xref: /dpdk/drivers/net/enic/enic_ethdev.c (revision 0d09cbc7)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2008-2017 Cisco Systems, Inc.  All rights reserved.
3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
4  */
5 
6 #include <stdio.h>
7 #include <stdint.h>
8 
9 #include <rte_dev.h>
10 #include <rte_pci.h>
11 #include <rte_bus_pci.h>
12 #include <rte_ethdev_driver.h>
13 #include <rte_ethdev_pci.h>
14 #include <rte_kvargs.h>
15 #include <rte_string_fns.h>
16 
17 #include "vnic_intr.h"
18 #include "vnic_cq.h"
19 #include "vnic_wq.h"
20 #include "vnic_rq.h"
21 #include "vnic_enet.h"
22 #include "enic.h"
23 
24 /*
25  * The set of PCI devices this driver supports
26  */
27 #define CISCO_PCI_VENDOR_ID 0x1137
28 static const struct rte_pci_id pci_id_enic_map[] = {
29 	{RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET)},
30 	{RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET_VF)},
31 	{RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET_SN)},
32 	{.vendor_id = 0, /* sentinel */},
33 };
34 
35 /* Supported link speeds of production VIC models */
36 static const struct vic_speed_capa {
37 	uint16_t sub_devid;
38 	uint32_t capa;
39 } vic_speed_capa_map[] = {
40 	{ 0x0043, ETH_LINK_SPEED_10G }, /* VIC */
41 	{ 0x0047, ETH_LINK_SPEED_10G }, /* P81E PCIe */
42 	{ 0x0048, ETH_LINK_SPEED_10G }, /* M81KR Mezz */
43 	{ 0x004f, ETH_LINK_SPEED_10G }, /* 1280 Mezz */
44 	{ 0x0084, ETH_LINK_SPEED_10G }, /* 1240 MLOM */
45 	{ 0x0085, ETH_LINK_SPEED_10G }, /* 1225 PCIe */
46 	{ 0x00cd, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1285 PCIe */
47 	{ 0x00ce, ETH_LINK_SPEED_10G }, /* 1225T PCIe */
48 	{ 0x012a, ETH_LINK_SPEED_40G }, /* M4308 */
49 	{ 0x012c, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1340 MLOM */
50 	{ 0x012e, ETH_LINK_SPEED_10G }, /* 1227 PCIe */
51 	{ 0x0137, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1380 Mezz */
52 	{ 0x014d, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1385 PCIe */
53 	{ 0x015d, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1387 MLOM */
54 	{ 0x0215, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G |
55 		  ETH_LINK_SPEED_40G }, /* 1440 Mezz */
56 	{ 0x0216, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G |
57 		  ETH_LINK_SPEED_40G }, /* 1480 MLOM */
58 	{ 0x0217, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G }, /* 1455 PCIe */
59 	{ 0x0218, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G }, /* 1457 MLOM */
60 	{ 0x0219, ETH_LINK_SPEED_40G }, /* 1485 PCIe */
61 	{ 0x021a, ETH_LINK_SPEED_40G }, /* 1487 MLOM */
62 	{ 0x024a, ETH_LINK_SPEED_40G | ETH_LINK_SPEED_100G }, /* 1495 PCIe */
63 	{ 0x024b, ETH_LINK_SPEED_40G | ETH_LINK_SPEED_100G }, /* 1497 MLOM */
64 	{ 0, 0 }, /* End marker */
65 };
66 
67 #define ENIC_DEVARG_DISABLE_OVERLAY "disable-overlay"
68 #define ENIC_DEVARG_ENABLE_AVX2_RX "enable-avx2-rx"
69 #define ENIC_DEVARG_GENEVE_OPT "geneve-opt"
70 #define ENIC_DEVARG_IG_VLAN_REWRITE "ig-vlan-rewrite"
71 #define ENIC_DEVARG_REPRESENTOR "representor"
72 
73 RTE_LOG_REGISTER(enic_pmd_logtype, pmd.net.enic, INFO);
74 
75 static int
76 enicpmd_fdir_ctrl_func(struct rte_eth_dev *eth_dev,
77 			enum rte_filter_op filter_op, void *arg)
78 {
79 	struct enic *enic = pmd_priv(eth_dev);
80 	int ret = 0;
81 
82 	ENICPMD_FUNC_TRACE();
83 	if (filter_op == RTE_ETH_FILTER_NOP)
84 		return 0;
85 
86 	if (arg == NULL && filter_op != RTE_ETH_FILTER_FLUSH)
87 		return -EINVAL;
88 
89 	switch (filter_op) {
90 	case RTE_ETH_FILTER_ADD:
91 	case RTE_ETH_FILTER_UPDATE:
92 		ret = enic_fdir_add_fltr(enic,
93 			(struct rte_eth_fdir_filter *)arg);
94 		break;
95 
96 	case RTE_ETH_FILTER_DELETE:
97 		ret = enic_fdir_del_fltr(enic,
98 			(struct rte_eth_fdir_filter *)arg);
99 		break;
100 
101 	case RTE_ETH_FILTER_STATS:
102 		enic_fdir_stats_get(enic, (struct rte_eth_fdir_stats *)arg);
103 		break;
104 
105 	case RTE_ETH_FILTER_FLUSH:
106 		dev_warning(enic, "unsupported operation %u", filter_op);
107 		ret = -ENOTSUP;
108 		break;
109 	case RTE_ETH_FILTER_INFO:
110 		enic_fdir_info_get(enic, (struct rte_eth_fdir_info *)arg);
111 		break;
112 	default:
113 		dev_err(enic, "unknown operation %u", filter_op);
114 		ret = -EINVAL;
115 		break;
116 	}
117 	return ret;
118 }
119 
120 static int
121 enicpmd_dev_filter_ctrl(struct rte_eth_dev *dev,
122 		     enum rte_filter_type filter_type,
123 		     enum rte_filter_op filter_op,
124 		     void *arg)
125 {
126 	struct enic *enic = pmd_priv(dev);
127 	int ret = 0;
128 
129 	ENICPMD_FUNC_TRACE();
130 
131 	/*
132 	 * Currently, when Geneve with options offload is enabled, host
133 	 * cannot insert match-action rules.
