xref: /dpdk/drivers/net/thunderx/nicvf_ethdev.c (revision bb44fb6fe7713ddcd023d5b9bacadf074d68092e)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2016 Cavium, Inc
3  */
4 
5 #include <assert.h>
6 #include <stdio.h>
7 #include <stdbool.h>
8 #include <errno.h>
9 #include <stdint.h>
10 #include <string.h>
11 #include <unistd.h>
12 #include <stdarg.h>
13 #include <inttypes.h>
14 #include <netinet/in.h>
15 #include <sys/queue.h>
16 
17 #include <rte_alarm.h>
18 #include <rte_branch_prediction.h>
19 #include <rte_byteorder.h>
20 #include <rte_common.h>
21 #include <rte_cycles.h>
22 #include <rte_debug.h>
23 #include <rte_dev.h>
24 #include <rte_eal.h>
25 #include <rte_ether.h>
26 #include <rte_ethdev_driver.h>
27 #include <rte_ethdev_pci.h>
28 #include <rte_interrupts.h>
29 #include <rte_log.h>
30 #include <rte_memory.h>
31 #include <rte_memzone.h>
32 #include <rte_malloc.h>
33 #include <rte_random.h>
34 #include <rte_pci.h>
35 #include <rte_bus_pci.h>
36 #include <rte_tailq.h>
37 #include <rte_devargs.h>
38 #include <rte_kvargs.h>
39 
40 #include "base/nicvf_plat.h"
41 
42 #include "nicvf_ethdev.h"
43 #include "nicvf_rxtx.h"
44 #include "nicvf_svf.h"
45 #include "nicvf_logs.h"
46 
47 int nicvf_logtype_mbox;
48 int nicvf_logtype_init;
49 int nicvf_logtype_driver;
50 
51 static void nicvf_dev_stop(struct rte_eth_dev *dev);
52 static void nicvf_dev_stop_cleanup(struct rte_eth_dev *dev, bool cleanup);
53 static void nicvf_vf_stop(struct rte_eth_dev *dev, struct nicvf *nic,
54 			  bool cleanup);
55 
56 RTE_INIT(nicvf_init_log)
57 {
58 	nicvf_logtype_mbox = rte_log_register("pmd.net.thunderx.mbox");
59 	if (nicvf_logtype_mbox >= 0)
60 		rte_log_set_level(nicvf_logtype_mbox, RTE_LOG_NOTICE);
61 
62 	nicvf_logtype_init = rte_log_register("pmd.net.thunderx.init");
63 	if (nicvf_logtype_init >= 0)
64 		rte_log_set_level(nicvf_logtype_init, RTE_LOG_NOTICE);
65 
66 	nicvf_logtype_driver = rte_log_register("pmd.net.thunderx.driver");
67 	if (nicvf_logtype_driver >= 0)
68 		rte_log_set_level(nicvf_logtype_driver, RTE_LOG_NOTICE);
69 }
70 
71 static void
72 nicvf_link_status_update(struct nicvf *nic,
73 			 struct rte_eth_link *link)
74 {
75 	memset(link, 0, sizeof(*link));
76 
77 	link->link_status = nic->link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
78 
79 	if (nic->duplex == NICVF_HALF_DUPLEX)
80 		link->link_duplex = ETH_LINK_HALF_DUPLEX;
81 	else if (nic->duplex == NICVF_FULL_DUPLEX)
82 		link->link_duplex = ETH_LINK_FULL_DUPLEX;
83 	link->link_speed = nic->speed;
84 	link->link_autoneg = ETH_LINK_AUTONEG;
85 }
86 
87 static void
88 nicvf_interrupt(void *arg)
89 {
90 	struct rte_eth_dev *dev = arg;
91 	struct nicvf *nic = nicvf_pmd_priv(dev);
92 	struct rte_eth_link link;
93 
94 	if (nicvf_reg_poll_interrupts(nic) == NIC_MBOX_MSG_BGX_LINK_CHANGE) {
95 		if (dev->data->dev_conf.intr_conf.lsc) {
96 			nicvf_link_status_update(nic, &link);
97 			rte_eth_linkstatus_set(dev, &link);
98 
99 			_rte_eth_dev_callback_process(dev,
100 						      RTE_ETH_EVENT_INTR_LSC,
101 						      NULL);
102 		}
103 	}
104 
105 	rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
106 				nicvf_interrupt, dev);
107 }
108 
109 static void
110 nicvf_vf_interrupt(void *arg)
111 {
112 	struct nicvf *nic = arg;
113 
114 	nicvf_reg_poll_interrupts(nic);
115 
116 	rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
117 				nicvf_vf_interrupt, nic);
118 }
119 
120 static int
121 nicvf_periodic_alarm_start(void (fn)(void *), void *arg)
122 {
123 	return rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000, fn, arg);
124 }
125 
126 static int
127 nicvf_periodic_alarm_stop(void (fn)(void *), void *arg)
