748 }
749 
750 static int
751 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
752 {
753 	struct sfc_adapter *sa = dev->data->dev_private;
754 	unsigned int wanted_fc, link_fc;
755 
756 	memset(fc_conf, 0, sizeof(*fc_conf));
757 
758 	sfc_adapter_lock(sa);
759 
760 	if (sa->state == SFC_ADAPTER_STARTED)
761 		efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
762 	else
763 		link_fc = sa->port.flow_ctrl;
764 
765 	switch (link_fc) {
766 	case 0:
767 		fc_conf->mode = RTE_FC_NONE;
768 		break;
769 	case EFX_FCNTL_RESPOND:
770 		fc_conf->mode = RTE_FC_RX_PAUSE;
771 		break;
772 	case EFX_FCNTL_GENERATE:
773 		fc_conf->mode = RTE_FC_TX_PAUSE;
774 		break;
775 	case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
776 		fc_conf->mode = RTE_FC_FULL;
777 		break;
778 	default:
779 		sfc_err(sa, "%s: unexpected flow control value %#x",
780 			__func__, link_fc);
781 	}
782 
783 	fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
784 
785 	sfc_adapter_unlock(sa);
786 
787 	return 0;
788 }
789 
790 static int
791 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
792 {
793 	struct sfc_adapter *sa = dev->data->dev_private;
794 	struct sfc_port *port = &sa->port;
795 	unsigned int fcntl;
796 	int rc;
797 
798 	if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
799 	    fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
800 	    fc_conf->mac_ctrl_frame_fwd != 0) {
801 		sfc_err(sa, "unsupported flow control settings specified");
802 		rc = EINVAL;
803 		goto fail_inval;
804 	}
805 
806 	switch (fc_conf->mode) {
807 	case RTE_FC_NONE:
808 		fcntl = 0;
809 		break;
810 	case RTE_FC_RX_PAUSE:
811 		fcntl = EFX_FCNTL_RESPOND;
812 		break;
813 	case RTE_FC_TX_PAUSE:
814 		fcntl = EFX_FCNTL_GENERATE;
815 		break;
816 	case RTE_FC_FULL:
817 		fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
818 		break;
819 	default:
820 		rc = EINVAL;
821 		goto fail_inval;
822 	}
823 
824 	sfc_adapter_lock(sa);
825 
826 	if (sa->state == SFC_ADAPTER_STARTED) {
827 		rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
828 		if (rc != 0)
829 			goto fail_mac_fcntl_set;
830 	}
831 
832 	port->flow_ctrl = fcntl;
833 	port->flow_ctrl_autoneg = fc_conf->autoneg;
834 
835 	sfc_adapter_unlock(sa);
836 
837 	return 0;
838 
839 fail_mac_fcntl_set:
840 	sfc_adapter_unlock(sa);
841 fail_inval:
842 	SFC_ASSERT(rc > 0);
843 	return -rc;
844 }
845 
846 static int
847 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
848 {
849 	struct sfc_adapter *sa = dev->data->dev_private;
850 	size_t pdu = EFX_MAC_PDU(mtu);
851 	size_t old_pdu;
852 	int rc;
853 
854 	sfc_log_init(sa, "mtu=%u", mtu);
855 
856 	rc = EINVAL;
857 	if (pdu < EFX_MAC_PDU_MIN) {
858 		sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
859 			(unsigned int)mtu, (unsigned int)pdu,
860 			EFX_MAC_PDU_MIN);
861 		goto fail_inval;
862 	}
863 	if (pdu > EFX_MAC_PDU_MAX) {
864 		sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
865 			(unsigned int)mtu, (unsigned int)pdu,
866 			EFX_MAC_PDU_MAX);
867 		goto fail_inval;
868 	}
869 
870 	sfc_adapter_lock(sa);
871 
872 	if (pdu != sa->port.pdu) {
873 		if (sa->state == SFC_ADAPTER_STARTED) {
874 			sfc_stop(sa);
875 
876 			old_pdu = sa->port.pdu;
877 			sa->port.pdu = pdu;
878 			rc = sfc_start(sa);
879 			if (rc != 0)
880 				goto fail_start;
881 		} else {
882 			sa->port.pdu = pdu;
883 		}
884 	}
885 
886 	/*
887 	 * The driver does not use it, but other PMDs update jumbo_frame
888 	 * flag and max_rx_pkt_len when MTU is set.
889 	 */
890 	if (mtu > ETHER_MAX_LEN) {
891 		struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
892 
893 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
894 		rxmode->jumbo_frame = 1;
895 	}
896 
897 	dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
898 
899 	sfc_adapter_unlock(sa);
900 
901 	sfc_log_init(sa, "done");
902 	return 0;
903 
904 fail_start:
905 	sa->port.pdu = old_pdu;
906 	if (sfc_start(sa) != 0)
907 		sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
908 			"PDU max size - port is stopped",
909 			(unsigned int)pdu, (unsigned int)old_pdu);
910 	sfc_adapter_unlock(sa);
911 
912 fail_inval:
913 	sfc_log_init(sa, "failed %d", rc);
914 	SFC_ASSERT(rc > 0);
915 	return -rc;
916 }
917 static void
918 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
919 {
920 	struct sfc_adapter *sa = dev->data->dev_private;
921 	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
922 	struct sfc_port *port = &sa->port;
923 	int rc;
924 
925 	sfc_adapter_lock(sa);
926 
927 	/*
928 	 * Copy the address to the device private data so that
929 	 * it could be recalled in the case of adapter restart.