134 	 */
135 	if (enic->geneve_opt_enabled)
136 		return -ENOTSUP;
137 	switch (filter_type) {
138 	case RTE_ETH_FILTER_GENERIC:
139 		if (filter_op != RTE_ETH_FILTER_GET)
140 			return -EINVAL;
141 		if (enic->flow_filter_mode == FILTER_FLOWMAN)
142 			*(const void **)arg = &enic_fm_flow_ops;
143 		else
144 			*(const void **)arg = &enic_flow_ops;
145 		break;
146 	case RTE_ETH_FILTER_FDIR:
147 		ret = enicpmd_fdir_ctrl_func(dev, filter_op, arg);
148 		break;
149 	default:
150 		dev_warning(enic, "Filter type (%d) not supported",
151 			filter_type);
152 		ret = -EINVAL;
153 		break;
154 	}
155 
156 	return ret;
157 }
158 
159 static void enicpmd_dev_tx_queue_release(void *txq)
160 {
161 	ENICPMD_FUNC_TRACE();
162 
163 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
164 		return;
165 
166 	enic_free_wq(txq);
167 }
168 
169 static int enicpmd_dev_setup_intr(struct enic *enic)
170 {
171 	int ret;
172 	unsigned int index;
173 
174 	ENICPMD_FUNC_TRACE();
175 
176 	/* Are we done with the init of all the queues? */
177 	for (index = 0; index < enic->cq_count; index++) {
178 		if (!enic->cq[index].ctrl)
179 			break;
180 	}
181 	if (enic->cq_count != index)
182 		return 0;
183 	for (index = 0; index < enic->wq_count; index++) {
184 		if (!enic->wq[index].ctrl)
185 			break;
186 	}
187 	if (enic->wq_count != index)
188 		return 0;
189 	/* check start of packet (SOP) RQs only in case scatter is disabled. */
190 	for (index = 0; index < enic->rq_count; index++) {
191 		if (!enic->rq[enic_rte_rq_idx_to_sop_idx(index)].ctrl)
192 			break;
193 	}
194 	if (enic->rq_count != index)
195 		return 0;
196 
197 	ret = enic_alloc_intr_resources(enic);
198 	if (ret) {
199 		dev_err(enic, "alloc intr failed\n");
200 		return ret;
201 	}
202 	enic_init_vnic_resources(enic);
203 
204 	ret = enic_setup_finish(enic);
205 	if (ret)
206 		dev_err(enic, "setup could not be finished\n");
207 
208 	return ret;
209 }
210 
211 static int enicpmd_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
212 	uint16_t queue_idx,
213 	uint16_t nb_desc,
214 	unsigned int socket_id,
215 	const struct rte_eth_txconf *tx_conf)
216 {
217 	int ret;
218 	struct enic *enic = pmd_priv(eth_dev);
219 	struct vnic_wq *wq;
220 
221 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
222 		return -E_RTE_SECONDARY;
223 
224 	ENICPMD_FUNC_TRACE();
225 	RTE_ASSERT(queue_idx < enic->conf_wq_count);
226 	wq = &enic->wq[queue_idx];
227 	wq->offloads = tx_conf->offloads |
228 		eth_dev->data->dev_conf.txmode.offloads;
229 	eth_dev->data->tx_queues[queue_idx] = (void *)wq;
230 
231 	ret = enic_alloc_wq(enic, queue_idx, socket_id, nb_desc);
232 	if (ret) {
233 		dev_err(enic, "error in allocating wq\n");
234 		return ret;
235 	}
236 
237 	return enicpmd_dev_setup_intr(enic);
238 }
239 
240 static int enicpmd_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
241 	uint16_t queue_idx)
242 {
243 	struct enic *enic = pmd_priv(eth_dev);
244 
245 	ENICPMD_FUNC_TRACE();
246 
247 	enic_start_wq(enic, queue_idx);
248 
249 	return 0;
250 }
251 
252 static int enicpmd_dev_tx_queue_stop(struct rte_eth_dev *eth_dev,
253 	uint16_t queue_idx)
254 {
255 	int ret;
256 	struct enic *enic = pmd_priv(eth_dev);
257 
258 	ENICPMD_FUNC_TRACE();
259 
260 	ret = enic_stop_wq(enic, queue_idx);
261 	if (ret)
262 		dev_err(enic, "error in stopping wq %d\n", queue_idx);
263 
264 	return ret;
265 }
266 
267 static int enicpmd_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
268 	uint16_t queue_idx)
269 {
270 	struct enic *enic = pmd_priv(eth_dev);
271 
272 	ENICPMD_FUNC_TRACE();
273 
274 	enic_start_rq(enic, queue_idx);
275 
276 	return 0;
277 }
278 
279 static int enicpmd_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
280 	uint16_t queue_idx)
281 {
282 	int ret;
283 	struct enic *enic = pmd_priv(eth_dev);
284 
285 	ENICPMD_FUNC_TRACE();
286 
287 	ret = enic_stop_rq(enic, queue_idx);
288 	if (ret)
289 		dev_err(enic, "error in stopping rq %d\n", queue_idx);
290 
291 	return ret;
292 }
293 
294 static void enicpmd_dev_rx_queue_release(void *rxq)
295 {
296 	ENICPMD_FUNC_TRACE();
297 
298 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
299 		return;
300 
301 	enic_free_rq(rxq);
302 }
303 
304 static uint32_t enicpmd_dev_rx_queue_count(struct rte_eth_dev *dev,
305 					   uint16_t rx_queue_id)
306 {
307 	struct enic *enic = pmd_priv(dev);
308 	uint32_t queue_count = 0;
309 	struct vnic_cq *cq;
310 	uint32_t cq_tail;
311 	uint16_t cq_idx;
312 	int rq_num;
313 
314 	rq_num = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
315 	cq = &enic->cq[enic_cq_rq(enic, rq_num)];
316 	cq_idx = cq->to_clean;
317 
318 	cq_tail = ioread32(&cq->ctrl->cq_tail);
319 
320 	if (cq_tail < cq_idx)
321 		cq_tail += cq->ring.desc_count;
322 
323 	queue_count = cq_tail - cq_idx;
324 
325 	return queue_count;
326 }
327 
328 static int enicpmd_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
329 	uint16_t queue_idx,
330 	uint16_t nb_desc,
331 	unsigned int socket_id,
332 	const struct rte_eth_rxconf *rx_conf,
333 	struct rte_mempool *mp)
334 {
335 	int ret;
336 	struct enic *enic = pmd_priv(eth_dev);
337 
338 	ENICPMD_FUNC_TRACE();
339 
340 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
341 		return -E_RTE_SECONDARY;
342 	RTE_ASSERT(enic_rte_rq_idx_to_sop_idx(queue_idx) < enic->conf_rq_count);
343 	eth_dev->data->rx_queues[queue_idx] =
344 		(void *)&enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
345 
346 	ret = enic_alloc_rq(enic, queue_idx, socket_id, mp, nb_desc,
347 			    rx_conf->rx_free_thresh);
348 	if (ret) {
349 		dev_err(enic, "error in allocating rq\n");
350 		return ret;
351 	}
352 
353 	return enicpmd_dev_setup_intr(enic);
354 }
355 
356 static int enicpmd_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
357 {
358 	struct enic *enic = pmd_priv(eth_dev);
359 	uint64_t offloads;
360 
361 	ENICPMD_FUNC_TRACE();
362 
363 	offloads = eth_dev->data->dev_conf.rxmode.offloads;
364 	if (mask & ETH_VLAN_STRIP_MASK) {
365 		if (offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
366 			enic->ig_vlan_strip_en = 1;
367 		else
368 			enic->ig_vlan_strip_en = 0;
369 	}
370 
371 	return enic_set_vlan_strip(enic);
372 }
373 
374 static int enicpmd_dev_configure(struct rte_eth_dev *eth_dev)
375 {
376 	int ret;
377 	int mask;
378 	struct enic *enic = pmd_priv(eth_dev);
379 
380 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
381 		return -E_RTE_SECONDARY;
382 
383 	ENICPMD_FUNC_TRACE();
384 	ret = enic_set_vnic_res(enic);
385 	if (ret) {
386 		dev_err(enic, "Set vNIC resource num  failed, aborting\n");
387 		return ret;
388 	}
389 
390 	if (eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
391 		eth_dev->data->dev_conf.rxmode.offloads |=
392 			DEV_RX_OFFLOAD_RSS_HASH;
393 
394 	enic->mc_count = 0;
395 	enic->hw_ip_checksum = !!(eth_dev->data->dev_conf.rxmode.offloads &
396 				  DEV_RX_OFFLOAD_CHECKSUM);
397 	/* All vlan offload masks to apply the current settings */
398 	mask = ETH_VLAN_STRIP_MASK |
399 		ETH_VLAN_FILTER_MASK |
400 		ETH_VLAN_EXTEND_MASK;
401 	ret = enicpmd_vlan_offload_set(eth_dev, mask);
402 	if (ret) {
403 		dev_err(enic, "Failed to configure VLAN offloads\n");
404 		return ret;
405 	}
406 	/*
407 	 * Initialize RSS with the default reta and key. If the user key is
408 	 * given (rx_adv_conf.rss_conf.rss_key), will use that instead of the
409 	 * default key.