128 {
129 	return rte_eal_alarm_cancel(fn, arg);
130 }
131 
132 /*
133  * Return 0 means link status changed, -1 means not changed
134  */
135 static int
136 nicvf_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
137 {
138 #define CHECK_INTERVAL 100  /* 100ms */
139 #define MAX_CHECK_TIME 90   /* 9s (90 * 100ms) in total */
140 	struct rte_eth_link link;
141 	struct nicvf *nic = nicvf_pmd_priv(dev);
142 	int i;
143 
144 	PMD_INIT_FUNC_TRACE();
145 
146 	if (wait_to_complete) {
147 		/* rte_eth_link_get() might need to wait up to 9 seconds */
148 		for (i = 0; i < MAX_CHECK_TIME; i++) {
149 			nicvf_link_status_update(nic, &link);
150 			if (link.link_status == ETH_LINK_UP)
151 				break;
152 			rte_delay_ms(CHECK_INTERVAL);
153 		}
154 	} else {
155 		nicvf_link_status_update(nic, &link);
156 	}
157 
158 	return rte_eth_linkstatus_set(dev, &link);
159 }
160 
161 static int
162 nicvf_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
163 {
164 	struct nicvf *nic = nicvf_pmd_priv(dev);
165 	uint32_t buffsz, frame_size = mtu + NIC_HW_L2_OVERHEAD;
166 	size_t i;
167 	struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
168 
169 	PMD_INIT_FUNC_TRACE();
170 
171 	if (frame_size > NIC_HW_MAX_FRS)
172 		return -EINVAL;
173 
174 	if (frame_size < NIC_HW_MIN_FRS)
175 		return -EINVAL;
176 
177 	buffsz = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
178 
179 	/*
180 	 * Refuse mtu that requires the support of scattered packets
181 	 * when this feature has not been enabled before.
182 	 */
183 	if (dev->data->dev_started && !dev->data->scattered_rx &&
184 		(frame_size + 2 * VLAN_TAG_SIZE > buffsz))
185 		return -EINVAL;
186 
187 	/* check <seg size> * <max_seg>  >= max_frame */
188 	if (dev->data->scattered_rx &&
189 		(frame_size + 2 * VLAN_TAG_SIZE > buffsz * NIC_HW_MAX_SEGS))
190 		return -EINVAL;
191 
192 	if (frame_size > ETHER_MAX_LEN)
193 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
194 	else
195 		rxmode->offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
196 
197 	if (nicvf_mbox_update_hw_max_frs(nic, mtu))
198 		return -EINVAL;
199 
200 	/* Update max_rx_pkt_len */
201 	rxmode->max_rx_pkt_len = mtu + ETHER_HDR_LEN;
202 	nic->mtu = mtu;
203 
204 	for (i = 0; i < nic->sqs_count; i++)
205 		nic->snicvf[i]->mtu = mtu;
206 
207 	return 0;
208 }
209 
210 static int
211 nicvf_dev_get_regs(struct rte_eth_dev *dev, struct rte_dev_reg_info *regs)
212 {
213 	uint64_t *data = regs->data;
214 	struct nicvf *nic = nicvf_pmd_priv(dev);
215 
216 	if (data == NULL) {
217 		regs->length = nicvf_reg_get_count();
218 		regs->width = THUNDERX_REG_BYTES;
219 		return 0;
220 	}
221 
222 	/* Support only full register dump */
223 	if ((regs->length == 0) ||
224 		(regs->length == (uint32_t)nicvf_reg_get_count())) {
225 		regs->version = nic->vendor_id << 16 | nic->device_id;
226 		nicvf_reg_dump(nic, data);
227 		return 0;
228 	}
229 	return -ENOTSUP;
230 }
231 
232 static int
233 nicvf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
234 {
235 	uint16_t qidx;
236 	struct nicvf_hw_rx_qstats rx_qstats;
237 	struct nicvf_hw_tx_qstats tx_qstats;
238 	struct nicvf_hw_stats port_stats;
239 	struct nicvf *nic = nicvf_pmd_priv(dev);
240 	uint16_t rx_start, rx_end;
241 	uint16_t tx_start, tx_end;
242 	size_t i;
243 
244 	/* RX queue indices for the first VF */
245 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
246 
247 	/* Reading per RX ring stats */
248 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
249 		if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
250 			break;
251 
252 		nicvf_hw_get_rx_qstats(nic, &rx_qstats, qidx);
