xref: /dpdk/drivers/net/enic/enic_ethdev.c (revision 081e42dab11d1add2d038fdf2bd4c86b20043d08)
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 <ethdev_driver.h>
13 #include <ethdev_pci.h>
14 #include <rte_geneve.h>
15 #include <rte_kvargs.h>
16 #include <rte_string_fns.h>
17 
18 #include "vnic_intr.h"
19 #include "vnic_cq.h"
20 #include "vnic_wq.h"
21 #include "vnic_rq.h"
22 #include "vnic_enet.h"
23 #include "enic.h"
24 
25 /*
26  * The set of PCI devices this driver supports
27  */
28 #define CISCO_PCI_VENDOR_ID 0x1137
29 static const struct rte_pci_id pci_id_enic_map[] = {
30 	{RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET)},
31 	{RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET_VF)},
32 	{RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET_SN)},
33 	{.vendor_id = 0, /* sentinel */},
34 };
35 
36 /* Supported link speeds of production VIC models */
37 static const struct vic_speed_capa {
38 	uint16_t sub_devid;
39 	uint32_t capa;
40 } vic_speed_capa_map[] = {
41 	{ 0x0043, ETH_LINK_SPEED_10G }, /* VIC */
42 	{ 0x0047, ETH_LINK_SPEED_10G }, /* P81E PCIe */
43 	{ 0x0048, ETH_LINK_SPEED_10G }, /* M81KR Mezz */
44 	{ 0x004f, ETH_LINK_SPEED_10G }, /* 1280 Mezz */
45 	{ 0x0084, ETH_LINK_SPEED_10G }, /* 1240 MLOM */
46 	{ 0x0085, ETH_LINK_SPEED_10G }, /* 1225 PCIe */
47 	{ 0x00cd, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1285 PCIe */
48 	{ 0x00ce, ETH_LINK_SPEED_10G }, /* 1225T PCIe */
49 	{ 0x012a, ETH_LINK_SPEED_40G }, /* M4308 */
50 	{ 0x012c, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1340 MLOM */
51 	{ 0x012e, ETH_LINK_SPEED_10G }, /* 1227 PCIe */
52 	{ 0x0137, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1380 Mezz */
53 	{ 0x014d, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1385 PCIe */
54 	{ 0x015d, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1387 MLOM */
55 	{ 0x0215, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G |
56 		  ETH_LINK_SPEED_40G }, /* 1440 Mezz */
57 	{ 0x0216, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G |
58 		  ETH_LINK_SPEED_40G }, /* 1480 MLOM */
59 	{ 0x0217, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G }, /* 1455 PCIe */
60 	{ 0x0218, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G }, /* 1457 MLOM */
61 	{ 0x0219, ETH_LINK_SPEED_40G }, /* 1485 PCIe */
62 	{ 0x021a, ETH_LINK_SPEED_40G }, /* 1487 MLOM */
63 	{ 0x024a, ETH_LINK_SPEED_40G | ETH_LINK_SPEED_100G }, /* 1495 PCIe */
64 	{ 0x024b, ETH_LINK_SPEED_40G | ETH_LINK_SPEED_100G }, /* 1497 MLOM */
65 	{ 0, 0 }, /* End marker */
66 };
67 
68 #define ENIC_DEVARG_CQ64 "cq64"
69 #define ENIC_DEVARG_DISABLE_OVERLAY "disable-overlay"
70 #define ENIC_DEVARG_ENABLE_AVX2_RX "enable-avx2-rx"
71 #define ENIC_DEVARG_IG_VLAN_REWRITE "ig-vlan-rewrite"
72 #define ENIC_DEVARG_REPRESENTOR "representor"
73 
74 RTE_LOG_REGISTER_DEFAULT(enic_pmd_logtype, INFO);
75 
76 static int
77 enicpmd_dev_flow_ops_get(struct rte_eth_dev *dev,
78 			 const struct rte_flow_ops **ops)
79 {
80 	struct enic *enic = pmd_priv(dev);
81 
82 	ENICPMD_FUNC_TRACE();
83 
84 	if (enic->flow_filter_mode == FILTER_FLOWMAN)
85 		*ops = &enic_fm_flow_ops;
86 	else
87 		*ops = &enic_flow_ops;
88 	return 0;
89 }
90 
91 static void enicpmd_dev_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
92 {
93 	void *txq = dev->data->tx_queues[qid];
94 
95 	ENICPMD_FUNC_TRACE();
96 
97 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
98 		return;
99 
100 	enic_free_wq(txq);
101 }
102 
103 static int enicpmd_dev_setup_intr(struct enic *enic)
104 {
105 	int ret;
106 	unsigned int index;
107 
108 	ENICPMD_FUNC_TRACE();
109 
110 	/* Are we done with the init of all the queues? */
111 	for (index = 0; index < enic->cq_count; index++) {
112 		if (!enic->cq[index].ctrl)
113 			break;
114 	}
115 	if (enic->cq_count != index)
116 		return 0;
117 	for (index = 0; index < enic->wq_count; index++) {
118 		if (!enic->wq[index].ctrl)
119 			break;
120 	}
121 	if (enic->wq_count != index)
122 		return 0;
123 	/* check start of packet (SOP) RQs only in case scatter is disabled. */
124 	for (index = 0; index < enic->rq_count; index++) {
125 		if (!enic->rq[enic_rte_rq_idx_to_sop_idx(index)].ctrl)
126 			break;
127 	}
128 	if (enic->rq_count != index)
129 		return 0;
130 
131 	ret = enic_alloc_intr_resources(enic);
132 	if (ret) {
133 		dev_err(enic, "alloc intr failed\n");
134 		return ret;
135 	}
136 	enic_init_vnic_resources(enic);
137 
138 	ret = enic_setup_finish(enic);
139 	if (ret)
140 		dev_err(enic, "setup could not be finished\n");
141 
142 	return ret;
143 }
144 
145 static int enicpmd_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
146 	uint16_t queue_idx,
147 	uint16_t nb_desc,
148 	unsigned int socket_id,
149 	const struct rte_eth_txconf *tx_conf)
150 {
151 	int ret;
152 	struct enic *enic = pmd_priv(eth_dev);
153 	struct vnic_wq *wq;
154 
155 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
156 		return -E_RTE_SECONDARY;
157 
158 	ENICPMD_FUNC_TRACE();
159 	RTE_ASSERT(queue_idx < enic->conf_wq_count);
160 	wq = &enic->wq[queue_idx];
161 	wq->offloads = tx_conf->offloads |
162 		eth_dev->data->dev_conf.txmode.offloads;
163 	eth_dev->data->tx_queues[queue_idx] = (void *)wq;
164 
165 	ret = enic_alloc_wq(enic, queue_idx, socket_id, nb_desc);
166 	if (ret) {
167 		dev_err(enic, "error in allocating wq\n");
168 		return ret;
169 	}
170 
171 	return enicpmd_dev_setup_intr(enic);
172 }
173 
174 static int enicpmd_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