410 	 */
411 	return enic_init_rss_nic_cfg(enic);
412 }
413 
414 /* Start the device.
415  * It returns 0 on success.
416  */
417 static int enicpmd_dev_start(struct rte_eth_dev *eth_dev)
418 {
419 	struct enic *enic = pmd_priv(eth_dev);
420 
421 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
422 		return -E_RTE_SECONDARY;
423 
424 	ENICPMD_FUNC_TRACE();
425 	return enic_enable(enic);
426 }
427 
428 /*
429  * Stop device: disable rx and tx functions to allow for reconfiguring.
430  */
431 static void enicpmd_dev_stop(struct rte_eth_dev *eth_dev)
432 {
433 	struct rte_eth_link link;
434 	struct enic *enic = pmd_priv(eth_dev);
435 
436 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
437 		return;
438 
439 	ENICPMD_FUNC_TRACE();
440 	enic_disable(enic);
441 
442 	memset(&link, 0, sizeof(link));
443 	rte_eth_linkstatus_set(eth_dev, &link);
444 }
445 
446 /*
447  * Stop device.
448  */
449 static void enicpmd_dev_close(struct rte_eth_dev *eth_dev)
450 {
451 	struct enic *enic = pmd_priv(eth_dev);
452 
453 	ENICPMD_FUNC_TRACE();
454 	enic_remove(enic);
455 }
456 
457 static int enicpmd_dev_link_update(struct rte_eth_dev *eth_dev,
458 	__rte_unused int wait_to_complete)
459 {
460 	ENICPMD_FUNC_TRACE();
461 	return enic_link_update(eth_dev);
462 }
463 
464 static int enicpmd_dev_stats_get(struct rte_eth_dev *eth_dev,
465 	struct rte_eth_stats *stats)
466 {
467 	struct enic *enic = pmd_priv(eth_dev);
468 
469 	ENICPMD_FUNC_TRACE();
470 	return enic_dev_stats_get(enic, stats);
471 }
472 
473 static int enicpmd_dev_stats_reset(struct rte_eth_dev *eth_dev)
474 {
475 	struct enic *enic = pmd_priv(eth_dev);
476 
477 	ENICPMD_FUNC_TRACE();
478 	return enic_dev_stats_clear(enic);
479 }
480 
481 static uint32_t speed_capa_from_pci_id(struct rte_eth_dev *eth_dev)
482 {
483 	const struct vic_speed_capa *m;
484 	struct rte_pci_device *pdev;
485 	uint16_t id;
486 
487 	pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
488 	id = pdev->id.subsystem_device_id;
489 	for (m = vic_speed_capa_map; m->sub_devid != 0; m++) {
490 		if (m->sub_devid == id)
491 			return m->capa;
492 	}
493 	/* 1300 and later models are at least 40G */
494 	if (id >= 0x0100)
495 		return ETH_LINK_SPEED_40G;
496 	/* VFs have subsystem id 0, check device id */
497 	if (id == 0) {
498 		/* Newer VF implies at least 40G model */
499 		if (pdev->id.device_id == PCI_DEVICE_ID_CISCO_VIC_ENET_SN)
500 			return ETH_LINK_SPEED_40G;
501 	}
502 	return ETH_LINK_SPEED_10G;
503 }
504 
505 static int enicpmd_dev_info_get(struct rte_eth_dev *eth_dev,
506 	struct rte_eth_dev_info *device_info)
507 {
508 	struct enic *enic = pmd_priv(eth_dev);
509 
510 	ENICPMD_FUNC_TRACE();
511 	/* Scattered Rx uses two receive queues per rx queue exposed to dpdk */
512 	device_info->max_rx_queues = enic->conf_rq_count / 2;
513 	device_info->max_tx_queues = enic->conf_wq_count;
514 	device_info->min_rx_bufsize = ENIC_MIN_MTU;
515 	/* "Max" mtu is not a typo. HW receives packet sizes up to the
516 	 * max mtu regardless of the current mtu (vNIC's mtu). vNIC mtu is
517 	 * a hint to the driver to size receive buffers accordingly so that
518 	 * larger-than-vnic-mtu packets get truncated.. For DPDK, we let
519 	 * the user decide the buffer size via rxmode.max_rx_pkt_len, basically
520 	 * ignoring vNIC mtu.
521 	 */
522 	device_info->max_rx_pktlen = enic_mtu_to_max_rx_pktlen(enic->max_mtu);
523 	device_info->max_mac_addrs = ENIC_UNICAST_PERFECT_FILTERS;
524 	device_info->min_mtu = ENIC_MIN_MTU;
525 	device_info->max_mtu = enic->max_mtu;
526 	device_info->rx_offload_capa = enic->rx_offload_capa;
527 	device_info->tx_offload_capa = enic->tx_offload_capa;
528 	device_info->tx_queue_offload_capa = enic->tx_queue_offload_capa;
529 	device_info->default_rxconf = (struct rte_eth_rxconf) {
530 		.rx_free_thresh = ENIC_DEFAULT_RX_FREE_THRESH
531 	};
532 	device_info->reta_size = enic->reta_size;
533 	device_info->hash_key_size = enic->hash_key_size;
534 	device_info->flow_type_rss_offloads = enic->flow_type_rss_offloads;
535 	device_info->rx_desc_lim = (struct rte_eth_desc_lim) {
536 		.nb_max = enic->config.rq_desc_count,
537 		.nb_min = ENIC_MIN_RQ_DESCS,
538 		.nb_align = ENIC_ALIGN_DESCS,
539 	};
540 	device_info->tx_desc_lim = (struct rte_eth_desc_lim) {
541 		.nb_max = enic->config.wq_desc_count,
542 		.nb_min = ENIC_MIN_WQ_DESCS,
543 		.nb_align = ENIC_ALIGN_DESCS,
544 		.nb_seg_max = ENIC_TX_XMIT_MAX,
545 		.nb_mtu_seg_max = ENIC_NON_TSO_MAX_DESC,
546 	};
547 	device_info->default_rxportconf = (struct rte_eth_dev_portconf) {
548 		.burst_size = ENIC_DEFAULT_RX_BURST,
549 		.ring_size = RTE_MIN(device_info->rx_desc_lim.nb_max,
550 			ENIC_DEFAULT_RX_RING_SIZE),
551 		.nb_queues = ENIC_DEFAULT_RX_RINGS,
552 	};
553 	device_info->default_txportconf = (struct rte_eth_dev_portconf) {
554 		.burst_size = ENIC_DEFAULT_TX_BURST,
555 		.ring_size = RTE_MIN(device_info->tx_desc_lim.nb_max,
556 			ENIC_DEFAULT_TX_RING_SIZE),
557 		.nb_queues = ENIC_DEFAULT_TX_RINGS,
558 	};
559 	device_info->speed_capa = speed_capa_from_pci_id(eth_dev);
560 
561 	return 0;
562 }
563 
564 static const uint32_t *enicpmd_dev_supported_ptypes_get(struct rte_eth_dev *dev)
565 {
566 	static const uint32_t ptypes[] = {
567 		RTE_PTYPE_L2_ETHER,
568 		RTE_PTYPE_L2_ETHER_VLAN,
569 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
570 		RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
571 		RTE_PTYPE_L4_TCP,
572 		RTE_PTYPE_L4_UDP,
573 		RTE_PTYPE_L4_FRAG,
574 		RTE_PTYPE_L4_NONFRAG,
575 		RTE_PTYPE_UNKNOWN
576 	};
577 	static const uint32_t ptypes_overlay[] = {