253 		stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
254 		stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
255 	}
256 
257 	/* TX queue indices for the first VF */
258 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
259 
260 	/* Reading per TX ring stats */
261 	for (qidx = tx_start; qidx <= tx_end; qidx++) {
262 		if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
263 			break;
264 
265 		nicvf_hw_get_tx_qstats(nic, &tx_qstats, qidx);
266 		stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
267 		stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
268 	}
269 
270 	for (i = 0; i < nic->sqs_count; i++) {
271 		struct nicvf *snic = nic->snicvf[i];
272 
273 		if (snic == NULL)
274 			break;
275 
276 		/* RX queue indices for a secondary VF */
277 		nicvf_rx_range(dev, snic, &rx_start, &rx_end);
278 
279 		/* Reading per RX ring stats */
280 		for (qidx = rx_start; qidx <= rx_end; qidx++) {
281 			if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
282 				break;
283 
284 			nicvf_hw_get_rx_qstats(snic, &rx_qstats,
285 					       qidx % MAX_RCV_QUEUES_PER_QS);
286 			stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
287 			stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
288 		}
289 
290 		/* TX queue indices for a secondary VF */
291 		nicvf_tx_range(dev, snic, &tx_start, &tx_end);
292 		/* Reading per TX ring stats */
293 		for (qidx = tx_start; qidx <= tx_end; qidx++) {
294 			if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
295 				break;
296 
297 			nicvf_hw_get_tx_qstats(snic, &tx_qstats,
298 					       qidx % MAX_SND_QUEUES_PER_QS);
299 			stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
300 			stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
301 		}
302 	}
303 
304 	nicvf_hw_get_stats(nic, &port_stats);
305 	stats->ibytes = port_stats.rx_bytes;
306 	stats->ipackets = port_stats.rx_ucast_frames;
307 	stats->ipackets += port_stats.rx_bcast_frames;
308 	stats->ipackets += port_stats.rx_mcast_frames;
309 	stats->ierrors = port_stats.rx_l2_errors;
310 	stats->imissed = port_stats.rx_drop_red;
311 	stats->imissed += port_stats.rx_drop_overrun;
312 	stats->imissed += port_stats.rx_drop_bcast;
313 	stats->imissed += port_stats.rx_drop_mcast;
314 	stats->imissed += port_stats.rx_drop_l3_bcast;
315 	stats->imissed += port_stats.rx_drop_l3_mcast;
316 
317 	stats->obytes = port_stats.tx_bytes_ok;
318 	stats->opackets = port_stats.tx_ucast_frames_ok;
319 	stats->opackets += port_stats.tx_bcast_frames_ok;
320 	stats->opackets += port_stats.tx_mcast_frames_ok;
321 	stats->oerrors = port_stats.tx_drops;
322 
323 	return 0;
324 }
325 
326 static const uint32_t *
327 nicvf_dev_supported_ptypes_get(struct rte_eth_dev *dev)
328 {
329 	size_t copied;
330 	static uint32_t ptypes[32];
331 	struct nicvf *nic = nicvf_pmd_priv(dev);
332 	static const uint32_t ptypes_common[] = {
333 		RTE_PTYPE_L3_IPV4,
334 		RTE_PTYPE_L3_IPV4_EXT,
335 		RTE_PTYPE_L3_IPV6,
336 		RTE_PTYPE_L3_IPV6_EXT,
337 		RTE_PTYPE_L4_TCP,
338 		RTE_PTYPE_L4_UDP,
339 		RTE_PTYPE_L4_FRAG,
340 	};
341 	static const uint32_t ptypes_tunnel[] = {
342 		RTE_PTYPE_TUNNEL_GRE,
343 		RTE_PTYPE_TUNNEL_GENEVE,
344 		RTE_PTYPE_TUNNEL_VXLAN,
345 		RTE_PTYPE_TUNNEL_NVGRE,
346 	};
347 	static const uint32_t ptypes_end = RTE_PTYPE_UNKNOWN;
348 
349 	copied = sizeof(ptypes_common);
350 	memcpy(ptypes, ptypes_common, copied);
351 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
352 		memcpy((char *)ptypes + copied, ptypes_tunnel,
353 			sizeof(ptypes_tunnel));
354 		copied += sizeof(ptypes_tunnel);
355 	}
356 
357 	memcpy((char *)ptypes + copied, &ptypes_end, sizeof(ptypes_end));