175 	uint16_t queue_idx)
176 {
177 	struct enic *enic = pmd_priv(eth_dev);
178 
179 	ENICPMD_FUNC_TRACE();
180 
181 	enic_start_wq(enic, queue_idx);
182 
183 	return 0;
184 }
185 
186 static int enicpmd_dev_tx_queue_stop(struct rte_eth_dev *eth_dev,
187 	uint16_t queue_idx)
188 {
189 	int ret;
190 	struct enic *enic = pmd_priv(eth_dev);
191 
192 	ENICPMD_FUNC_TRACE();
193 
194 	ret = enic_stop_wq(enic, queue_idx);
195 	if (ret)
196 		dev_err(enic, "error in stopping wq %d\n", queue_idx);
197 
198 	return ret;
199 }
200 
201 static int enicpmd_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
202 	uint16_t queue_idx)
203 {
204 	struct enic *enic = pmd_priv(eth_dev);
205 
206 	ENICPMD_FUNC_TRACE();
207 
208 	enic_start_rq(enic, queue_idx);
209 
210 	return 0;
211 }
212 
213 static int enicpmd_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
214 	uint16_t queue_idx)
215 {
216 	int ret;
217 	struct enic *enic = pmd_priv(eth_dev);
218 
219 	ENICPMD_FUNC_TRACE();
220 
221 	ret = enic_stop_rq(enic, queue_idx);
222 	if (ret)
223 		dev_err(enic, "error in stopping rq %d\n", queue_idx);
224 
225 	return ret;
226 }
227 
228 static void enicpmd_dev_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
229 {
230 	void *rxq = dev->data->rx_queues[qid];
231 
232 	ENICPMD_FUNC_TRACE();
233 
234 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
235 		return;
236 
237 	enic_free_rq(rxq);
238 }
239 
240 static uint32_t enicpmd_dev_rx_queue_count(void *rx_queue)
241 {
242 	struct enic *enic;
243 	struct vnic_rq *sop_rq;
244 	uint32_t queue_count = 0;
245 	struct vnic_cq *cq;
246 	uint32_t cq_tail;
247 	uint16_t cq_idx;
248 
249 	sop_rq = rx_queue;
250 	enic = vnic_dev_priv(sop_rq->vdev);
251 	cq = &enic->cq[enic_cq_rq(enic, sop_rq->index)];
252 	cq_idx = cq->to_clean;
253 
254 	cq_tail = ioread32(&cq->ctrl->cq_tail);
255 
256 	if (cq_tail < cq_idx)
257 		cq_tail += cq->ring.desc_count;
258 
259 	queue_count = cq_tail - cq_idx;
260 
261 	return queue_count;
262 }
263 
264 static int enicpmd_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
265 	uint16_t queue_idx,
266 	uint16_t nb_desc,
267 	unsigned int socket_id,
268 	const struct rte_eth_rxconf *rx_conf,
269 	struct rte_mempool *mp)
270 {
271 	int ret;
272 	struct enic *enic = pmd_priv(eth_dev);
273 
274 	ENICPMD_FUNC_TRACE();
275 
276 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
277 		return -E_RTE_SECONDARY;
278 	RTE_ASSERT(enic_rte_rq_idx_to_sop_idx(queue_idx) < enic->conf_rq_count);
279 	eth_dev->data->rx_queues[queue_idx] =
280 		(void *)&enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
281 
282 	ret = enic_alloc_rq(enic, queue_idx, socket_id, mp, nb_desc,
283 			    rx_conf->rx_free_thresh);
284 	if (ret) {
285 		dev_err(enic, "error in allocating rq\n");
286 		return ret;
287 	}
288 
289 	return enicpmd_dev_setup_intr(enic);
290 }
291 
292 static int enicpmd_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
293 {
294 	struct enic *enic = pmd_priv(eth_dev);
295 	uint64_t offloads;
296 
297 	ENICPMD_FUNC_TRACE();
298 
299 	offloads = eth_dev->data->dev_conf.rxmode.offloads;
300 	if (mask & ETH_VLAN_STRIP_MASK) {
301 		if (offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
302 			enic->ig_vlan_strip_en = 1;
303 		else
304 			enic->ig_vlan_strip_en = 0;
305 	}
306 
307 	return enic_set_vlan_strip(enic);
308 }
309 
310 static int enicpmd_dev_configure(struct rte_eth_dev *eth_dev)
311 {
312 	int ret;
313 	int mask;
314 	struct enic *enic = pmd_priv(eth_dev);
315 
316 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
317 		return -E_RTE_SECONDARY;
318 
319 	ENICPMD_FUNC_TRACE();
320 	ret = enic_set_vnic_res(enic);
321 	if (ret) {
322 		dev_err(enic, "Set vNIC resource num  failed, aborting\n");
323 		return ret;
324 	}
325 
326 	if (eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
327 		eth_dev->data->dev_conf.rxmode.offloads |=
328 			DEV_RX_OFFLOAD_RSS_HASH;
329 
330 	enic->mc_count = 0;
331 	enic->hw_ip_checksum = !!(eth_dev->data->dev_conf.rxmode.offloads &
332 				  DEV_RX_OFFLOAD_CHECKSUM);
333 	/* All vlan offload masks to apply the current settings */
334 	mask = ETH_VLAN_STRIP_MASK |
335 		ETH_VLAN_FILTER_MASK |
336 		ETH_VLAN_EXTEND_MASK;
337 	ret = enicpmd_vlan_offload_set(eth_dev, mask);
338 	if (ret) {
339 		dev_err(enic, "Failed to configure VLAN offloads\n");
340 		return ret;
341 	}
342 	/*
343 	 * Initialize RSS with the default reta and key. If the user key is
344 	 * given (rx_adv_conf.rss_conf.rss_key), will use that instead of the
345 	 * default key.
346 	 */
347 	return enic_init_rss_nic_cfg(enic);
348 }
349 
350 /* Start the device.
351  * It returns 0 on success.
352  */
353 static int enicpmd_dev_start(struct rte_eth_dev *eth_dev)
354 {
355 	struct enic *enic = pmd_priv(eth_dev);
356 
357 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
358 		return -E_RTE_SECONDARY;
359 
360 	ENICPMD_FUNC_TRACE();
361 	return enic_enable(enic);
362 }
363 
364 /*
365  * Stop device: disable rx and tx functions to allow for reconfiguring.
366  */
367 static int enicpmd_dev_stop(struct rte_eth_dev *eth_dev)
368 {
369 	struct rte_eth_link link;
370 	struct enic *enic = pmd_priv(eth_dev);
371 
372 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
373 		return 0;
374 
375 	ENICPMD_FUNC_TRACE();
376 	enic_disable(enic);
377 
378 	memset(&link, 0, sizeof(link));
379 	rte_eth_linkstatus_set(eth_dev, &link);
380 
381 	return 0;
382 }
383 
384 /*
385  * Stop device.