930 	 */
931 	ether_addr_copy(mac_addr, &port->default_mac_addr);
932 
933 	if (port->isolated) {
934 		sfc_err(sa, "isolated mode is active on the port");
935 		sfc_err(sa, "will not set MAC address");
936 		goto unlock;
937 	}
938 
939 	if (sa->state != SFC_ADAPTER_STARTED) {
940 		sfc_notice(sa, "the port is not started");
941 		sfc_notice(sa, "the new MAC address will be set on port start");
942 
943 		goto unlock;
944 	}
945 
946 	if (encp->enc_allow_set_mac_with_installed_filters) {
947 		rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
948 		if (rc != 0) {
949 			sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
950 			goto unlock;
951 		}
952 
953 		/*
954 		 * Changing the MAC address by means of MCDI request
955 		 * has no effect on received traffic, therefore
956 		 * we also need to update unicast filters
957 		 */
958 		rc = sfc_set_rx_mode(sa);
959 		if (rc != 0)
960 			sfc_err(sa, "cannot set filter (rc = %u)", rc);
961 	} else {
962 		sfc_warn(sa, "cannot set MAC address with filters installed");
963 		sfc_warn(sa, "adapter will be restarted to pick the new MAC");
964 		sfc_warn(sa, "(some traffic may be dropped)");
965 
966 		/*
967 		 * Since setting MAC address with filters installed is not
968 		 * allowed on the adapter, the new MAC address will be set
969 		 * by means of adapter restart. sfc_start() shall retrieve
970 		 * the new address from the device private data and set it.
971 		 */
972 		sfc_stop(sa);
973 		rc = sfc_start(sa);
974 		if (rc != 0)
975 			sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
976 	}
977 
978 unlock:
979 	/*
980 	 * In the case of failure sa->port->default_mac_addr does not
981 	 * need rollback since no error code is returned, and the upper
982 	 * API will anyway update the external MAC address storage.
983 	 * To be consistent with that new value it is better to keep
984 	 * the device private value the same.
985 	 */
986 	sfc_adapter_unlock(sa);
987 }
988 
989 
990 static int
991 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
992 		     uint32_t nb_mc_addr)
993 {
994 	struct sfc_adapter *sa = dev->data->dev_private;
995 	struct sfc_port *port = &sa->port;
996 	uint8_t *mc_addrs = port->mcast_addrs;
997 	int rc;
998 	unsigned int i;
999 
1000 	if (port->isolated) {
1001 		sfc_err(sa, "isolated mode is active on the port");
1002 		sfc_err(sa, "will not set multicast address list");
1003 		return -ENOTSUP;
1004 	}
1005 
1006 	if (mc_addrs == NULL)
1007 		return -ENOBUFS;
1008 
1009 	if (nb_mc_addr > port->max_mcast_addrs) {
1010 		sfc_err(sa, "too many multicast addresses: %u > %u",
1011 			 nb_mc_addr, port->max_mcast_addrs);
1012 		return -EINVAL;
1013 	}
1014 
1015 	for (i = 0; i < nb_mc_addr; ++i) {
1016 		rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1017 				 EFX_MAC_ADDR_LEN);
1018 		mc_addrs += EFX_MAC_ADDR_LEN;
1019 	}
1020 
1021 	port->nb_mcast_addrs = nb_mc_addr;
1022 
1023 	if (sa->state != SFC_ADAPTER_STARTED)
1024 		return 0;
1025 
1026 	rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1027 					port->nb_mcast_addrs);
1028 	if (rc != 0)
1029 		sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1030 
1031 	SFC_ASSERT(rc > 0);
1032 	return -rc;
1033 }
1034 
1035 /*
1036  * The function is used by the secondary process as well. It must not
1037  * use any process-local pointers from the adapter data.
1038  */
1039 static void
1040 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1041 		      struct rte_eth_rxq_info *qinfo)
1042 {
1043 	struct sfc_adapter *sa = dev->data->dev_private;
1044 	struct sfc_rxq_info *rxq_info;
1045 	struct sfc_rxq *rxq;
1046 
1047 	sfc_adapter_lock(sa);
1048 
1049 	SFC_ASSERT(rx_queue_id < sa->rxq_count);
1050 
1051 	rxq_info = &sa->rxq_info[rx_queue_id];
1052 	rxq = rxq_info->rxq;
1053 	SFC_ASSERT(rxq != NULL);
1054 
1055 	qinfo->mp = rxq->refill_mb_pool;
1056 	qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1057 	qinfo->conf.rx_drop_en = 1;
1058 	qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1059 	qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1060 			       DEV_RX_OFFLOAD_UDP_CKSUM |
1061 			       DEV_RX_OFFLOAD_TCP_CKSUM;
1062 	if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1063 		qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1064 		qinfo->scattered_rx = 1;
1065 	}
1066 	qinfo->nb_desc = rxq_info->entries;
1067 
1068 	sfc_adapter_unlock(sa);
1069 }
1070 
1071 /*
1072  * The function is used by the secondary process as well. It must not
1073  * use any process-local pointers from the adapter data.