358 	if (dev->rx_pkt_burst == nicvf_recv_pkts ||
359 		dev->rx_pkt_burst == nicvf_recv_pkts_multiseg)
360 		return ptypes;
361 
362 	return NULL;
363 }
364 
365 static void
366 nicvf_dev_stats_reset(struct rte_eth_dev *dev)
367 {
368 	int i;
369 	uint16_t rxqs = 0, txqs = 0;
370 	struct nicvf *nic = nicvf_pmd_priv(dev);
371 	uint16_t rx_start, rx_end;
372 	uint16_t tx_start, tx_end;
373 
374 	/* Reset all primary nic counters */
375 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
376 	for (i = rx_start; i <= rx_end; i++)
377 		rxqs |= (0x3 << (i * 2));
378 
379 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
380 	for (i = tx_start; i <= tx_end; i++)
381 		txqs |= (0x3 << (i * 2));
382 
383 	nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, rxqs, txqs);
384 
385 	/* Reset secondary nic queue counters */
386 	for (i = 0; i < nic->sqs_count; i++) {
387 		struct nicvf *snic = nic->snicvf[i];
388 		if (snic == NULL)
389 			break;
390 
391 		nicvf_rx_range(dev, snic, &rx_start, &rx_end);
392 		for (i = rx_start; i <= rx_end; i++)
393 			rxqs |= (0x3 << ((i % MAX_CMP_QUEUES_PER_QS) * 2));
394 
395 		nicvf_tx_range(dev, snic, &tx_start, &tx_end);
396 		for (i = tx_start; i <= tx_end; i++)
397 			txqs |= (0x3 << ((i % MAX_SND_QUEUES_PER_QS) * 2));
398 
399 		nicvf_mbox_reset_stat_counters(snic, 0, 0, rxqs, txqs);
400 	}
401 }
402 
403 /* Promiscuous mode enabled by default in LMAC to VF 1:1 map configuration */
404 static void
405 nicvf_dev_promisc_enable(struct rte_eth_dev *dev __rte_unused)
406 {
407 }
408 
409 static inline uint64_t
410 nicvf_rss_ethdev_to_nic(struct nicvf *nic, uint64_t ethdev_rss)
411 {
412 	uint64_t nic_rss = 0;
413 
414 	if (ethdev_rss & ETH_RSS_IPV4)
415 		nic_rss |= RSS_IP_ENA;
416 
417 	if (ethdev_rss & ETH_RSS_IPV6)
418 		nic_rss |= RSS_IP_ENA;
419 
420 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_UDP)
421 		nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
422 
423 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_TCP)
424 		nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
425 
426 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_UDP)
427 		nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
428 
429 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_TCP)
430 		nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
431 
432 	if (ethdev_rss & ETH_RSS_PORT)
433 		nic_rss |= RSS_L2_EXTENDED_HASH_ENA;
434 
435 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
436 		if (ethdev_rss & ETH_RSS_VXLAN)
437 			nic_rss |= RSS_TUN_VXLAN_ENA;
438 
439 		if (ethdev_rss & ETH_RSS_GENEVE)
440 			nic_rss |= RSS_TUN_GENEVE_ENA;
441 
442 		if (ethdev_rss & ETH_RSS_NVGRE)
443 			nic_rss |= RSS_TUN_NVGRE_ENA;
444 	}
445 
446 	return nic_rss;
447 }
448 
449 static inline uint64_t
450 nicvf_rss_nic_to_ethdev(struct nicvf *nic,  uint64_t nic_rss)
451 {
452 	uint64_t ethdev_rss = 0;
453 
454 	if (nic_rss & RSS_IP_ENA)
455 		ethdev_rss |= (ETH_RSS_IPV4 | ETH_RSS_IPV6);
456 
457 	if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_TCP_ENA))
458 		ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_TCP |
459 				ETH_RSS_NONFRAG_IPV6_TCP);
460 
461 	if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_UDP_ENA))
462 		ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_UDP |
463 				ETH_RSS_NONFRAG_IPV6_UDP);
464 
465 	if (nic_rss & RSS_L2_EXTENDED_HASH_ENA)
466 		ethdev_rss |= ETH_RSS_PORT;
467 
468 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
469 		if (nic_rss & RSS_TUN_VXLAN_ENA)
470 			ethdev_rss |= ETH_RSS_VXLAN;
471 
472 		if (nic_rss & RSS_TUN_GENEVE_ENA)
473 			ethdev_rss |= ETH_RSS_GENEVE;
474 
475 		if (nic_rss & RSS_TUN_NVGRE_ENA)