386  */
387 static int enicpmd_dev_close(struct rte_eth_dev *eth_dev)
388 {
389 	struct enic *enic = pmd_priv(eth_dev);
390 
391 	ENICPMD_FUNC_TRACE();
392 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
393 		return 0;
394 
395 	enic_remove(enic);
396 
397 	return 0;
398 }
399 
400 static int enicpmd_dev_link_update(struct rte_eth_dev *eth_dev,
401 	__rte_unused int wait_to_complete)
402 {
403 	ENICPMD_FUNC_TRACE();
404 	return enic_link_update(eth_dev);
405 }
406 
407 static int enicpmd_dev_stats_get(struct rte_eth_dev *eth_dev,
408 	struct rte_eth_stats *stats)
409 {
410 	struct enic *enic = pmd_priv(eth_dev);
411 
412 	ENICPMD_FUNC_TRACE();
413 	return enic_dev_stats_get(enic, stats);
414 }
415 
416 static int enicpmd_dev_stats_reset(struct rte_eth_dev *eth_dev)
417 {
418 	struct enic *enic = pmd_priv(eth_dev);
419 
420 	ENICPMD_FUNC_TRACE();
421 	return enic_dev_stats_clear(enic);
422 }
423 
424 static uint32_t speed_capa_from_pci_id(struct rte_eth_dev *eth_dev)
425 {
426 	const struct vic_speed_capa *m;
427 	struct rte_pci_device *pdev;
428 	uint16_t id;
429 
430 	pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
431 	id = pdev->id.subsystem_device_id;
432 	for (m = vic_speed_capa_map; m->sub_devid != 0; m++) {
433 		if (m->sub_devid == id)
434 			return m->capa;
435 	}
436 	/* 1300 and later models are at least 40G */
437 	if (id >= 0x0100)
438 		return ETH_LINK_SPEED_40G;
439 	/* VFs have subsystem id 0, check device id */
440 	if (id == 0) {
441 		/* Newer VF implies at least 40G model */
442 		if (pdev->id.device_id == PCI_DEVICE_ID_CISCO_VIC_ENET_SN)
443 			return ETH_LINK_SPEED_40G;
444 	}
445 	return ETH_LINK_SPEED_10G;
446 }
447 
448 static int enicpmd_dev_info_get(struct rte_eth_dev *eth_dev,
449 	struct rte_eth_dev_info *device_info)
450 {
451 	struct enic *enic = pmd_priv(eth_dev);
452 
453 	ENICPMD_FUNC_TRACE();
454 	/* Scattered Rx uses two receive queues per rx queue exposed to dpdk */
455 	device_info->max_rx_queues = enic->conf_rq_count / 2;
456 	device_info->max_tx_queues = enic->conf_wq_count;
457 	device_info->min_rx_bufsize = ENIC_MIN_MTU;
458 	/* "Max" mtu is not a typo. HW receives packet sizes up to the
459 	 * max mtu regardless of the current mtu (vNIC's mtu). vNIC mtu is
460 	 * a hint to the driver to size receive buffers accordingly so that
461 	 * larger-than-vnic-mtu packets get truncated.. For DPDK, we let
462 	 * the user decide the buffer size via rxmode.max_rx_pkt_len, basically
463 	 * ignoring vNIC mtu.
464 	 */
465 	device_info->max_rx_pktlen = enic_mtu_to_max_rx_pktlen(enic->max_mtu);
466 	device_info->max_mac_addrs = ENIC_UNICAST_PERFECT_FILTERS;
467 	device_info->min_mtu = ENIC_MIN_MTU;
468 	device_info->max_mtu = enic->max_mtu;
469 	device_info->rx_offload_capa = enic->rx_offload_capa;
470 	device_info->tx_offload_capa = enic->tx_offload_capa;
471 	device_info->tx_queue_offload_capa = enic->tx_queue_offload_capa;
472 	device_info->default_rxconf = (struct rte_eth_rxconf) {
473 		.rx_free_thresh = ENIC_DEFAULT_RX_FREE_THRESH
474 	};
475 	device_info->reta_size = enic->reta_size;
476 	device_info->hash_key_size = enic->hash_key_size;
477 	device_info->flow_type_rss_offloads = enic->flow_type_rss_offloads;
478 	device_info->rx_desc_lim = (struct rte_eth_desc_lim) {
479 		.nb_max = enic->config.rq_desc_count,
480 		.nb_min = ENIC_MIN_RQ_DESCS,
481 		.nb_align = ENIC_ALIGN_DESCS,
482 	};
483 	device_info->tx_desc_lim = (struct rte_eth_desc_lim) {
484 		.nb_max = enic->config.wq_desc_count,
485 		.nb_min = ENIC_MIN_WQ_DESCS,
486 		.nb_align = ENIC_ALIGN_DESCS,
487 		.nb_seg_max = ENIC_TX_XMIT_MAX,
488 		.nb_mtu_seg_max = ENIC_NON_TSO_MAX_DESC,
489 	};
490 	device_info->default_rxportconf = (struct rte_eth_dev_portconf) {
491 		.burst_size = ENIC_DEFAULT_RX_BURST,
492 		.ring_size = RTE_MIN(device_info->rx_desc_lim.nb_max,
493 			ENIC_DEFAULT_RX_RING_SIZE),
494 		.nb_queues = ENIC_DEFAULT_RX_RINGS,
495 	};
496 	device_info->default_txportconf = (struct rte_eth_dev_portconf) {
497 		.burst_size = ENIC_DEFAULT_TX_BURST,
498 		.ring_size = RTE_MIN(device_info->tx_desc_lim.nb_max,
499 			ENIC_DEFAULT_TX_RING_SIZE),
500 		.nb_queues = ENIC_DEFAULT_TX_RINGS,
501 	};
502 	device_info->speed_capa = speed_capa_from_pci_id(eth_dev);
503 
504 	return 0;
505 }
506 
507 static const uint32_t *enicpmd_dev_supported_ptypes_get(struct rte_eth_dev *dev)
508 {
509 	static const uint32_t ptypes[] = {
510 		RTE_PTYPE_L2_ETHER,
511 		RTE_PTYPE_L2_ETHER_VLAN,
512 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
513 		RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
514 		RTE_PTYPE_L4_TCP,
515 		RTE_PTYPE_L4_UDP,
516 		RTE_PTYPE_L4_FRAG,
517 		RTE_PTYPE_L4_NONFRAG,
518 		RTE_PTYPE_UNKNOWN
519 	};
520 	static const uint32_t ptypes_overlay[] = {
521 		RTE_PTYPE_L2_ETHER,
522 		RTE_PTYPE_L2_ETHER_VLAN,
523 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
524 		RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
525 		RTE_PTYPE_L4_TCP,
526 		RTE_PTYPE_L4_UDP,
527 		RTE_PTYPE_L4_FRAG,
528 		RTE_PTYPE_L4_NONFRAG,
529 		RTE_PTYPE_TUNNEL_GRENAT,
530 		RTE_PTYPE_INNER_L2_ETHER,
531 		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
532 		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
533 		RTE_PTYPE_INNER_L4_TCP,
534 		RTE_PTYPE_INNER_L4_UDP,
535 		RTE_PTYPE_INNER_L4_FRAG,
536 		RTE_PTYPE_INNER_L4_NONFRAG,
537 		RTE_PTYPE_UNKNOWN
538 	};
539 
540 	if (dev->rx_pkt_burst != enic_dummy_recv_pkts &&
541 	    dev->rx_pkt_burst != NULL) {
542 		struct enic *enic = pmd_priv(dev);
543 		if (enic->overlay_offload)
544 			return ptypes_overlay;
545 		else
546 			return ptypes;
547 	}
548 	return NULL;
549 }
550 
551 static int enicpmd_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
552 {
553 	struct enic *enic = pmd_priv(eth_dev);
554 	int ret;
555 
556 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
557 		return -E_RTE_SECONDARY;
558 
559 	ENICPMD_FUNC_TRACE();
560 
561 	enic->promisc = 1;
562 	ret = enic_add_packet_filter(enic);
563 	if (ret != 0)
564 		enic->promisc = 0;
565 
566 	return ret;
567 }
568 
569 static int enicpmd_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
570 {
571 	struct enic *enic = pmd_priv(eth_dev);
572 	int ret;
573 
574 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
575 		return -E_RTE_SECONDARY;
576 
577 	ENICPMD_FUNC_TRACE();
578 	enic->promisc = 0;
579 	ret = enic_add_packet_filter(enic);
580 	if (ret != 0)
581 		enic->promisc = 1;
582 
583 	return ret;
584 }
585 
586 static int enicpmd_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
587 {