1074  */
1075 static void
1076 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1077 		      struct rte_eth_txq_info *qinfo)
1078 {
1079 	struct sfc_adapter *sa = dev->data->dev_private;
1080 	struct sfc_txq_info *txq_info;
1081 
1082 	sfc_adapter_lock(sa);
1083 
1084 	SFC_ASSERT(tx_queue_id < sa->txq_count);
1085 
1086 	txq_info = &sa->txq_info[tx_queue_id];
1087 	SFC_ASSERT(txq_info->txq != NULL);
1088 
1089 	memset(qinfo, 0, sizeof(*qinfo));
1090 
1091 	qinfo->conf.txq_flags = txq_info->txq->flags;
1092 	qinfo->conf.offloads = txq_info->txq->offloads;
1093 	qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1094 	qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1095 	qinfo->nb_desc = txq_info->entries;
1096 
1097 	sfc_adapter_unlock(sa);
1098 }
1099 
1100 static uint32_t
1101 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1102 {
1103 	struct sfc_adapter *sa = dev->data->dev_private;
1104 
1105 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1106 
1107 	return sfc_rx_qdesc_npending(sa, rx_queue_id);
1108 }
1109 
1110 static int
1111 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1112 {
1113 	struct sfc_dp_rxq *dp_rxq = queue;
1114 
1115 	return sfc_rx_qdesc_done(dp_rxq, offset);
1116 }
1117 
1118 static int
1119 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1120 {
1121 	struct sfc_dp_rxq *dp_rxq = queue;
1122 	struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1123 
1124 	return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1125 }
1126 
1127 static int
1128 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1129 {
1130 	struct sfc_dp_txq *dp_txq = queue;
1131 	struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1132 
1133 	return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1134 }
1135 
1136 static int
1137 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1138 {
1139 	struct sfc_adapter *sa = dev->data->dev_private;
1140 	int rc;
1141 
1142 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1143 
1144 	sfc_adapter_lock(sa);
1145 
1146 	rc = EINVAL;
1147 	if (sa->state != SFC_ADAPTER_STARTED)
1148 		goto fail_not_started;
1149 
1150 	rc = sfc_rx_qstart(sa, rx_queue_id);
1151 	if (rc != 0)
1152 		goto fail_rx_qstart;
1153 
1154 	sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1155 
1156 	sfc_adapter_unlock(sa);
1157 
1158 	return 0;
1159 
1160 fail_rx_qstart:
1161 fail_not_started:
1162 	sfc_adapter_unlock(sa);
1163 	SFC_ASSERT(rc > 0);
1164 	return -rc;
1165 }
1166 
1167 static int
1168 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1169 {
1170 	struct sfc_adapter *sa = dev->data->dev_private;
1171 
1172 	sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1173 
1174 	sfc_adapter_lock(sa);
1175 	sfc_rx_qstop(sa, rx_queue_id);
1176 
1177 	sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1178 
1179 	sfc_adapter_unlock(sa);
1180 
1181 	return 0;
1182 }
1183 
1184 static int
1185 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1186 {
1187 	struct sfc_adapter *sa = dev->data->dev_private;
1188 	int rc;
1189 
1190 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1191 
1192 	sfc_adapter_lock(sa);
1193 
1194 	rc = EINVAL;
1195 	if (sa->state != SFC_ADAPTER_STARTED)
1196 		goto fail_not_started;
1197 
1198 	rc = sfc_tx_qstart(sa, tx_queue_id);
1199 	if (rc != 0)
1200 		goto fail_tx_qstart;
1201 
1202 	sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1203 
1204 	sfc_adapter_unlock(sa);
1205 	return 0;
1206 
1207 fail_tx_qstart:
1208 
1209 fail_not_started:
1210 	sfc_adapter_unlock(sa);
1211 	SFC_ASSERT(rc > 0);
1212 	return -rc;
1213 }
1214 
1215 static int
1216 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1217 {
1218 	struct sfc_adapter *sa = dev->data->dev_private;
1219 
1220 	sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1221 
1222 	sfc_adapter_lock(sa);
1223 
1224 	sfc_tx_qstop(sa, tx_queue_id);
1225 
1226 	sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1227 
1228 	sfc_adapter_unlock(sa);
1229 	return 0;
1230 }
1231 
1232 static efx_tunnel_protocol_t
1233 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1234 {
1235 	switch (rte_type) {
1236 	case RTE_TUNNEL_TYPE_VXLAN:
1237 		return EFX_TUNNEL_PROTOCOL_VXLAN;
1238 	case RTE_TUNNEL_TYPE_GENEVE:
1239 		return EFX_TUNNEL_PROTOCOL_GENEVE;
1240 	default:
1241 		return EFX_TUNNEL_NPROTOS;
1242 	}
1243 }
1244 
1245 enum sfc_udp_tunnel_op_e {
1246 	SFC_UDP_TUNNEL_ADD_PORT,
1247 	SFC_UDP_TUNNEL_DEL_PORT,
1248 };
1249 
1250 static int
1251 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1252 		      struct rte_eth_udp_tunnel *tunnel_udp,
1253 		      enum sfc_udp_tunnel_op_e op)
1254 {
1255 	struct sfc_adapter *sa = dev->data->dev_private;
1256 	efx_tunnel_protocol_t tunnel_proto;
1257 	int rc;
1258 
1259 	sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1260 		     (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1261 		     (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1262 		     tunnel_udp->udp_port, tunnel_udp->prot_type);
1263 
1264 	tunnel_proto =
1265 		sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1266 	if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1267 		rc = ENOTSUP;
1268 		goto fail_bad_proto;
1269 	}
1270 
1271 	sfc_adapter_lock(sa);
1272 
1273 	switch (op) {
1274 	case SFC_UDP_TUNNEL_ADD_PORT:
1275 		rc = efx_tunnel_config_udp_add(sa->nic,
1276 					       tunnel_udp->udp_port,
1277 					       tunnel_proto);
1278 		break;
1279 	case SFC_UDP_TUNNEL_DEL_PORT:
1280 		rc = efx_tunnel_config_udp_remove(sa->nic,
1281 						  tunnel_udp->udp_port,
1282 						  tunnel_proto);
1283 		break;
1284 	default:
1285 		rc = EINVAL;
1286 		goto fail_bad_op;
1287 	}
1288 
1289 	if (rc != 0)
1290 		goto fail_op;
1291 
1292 	if (sa->state == SFC_ADAPTER_STARTED) {
1293 		rc = efx_tunnel_reconfigure(sa->nic);
1294 		if (rc == EAGAIN) {
1295 			/*
1296 			 * Configuration is accepted by FW and MC reboot
1297 			 * is initiated to apply the changes. MC reboot
1298 			 * will be handled in a usual way (MC reboot
1299 			 * event on management event queue and adapter
1300 			 * restart).