578 		RTE_PTYPE_L2_ETHER,
579 		RTE_PTYPE_L2_ETHER_VLAN,
580 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
581 		RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
582 		RTE_PTYPE_L4_TCP,
583 		RTE_PTYPE_L4_UDP,
584 		RTE_PTYPE_L4_FRAG,
585 		RTE_PTYPE_L4_NONFRAG,
586 		RTE_PTYPE_TUNNEL_GRENAT,
587 		RTE_PTYPE_INNER_L2_ETHER,
588 		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
589 		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
590 		RTE_PTYPE_INNER_L4_TCP,
591 		RTE_PTYPE_INNER_L4_UDP,
592 		RTE_PTYPE_INNER_L4_FRAG,
593 		RTE_PTYPE_INNER_L4_NONFRAG,
594 		RTE_PTYPE_UNKNOWN
595 	};
596 
597 	if (dev->rx_pkt_burst != enic_dummy_recv_pkts &&
598 	    dev->rx_pkt_burst != NULL) {
599 		struct enic *enic = pmd_priv(dev);
600 		if (enic->overlay_offload)
601 			return ptypes_overlay;
602 		else
603 			return ptypes;
604 	}
605 	return NULL;
606 }
607 
608 static int enicpmd_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
609 {
610 	struct enic *enic = pmd_priv(eth_dev);
611 	int ret;
612 
613 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
614 		return -E_RTE_SECONDARY;
615 
616 	ENICPMD_FUNC_TRACE();
617 
618 	enic->promisc = 1;
619 	ret = enic_add_packet_filter(enic);
620 	if (ret != 0)
621 		enic->promisc = 0;
622 
623 	return ret;
624 }
625 
626 static int enicpmd_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
627 {
628 	struct enic *enic = pmd_priv(eth_dev);
629 	int ret;
630 
631 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
632 		return -E_RTE_SECONDARY;
633 
634 	ENICPMD_FUNC_TRACE();
635 	enic->promisc = 0;
636 	ret = enic_add_packet_filter(enic);
637 	if (ret != 0)
638 		enic->promisc = 1;
639 
640 	return ret;
641 }
642 
643 static int enicpmd_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
644 {
645 	struct enic *enic = pmd_priv(eth_dev);
646 	int ret;
647 
648 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
649 		return -E_RTE_SECONDARY;
650 
651 	ENICPMD_FUNC_TRACE();
652 	enic->allmulti = 1;
653 	ret = enic_add_packet_filter(enic);
654 	if (ret != 0)
655 		enic->allmulti = 0;
656 
657 	return ret;
658 }
659 
660 static int enicpmd_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
661 {
662 	struct enic *enic = pmd_priv(eth_dev);
663 	int ret;
664 
665 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
666 		return -E_RTE_SECONDARY;
667 
668 	ENICPMD_FUNC_TRACE();
669 	enic->allmulti = 0;
670 	ret = enic_add_packet_filter(enic);
671 	if (ret != 0)
672 		enic->allmulti = 1;
673 
674 	return ret;
675 }
676 
677 static int enicpmd_add_mac_addr(struct rte_eth_dev *eth_dev,
678 	struct rte_ether_addr *mac_addr,
679 	__rte_unused uint32_t index, __rte_unused uint32_t pool)
680 {
681 	struct enic *enic = pmd_priv(eth_dev);
682 
683 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
684 		return -E_RTE_SECONDARY;
685 
686 	ENICPMD_FUNC_TRACE();
687 	return enic_set_mac_address(enic, mac_addr->addr_bytes);
688 }
689 
690 static void enicpmd_remove_mac_addr(struct rte_eth_dev *eth_dev, uint32_t index)
691 {
692 	struct enic *enic = pmd_priv(eth_dev);
693 
694 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
695 		return;
696 
697 	ENICPMD_FUNC_TRACE();
698 	if (enic_del_mac_address(enic, index))
699 		dev_err(enic, "del mac addr failed\n");
700 }
701 
702 static int enicpmd_set_mac_addr(struct rte_eth_dev *eth_dev,
703 				struct rte_ether_addr *addr)
704 {
705 	struct enic *enic = pmd_priv(eth_dev);
706 	int ret;
707 
708 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
709 		return -E_RTE_SECONDARY;
710 
711 	ENICPMD_FUNC_TRACE();
712 	ret = enic_del_mac_address(enic, 0);
713 	if (ret)
714 		return ret;
715 	return enic_set_mac_address(enic, addr->addr_bytes);
716 }
717 
718 static void debug_log_add_del_addr(struct rte_ether_addr *addr, bool add)
719 {
720 	char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
721 
722 	rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE, addr);
723 	ENICPMD_LOG(DEBUG, " %s address %s\n",
724 		     add ? "add" : "remove", mac_str);
725 }
726 
727 static int enicpmd_set_mc_addr_list(struct rte_eth_dev *eth_dev,
728 				    struct rte_ether_addr *mc_addr_set,
729 				    uint32_t nb_mc_addr)
730 {
731 	struct enic *enic = pmd_priv(eth_dev);
732 	char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
733 	struct rte_ether_addr *addr;
734 	uint32_t i, j;
735 	int ret;
736 
737 	ENICPMD_FUNC_TRACE();
738 
739 	/* Validate the given addresses first */
740 	for (i = 0; i < nb_mc_addr && mc_addr_set != NULL; i++) {
741 		addr = &mc_addr_set[i];
742 		if (!rte_is_multicast_ether_addr(addr) ||
743 		    rte_is_broadcast_ether_addr(addr)) {
744 			rte_ether_format_addr(mac_str,
745 					RTE_ETHER_ADDR_FMT_SIZE, addr);
746 			ENICPMD_LOG(ERR, " invalid multicast address %s\n",
747 				     mac_str);
748 			return -EINVAL;
749 		}
750 	}
751 
752 	/* Flush all if requested */
753 	if (nb_mc_addr == 0 || mc_addr_set == NULL) {
754 		ENICPMD_LOG(DEBUG, " flush multicast addresses\n");
755 		for (i = 0; i < enic->mc_count; i++) {
756 			addr = &enic->mc_addrs[i];
757 			debug_log_add_del_addr(addr, false);
758 			ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
759 			if (ret)
760 				return ret;
761 		}
762 		enic->mc_count = 0;
763 		return 0;
764 	}
765 
766 	if (nb_mc_addr > ENIC_MULTICAST_PERFECT_FILTERS) {
767 		ENICPMD_LOG(ERR, " too many multicast addresses: max=%d\n",
768 			     ENIC_MULTICAST_PERFECT_FILTERS);
769 		return -ENOSPC;
770 	}
771 	/*
772 	 * devcmd is slow, so apply the difference instead of flushing and
773 	 * adding everything.