588 	struct enic *enic = pmd_priv(eth_dev);
589 	int ret;
590 
591 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
592 		return -E_RTE_SECONDARY;
593 
594 	ENICPMD_FUNC_TRACE();
595 	enic->allmulti = 1;
596 	ret = enic_add_packet_filter(enic);
597 	if (ret != 0)
598 		enic->allmulti = 0;
599 
600 	return ret;
601 }
602 
603 static int enicpmd_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
604 {
605 	struct enic *enic = pmd_priv(eth_dev);
606 	int ret;
607 
608 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
609 		return -E_RTE_SECONDARY;
610 
611 	ENICPMD_FUNC_TRACE();
612 	enic->allmulti = 0;
613 	ret = enic_add_packet_filter(enic);
614 	if (ret != 0)
615 		enic->allmulti = 1;
616 
617 	return ret;
618 }
619 
620 static int enicpmd_add_mac_addr(struct rte_eth_dev *eth_dev,
621 	struct rte_ether_addr *mac_addr,
622 	__rte_unused uint32_t index, __rte_unused uint32_t pool)
623 {
624 	struct enic *enic = pmd_priv(eth_dev);
625 
626 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
627 		return -E_RTE_SECONDARY;
628 
629 	ENICPMD_FUNC_TRACE();
630 	return enic_set_mac_address(enic, mac_addr->addr_bytes);
631 }
632 
633 static void enicpmd_remove_mac_addr(struct rte_eth_dev *eth_dev, uint32_t index)
634 {
635 	struct enic *enic = pmd_priv(eth_dev);
636 
637 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
638 		return;
639 
640 	ENICPMD_FUNC_TRACE();
641 	if (enic_del_mac_address(enic, index))
642 		dev_err(enic, "del mac addr failed\n");
643 }
644 
645 static int enicpmd_set_mac_addr(struct rte_eth_dev *eth_dev,
646 				struct rte_ether_addr *addr)
647 {
648 	struct enic *enic = pmd_priv(eth_dev);
649 	int ret;
650 
651 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
652 		return -E_RTE_SECONDARY;
653 
654 	ENICPMD_FUNC_TRACE();
655 	ret = enic_del_mac_address(enic, 0);
656 	if (ret)
657 		return ret;
658 	return enic_set_mac_address(enic, addr->addr_bytes);
659 }
660 
661 static void debug_log_add_del_addr(struct rte_ether_addr *addr, bool add)
662 {
663 	char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
664 
665 	rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE, addr);
666 	ENICPMD_LOG(DEBUG, " %s address %s\n",
667 		     add ? "add" : "remove", mac_str);
668 }
669 
670 static int enicpmd_set_mc_addr_list(struct rte_eth_dev *eth_dev,
671 				    struct rte_ether_addr *mc_addr_set,
672 				    uint32_t nb_mc_addr)
673 {
674 	struct enic *enic = pmd_priv(eth_dev);
675 	char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
676 	struct rte_ether_addr *addr;
677 	uint32_t i, j;
678 	int ret;
679 
680 	ENICPMD_FUNC_TRACE();
681 
682 	/* Validate the given addresses first */
683 	for (i = 0; i < nb_mc_addr && mc_addr_set != NULL; i++) {
684 		addr = &mc_addr_set[i];
685 		if (!rte_is_multicast_ether_addr(addr) ||
686 		    rte_is_broadcast_ether_addr(addr)) {
687 			rte_ether_format_addr(mac_str,
688 					RTE_ETHER_ADDR_FMT_SIZE, addr);
689 			ENICPMD_LOG(ERR, " invalid multicast address %s\n",
690 				     mac_str);
691 			return -EINVAL;
692 		}
693 	}
694 
695 	/* Flush all if requested */
696 	if (nb_mc_addr == 0 || mc_addr_set == NULL) {
697 		ENICPMD_LOG(DEBUG, " flush multicast addresses\n");
698 		for (i = 0; i < enic->mc_count; i++) {
699 			addr = &enic->mc_addrs[i];
700 			debug_log_add_del_addr(addr, false);
701 			ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
702 			if (ret)
703 				return ret;
704 		}
705 		enic->mc_count = 0;
706 		return 0;
707 	}
708 
709 	if (nb_mc_addr > ENIC_MULTICAST_PERFECT_FILTERS) {
710 		ENICPMD_LOG(ERR, " too many multicast addresses: max=%d\n",
711 			     ENIC_MULTICAST_PERFECT_FILTERS);
712 		return -ENOSPC;
713 	}
714 	/*
715 	 * devcmd is slow, so apply the difference instead of flushing and
716 	 * adding everything.
717 	 * 1. Delete addresses on the NIC but not on the host
718 	 */
719 	for (i = 0; i < enic->mc_count; i++) {
720 		addr = &enic->mc_addrs[i];
721 		for (j = 0; j < nb_mc_addr; j++) {
722 			if (rte_is_same_ether_addr(addr, &mc_addr_set[j]))
723 				break;
724 		}
725 		if (j < nb_mc_addr)
726 			continue;
727 		debug_log_add_del_addr(addr, false);
728 		ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
729 		if (ret)
730 			return ret;
731 	}
732 	/* 2. Add addresses on the host but not on the NIC */
733 	for (i = 0; i < nb_mc_addr; i++) {
734 		addr = &mc_addr_set[i];
735 		for (j = 0; j < enic->mc_count; j++) {
736 			if (rte_is_same_ether_addr(addr, &enic->mc_addrs[j]))
737 				break;
738 		}
739 		if (j < enic->mc_count)
740 			continue;
741 		debug_log_add_del_addr(addr, true);
742 		ret = vnic_dev_add_addr(enic->vdev, addr->addr_bytes);
743 		if (ret)
744 			return ret;
745 	}
746 	/* Keep a copy so we can flush/apply later on.. */
747 	memcpy(enic->mc_addrs, mc_addr_set,
748 	       nb_mc_addr * sizeof(struct rte_ether_addr));
749 	enic->mc_count = nb_mc_addr;
750 	return 0;
751 }
752 
753 static int enicpmd_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
754 {
755 	struct enic *enic = pmd_priv(eth_dev);
756 
757 	ENICPMD_FUNC_TRACE();
758 	return enic_set_mtu(enic, mtu);
759 }
760 
761 static int enicpmd_dev_rss_reta_query(struct rte_eth_dev *dev,
762 				      struct rte_eth_rss_reta_entry64
763 				      *reta_conf,
764 				      uint16_t reta_size)
765 {
766 	struct enic *enic = pmd_priv(dev);
767 	uint16_t i, idx, shift;
768 
769 	ENICPMD_FUNC_TRACE();
770 	if (reta_size != ENIC_RSS_RETA_SIZE) {
771 		dev_err(enic, "reta_query: wrong reta_size. given=%u expected=%u\n",
772 			reta_size, ENIC_RSS_RETA_SIZE);
773 		return -EINVAL;
774 	}
775 
776 	for (i = 0; i < reta_size; i++) {
777 		idx = i / RTE_RETA_GROUP_SIZE;
778 		shift = i % RTE_RETA_GROUP_SIZE;
779 		if (reta_conf[idx].mask & (1ULL << shift))
780 			reta_conf[idx].reta[shift] = enic_sop_rq_idx_to_rte_idx(
781 				enic->rss_cpu.cpu[i / 4].b[i % 4]);
782 	}
783 
784 	return 0;
785 }
786 
787 static int enicpmd_dev_rss_reta_update(struct rte_eth_dev *dev,
788 				       struct rte_eth_rss_reta_entry64
789 				       *reta_conf,
790 				       uint16_t reta_size)
791 {
792 	struct enic *enic = pmd_priv(dev);
793 	union vnic_rss_cpu rss_cpu;
794 	uint16_t i, idx, shift;
795 
796 	ENICPMD_FUNC_TRACE();
797 	if (reta_size != ENIC_RSS_RETA_SIZE) {
798 		dev_err(enic, "reta_update: wrong reta_size. given=%u"
799 			" expected=%u\n",
800 			reta_size, ENIC_RSS_RETA_SIZE);
801 		return -EINVAL;
802 	}
803 	/*
804 	 * Start with the current reta and modify it per reta_conf, as we
805 	 * need to push the entire reta even if we only modify one entry.