1301 			 */
1302 			rc = 0;
1303 		} else if (rc != 0) {
1304 			goto fail_reconfigure;
1305 		}
1306 	}
1307 
1308 	sfc_adapter_unlock(sa);
1309 	return 0;
1310 
1311 fail_reconfigure:
1312 	/* Remove/restore entry since the change makes the trouble */
1313 	switch (op) {
1314 	case SFC_UDP_TUNNEL_ADD_PORT:
1315 		(void)efx_tunnel_config_udp_remove(sa->nic,
1316 						   tunnel_udp->udp_port,
1317 						   tunnel_proto);
1318 		break;
1319 	case SFC_UDP_TUNNEL_DEL_PORT:
1320 		(void)efx_tunnel_config_udp_add(sa->nic,
1321 						tunnel_udp->udp_port,
1322 						tunnel_proto);
1323 		break;
1324 	}
1325 
1326 fail_op:
1327 fail_bad_op:
1328 	sfc_adapter_unlock(sa);
1329 
1330 fail_bad_proto:
1331 	SFC_ASSERT(rc > 0);
1332 	return -rc;
1333 }
1334 
1335 static int
1336 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1337 			    struct rte_eth_udp_tunnel *tunnel_udp)
1338 {
1339 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1340 }
1341 
1342 static int
1343 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1344 			    struct rte_eth_udp_tunnel *tunnel_udp)
1345 {
1346 	return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1347 }
1348 
1349 #if EFSYS_OPT_RX_SCALE
1350 static int
1351 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1352 			  struct rte_eth_rss_conf *rss_conf)
1353 {
1354 	struct sfc_adapter *sa = dev->data->dev_private;
1355 	struct sfc_port *port = &sa->port;
1356 
1357 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1358 		return -ENOTSUP;
1359 
1360 	if (sa->rss_channels == 0)
1361 		return -EINVAL;
1362 
1363 	sfc_adapter_lock(sa);
1364 
1365 	/*
1366 	 * Mapping of hash configuration between RTE and EFX is not one-to-one,
1367 	 * hence, conversion is done here to derive a correct set of ETH_RSS
1368 	 * flags which corresponds to the active EFX configuration stored
1369 	 * locally in 'sfc_adapter' and kept up-to-date
1370 	 */
1371 	rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1372 	rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1373 	if (rss_conf->rss_key != NULL)
1374 		rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1375 
1376 	sfc_adapter_unlock(sa);
1377 
1378 	return 0;
1379 }
1380 
1381 static int
1382 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1383 			struct rte_eth_rss_conf *rss_conf)
1384 {
1385 	struct sfc_adapter *sa = dev->data->dev_private;
1386 	struct sfc_port *port = &sa->port;
1387 	unsigned int efx_hash_types;
1388 	int rc = 0;
1389 
1390 	if (port->isolated)
1391 		return -ENOTSUP;
1392 
1393 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1394 		sfc_err(sa, "RSS is not available");
1395 		return -ENOTSUP;
1396 	}
1397 
1398 	if (sa->rss_channels == 0) {
1399 		sfc_err(sa, "RSS is not configured");
1400 		return -EINVAL;
1401 	}
1402 
1403 	if ((rss_conf->rss_key != NULL) &&
1404 	    (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1405 		sfc_err(sa, "RSS key size is wrong (should be %lu)",
1406 			sizeof(sa->rss_key));
1407 		return -EINVAL;
1408 	}
1409 
1410 	if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1411 		sfc_err(sa, "unsupported hash functions requested");
1412 		return -EINVAL;
1413 	}
1414 
1415 	sfc_adapter_lock(sa);
1416 
1417 	efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1418 
1419 	rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1420 				   EFX_RX_HASHALG_TOEPLITZ,
1421 				   efx_hash_types, B_TRUE);
1422 	if (rc != 0)
1423 		goto fail_scale_mode_set;
1424 
1425 	if (rss_conf->rss_key != NULL) {
1426 		if (sa->state == SFC_ADAPTER_STARTED) {
1427 			rc = efx_rx_scale_key_set(sa->nic,
1428 						  EFX_RSS_CONTEXT_DEFAULT,
1429 						  rss_conf->rss_key,
1430 						  sizeof(sa->rss_key));
1431 			if (rc != 0)
1432 				goto fail_scale_key_set;
1433 		}
1434 
1435 		rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1436 	}
1437 
1438 	sa->rss_hash_types = efx_hash_types;
1439 
1440 	sfc_adapter_unlock(sa);
1441 
1442 	return 0;
1443 
1444 fail_scale_key_set:
1445 	if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1446 				  EFX_RX_HASHALG_TOEPLITZ,
1447 				  sa->rss_hash_types, B_TRUE) != 0)
1448 		sfc_err(sa, "failed to restore RSS mode");
1449 
1450 fail_scale_mode_set:
1451 	sfc_adapter_unlock(sa);
1452 	return -rc;
1453 }
1454 
1455 static int
1456 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1457 		       struct rte_eth_rss_reta_entry64 *reta_conf,
1458 		       uint16_t reta_size)
1459 {
1460 	struct sfc_adapter *sa = dev->data->dev_private;
1461 	struct sfc_port *port = &sa->port;
1462 	int entry;
1463 