476 			ethdev_rss |= ETH_RSS_NVGRE;
477 	}
478 	return ethdev_rss;
479 }
480 
481 static int
482 nicvf_dev_reta_query(struct rte_eth_dev *dev,
483 		     struct rte_eth_rss_reta_entry64 *reta_conf,
484 		     uint16_t reta_size)
485 {
486 	struct nicvf *nic = nicvf_pmd_priv(dev);
487 	uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
488 	int ret, i, j;
489 
490 	if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
491 		RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
492 			"(%d) doesn't match the number hardware can supported "
493 			"(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
494 		return -EINVAL;
495 	}
496 
497 	ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
498 	if (ret)
499 		return ret;
500 
501 	/* Copy RETA table */
502 	for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
503 		for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
504 			if ((reta_conf[i].mask >> j) & 0x01)
505 				reta_conf[i].reta[j] = tbl[j];
506 	}
507 
508 	return 0;
509 }
510 
511 static int
512 nicvf_dev_reta_update(struct rte_eth_dev *dev,
513 		      struct rte_eth_rss_reta_entry64 *reta_conf,
514 		      uint16_t reta_size)
515 {
516 	struct nicvf *nic = nicvf_pmd_priv(dev);
517 	uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
518 	int ret, i, j;
519 
520 	if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
521 		RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
522 			"(%d) doesn't match the number hardware can supported "
523 			"(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
524 		return -EINVAL;
525 	}
526 
527 	ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
528 	if (ret)
529 		return ret;
530 
531 	/* Copy RETA table */
532 	for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
533 		for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
534 			if ((reta_conf[i].mask >> j) & 0x01)
535 				tbl[j] = reta_conf[i].reta[j];
536 	}
537 
538 	return nicvf_rss_reta_update(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
539 }
540 
541 static int
542 nicvf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
543 			    struct rte_eth_rss_conf *rss_conf)
544 {
545 	struct nicvf *nic = nicvf_pmd_priv(dev);
546 
547 	if (rss_conf->rss_key)
548 		nicvf_rss_get_key(nic, rss_conf->rss_key);
549 
550 	rss_conf->rss_key_len =  RSS_HASH_KEY_BYTE_SIZE;
551 	rss_conf->rss_hf = nicvf_rss_nic_to_ethdev(nic, nicvf_rss_get_cfg(nic));
552 	return 0;
553 }
554 
555 static int
556 nicvf_dev_rss_hash_update(struct rte_eth_dev *dev,
557 			  struct rte_eth_rss_conf *rss_conf)
558 {
559 	struct nicvf *nic = nicvf_pmd_priv(dev);
560 	uint64_t nic_rss;
561 
562 	if (rss_conf->rss_key &&
563 		rss_conf->rss_key_len != RSS_HASH_KEY_BYTE_SIZE) {
564 		RTE_LOG(ERR, PMD, "Hash key size mismatch %d",
565 				rss_conf->rss_key_len);
566 		return -EINVAL;
567 	}
568 
569 	if (rss_conf->rss_key)
570 		nicvf_rss_set_key(nic, rss_conf->rss_key);
571 
572 	nic_rss = nicvf_rss_ethdev_to_nic(nic, rss_conf->rss_hf);
573 	nicvf_rss_set_cfg(nic, nic_rss);
574 	return 0;
575 }
576 
577 static int
578 nicvf_qset_cq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
579 		    struct nicvf_rxq *rxq, uint16_t qidx, uint32_t desc_cnt)
580 {
581 	const struct rte_memzone *rz;
582 	uint32_t ring_size = CMP_QUEUE_SZ_MAX * sizeof(union cq_entry_t);
583 
584 	rz = rte_eth_dma_zone_reserve(dev, "cq_ring",
585 				      nicvf_netdev_qidx(nic, qidx), ring_size,
586 				      NICVF_CQ_BASE_ALIGN_BYTES, nic->node);
587 	if (rz == NULL) {
588 		PMD_INIT_LOG(ERR, "Failed to allocate mem for cq hw ring");
589 		return -ENOMEM;
590 	}
591 
592 	memset(rz->addr, 0, ring_size);