806 	 */
807 	rss_cpu = enic->rss_cpu;
808 	for (i = 0; i < reta_size; i++) {
809 		idx = i / RTE_RETA_GROUP_SIZE;
810 		shift = i % RTE_RETA_GROUP_SIZE;
811 		if (reta_conf[idx].mask & (1ULL << shift))
812 			rss_cpu.cpu[i / 4].b[i % 4] =
813 				enic_rte_rq_idx_to_sop_idx(
814 					reta_conf[idx].reta[shift]);
815 	}
816 	return enic_set_rss_reta(enic, &rss_cpu);
817 }
818 
819 static int enicpmd_dev_rss_hash_update(struct rte_eth_dev *dev,
820 				       struct rte_eth_rss_conf *rss_conf)
821 {
822 	struct enic *enic = pmd_priv(dev);
823 
824 	ENICPMD_FUNC_TRACE();
825 	return enic_set_rss_conf(enic, rss_conf);
826 }
827 
828 static int enicpmd_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
829 					 struct rte_eth_rss_conf *rss_conf)
830 {
831 	struct enic *enic = pmd_priv(dev);
832 
833 	ENICPMD_FUNC_TRACE();
834 	if (rss_conf == NULL)
835 		return -EINVAL;
836 	if (rss_conf->rss_key != NULL &&
837 	    rss_conf->rss_key_len < ENIC_RSS_HASH_KEY_SIZE) {
838 		dev_err(enic, "rss_hash_conf_get: wrong rss_key_len. given=%u"
839 			" expected=%u+\n",
840 			rss_conf->rss_key_len, ENIC_RSS_HASH_KEY_SIZE);
841 		return -EINVAL;
842 	}
843 	rss_conf->rss_hf = enic->rss_hf;
844 	if (rss_conf->rss_key != NULL) {
845 		int i;
846 		for (i = 0; i < ENIC_RSS_HASH_KEY_SIZE; i++) {
847 			rss_conf->rss_key[i] =
848 				enic->rss_key.key[i / 10].b[i % 10];
849 		}
850 		rss_conf->rss_key_len = ENIC_RSS_HASH_KEY_SIZE;
851 	}
852 	return 0;
853 }
854 
855 static void enicpmd_dev_rxq_info_get(struct rte_eth_dev *dev,
856 				     uint16_t rx_queue_id,
857 				     struct rte_eth_rxq_info *qinfo)
858 {
859 	struct enic *enic = pmd_priv(dev);
860 	struct vnic_rq *rq_sop;
861 	struct vnic_rq *rq_data;
862 	struct rte_eth_rxconf *conf;
863 	uint16_t sop_queue_idx;
864 	uint16_t data_queue_idx;
865 
866 	ENICPMD_FUNC_TRACE();
867 	sop_queue_idx = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
868 	data_queue_idx = enic_rte_rq_idx_to_data_idx(rx_queue_id, enic);
869 	rq_sop = &enic->rq[sop_queue_idx];
870 	rq_data = &enic->rq[data_queue_idx]; /* valid if data_queue_enable */
871 	qinfo->mp = rq_sop->mp;
872 	qinfo->scattered_rx = rq_sop->data_queue_enable;
873 	qinfo->nb_desc = rq_sop->ring.desc_count;
874 	if (qinfo->scattered_rx)
875 		qinfo->nb_desc += rq_data->ring.desc_count;
876 	conf = &qinfo->conf;
877 	memset(conf, 0, sizeof(*conf));
878 	conf->rx_free_thresh = rq_sop->rx_free_thresh;
879 	conf->rx_drop_en = 1;
880 	/*
881 	 * Except VLAN stripping (port setting), all the checksum offloads
882 	 * are always enabled.
883 	 */
884 	conf->offloads = enic->rx_offload_capa;
885 	if (!enic->ig_vlan_strip_en)
886 		conf->offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
887 	/* rx_thresh and other fields are not applicable for enic */
888 }
889 
890 static void enicpmd_dev_txq_info_get(struct rte_eth_dev *dev,
891 				     uint16_t tx_queue_id,
892 				     struct rte_eth_txq_info *qinfo)
893 {
894 	struct enic *enic = pmd_priv(dev);
895 	struct vnic_wq *wq = &enic->wq[tx_queue_id];
896 
897 	ENICPMD_FUNC_TRACE();
898 	qinfo->nb_desc = wq->ring.desc_count;
899 	memset(&qinfo->conf, 0, sizeof(qinfo->conf));
900 	qinfo->conf.offloads = wq->offloads;
901 	/* tx_thresh, and all the other fields are not applicable for enic */
902 }
903 
904 static int enicpmd_dev_rx_burst_mode_get(struct rte_eth_dev *dev,
905 					 __rte_unused uint16_t queue_id,
906 					 struct rte_eth_burst_mode *mode)
907 {
908 	eth_rx_burst_t pkt_burst = dev->rx_pkt_burst;
909 	struct enic *enic = pmd_priv(dev);
910 	const char *info_str = NULL;
911 	int ret = -EINVAL;
912 
913 	ENICPMD_FUNC_TRACE();
914 	if (enic->use_noscatter_vec_rx_handler)
915 		info_str = "Vector AVX2 No Scatter";
916 	else if (pkt_burst == enic_noscatter_recv_pkts)
917 		info_str = "Scalar No Scatter";
918 	else if (pkt_burst == enic_recv_pkts)
919 		info_str = "Scalar";
920 	else if (pkt_burst == enic_recv_pkts_64)
921 		info_str = "Scalar 64B Completion";
922 	if (info_str) {
923 		strlcpy(mode->info, info_str, sizeof(mode->info));
924 		ret = 0;
925 	}
926 	return ret;
927 }
928 
929 static int enicpmd_dev_tx_burst_mode_get(struct rte_eth_dev *dev,
930 					 __rte_unused uint16_t queue_id,
931 					 struct rte_eth_burst_mode *mode)
932 {
933 	eth_tx_burst_t pkt_burst = dev->tx_pkt_burst;
934 	const char *info_str = NULL;
935 	int ret = -EINVAL;
936 
937 	ENICPMD_FUNC_TRACE();
938 	if (pkt_burst == enic_simple_xmit_pkts)
939 		info_str = "Scalar Simplified";
940 	else if (pkt_burst == enic_xmit_pkts)
941 		info_str = "Scalar";
942 	if (info_str) {
943 		strlcpy(mode->info, info_str, sizeof(mode->info));
944 		ret = 0;
945 	}
946 	return ret;
947 }
948 
949 static int enicpmd_dev_rx_queue_intr_enable(struct rte_eth_dev *eth_dev,
950 					    uint16_t rx_queue_id)
951 {
952 	struct enic *enic = pmd_priv(eth_dev);
953 
954 	ENICPMD_FUNC_TRACE();
955 	vnic_intr_unmask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
956 	return 0;
957 }
958 
959 static int enicpmd_dev_rx_queue_intr_disable(struct rte_eth_dev *eth_dev,
960 					     uint16_t rx_queue_id)
961 {
962 	struct enic *enic = pmd_priv(eth_dev);
963 
964 	ENICPMD_FUNC_TRACE();
965 	vnic_intr_mask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