1464 	if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1465 		return -ENOTSUP;
1466 
1467 	if (sa->rss_channels == 0)
1468 		return -EINVAL;
1469 
1470 	if (reta_size != EFX_RSS_TBL_SIZE)
1471 		return -EINVAL;
1472 
1473 	sfc_adapter_lock(sa);
1474 
1475 	for (entry = 0; entry < reta_size; entry++) {
1476 		int grp = entry / RTE_RETA_GROUP_SIZE;
1477 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1478 
1479 		if ((reta_conf[grp].mask >> grp_idx) & 1)
1480 			reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1481 	}
1482 
1483 	sfc_adapter_unlock(sa);
1484 
1485 	return 0;
1486 }
1487 
1488 static int
1489 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1490 			struct rte_eth_rss_reta_entry64 *reta_conf,
1491 			uint16_t reta_size)
1492 {
1493 	struct sfc_adapter *sa = dev->data->dev_private;
1494 	struct sfc_port *port = &sa->port;
1495 	unsigned int *rss_tbl_new;
1496 	uint16_t entry;
1497 	int rc = 0;
1498 
1499 
1500 	if (port->isolated)
1501 		return -ENOTSUP;
1502 
1503 	if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1504 		sfc_err(sa, "RSS is not available");
1505 		return -ENOTSUP;
1506 	}
1507 
1508 	if (sa->rss_channels == 0) {
1509 		sfc_err(sa, "RSS is not configured");
1510 		return -EINVAL;
1511 	}
1512 
1513 	if (reta_size != EFX_RSS_TBL_SIZE) {
1514 		sfc_err(sa, "RETA size is wrong (should be %u)",
1515 			EFX_RSS_TBL_SIZE);
1516 		return -EINVAL;
1517 	}
1518 
1519 	rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1520 	if (rss_tbl_new == NULL)
1521 		return -ENOMEM;
1522 
1523 	sfc_adapter_lock(sa);
1524 
1525 	rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1526 
1527 	for (entry = 0; entry < reta_size; entry++) {
1528 		int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1529 		struct rte_eth_rss_reta_entry64 *grp;
1530 
1531 		grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1532 
1533 		if (grp->mask & (1ull << grp_idx)) {
1534 			if (grp->reta[grp_idx] >= sa->rss_channels) {
1535 				rc = EINVAL;
1536 				goto bad_reta_entry;
1537 			}
1538 			rss_tbl_new[entry] = grp->reta[grp_idx];
1539 		}
1540 	}
1541 
1542 	if (sa->state == SFC_ADAPTER_STARTED) {
1543 		rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1544 					  rss_tbl_new, EFX_RSS_TBL_SIZE);
1545 		if (rc != 0)
1546 			goto fail_scale_tbl_set;
1547 	}
1548 
1549 	rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1550 
1551 fail_scale_tbl_set:
1552 bad_reta_entry:
1553 	sfc_adapter_unlock(sa);
1554 
1555 	rte_free(rss_tbl_new);
1556 
1557 	SFC_ASSERT(rc >= 0);
1558 	return -rc;
1559 }
1560 #endif
1561 
1562 static int
1563 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1564 		    enum rte_filter_op filter_op,
1565 		    void *arg)
1566 {
1567 	struct sfc_adapter *sa = dev->data->dev_private;
1568 	int rc = ENOTSUP;
1569 
1570 	sfc_log_init(sa, "entry");
1571 
1572 	switch (filter_type) {
1573 	case RTE_ETH_FILTER_NONE:
1574 		sfc_err(sa, "Global filters configuration not supported");
1575 		break;
1576 	case RTE_ETH_FILTER_MACVLAN:
1577 		sfc_err(sa, "MACVLAN filters not supported");
1578 		break;
1579 	case RTE_ETH_FILTER_ETHERTYPE:
1580 		sfc_err(sa, "EtherType filters not supported");
1581 		break;
1582 	case RTE_ETH_FILTER_FLEXIBLE:
1583 		sfc_err(sa, "Flexible filters not supported");
1584 		break;
1585 	case RTE_ETH_FILTER_SYN:
1586 		sfc_err(sa, "SYN filters not supported");
1587 		break;
1588 	case RTE_ETH_FILTER_NTUPLE:
1589 		sfc_err(sa, "NTUPLE filters not supported");
1590 		break;
1591 	case RTE_ETH_FILTER_TUNNEL:
1592 		sfc_err(sa, "Tunnel filters not supported");
1593 		break;
1594 	case RTE_ETH_FILTER_FDIR:
1595 		sfc_err(sa, "Flow Director filters not supported");
1596 		break;
1597 	case RTE_ETH_FILTER_HASH:
1598 		sfc_err(sa, "Hash filters not supported");
1599 		break;
1600 	case RTE_ETH_FILTER_GENERIC:
1601 		if (filter_op != RTE_ETH_FILTER_GET) {
1602 			rc = EINVAL;
1603 		} else {
1604 			*(const void **)arg = &sfc_flow_ops;
1605 			rc = 0;
1606 		}
1607 		break;
1608 	default:
1609 		sfc_err(sa, "Unknown filter type %u", filter_type);
1610 		break;
1611 	}
1612 
1613 	sfc_log_init(sa, "exit: %d", -rc);
1614 	SFC_ASSERT(rc >= 0);
1615 	return -rc;
1616 }
1617 
1618 static const struct eth_dev_ops sfc_eth_dev_ops = {
1619 	.dev_configure			= sfc_dev_configure,
1620 	.dev_start			= sfc_dev_start,
1621 	.dev_stop			= sfc_dev_stop,
1622 	.dev_set_link_up		= sfc_dev_set_link_up,
1623 	.dev_set_link_down		= sfc_dev_set_link_down,
1624 	.dev_close			= sfc_dev_close,
1625 	.promiscuous_enable		= sfc_dev_promisc_enable,