774 	 * 1. Delete addresses on the NIC but not on the host
775 	 */
776 	for (i = 0; i < enic->mc_count; i++) {
777 		addr = &enic->mc_addrs[i];
778 		for (j = 0; j < nb_mc_addr; j++) {
779 			if (rte_is_same_ether_addr(addr, &mc_addr_set[j]))
780 				break;
781 		}
782 		if (j < nb_mc_addr)
783 			continue;
784 		debug_log_add_del_addr(addr, false);
785 		ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
786 		if (ret)
787 			return ret;
788 	}
789 	/* 2. Add addresses on the host but not on the NIC */
790 	for (i = 0; i < nb_mc_addr; i++) {
791 		addr = &mc_addr_set[i];
792 		for (j = 0; j < enic->mc_count; j++) {
793 			if (rte_is_same_ether_addr(addr, &enic->mc_addrs[j]))
794 				break;
795 		}
796 		if (j < enic->mc_count)
797 			continue;
798 		debug_log_add_del_addr(addr, true);
799 		ret = vnic_dev_add_addr(enic->vdev, addr->addr_bytes);
800 		if (ret)
801 			return ret;
802 	}
803 	/* Keep a copy so we can flush/apply later on.. */
804 	memcpy(enic->mc_addrs, mc_addr_set,
805 	       nb_mc_addr * sizeof(struct rte_ether_addr));
806 	enic->mc_count = nb_mc_addr;
807 	return 0;
808 }
809 
810 static int enicpmd_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
811 {
812 	struct enic *enic = pmd_priv(eth_dev);
813 
814 	ENICPMD_FUNC_TRACE();
815 	return enic_set_mtu(enic, mtu);
816 }
817 
818 static int enicpmd_dev_rss_reta_query(struct rte_eth_dev *dev,
819 				      struct rte_eth_rss_reta_entry64
820 				      *reta_conf,
821 				      uint16_t reta_size)
822 {
823 	struct enic *enic = pmd_priv(dev);
824 	uint16_t i, idx, shift;
825 
826 	ENICPMD_FUNC_TRACE();
827 	if (reta_size != ENIC_RSS_RETA_SIZE) {
828 		dev_err(enic, "reta_query: wrong reta_size. given=%u expected=%u\n",
829 			reta_size, ENIC_RSS_RETA_SIZE);
830 		return -EINVAL;
831 	}
832 
833 	for (i = 0; i < reta_size; i++) {
834 		idx = i / RTE_RETA_GROUP_SIZE;
835 		shift = i % RTE_RETA_GROUP_SIZE;
836 		if (reta_conf[idx].mask & (1ULL << shift))
837 			reta_conf[idx].reta[shift] = enic_sop_rq_idx_to_rte_idx(
838 				enic->rss_cpu.cpu[i / 4].b[i % 4]);
839 	}
840 
841 	return 0;
842 }
843 
844 static int enicpmd_dev_rss_reta_update(struct rte_eth_dev *dev,
845 				       struct rte_eth_rss_reta_entry64
846 				       *reta_conf,
847 				       uint16_t reta_size)
848 {
849 	struct enic *enic = pmd_priv(dev);
850 	union vnic_rss_cpu rss_cpu;
851 	uint16_t i, idx, shift;
852 
853 	ENICPMD_FUNC_TRACE();
854 	if (reta_size != ENIC_RSS_RETA_SIZE) {
855 		dev_err(enic, "reta_update: wrong reta_size. given=%u"
856 			" expected=%u\n",
857 			reta_size, ENIC_RSS_RETA_SIZE);
858 		return -EINVAL;
859 	}
860 	/*
861 	 * Start with the current reta and modify it per reta_conf, as we
862 	 * need to push the entire reta even if we only modify one entry.
863 	 */
864 	rss_cpu = enic->rss_cpu;
865 	for (i = 0; i < reta_size; i++) {
866 		idx = i / RTE_RETA_GROUP_SIZE;
867 		shift = i % RTE_RETA_GROUP_SIZE;
868 		if (reta_conf[idx].mask & (1ULL << shift))
869 			rss_cpu.cpu[i / 4].b[i % 4] =
870 				enic_rte_rq_idx_to_sop_idx(
871 					reta_conf[idx].reta[shift]);
872 	}
873 	return enic_set_rss_reta(enic, &rss_cpu);
874 }
875 
876 static int enicpmd_dev_rss_hash_update(struct rte_eth_dev *dev,
877 				       struct rte_eth_rss_conf *rss_conf)
878 {
879 	struct enic *enic = pmd_priv(dev);
880 
881 	ENICPMD_FUNC_TRACE();
882 	return enic_set_rss_conf(enic, rss_conf);
883 }
884 
885 static int enicpmd_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
886 					 struct rte_eth_rss_conf *rss_conf)
887 {
888 	struct enic *enic = pmd_priv(dev);
889 
890 	ENICPMD_FUNC_TRACE();
891 	if (rss_conf == NULL)
892 		return -EINVAL;
893 	if (rss_conf->rss_key != NULL &&
894 	    rss_conf->rss_key_len < ENIC_RSS_HASH_KEY_SIZE) {
895 		dev_err(enic, "rss_hash_conf_get: wrong rss_key_len. given=%u"
896 			" expected=%u+\n",
897 			rss_conf->rss_key_len, ENIC_RSS_HASH_KEY_SIZE);
898 		return -EINVAL;
899 	}
900 	rss_conf->rss_hf = enic->rss_hf;
901 	if (rss_conf->rss_key != NULL) {
902 		int i;
903 		for (i = 0; i < ENIC_RSS_HASH_KEY_SIZE; i++) {
904 			rss_conf->rss_key[i] =
905 				enic->rss_key.key[i / 10].b[i % 10];
906 		}
907 		rss_conf->rss_key_len = ENIC_RSS_HASH_KEY_SIZE;
908 	}
909 	return 0;
910 }
911 
912 static void enicpmd_dev_rxq_info_get(struct rte_eth_dev *dev,
913 				     uint16_t rx_queue_id,
914 				     struct rte_eth_rxq_info *qinfo)
915 {
916 	struct enic *enic = pmd_priv(dev);
917 	struct vnic_rq *rq_sop;
918 	struct vnic_rq *rq_data;
919 	struct rte_eth_rxconf *conf;
920 	uint16_t sop_queue_idx;
921 	uint16_t data_queue_idx;
922 
923 	ENICPMD_FUNC_TRACE();
924 	sop_queue_idx = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
925 	data_queue_idx = enic_rte_rq_idx_to_data_idx(rx_queue_id, enic);
926 	rq_sop = &enic->rq[sop_queue_idx];
927 	rq_data = &enic->rq[data_queue_idx]; /* valid if data_queue_enable */
928 	qinfo->mp = rq_sop->mp;
929 	qinfo->scattered_rx = rq_sop->data_queue_enable;
930 	qinfo->nb_desc = rq_sop->ring.desc_count;
931 	if (qinfo->scattered_rx)
932 		qinfo->nb_desc += rq_data->ring.desc_count;
933 	conf = &qinfo->conf;
934 	memset(conf, 0, sizeof(*conf));
935 	conf->rx_free_thresh = rq_sop->rx_free_thresh;
936 	conf->rx_drop_en = 1;
937 	/*
938 	 * Except VLAN stripping (port setting), all the checksum offloads
939 	 * are always enabled.