966 	return 0;
967 }
968 
969 static int udp_tunnel_common_check(struct enic *enic,
970 				   struct rte_eth_udp_tunnel *tnl)
971 {
972 	if (tnl->prot_type != RTE_TUNNEL_TYPE_VXLAN &&
973 	    tnl->prot_type != RTE_TUNNEL_TYPE_GENEVE)
974 		return -ENOTSUP;
975 	if (!enic->overlay_offload) {
976 		ENICPMD_LOG(DEBUG, " overlay offload is not supported\n");
977 		return -ENOTSUP;
978 	}
979 	return 0;
980 }
981 
982 static int update_tunnel_port(struct enic *enic, uint16_t port, bool vxlan)
983 {
984 	uint8_t cfg;
985 
986 	cfg = vxlan ? OVERLAY_CFG_VXLAN_PORT_UPDATE :
987 		OVERLAY_CFG_GENEVE_PORT_UPDATE;
988 	if (vnic_dev_overlay_offload_cfg(enic->vdev, cfg, port)) {
989 		ENICPMD_LOG(DEBUG, " failed to update tunnel port\n");
990 		return -EINVAL;
991 	}
992 	ENICPMD_LOG(DEBUG, " updated %s port to %u\n",
993 		    vxlan ? "vxlan" : "geneve", port);
994 	if (vxlan)
995 		enic->vxlan_port = port;
996 	else
997 		enic->geneve_port = port;
998 	return 0;
999 }
1000 
1001 static int enicpmd_dev_udp_tunnel_port_add(struct rte_eth_dev *eth_dev,
1002 					   struct rte_eth_udp_tunnel *tnl)
1003 {
1004 	struct enic *enic = pmd_priv(eth_dev);
1005 	uint16_t port;
1006 	bool vxlan;
1007 	int ret;
1008 
1009 	ENICPMD_FUNC_TRACE();
1010 	ret = udp_tunnel_common_check(enic, tnl);
1011 	if (ret)
1012 		return ret;
1013 	vxlan = (tnl->prot_type == RTE_TUNNEL_TYPE_VXLAN);
1014 	if (vxlan)
1015 		port = enic->vxlan_port;
1016 	else
1017 		port = enic->geneve_port;
1018 	/*
1019 	 * The NIC has 1 configurable port number per tunnel type.
1020 	 * "Adding" a new port number replaces it.
1021 	 */
1022 	if (tnl->udp_port == port || tnl->udp_port == 0) {
1023 		ENICPMD_LOG(DEBUG, " %u is already configured or invalid\n",
1024 			     tnl->udp_port);
1025 		return -EINVAL;
1026 	}
1027 	return update_tunnel_port(enic, tnl->udp_port, vxlan);
1028 }
1029 
1030 static int enicpmd_dev_udp_tunnel_port_del(struct rte_eth_dev *eth_dev,
1031 					   struct rte_eth_udp_tunnel *tnl)
1032 {
1033 	struct enic *enic = pmd_priv(eth_dev);
1034 	uint16_t port;
1035 	bool vxlan;
1036 	int ret;
1037 
1038 	ENICPMD_FUNC_TRACE();
1039 	ret = udp_tunnel_common_check(enic, tnl);
1040 	if (ret)
1041 		return ret;
1042 	vxlan = (tnl->prot_type == RTE_TUNNEL_TYPE_VXLAN);
1043 	if (vxlan)
1044 		port = enic->vxlan_port;
1045 	else
1046 		port = enic->geneve_port;
1047 	/*
1048 	 * Clear the previously set port number and restore the
1049 	 * hardware default port number. Some drivers disable VXLAN
1050 	 * offloads when there are no configured port numbers. But
1051 	 * enic does not do that as VXLAN is part of overlay offload,
1052 	 * which is tied to inner RSS and TSO.
1053 	 */
1054 	if (tnl->udp_port != port) {
1055 		ENICPMD_LOG(DEBUG, " %u is not a configured tunnel port\n",
1056 			     tnl->udp_port);
1057 		return -EINVAL;
1058 	}
1059 	port = vxlan ? RTE_VXLAN_DEFAULT_PORT : RTE_GENEVE_DEFAULT_PORT;
1060 	return update_tunnel_port(enic, port, vxlan);
1061 }
1062 
1063 static int enicpmd_dev_fw_version_get(struct rte_eth_dev *eth_dev,
1064 				      char *fw_version, size_t fw_size)
1065 {
1066 	struct vnic_devcmd_fw_info *info;
1067 	struct enic *enic;
1068 	int ret;
1069 
1070 	ENICPMD_FUNC_TRACE();
1071 
1072 	enic = pmd_priv(eth_dev);
1073 	ret = vnic_dev_fw_info(enic->vdev, &info);
1074 	if (ret)
1075 		return ret;
1076 	ret = snprintf(fw_version, fw_size, "%s %s",
1077 		 info->fw_version, info->fw_build);
1078 	if (ret < 0)
1079 		return -EINVAL;
1080 
1081 	ret += 1; /* add the size of '\0' */
1082 	if (fw_size < (size_t)ret)
1083 		return ret;
1084 	else
1085 		return 0;
1086 }
1087 
1088 static const struct eth_dev_ops enicpmd_eth_dev_ops = {
1089 	.dev_configure        = enicpmd_dev_configure,
1090 	.dev_start            = enicpmd_dev_start,
1091 	.dev_stop             = enicpmd_dev_stop,
1092 	.dev_set_link_up      = NULL,
1093 	.dev_set_link_down    = NULL,
1094 	.dev_close            = enicpmd_dev_close,
1095 	.promiscuous_enable   = enicpmd_dev_promiscuous_enable,
1096 	.promiscuous_disable  = enicpmd_dev_promiscuous_disable,
1097 	.allmulticast_enable  = enicpmd_dev_allmulticast_enable,
1098 	.allmulticast_disable = enicpmd_dev_allmulticast_disable,
1099 	.link_update          = enicpmd_dev_link_update,
1100 	.stats_get            = enicpmd_dev_stats_get,
1101 	.stats_reset          = enicpmd_dev_stats_reset,
1102 	.queue_stats_mapping_set = NULL,
1103 	.dev_infos_get        = enicpmd_dev_info_get,
1104 	.dev_supported_ptypes_get = enicpmd_dev_supported_ptypes_get,
1105 	.mtu_set              = enicpmd_mtu_set,
1106 	.vlan_filter_set      = NULL,
1107 	.vlan_tpid_set        = NULL,
1108 	.vlan_offload_set     = enicpmd_vlan_offload_set,
1109 	.vlan_strip_queue_set = NULL,
1110 	.rx_queue_start       = enicpmd_dev_rx_queue_start,
1111 	.rx_queue_stop        = enicpmd_dev_rx_queue_stop,
1112 	.tx_queue_start       = enicpmd_dev_tx_queue_start,
1113 	.tx_queue_stop        = enicpmd_dev_tx_queue_stop,
1114 	.rx_queue_setup       = enicpmd_dev_rx_queue_setup,
1115 	.rx_queue_release     = enicpmd_dev_rx_queue_release,
1116 	.tx_queue_setup       = enicpmd_dev_tx_queue_setup,
1117 	.tx_queue_release     = enicpmd_dev_tx_queue_release,