1626 	.promiscuous_disable		= sfc_dev_promisc_disable,
1627 	.allmulticast_enable		= sfc_dev_allmulti_enable,
1628 	.allmulticast_disable		= sfc_dev_allmulti_disable,
1629 	.link_update			= sfc_dev_link_update,
1630 	.stats_get			= sfc_stats_get,
1631 	.stats_reset			= sfc_stats_reset,
1632 	.xstats_get			= sfc_xstats_get,
1633 	.xstats_reset			= sfc_stats_reset,
1634 	.xstats_get_names		= sfc_xstats_get_names,
1635 	.dev_infos_get			= sfc_dev_infos_get,
1636 	.dev_supported_ptypes_get	= sfc_dev_supported_ptypes_get,
1637 	.mtu_set			= sfc_dev_set_mtu,
1638 	.rx_queue_start			= sfc_rx_queue_start,
1639 	.rx_queue_stop			= sfc_rx_queue_stop,
1640 	.tx_queue_start			= sfc_tx_queue_start,
1641 	.tx_queue_stop			= sfc_tx_queue_stop,
1642 	.rx_queue_setup			= sfc_rx_queue_setup,
1643 	.rx_queue_release		= sfc_rx_queue_release,
1644 	.rx_queue_count			= sfc_rx_queue_count,
1645 	.rx_descriptor_done		= sfc_rx_descriptor_done,
1646 	.rx_descriptor_status		= sfc_rx_descriptor_status,
1647 	.tx_descriptor_status		= sfc_tx_descriptor_status,
1648 	.tx_queue_setup			= sfc_tx_queue_setup,
1649 	.tx_queue_release		= sfc_tx_queue_release,
1650 	.flow_ctrl_get			= sfc_flow_ctrl_get,
1651 	.flow_ctrl_set			= sfc_flow_ctrl_set,
1652 	.mac_addr_set			= sfc_mac_addr_set,
1653 	.udp_tunnel_port_add		= sfc_dev_udp_tunnel_port_add,
1654 	.udp_tunnel_port_del		= sfc_dev_udp_tunnel_port_del,
1655 #if EFSYS_OPT_RX_SCALE
1656 	.reta_update			= sfc_dev_rss_reta_update,
1657 	.reta_query			= sfc_dev_rss_reta_query,
1658 	.rss_hash_update		= sfc_dev_rss_hash_update,
1659 	.rss_hash_conf_get		= sfc_dev_rss_hash_conf_get,
1660 #endif
1661 	.filter_ctrl			= sfc_dev_filter_ctrl,
1662 	.set_mc_addr_list		= sfc_set_mc_addr_list,
1663 	.rxq_info_get			= sfc_rx_queue_info_get,
1664 	.txq_info_get			= sfc_tx_queue_info_get,
1665 	.fw_version_get			= sfc_fw_version_get,
1666 	.xstats_get_by_id		= sfc_xstats_get_by_id,
1667 	.xstats_get_names_by_id		= sfc_xstats_get_names_by_id,
1668 };
1669 
1670 /**
1671  * Duplicate a string in potentially shared memory required for
1672  * multi-process support.
1673  *
1674  * strdup() allocates from process-local heap/memory.
1675  */
1676 static char *
1677 sfc_strdup(const char *str)
1678 {
1679 	size_t size;
1680 	char *copy;
1681 
1682 	if (str == NULL)
1683 		return NULL;
1684 
1685 	size = strlen(str) + 1;
1686 	copy = rte_malloc(__func__, size, 0);
1687 	if (copy != NULL)
1688 		rte_memcpy(copy, str, size);
1689 
1690 	return copy;
1691 }
1692 
1693 static int
1694 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1695 {
1696 	struct sfc_adapter *sa = dev->data->dev_private;
1697 	unsigned int avail_caps = 0;
1698 	const char *rx_name = NULL;
1699 	const char *tx_name = NULL;
1700 	int rc;
1701 
1702 	switch (sa->family) {
1703 	case EFX_FAMILY_HUNTINGTON:
1704 	case EFX_FAMILY_MEDFORD:
1705 	case EFX_FAMILY_MEDFORD2:
1706 		avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1707 		break;
1708 	default:
1709 		break;
1710 	}
1711 
1712 	rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1713 				sfc_kvarg_string_handler, &rx_name);
1714 	if (rc != 0)
1715 		goto fail_kvarg_rx_datapath;
1716 
1717 	if (rx_name != NULL) {
1718 		sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1719 		if (sa->dp_rx == NULL) {
1720 			sfc_err(sa, "Rx datapath %s not found", rx_name);
1721 			rc = ENOENT;
1722 			goto fail_dp_rx;
1723 		}
1724 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1725 			sfc_err(sa,
1726 				"Insufficient Hw/FW capabilities to use Rx datapath %s",
1727 				rx_name);
1728 			rc = EINVAL;
1729 			goto fail_dp_rx_caps;
1730 		}
1731 	} else {
1732 		sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1733 		if (sa->dp_rx == NULL) {
1734 			sfc_err(sa, "Rx datapath by caps %#x not found",
1735 				avail_caps);
1736 			rc = ENOENT;
1737 			goto fail_dp_rx;
1738 		}
1739 	}
1740 
1741 	sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1742 	if (sa->dp_rx_name == NULL) {
1743 		rc = ENOMEM;
1744 		goto fail_dp_rx_name;
1745 	}
1746 
1747 	sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1748 
1749 	dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1750 
1751 	rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1752 				sfc_kvarg_string_handler, &tx_name);
1753 	if (rc != 0)
1754 		goto fail_kvarg_tx_datapath;
1755 
1756 	if (tx_name != NULL) {
1757 		sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1758 		if (sa->dp_tx == NULL) {
1759 			sfc_err(sa, "Tx datapath %s not found", tx_name);
1760 			rc = ENOENT;
1761 			goto fail_dp_tx;
1762 		}