593 
594 	rxq->phys = rz->iova;
595 	rxq->desc = rz->addr;
596 	rxq->qlen_mask = desc_cnt - 1;
597 
598 	return 0;
599 }
600 
601 static int
602 nicvf_qset_sq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
603 		    struct nicvf_txq *sq, uint16_t qidx, uint32_t desc_cnt)
604 {
605 	const struct rte_memzone *rz;
606 	uint32_t ring_size = SND_QUEUE_SZ_MAX * sizeof(union sq_entry_t);
607 
608 	rz = rte_eth_dma_zone_reserve(dev, "sq",
609 				      nicvf_netdev_qidx(nic, qidx), ring_size,
610 				      NICVF_SQ_BASE_ALIGN_BYTES, nic->node);
611 	if (rz == NULL) {
612 		PMD_INIT_LOG(ERR, "Failed allocate mem for sq hw ring");
613 		return -ENOMEM;
614 	}
615 
616 	memset(rz->addr, 0, ring_size);
617 
618 	sq->phys = rz->iova;
619 	sq->desc = rz->addr;
620 	sq->qlen_mask = desc_cnt - 1;
621 
622 	return 0;
623 }
624 
625 static int
626 nicvf_qset_rbdr_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
627 		      uint32_t desc_cnt, uint32_t buffsz)
628 {
629 	struct nicvf_rbdr *rbdr;
630 	const struct rte_memzone *rz;
631 	uint32_t ring_size;
632 
633 	assert(nic->rbdr == NULL);
634 	rbdr = rte_zmalloc_socket("rbdr", sizeof(struct nicvf_rbdr),
635 				  RTE_CACHE_LINE_SIZE, nic->node);
636 	if (rbdr == NULL) {
637 		PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr");
638 		return -ENOMEM;
639 	}
640 
641 	ring_size = sizeof(struct rbdr_entry_t) * RBDR_QUEUE_SZ_MAX;
642 	rz = rte_eth_dma_zone_reserve(dev, "rbdr",
643 				      nicvf_netdev_qidx(nic, 0), ring_size,
644 				      NICVF_RBDR_BASE_ALIGN_BYTES, nic->node);
645 	if (rz == NULL) {
646 		PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr desc ring");
647 		return -ENOMEM;
648 	}
649 
650 	memset(rz->addr, 0, ring_size);
651 
652 	rbdr->phys = rz->iova;
653 	rbdr->tail = 0;
654 	rbdr->next_tail = 0;
655 	rbdr->desc = rz->addr;
656 	rbdr->buffsz = buffsz;
657 	rbdr->qlen_mask = desc_cnt - 1;
658 	rbdr->rbdr_status =
659 		nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_STATUS0;
660 	rbdr->rbdr_door =
661 		nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_DOOR;
662 
663 	nic->rbdr = rbdr;
664 	return 0;
665 }
666 
667 static void
668 nicvf_rbdr_release_mbuf(struct rte_eth_dev *dev, struct nicvf *nic,
669 			nicvf_iova_addr_t phy)
670 {
671 	uint16_t qidx;
672 	void *obj;
673 	struct nicvf_rxq *rxq;
674 	uint16_t rx_start, rx_end;
675 
676 	/* Get queue ranges for this VF */
677 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
678 
679 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
680 		rxq = dev->data->rx_queues[qidx];
681 		if (rxq->precharge_cnt) {
682 			obj = (void *)nicvf_mbuff_phy2virt(phy,
683 							   rxq->mbuf_phys_off);
684 			rte_mempool_put(rxq->pool, obj);
685 			rxq->precharge_cnt--;
686 			break;
687 		}
688 	}
689 }
690 
691 static inline void
692 nicvf_rbdr_release_mbufs(struct rte_eth_dev *dev, struct nicvf *nic)
693 {
694 	uint32_t qlen_mask, head;
695 	struct rbdr_entry_t *entry;
696 	struct nicvf_rbdr *rbdr = nic->rbdr;
697 
698 	qlen_mask = rbdr->qlen_mask;
699 	head = rbdr->head;
700 	while (head != rbdr->tail) {
701 		entry = rbdr->desc + head;
702 		nicvf_rbdr_release_mbuf(dev, nic, entry->full_addr);
703 		head++;
704 		head = head & qlen_mask;
705 	}
706 }
707 
708 static inline void
709 nicvf_tx_queue_release_mbufs(struct nicvf_txq *txq)
710 {
711 	uint32_t head;
712 
713 	head = txq->head;
714 	while (head != txq->tail) {
715 		if (txq->txbuffs[head]) {
716 			rte_pktmbuf_free_seg(txq->txbuffs[head]);
717 			txq->txbuffs[head] = NULL;
718 		}
719 		head++;
720 		head = head & txq->qlen_mask;
721 	}
722 }
723 
724 static void
725 nicvf_tx_queue_reset(struct nicvf_txq *txq)