1118 	.rx_queue_intr_enable = enicpmd_dev_rx_queue_intr_enable,
1119 	.rx_queue_intr_disable = enicpmd_dev_rx_queue_intr_disable,
1120 	.rxq_info_get         = enicpmd_dev_rxq_info_get,
1121 	.txq_info_get         = enicpmd_dev_txq_info_get,
1122 	.rx_burst_mode_get    = enicpmd_dev_rx_burst_mode_get,
1123 	.tx_burst_mode_get    = enicpmd_dev_tx_burst_mode_get,
1124 	.dev_led_on           = NULL,
1125 	.dev_led_off          = NULL,
1126 	.flow_ctrl_get        = NULL,
1127 	.flow_ctrl_set        = NULL,
1128 	.priority_flow_ctrl_set = NULL,
1129 	.mac_addr_add         = enicpmd_add_mac_addr,
1130 	.mac_addr_remove      = enicpmd_remove_mac_addr,
1131 	.mac_addr_set         = enicpmd_set_mac_addr,
1132 	.set_mc_addr_list     = enicpmd_set_mc_addr_list,
1133 	.flow_ops_get         = enicpmd_dev_flow_ops_get,
1134 	.reta_query           = enicpmd_dev_rss_reta_query,
1135 	.reta_update          = enicpmd_dev_rss_reta_update,
1136 	.rss_hash_conf_get    = enicpmd_dev_rss_hash_conf_get,
1137 	.rss_hash_update      = enicpmd_dev_rss_hash_update,
1138 	.udp_tunnel_port_add  = enicpmd_dev_udp_tunnel_port_add,
1139 	.udp_tunnel_port_del  = enicpmd_dev_udp_tunnel_port_del,
1140 	.fw_version_get       = enicpmd_dev_fw_version_get,
1141 };
1142 
1143 static int enic_parse_zero_one(const char *key,
1144 			       const char *value,
1145 			       void *opaque)
1146 {
1147 	struct enic *enic;
1148 	bool b;
1149 
1150 	enic = (struct enic *)opaque;
1151 	if (strcmp(value, "0") == 0) {
1152 		b = false;
1153 	} else if (strcmp(value, "1") == 0) {
1154 		b = true;
1155 	} else {
1156 		dev_err(enic, "Invalid value for %s"
1157 			": expected=0|1 given=%s\n", key, value);
1158 		return -EINVAL;
1159 	}
1160 	if (strcmp(key, ENIC_DEVARG_CQ64) == 0)
1161 		enic->cq64_request = b;
1162 	if (strcmp(key, ENIC_DEVARG_DISABLE_OVERLAY) == 0)
1163 		enic->disable_overlay = b;
1164 	if (strcmp(key, ENIC_DEVARG_ENABLE_AVX2_RX) == 0)
1165 		enic->enable_avx2_rx = b;
1166 	return 0;
1167 }
1168 
1169 static int enic_parse_ig_vlan_rewrite(__rte_unused const char *key,
1170 				      const char *value,
1171 				      void *opaque)
1172 {
1173 	struct enic *enic;
1174 
1175 	enic = (struct enic *)opaque;
1176 	if (strcmp(value, "trunk") == 0) {
1177 		/* Trunk mode: always tag */
1178 		enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_DEFAULT_TRUNK;
1179 	} else if (strcmp(value, "untag") == 0) {
1180 		/* Untag default VLAN mode: untag if VLAN = default VLAN */
1181 		enic->ig_vlan_rewrite_mode =
1182 			IG_VLAN_REWRITE_MODE_UNTAG_DEFAULT_VLAN;
1183 	} else if (strcmp(value, "priority") == 0) {
1184 		/*
1185 		 * Priority-tag default VLAN mode: priority tag (VLAN header
1186 		 * with ID=0) if VLAN = default
1187 		 */
1188 		enic->ig_vlan_rewrite_mode =
1189 			IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN;
1190 	} else if (strcmp(value, "pass") == 0) {
1191 		/* Pass through mode: do not touch tags */
1192 		enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1193 	} else {
1194 		dev_err(enic, "Invalid value for " ENIC_DEVARG_IG_VLAN_REWRITE
1195 			": expected=trunk|untag|priority|pass given=%s\n",
1196 			value);
1197 		return -EINVAL;
1198 	}
1199 	return 0;
1200 }
1201 
1202 static int enic_check_devargs(struct rte_eth_dev *dev)
1203 {
1204 	static const char *const valid_keys[] = {
1205 		ENIC_DEVARG_CQ64,
1206 		ENIC_DEVARG_DISABLE_OVERLAY,
1207 		ENIC_DEVARG_ENABLE_AVX2_RX,
1208 		ENIC_DEVARG_IG_VLAN_REWRITE,
1209 		ENIC_DEVARG_REPRESENTOR,
1210 		NULL};
1211 	struct enic *enic = pmd_priv(dev);
1212 	struct rte_kvargs *kvlist;
1213 
1214 	ENICPMD_FUNC_TRACE();
1215 
1216 	enic->cq64_request = true; /* Use 64B entry if available */
1217 	enic->disable_overlay = false;
1218 	enic->enable_avx2_rx = false;
1219 	enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1220 	if (!dev->device->devargs)
1221 		return 0;
1222 	kvlist = rte_kvargs_parse(dev->device->devargs->args, valid_keys);
1223 	if (!kvlist)
1224 		return -EINVAL;
1225 	if (rte_kvargs_process(kvlist, ENIC_DEVARG_CQ64,
1226 			       enic_parse_zero_one, enic) < 0 ||
1227 	    rte_kvargs_process(kvlist, ENIC_DEVARG_DISABLE_OVERLAY,
1228 			       enic_parse_zero_one, enic) < 0 ||
1229 	    rte_kvargs_process(kvlist, ENIC_DEVARG_ENABLE_AVX2_RX,
1230 			       enic_parse_zero_one, enic) < 0 ||
1231 	    rte_kvargs_process(kvlist, ENIC_DEVARG_IG_VLAN_REWRITE,
1232 			       enic_parse_ig_vlan_rewrite, enic) < 0) {
1233 		rte_kvargs_free(kvlist);
1234 		return -EINVAL;
1235 	}
1236 	rte_kvargs_free(kvlist);
1237 	return 0;
1238 }
1239 
1240 /* Initialize the driver for PF */
1241 static int eth_enic_dev_init(struct rte_eth_dev *eth_dev,
1242 			     void *init_params __rte_unused)
1243 {
1244 	struct rte_pci_device *pdev;
1245 	struct rte_pci_addr *addr;
1246 	struct enic *enic = pmd_priv(eth_dev);
1247 	int err;
1248 
1249 	ENICPMD_FUNC_TRACE();
1250 	eth_dev->dev_ops = &enicpmd_eth_dev_ops;
1251 	eth_dev->rx_queue_count = enicpmd_dev_rx_queue_count;
1252 	eth_dev->rx_pkt_burst = &enic_recv_pkts;
1253 	eth_dev->tx_pkt_burst = &enic_xmit_pkts;
1254 	eth_dev->tx_pkt_prepare = &enic_prep_pkts;
1255 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1256 		enic_pick_tx_handler(eth_dev);