1763 		if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1764 			sfc_err(sa,
1765 				"Insufficient Hw/FW capabilities to use Tx datapath %s",
1766 				tx_name);
1767 			rc = EINVAL;
1768 			goto fail_dp_tx_caps;
1769 		}
1770 	} else {
1771 		sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1772 		if (sa->dp_tx == NULL) {
1773 			sfc_err(sa, "Tx datapath by caps %#x not found",
1774 				avail_caps);
1775 			rc = ENOENT;
1776 			goto fail_dp_tx;
1777 		}
1778 	}
1779 
1780 	sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1781 	if (sa->dp_tx_name == NULL) {
1782 		rc = ENOMEM;
1783 		goto fail_dp_tx_name;
1784 	}
1785 
1786 	sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1787 
1788 	dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1789 
1790 	dev->dev_ops = &sfc_eth_dev_ops;
1791 
1792 	return 0;
1793 
1794 fail_dp_tx_name:
1795 fail_dp_tx_caps:
1796 	sa->dp_tx = NULL;
1797 
1798 fail_dp_tx:
1799 fail_kvarg_tx_datapath:
1800 	rte_free(sa->dp_rx_name);
1801 	sa->dp_rx_name = NULL;
1802 
1803 fail_dp_rx_name:
1804 fail_dp_rx_caps:
1805 	sa->dp_rx = NULL;
1806 
1807 fail_dp_rx:
1808 fail_kvarg_rx_datapath:
1809 	return rc;
1810 }
1811 
1812 static void
1813 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1814 {
1815 	struct sfc_adapter *sa = dev->data->dev_private;
1816 
1817 	dev->dev_ops = NULL;
1818 	dev->rx_pkt_burst = NULL;
1819 	dev->tx_pkt_burst = NULL;
1820 
1821 	rte_free(sa->dp_tx_name);
1822 	sa->dp_tx_name = NULL;
1823 	sa->dp_tx = NULL;
1824 
1825 	rte_free(sa->dp_rx_name);
1826 	sa->dp_rx_name = NULL;
1827 	sa->dp_rx = NULL;
1828 }
1829 
1830 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1831 	.rxq_info_get			= sfc_rx_queue_info_get,
1832 	.txq_info_get			= sfc_tx_queue_info_get,
1833 };
1834 
1835 static int
1836 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1837 {
1838 	/*
1839 	 * Device private data has really many process-local pointers.
1840 	 * Below code should be extremely careful to use data located
1841 	 * in shared memory only.
1842 	 */
1843 	struct sfc_adapter *sa = dev->data->dev_private;
1844 	const struct sfc_dp_rx *dp_rx;
1845 	const struct sfc_dp_tx *dp_tx;
1846 	int rc;
1847 
1848 	dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1849 	if (dp_rx == NULL) {
1850 		sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1851 		rc = ENOENT;
1852 		goto fail_dp_rx;
1853 	}
1854 	if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1855 		sfc_err(sa, "%s Rx datapath does not support multi-process",
1856 			sa->dp_tx_name);
1857 		rc = EINVAL;
1858 		goto fail_dp_rx_multi_process;
1859 	}
1860 
1861 	dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1862 	if (dp_tx == NULL) {
1863 		sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1864 		rc = ENOENT;
1865 		goto fail_dp_tx;
1866 	}
1867 	if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1868 		sfc_err(sa, "%s Tx datapath does not support multi-process",
1869 			sa->dp_tx_name);
1870 		rc = EINVAL;
1871 		goto fail_dp_tx_multi_process;
1872 	}
1873 
1874 	dev->rx_pkt_burst = dp_rx->pkt_burst;
1875 	dev->tx_pkt_burst = dp_tx->pkt_burst;
1876 	dev->dev_ops = &sfc_eth_dev_secondary_ops;
1877 
1878 	return 0;
1879 
1880 fail_dp_tx_multi_process:
1881 fail_dp_tx:
1882 fail_dp_rx_multi_process:
1883 fail_dp_rx:
1884 	return rc;
1885 }
1886 
1887 static void
1888 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1889 {
1890 	dev->dev_ops = NULL;
1891 	dev->tx_pkt_burst = NULL;
1892 	dev->rx_pkt_burst = NULL;
1893 }
1894 
1895 static void
1896 sfc_register_dp(void)
1897 {
1898 	/* Register once */
1899 	if (TAILQ_EMPTY(&sfc_dp_head)) {
1900 		/* Prefer EF10 datapath */
1901 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1902 		sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1903 
1904 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1905 		sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1906 		sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1907 	}
1908 }
1909 
1910 static int
1911 sfc_eth_dev_init(struct rte_eth_dev *dev)
1912 {
1913 	struct sfc_adapter *sa = dev->data->dev_private;
1914 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1915 	int rc;
1916 	const efx_nic_cfg_t *encp;
1917 	const struct ether_addr *from;
1918 
1919 	sfc_register_dp();
1920 
1921 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1922 		return -sfc_eth_dev_secondary_set_ops(dev);
1923 
1924 	/* Required for logging */
1925 	sa->pci_addr = pci_dev->addr;
1926 	sa->port_id = dev->data->port_id;
1927 
1928 	sa->eth_dev = dev;
1929 