940 	 */
941 	conf->offloads = enic->rx_offload_capa;
942 	if (!enic->ig_vlan_strip_en)
943 		conf->offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
944 	/* rx_thresh and other fields are not applicable for enic */
945 }
946 
947 static void enicpmd_dev_txq_info_get(struct rte_eth_dev *dev,
948 				     uint16_t tx_queue_id,
949 				     struct rte_eth_txq_info *qinfo)
950 {
951 	struct enic *enic = pmd_priv(dev);
952 	struct vnic_wq *wq = &enic->wq[tx_queue_id];
953 
954 	ENICPMD_FUNC_TRACE();
955 	qinfo->nb_desc = wq->ring.desc_count;
956 	memset(&qinfo->conf, 0, sizeof(qinfo->conf));
957 	qinfo->conf.offloads = wq->offloads;
958 	/* tx_thresh, and all the other fields are not applicable for enic */
959 }
960 
961 static int enicpmd_dev_rx_burst_mode_get(struct rte_eth_dev *dev,
962 					 __rte_unused uint16_t queue_id,
963 					 struct rte_eth_burst_mode *mode)
964 {
965 	eth_rx_burst_t pkt_burst = dev->rx_pkt_burst;
966 	struct enic *enic = pmd_priv(dev);
967 	const char *info_str = NULL;
968 	int ret = -EINVAL;
969 
970 	ENICPMD_FUNC_TRACE();
971 	if (enic->use_noscatter_vec_rx_handler)
972 		info_str = "Vector AVX2 No Scatter";
973 	else if (pkt_burst == enic_noscatter_recv_pkts)
974 		info_str = "Scalar No Scatter";
975 	else if (pkt_burst == enic_recv_pkts)
976 		info_str = "Scalar";
977 	if (info_str) {
978 		strlcpy(mode->info, info_str, sizeof(mode->info));
979 		ret = 0;
980 	}
981 	return ret;
982 }
983 
984 static int enicpmd_dev_tx_burst_mode_get(struct rte_eth_dev *dev,
985 					 __rte_unused uint16_t queue_id,
986 					 struct rte_eth_burst_mode *mode)
987 {
988 	eth_tx_burst_t pkt_burst = dev->tx_pkt_burst;
989 	const char *info_str = NULL;
990 	int ret = -EINVAL;
991 
992 	ENICPMD_FUNC_TRACE();
993 	if (pkt_burst == enic_simple_xmit_pkts)
994 		info_str = "Scalar Simplified";
995 	else if (pkt_burst == enic_xmit_pkts)
996 		info_str = "Scalar";
997 	if (info_str) {
998 		strlcpy(mode->info, info_str, sizeof(mode->info));
999 		ret = 0;
1000 	}
1001 	return ret;
1002 }
1003 
1004 static int enicpmd_dev_rx_queue_intr_enable(struct rte_eth_dev *eth_dev,
1005 					    uint16_t rx_queue_id)
1006 {
1007 	struct enic *enic = pmd_priv(eth_dev);
1008 
1009 	ENICPMD_FUNC_TRACE();
1010 	vnic_intr_unmask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
1011 	return 0;
1012 }
1013 
1014 static int enicpmd_dev_rx_queue_intr_disable(struct rte_eth_dev *eth_dev,
1015 					     uint16_t rx_queue_id)
1016 {
1017 	struct enic *enic = pmd_priv(eth_dev);
1018 
1019 	ENICPMD_FUNC_TRACE();
1020 	vnic_intr_mask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
1021 	return 0;
1022 }
1023 
1024 static int udp_tunnel_common_check(struct enic *enic,
1025 				   struct rte_eth_udp_tunnel *tnl)
1026 {
1027 	if (tnl->prot_type != RTE_TUNNEL_TYPE_VXLAN)
1028 		return -ENOTSUP;
1029 	if (!enic->overlay_offload) {
1030 		ENICPMD_LOG(DEBUG, " vxlan (overlay offload) is not "
1031 			     "supported\n");
1032 		return -ENOTSUP;
1033 	}
1034 	return 0;
1035 }
1036 
1037 static int update_vxlan_port(struct enic *enic, uint16_t port)
1038 {
1039 	if (vnic_dev_overlay_offload_cfg(enic->vdev,
1040 					 OVERLAY_CFG_VXLAN_PORT_UPDATE,
1041 					 port)) {
1042 		ENICPMD_LOG(DEBUG, " failed to update vxlan port\n");
1043 		return -EINVAL;
1044 	}
1045 	ENICPMD_LOG(DEBUG, " updated vxlan port to %u\n", port);
1046 	enic->vxlan_port = port;
1047 	return 0;
1048 }
1049 
1050 static int enicpmd_dev_udp_tunnel_port_add(struct rte_eth_dev *eth_dev,
1051 					   struct rte_eth_udp_tunnel *tnl)
1052 {
1053 	struct enic *enic = pmd_priv(eth_dev);
1054 	int ret;
1055 
1056 	ENICPMD_FUNC_TRACE();
1057 	ret = udp_tunnel_common_check(enic, tnl);
1058 	if (ret)
1059 		return ret;
1060 	/*
1061 	 * The NIC has 1 configurable VXLAN port number. "Adding" a new port
1062 	 * number replaces it.
1063 	 */
1064 	if (tnl->udp_port == enic->vxlan_port || tnl->udp_port == 0) {
1065 		ENICPMD_LOG(DEBUG, " %u is already configured or invalid\n",
1066 			     tnl->udp_port);
1067 		return -EINVAL;
1068 	}
1069 	return update_vxlan_port(enic, tnl->udp_port);
1070 }
1071 
1072 static int enicpmd_dev_udp_tunnel_port_del(struct rte_eth_dev *eth_dev,
1073 					   struct rte_eth_udp_tunnel *tnl)
1074 {
1075 	struct enic *enic = pmd_priv(eth_dev);
1076 	int ret;
1077 
1078 	ENICPMD_FUNC_TRACE();
1079 	ret = udp_tunnel_common_check(enic, tnl);
1080 	if (ret)
1081 		return ret;
1082 	/*
1083 	 * Clear the previously set port number and restore the
1084 	 * hardware default port number. Some drivers disable VXLAN
1085 	 * offloads when there are no configured port numbers. But
1086 	 * enic does not do that as VXLAN is part of overlay offload,
1087 	 * which is tied to inner RSS and TSO.