1257 		enic_pick_rx_handler(eth_dev);
1258 		return 0;
1259 	}
1260 	/* Only the primary sets up adapter and other data in shared memory */
1261 	enic->port_id = eth_dev->data->port_id;
1262 	enic->rte_dev = eth_dev;
1263 	enic->dev_data = eth_dev->data;
1264 
1265 	pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
1266 	rte_eth_copy_pci_info(eth_dev, pdev);
1267 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1268 	enic->pdev = pdev;
1269 	addr = &pdev->addr;
1270 
1271 	snprintf(enic->bdf_name, ENICPMD_BDF_LENGTH, "%04x:%02x:%02x.%x",
1272 		addr->domain, addr->bus, addr->devid, addr->function);
1273 
1274 	err = enic_check_devargs(eth_dev);
1275 	if (err)
1276 		return err;
1277 	err = enic_probe(enic);
1278 	if (!err && enic->fm) {
1279 		err = enic_fm_allocate_switch_domain(enic);
1280 		if (err)
1281 			ENICPMD_LOG(ERR, "failed to allocate switch domain id");
1282 	}
1283 	return err;
1284 }
1285 
1286 static int eth_enic_dev_uninit(struct rte_eth_dev *eth_dev)
1287 {
1288 	struct enic *enic = pmd_priv(eth_dev);
1289 	int err;
1290 
1291 	ENICPMD_FUNC_TRACE();
1292 	eth_dev->device = NULL;
1293 	eth_dev->intr_handle = NULL;
1294 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1295 		return 0;
1296 	err = rte_eth_switch_domain_free(enic->switch_domain_id);
1297 	if (err)
1298 		ENICPMD_LOG(WARNING, "failed to free switch domain: %d", err);
1299 	return 0;
1300 }
1301 
1302 static int eth_enic_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1303 	struct rte_pci_device *pci_dev)
1304 {
1305 	char name[RTE_ETH_NAME_MAX_LEN];
1306 	struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
1307 	struct rte_eth_dev *pf_ethdev;
1308 	struct enic *pf_enic;
1309 	int i, retval;
1310 
1311 	ENICPMD_FUNC_TRACE();
1312 	if (pci_dev->device.devargs) {
1313 		retval = rte_eth_devargs_parse(pci_dev->device.devargs->args,
1314 				&eth_da);
1315 		if (retval)
1316 			return retval;
1317 	}
1318 	if (eth_da.nb_representor_ports > 0 &&
1319 	    eth_da.type != RTE_ETH_REPRESENTOR_VF) {
1320 		ENICPMD_LOG(ERR, "unsupported representor type: %s\n",
1321 			    pci_dev->device.devargs->args);
1322 		return -ENOTSUP;
1323 	}
1324 	retval = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
1325 		sizeof(struct enic),
1326 		eth_dev_pci_specific_init, pci_dev,
1327 		eth_enic_dev_init, NULL);
1328 	if (retval || eth_da.nb_representor_ports < 1)
1329 		return retval;
1330 
1331 	/* Probe VF representor */
1332 	pf_ethdev = rte_eth_dev_allocated(pci_dev->device.name);
1333 	if (pf_ethdev == NULL)
1334 		return -ENODEV;
1335 	/* Representors require flowman */
1336 	pf_enic = pmd_priv(pf_ethdev);
1337 	if (pf_enic->fm == NULL) {
1338 		ENICPMD_LOG(ERR, "VF representors require flowman");
1339 		return -ENOTSUP;
1340 	}
1341 	/*
1342 	 * For now representors imply switchdev, as firmware does not support
1343 	 * legacy mode SR-IOV
1344 	 */
1345 	pf_enic->switchdev_mode = 1;
1346 	/* Calculate max VF ID before initializing representor*/
1347 	pf_enic->max_vf_id = 0;
1348 	for (i = 0; i < eth_da.nb_representor_ports; i++) {
1349 		pf_enic->max_vf_id = RTE_MAX(pf_enic->max_vf_id,
1350 					     eth_da.representor_ports[i]);
1351 	}
1352 	for (i = 0; i < eth_da.nb_representor_ports; i++) {
1353 		struct enic_vf_representor representor;
1354 
1355 		representor.vf_id = eth_da.representor_ports[i];
1356 				representor.switch_domain_id =
1357 			pmd_priv(pf_ethdev)->switch_domain_id;
1358 		representor.pf = pmd_priv(pf_ethdev);
1359 		snprintf(name, sizeof(name), "net_%s_representor_%d",
1360 			pci_dev->device.name, eth_da.representor_ports[i]);
1361 		retval = rte_eth_dev_create(&pci_dev->device, name,
1362 			sizeof(struct enic_vf_representor), NULL, NULL,
1363 			enic_vf_representor_init, &representor);
1364 		if (retval) {
1365 			ENICPMD_LOG(ERR, "failed to create enic vf representor %s",
1366 				    name);
1367 			return retval;
1368 		}
1369 	}
1370 	return 0;
1371 }
1372 
1373 static int eth_enic_pci_remove(struct rte_pci_device *pci_dev)
1374 {
1375 	struct rte_eth_dev *ethdev;
1376 
1377 	ENICPMD_FUNC_TRACE();
1378 	ethdev = rte_eth_dev_allocated(pci_dev->device.name);
1379 	if (!ethdev)
1380 		return -ENODEV;
1381 	if (ethdev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
1382 		return rte_eth_dev_destroy(ethdev, enic_vf_representor_uninit);
1383 	else
1384 		return rte_eth_dev_destroy(ethdev, eth_enic_dev_uninit);
1385 }
1386 
1387 static struct rte_pci_driver rte_enic_pmd = {
1388 	.id_table = pci_id_enic_map,
1389 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1390 	.probe = eth_enic_pci_probe,
1391 	.remove = eth_enic_pci_remove,
1392 };
1393 
1394 int dev_is_enic(struct rte_eth_dev *dev)
1395 {
1396 	return dev->device->driver == &rte_enic_pmd.driver;
1397 }
1398 
1399 RTE_PMD_REGISTER_PCI(net_enic, rte_enic_pmd);
1400 RTE_PMD_REGISTER_PCI_TABLE(net_enic, pci_id_enic_map);
1401 RTE_PMD_REGISTER_KMOD_DEP(net_enic, "* igb_uio | uio_pci_generic | vfio-pci");
1402 RTE_PMD_REGISTER_PARAM_STRING(net_enic,
1403 	ENIC_DEVARG_CQ64 "=0|1"
1404 	ENIC_DEVARG_DISABLE_OVERLAY "=0|1 "
1405 	ENIC_DEVARG_ENABLE_AVX2_RX "=0|1 "
1406 	ENIC_DEVARG_IG_VLAN_REWRITE "=trunk|untag|priority|pass");
1407