1930 	/* Copy PCI device info to the dev->data */
1931 	rte_eth_copy_pci_info(dev, pci_dev);
1932 
1933 	sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1934 						RTE_LOG_NOTICE);
1935 
1936 	rc = sfc_kvargs_parse(sa);
1937 	if (rc != 0)
1938 		goto fail_kvargs_parse;
1939 
1940 	sfc_log_init(sa, "entry");
1941 
1942 	dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1943 	if (dev->data->mac_addrs == NULL) {
1944 		rc = ENOMEM;
1945 		goto fail_mac_addrs;
1946 	}
1947 
1948 	sfc_adapter_lock_init(sa);
1949 	sfc_adapter_lock(sa);
1950 
1951 	sfc_log_init(sa, "probing");
1952 	rc = sfc_probe(sa);
1953 	if (rc != 0)
1954 		goto fail_probe;
1955 
1956 	sfc_log_init(sa, "set device ops");
1957 	rc = sfc_eth_dev_set_ops(dev);
1958 	if (rc != 0)
1959 		goto fail_set_ops;
1960 
1961 	sfc_log_init(sa, "attaching");
1962 	rc = sfc_attach(sa);
1963 	if (rc != 0)
1964 		goto fail_attach;
1965 
1966 	encp = efx_nic_cfg_get(sa->nic);
1967 
1968 	/*
1969 	 * The arguments are really reverse order in comparison to
1970 	 * Linux kernel. Copy from NIC config to Ethernet device data.
1971 	 */
1972 	from = (const struct ether_addr *)(encp->enc_mac_addr);
1973 	ether_addr_copy(from, &dev->data->mac_addrs[0]);
1974 
1975 	sfc_adapter_unlock(sa);
1976 
1977 	sfc_log_init(sa, "done");
1978 	return 0;
1979 
1980 fail_attach:
1981 	sfc_eth_dev_clear_ops(dev);
1982 
1983 fail_set_ops:
1984 	sfc_unprobe(sa);
1985 
1986 fail_probe:
1987 	sfc_adapter_unlock(sa);
1988 	sfc_adapter_lock_fini(sa);
1989 	rte_free(dev->data->mac_addrs);
1990 	dev->data->mac_addrs = NULL;
1991 
1992 fail_mac_addrs:
1993 	sfc_kvargs_cleanup(sa);
1994 
1995 fail_kvargs_parse:
1996 	sfc_log_init(sa, "failed %d", rc);
1997 	SFC_ASSERT(rc > 0);
1998 	return -rc;
1999 }
2000 
2001 static int
2002 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2003 {
2004 	struct sfc_adapter *sa;
2005 
2006 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2007 		sfc_eth_dev_secondary_clear_ops(dev);
2008 		return 0;
2009 	}
2010 
2011 	sa = dev->data->dev_private;
2012 	sfc_log_init(sa, "entry");
2013 
2014 	sfc_adapter_lock(sa);
2015 
2016 	sfc_eth_dev_clear_ops(dev);
2017 
2018 	sfc_detach(sa);
2019 	sfc_unprobe(sa);
2020 
2021 	rte_free(dev->data->mac_addrs);
2022 	dev->data->mac_addrs = NULL;
2023 
2024 	sfc_kvargs_cleanup(sa);
2025 
2026 	sfc_adapter_unlock(sa);
2027 	sfc_adapter_lock_fini(sa);
2028 
2029 	sfc_log_init(sa, "done");
2030 
2031 	/* Required for logging, so cleanup last */
2032 	sa->eth_dev = NULL;
2033 	return 0;
2034 }
2035 
2036 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2037 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2038 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2039 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2040 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2041 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2042 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2043 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2044 	{ RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2045 	{ .vendor_id = 0 /* sentinel */ }
2046 };
2047 
2048 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2049 	struct rte_pci_device *pci_dev)
2050 {
2051 	return rte_eth_dev_pci_generic_probe(pci_dev,
2052 		sizeof(struct sfc_adapter), sfc_eth_dev_init);
2053 }
2054 
2055 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2056 {
2057 	return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2058 }
2059 
2060 static struct rte_pci_driver sfc_efx_pmd = {
2061 	.id_table = pci_id_sfc_efx_map,
2062 	.drv_flags =
2063 		RTE_PCI_DRV_INTR_LSC |
2064 		RTE_PCI_DRV_NEED_MAPPING,
2065 	.probe = sfc_eth_dev_pci_probe,
2066 	.remove = sfc_eth_dev_pci_remove,
2067 };
2068 
2069 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2070 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2071 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2072 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2073 	SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2074 	SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2075 	SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2076 	SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2077 	SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2078 
2079 RTE_INIT(sfc_driver_register_logtype);
2080 static void
2081 sfc_driver_register_logtype(void)
2082 {
2083 	int ret;
2084 
2085 	ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2086 						   RTE_LOG_NOTICE);
2087 	sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2088 }
2089