1088 	 */
1089 	if (tnl->udp_port != enic->vxlan_port) {
1090 		ENICPMD_LOG(DEBUG, " %u is not a configured vxlan port\n",
1091 			     tnl->udp_port);
1092 		return -EINVAL;
1093 	}
1094 	return update_vxlan_port(enic, RTE_VXLAN_DEFAULT_PORT);
1095 }
1096 
1097 static int enicpmd_dev_fw_version_get(struct rte_eth_dev *eth_dev,
1098 				      char *fw_version, size_t fw_size)
1099 {
1100 	struct vnic_devcmd_fw_info *info;
1101 	struct enic *enic;
1102 	int ret;
1103 
1104 	ENICPMD_FUNC_TRACE();
1105 	if (fw_version == NULL || fw_size <= 0)
1106 		return -EINVAL;
1107 	enic = pmd_priv(eth_dev);
1108 	ret = vnic_dev_fw_info(enic->vdev, &info);
1109 	if (ret)
1110 		return ret;
1111 	snprintf(fw_version, fw_size, "%s %s",
1112 		 info->fw_version, info->fw_build);
1113 	fw_version[fw_size - 1] = '\0';
1114 	return 0;
1115 }
1116 
1117 static const struct eth_dev_ops enicpmd_eth_dev_ops = {
1118 	.dev_configure        = enicpmd_dev_configure,
1119 	.dev_start            = enicpmd_dev_start,
1120 	.dev_stop             = enicpmd_dev_stop,
1121 	.dev_set_link_up      = NULL,
1122 	.dev_set_link_down    = NULL,
1123 	.dev_close            = enicpmd_dev_close,
1124 	.promiscuous_enable   = enicpmd_dev_promiscuous_enable,
1125 	.promiscuous_disable  = enicpmd_dev_promiscuous_disable,
1126 	.allmulticast_enable  = enicpmd_dev_allmulticast_enable,
1127 	.allmulticast_disable = enicpmd_dev_allmulticast_disable,
1128 	.link_update          = enicpmd_dev_link_update,
1129 	.stats_get            = enicpmd_dev_stats_get,
1130 	.stats_reset          = enicpmd_dev_stats_reset,
1131 	.queue_stats_mapping_set = NULL,
1132 	.dev_infos_get        = enicpmd_dev_info_get,
1133 	.dev_supported_ptypes_get = enicpmd_dev_supported_ptypes_get,
1134 	.mtu_set              = enicpmd_mtu_set,
1135 	.vlan_filter_set      = NULL,
1136 	.vlan_tpid_set        = NULL,
1137 	.vlan_offload_set     = enicpmd_vlan_offload_set,
1138 	.vlan_strip_queue_set = NULL,
1139 	.rx_queue_start       = enicpmd_dev_rx_queue_start,
1140 	.rx_queue_stop        = enicpmd_dev_rx_queue_stop,
1141 	.tx_queue_start       = enicpmd_dev_tx_queue_start,
1142 	.tx_queue_stop        = enicpmd_dev_tx_queue_stop,
1143 	.rx_queue_setup       = enicpmd_dev_rx_queue_setup,
1144 	.rx_queue_release     = enicpmd_dev_rx_queue_release,
1145 	.tx_queue_setup       = enicpmd_dev_tx_queue_setup,
1146 	.tx_queue_release     = enicpmd_dev_tx_queue_release,
1147 	.rx_queue_intr_enable = enicpmd_dev_rx_queue_intr_enable,
1148 	.rx_queue_intr_disable = enicpmd_dev_rx_queue_intr_disable,
1149 	.rxq_info_get         = enicpmd_dev_rxq_info_get,
1150 	.txq_info_get         = enicpmd_dev_txq_info_get,
1151 	.rx_burst_mode_get    = enicpmd_dev_rx_burst_mode_get,
1152 	.tx_burst_mode_get    = enicpmd_dev_tx_burst_mode_get,
1153 	.dev_led_on           = NULL,
1154 	.dev_led_off          = NULL,
1155 	.flow_ctrl_get        = NULL,
1156 	.flow_ctrl_set        = NULL,
1157 	.priority_flow_ctrl_set = NULL,
1158 	.mac_addr_add         = enicpmd_add_mac_addr,
1159 	.mac_addr_remove      = enicpmd_remove_mac_addr,
1160 	.mac_addr_set         = enicpmd_set_mac_addr,
1161 	.set_mc_addr_list     = enicpmd_set_mc_addr_list,
1162 	.filter_ctrl          = enicpmd_dev_filter_ctrl,
1163 	.reta_query           = enicpmd_dev_rss_reta_query,
1164 	.reta_update          = enicpmd_dev_rss_reta_update,
1165 	.rss_hash_conf_get    = enicpmd_dev_rss_hash_conf_get,
1166 	.rss_hash_update      = enicpmd_dev_rss_hash_update,
1167 	.udp_tunnel_port_add  = enicpmd_dev_udp_tunnel_port_add,
1168 	.udp_tunnel_port_del  = enicpmd_dev_udp_tunnel_port_del,
1169 	.fw_version_get       = enicpmd_dev_fw_version_get,
1170 };
1171 
1172 static int enic_parse_zero_one(const char *key,
1173 			       const char *value,
1174 			       void *opaque)
1175 {
1176 	struct enic *enic;
1177 	bool b;
1178 
1179 	enic = (struct enic *)opaque;
1180 	if (strcmp(value, "0") == 0) {
1181 		b = false;
1182 	} else if (strcmp(value, "1") == 0) {
1183 		b = true;
1184 	} else {
1185 		dev_err(enic, "Invalid value for %s"
1186 			": expected=0|1 given=%s\n", key, value);
1187 		return -EINVAL;
1188 	}
1189 	if (strcmp(key, ENIC_DEVARG_DISABLE_OVERLAY) == 0)
1190 		enic->disable_overlay = b;
1191 	if (strcmp(key, ENIC_DEVARG_ENABLE_AVX2_RX) == 0)
1192 		enic->enable_avx2_rx = b;
1193 	if (strcmp(key, ENIC_DEVARG_GENEVE_OPT) == 0)
1194 		enic->geneve_opt_request = b;
1195 	return 0;
1196 }
1197 
1198 static int enic_parse_ig_vlan_rewrite(__rte_unused const char *key,
1199 				      const char *value,
1200 				      void *opaque)
1201 {
1202 	struct enic *enic;
1203 
1204 	enic = (struct enic *)opaque;
1205 	if (strcmp(value, "trunk") == 0) {
1206 		/* Trunk mode: always tag */
1207 		enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_DEFAULT_TRUNK;
1208 	} else if (strcmp(value, "untag") == 0) {
1209 		/* Untag default VLAN mode: untag if VLAN = default VLAN */
1210 		enic->ig_vlan_rewrite_mode =
1211 			IG_VLAN_REWRITE_MODE_UNTAG_DEFAULT_VLAN;
1212 	} else if (strcmp(value, "priority") == 0) {
1213 		/*
1214 		 * Priority-tag default VLAN mode: priority tag (VLAN header
1215 		 * with ID=0) if VLAN = default
1216 		 */
1217 		enic->ig_vlan_rewrite_mode =
1218 			IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN;
1219 	} else if (strcmp(value, "pass") == 0) {
1220 		/* Pass through mode: do not touch tags */
1221 		enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1222 	} else {
1223 		dev_err(enic, "Invalid value for " ENIC_DEVARG_IG_VLAN_REWRITE
1224 			": expected=trunk|untag|priority|pass given=%s\n",
1225 			value);
1226 		return -EINVAL;
1227 	}
1228 	return 0;
1229 }
1230 
1231 static int enic_check_devargs(struct rte_eth_dev *dev)
1232 {
1233 	static const char *const valid_keys[] = {
1234 		ENIC_DEVARG_DISABLE_OVERLAY,
1235 		ENIC_DEVARG_ENABLE_AVX2_RX,
1236 		ENIC_DEVARG_GENEVE_OPT,
1237 		ENIC_DEVARG_IG_VLAN_REWRITE,
1238 		ENIC_DEVARG_REPRESENTOR,
1239 		NULL};
1240 	struct enic *enic = pmd_priv(dev);
1241 	struct rte_kvargs *kvlist;
1242 
1243 	ENICPMD_FUNC_TRACE();
1244 
1245 	enic->disable_overlay = false;
1246 	enic->enable_avx2_rx = false;
1247 	enic->geneve_opt_request = false;
1248 	enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1249 	if (!dev->device->devargs)
1250 		return 0;
1251 	kvlist = rte_kvargs_parse(dev->device->devargs->args, valid_keys);