726 {
727 	uint32_t txq_desc_cnt = txq->qlen_mask + 1;
728 
729 	memset(txq->desc, 0, sizeof(union sq_entry_t) * txq_desc_cnt);
730 	memset(txq->txbuffs, 0, sizeof(struct rte_mbuf *) * txq_desc_cnt);
731 	txq->tail = 0;
732 	txq->head = 0;
733 	txq->xmit_bufs = 0;
734 }
735 
736 static inline int
737 nicvf_vf_start_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
738 			uint16_t qidx)
739 {
740 	struct nicvf_txq *txq;
741 	int ret;
742 
743 	assert(qidx < MAX_SND_QUEUES_PER_QS);
744 
745 	if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
746 		RTE_ETH_QUEUE_STATE_STARTED)
747 		return 0;
748 
749 	txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
750 	txq->pool = NULL;
751 	ret = nicvf_qset_sq_config(nic, qidx, txq);
752 	if (ret) {
753 		PMD_INIT_LOG(ERR, "Failed to configure sq VF%d %d %d",
754 			     nic->vf_id, qidx, ret);
755 		goto config_sq_error;
756 	}
757 
758 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
759 		RTE_ETH_QUEUE_STATE_STARTED;
760 	return ret;
761 
762 config_sq_error:
763 	nicvf_qset_sq_reclaim(nic, qidx);
764 	return ret;
765 }
766 
767 static inline int
768 nicvf_vf_stop_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
769 		       uint16_t qidx)
770 {
771 	struct nicvf_txq *txq;
772 	int ret;
773 
774 	assert(qidx < MAX_SND_QUEUES_PER_QS);
775 
776 	if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
777 		RTE_ETH_QUEUE_STATE_STOPPED)
778 		return 0;
779 
780 	ret = nicvf_qset_sq_reclaim(nic, qidx);
781 	if (ret)
782 		PMD_INIT_LOG(ERR, "Failed to reclaim sq VF%d %d %d",
783 			     nic->vf_id, qidx, ret);
784 
785 	txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
786 	nicvf_tx_queue_release_mbufs(txq);
787 	nicvf_tx_queue_reset(txq);
788 
789 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
790 		RTE_ETH_QUEUE_STATE_STOPPED;
791 	return ret;
792 }
793 
794 static inline int
795 nicvf_configure_cpi(struct rte_eth_dev *dev)
796 {
797 	struct nicvf *nic = nicvf_pmd_priv(dev);
798 	uint16_t qidx, qcnt;
799 	int ret;
800 
801 	/* Count started rx queues */
802 	for (qidx = qcnt = 0; qidx < dev->data->nb_rx_queues; qidx++)
803 		if (dev->data->rx_queue_state[qidx] ==
804 		    RTE_ETH_QUEUE_STATE_STARTED)
805 			qcnt++;
806 
807 	nic->cpi_alg = CPI_ALG_NONE;
808 	ret = nicvf_mbox_config_cpi(nic, qcnt);
809 	if (ret)
810 		PMD_INIT_LOG(ERR, "Failed to configure CPI %d", ret);
811 
812 	return ret;
813 }
814 
815 static inline int
816 nicvf_configure_rss(struct rte_eth_dev *dev)
817 {
818 	struct nicvf *nic = nicvf_pmd_priv(dev);
819 	uint64_t rsshf;
820 	int ret = -EINVAL;
821 
822 	rsshf = nicvf_rss_ethdev_to_nic(nic,
823 			dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf);
824 	PMD_DRV_LOG(INFO, "mode=%d rx_queues=%d loopback=%d rsshf=0x%" PRIx64,
825 		    dev->data->dev_conf.rxmode.mq_mode,
826 		    dev->data->nb_rx_queues,
827 		    dev->data->dev_conf.lpbk_mode, rsshf);
828 
829 	if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_NONE)
830 		ret = nicvf_rss_term(nic);
831 	else if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS)
832 		ret = nicvf_rss_config(nic, dev->data->nb_rx_queues, rsshf);
833 	if (ret)
834 		PMD_INIT_LOG(ERR, "Failed to configure RSS %d", ret);
835 
836 	return ret;
837 }
838 
839 static int
840 nicvf_configure_rss_reta(struct rte_eth_dev *dev)
841 {
842 	struct nicvf *nic = nicvf_pmd_priv(dev);
843 	unsigned int idx, qmap_size;
844 	uint8_t qmap[RTE_MAX_QUEUES_PER_PORT];
845 	uint8_t default_reta[NIC_MAX_RSS_IDR_TBL_SIZE];
846 
847 	if (nic->cpi_alg != CPI_ALG_NONE)
848 		return -EINVAL;
849 
850 	/* Prepare queue map */