1252 	if (!kvlist)
1253 		return -EINVAL;
1254 	if (rte_kvargs_process(kvlist, ENIC_DEVARG_DISABLE_OVERLAY,
1255 			       enic_parse_zero_one, enic) < 0 ||
1256 	    rte_kvargs_process(kvlist, ENIC_DEVARG_ENABLE_AVX2_RX,
1257 			       enic_parse_zero_one, enic) < 0 ||
1258 	    rte_kvargs_process(kvlist, ENIC_DEVARG_GENEVE_OPT,
1259 			       enic_parse_zero_one, enic) < 0 ||
1260 	    rte_kvargs_process(kvlist, ENIC_DEVARG_IG_VLAN_REWRITE,
1261 			       enic_parse_ig_vlan_rewrite, enic) < 0) {
1262 		rte_kvargs_free(kvlist);
1263 		return -EINVAL;
1264 	}
1265 	rte_kvargs_free(kvlist);
1266 	return 0;
1267 }
1268 
1269 /* Initialize the driver for PF */
1270 static int eth_enic_dev_init(struct rte_eth_dev *eth_dev,
1271 			     void *init_params __rte_unused)
1272 {
1273 	struct rte_pci_device *pdev;
1274 	struct rte_pci_addr *addr;
1275 	struct enic *enic = pmd_priv(eth_dev);
1276 	int err;
1277 
1278 	ENICPMD_FUNC_TRACE();
1279 	eth_dev->dev_ops = &enicpmd_eth_dev_ops;
1280 	eth_dev->rx_queue_count = enicpmd_dev_rx_queue_count;
1281 	eth_dev->rx_pkt_burst = &enic_recv_pkts;
1282 	eth_dev->tx_pkt_burst = &enic_xmit_pkts;
1283 	eth_dev->tx_pkt_prepare = &enic_prep_pkts;
1284 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1285 		enic_pick_tx_handler(eth_dev);
1286 		enic_pick_rx_handler(eth_dev);
1287 		return 0;
1288 	}
1289 	/* Only the primary sets up adapter and other data in shared memory */
1290 	enic->port_id = eth_dev->data->port_id;
1291 	enic->rte_dev = eth_dev;
1292 	enic->dev_data = eth_dev->data;
1293 	/* Let rte_eth_dev_close() release the port resources */
1294 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1295 
1296 	pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
1297 	rte_eth_copy_pci_info(eth_dev, pdev);
1298 	enic->pdev = pdev;
1299 	addr = &pdev->addr;
1300 
1301 	snprintf(enic->bdf_name, ENICPMD_BDF_LENGTH, "%04x:%02x:%02x.%x",
1302 		addr->domain, addr->bus, addr->devid, addr->function);
1303 
1304 	err = enic_check_devargs(eth_dev);
1305 	if (err)
1306 		return err;
1307 	err = enic_probe(enic);
1308 	if (!err && enic->fm) {
1309 		err = enic_fm_allocate_switch_domain(enic);
1310 		if (err)
1311 			ENICPMD_LOG(ERR, "failed to allocate switch domain id");
1312 	}
1313 	return err;
1314 }
1315 
1316 static int eth_enic_dev_uninit(struct rte_eth_dev *eth_dev)
1317 {
1318 	struct enic *enic = pmd_priv(eth_dev);
1319 	int err;
1320 
1321 	ENICPMD_FUNC_TRACE();
1322 	eth_dev->device = NULL;
1323 	eth_dev->intr_handle = NULL;
1324 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1325 		return 0;
1326 	err = rte_eth_switch_domain_free(enic->switch_domain_id);
1327 	if (err)
1328 		ENICPMD_LOG(WARNING, "failed to free switch domain: %d", err);
1329 	return 0;
1330 }
1331 
1332 static int eth_enic_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1333 	struct rte_pci_device *pci_dev)
1334 {
1335 	char name[RTE_ETH_NAME_MAX_LEN];
1336 	struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
1337 	struct rte_eth_dev *pf_ethdev;
1338 	struct enic *pf_enic;
1339 	int i, retval;
1340 
1341 	ENICPMD_FUNC_TRACE();
1342 	if (pci_dev->device.devargs) {
1343 		retval = rte_eth_devargs_parse(pci_dev->device.devargs->args,
1344 				&eth_da);
1345 		if (retval)
1346 			return retval;
1347 	}
1348 	retval = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
1349 		sizeof(struct enic),
1350 		eth_dev_pci_specific_init, pci_dev,
1351 		eth_enic_dev_init, NULL);
1352 	if (retval || eth_da.nb_representor_ports < 1)
1353 		return retval;
1354 
1355 	/* Probe VF representor */
1356 	pf_ethdev = rte_eth_dev_allocated(pci_dev->device.name);
1357 	if (pf_ethdev == NULL)
1358 		return -ENODEV;
1359 	/* Representors require flowman */
1360 	pf_enic = pmd_priv(pf_ethdev);
1361 	if (pf_enic->fm == NULL) {
1362 		ENICPMD_LOG(ERR, "VF representors require flowman");
1363 		return -ENOTSUP;
1364 	}
1365 	/*
1366 	 * For now representors imply switchdev, as firmware does not support
1367 	 * legacy mode SR-IOV
1368 	 */
1369 	pf_enic->switchdev_mode = 1;
1370 	/* Calculate max VF ID before initializing representor*/
1371 	pf_enic->max_vf_id = 0;
1372 	for (i = 0; i < eth_da.nb_representor_ports; i++) {
1373 		pf_enic->max_vf_id = RTE_MAX(pf_enic->max_vf_id,
1374 					     eth_da.representor_ports[i]);
1375 	}
1376 	for (i = 0; i < eth_da.nb_representor_ports; i++) {
1377 		struct enic_vf_representor representor;
1378 
1379 		representor.vf_id = eth_da.representor_ports[i];
1380 				representor.switch_domain_id =
1381 			pmd_priv(pf_ethdev)->switch_domain_id;
1382 		representor.pf = pmd_priv(pf_ethdev);
1383 		snprintf(name, sizeof(name), "net_%s_representor_%d",
1384 			pci_dev->device.name, eth_da.representor_ports[i]);
1385 		retval = rte_eth_dev_create(&pci_dev->device, name,
1386 			sizeof(struct enic_vf_representor), NULL, NULL,
1387 			enic_vf_representor_init, &representor);
1388 		if (retval) {
1389 			ENICPMD_LOG(ERR, "failed to create enic vf representor %s",
1390 				    name);
1391 			return retval;
1392 		}
1393 	}
1394 	return 0;
1395 }
1396 
1397 static int eth_enic_pci_remove(struct rte_pci_device *pci_dev)
1398 {
1399 	struct rte_eth_dev *ethdev;
1400 
1401 	ENICPMD_FUNC_TRACE();
1402 	ethdev = rte_eth_dev_allocated(pci_dev->device.name);
1403 	if (!ethdev)
1404 		return -ENODEV;
1405 	if (ethdev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
1406 		return rte_eth_dev_destroy(ethdev, enic_vf_representor_uninit);
1407 	else
1408 		return rte_eth_dev_destroy(ethdev, eth_enic_dev_uninit);
1409 }
1410 
1411 static struct rte_pci_driver rte_enic_pmd = {
1412 	.id_table = pci_id_enic_map,
1413 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1414 	.probe = eth_enic_pci_probe,
1415 	.remove = eth_enic_pci_remove,
1416 };
1417 
1418 int dev_is_enic(struct rte_eth_dev *dev)
1419 {
1420 	return dev->device->driver == &rte_enic_pmd.driver;
1421 }
1422 
1423 RTE_PMD_REGISTER_PCI(net_enic, rte_enic_pmd);
1424 RTE_PMD_REGISTER_PCI_TABLE(net_enic, pci_id_enic_map);
1425 RTE_PMD_REGISTER_KMOD_DEP(net_enic, "* igb_uio | uio_pci_generic | vfio-pci");
1426 RTE_PMD_REGISTER_PARAM_STRING(net_enic,
1427 	ENIC_DEVARG_DISABLE_OVERLAY "=0|1 "
1428 	ENIC_DEVARG_ENABLE_AVX2_RX "=0|1 "
1429 	ENIC_DEVARG_GENEVE_OPT "=0|1 "
1430 	ENIC_DEVARG_IG_VLAN_REWRITE "=trunk|untag|priority|pass");
1431