851 	for (idx = 0, qmap_size = 0; idx < dev->data->nb_rx_queues; idx++) {
852 		if (dev->data->rx_queue_state[idx] ==
853 				RTE_ETH_QUEUE_STATE_STARTED)
854 			qmap[qmap_size++] = idx;
855 	}
856 
857 	/* Update default RSS RETA */
858 	for (idx = 0; idx < NIC_MAX_RSS_IDR_TBL_SIZE; idx++)
859 		default_reta[idx] = qmap[idx % qmap_size];
860 
861 	return nicvf_rss_reta_update(nic, default_reta,
862 				     NIC_MAX_RSS_IDR_TBL_SIZE);
863 }
864 
865 static void
866 nicvf_dev_tx_queue_release(void *sq)
867 {
868 	struct nicvf_txq *txq;
869 
870 	PMD_INIT_FUNC_TRACE();
871 
872 	txq = (struct nicvf_txq *)sq;
873 	if (txq) {
874 		if (txq->txbuffs != NULL) {
875 			nicvf_tx_queue_release_mbufs(txq);
876 			rte_free(txq->txbuffs);
877 			txq->txbuffs = NULL;
878 		}
879 		rte_free(txq);
880 	}
881 }
882 
883 static void
884 nicvf_set_tx_function(struct rte_eth_dev *dev)
885 {
886 	struct nicvf_txq *txq = NULL;
887 	size_t i;
888 	bool multiseg = false;
889 
890 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
891 		txq = dev->data->tx_queues[i];
892 		if (txq->offloads & DEV_TX_OFFLOAD_MULTI_SEGS) {
893 			multiseg = true;
894 			break;
895 		}
896 	}
897 
898 	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
899 	if (multiseg) {
900 		PMD_DRV_LOG(DEBUG, "Using multi-segment tx callback");
901 		dev->tx_pkt_burst = nicvf_xmit_pkts_multiseg;
902 	} else {
903 		PMD_DRV_LOG(DEBUG, "Using single-segment tx callback");
904 		dev->tx_pkt_burst = nicvf_xmit_pkts;
905 	}
906 
907 	if (!txq)
908 		return;
909 
910 	if (txq->pool_free == nicvf_single_pool_free_xmited_buffers)
911 		PMD_DRV_LOG(DEBUG, "Using single-mempool tx free method");
912 	else
913 		PMD_DRV_LOG(DEBUG, "Using multi-mempool tx free method");
914 }
915 
916 static void
917 nicvf_set_rx_function(struct rte_eth_dev *dev)
918 {
919 	if (dev->data->scattered_rx) {
920 		PMD_DRV_LOG(DEBUG, "Using multi-segment rx callback");
921 		dev->rx_pkt_burst = nicvf_recv_pkts_multiseg;
922 	} else {
923 		PMD_DRV_LOG(DEBUG, "Using single-segment rx callback");
924 		dev->rx_pkt_burst = nicvf_recv_pkts;
925 	}
926 }
927 
928 static int
929 nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
930 			 uint16_t nb_desc, unsigned int socket_id,
931 			 const struct rte_eth_txconf *tx_conf)
932 {
933 	uint16_t tx_free_thresh;
934 	bool is_single_pool;
935 	struct nicvf_txq *txq;
936 	struct nicvf *nic = nicvf_pmd_priv(dev);
937 	uint64_t offloads;
938 
939 	PMD_INIT_FUNC_TRACE();
940 
941 	if (qidx >= MAX_SND_QUEUES_PER_QS)
942 		nic = nic->snicvf[qidx / MAX_SND_QUEUES_PER_QS - 1];
943 
944 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
945 
946 	/* Socket id check */
947 	if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
948 		PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
949 		socket_id, nic->node);
950 
951 	/* Tx deferred start is not supported */
952 	if (tx_conf->tx_deferred_start) {
953 		PMD_INIT_LOG(ERR, "Tx deferred start not supported");
954 		return -EINVAL;
955 	}
956 
957 	/* Roundup nb_desc to available qsize and validate max number of desc */
958 	nb_desc = nicvf_qsize_sq_roundup(nb_desc);
959 	if (nb_desc == 0) {
960 		PMD_INIT_LOG(ERR, "Value of nb_desc beyond available sq qsize");
961 		return -EINVAL;
962 	}
963 
964 	/* Validate tx_free_thresh */
965 	tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
966 				tx_conf->tx_free_thresh :
967 				NICVF_DEFAULT_TX_FREE_THRESH);
968 
969 	if (tx_free_thresh > (nb_desc) ||
970 		tx_free_thresh > NICVF_MAX_TX_FREE_THRESH) {
971 		PMD_INIT_LOG(ERR,
972 			"tx_free_thresh must be less than the number of TX "
973 			"descriptors. (tx_free_thresh=%u port=%d "
974 			"queue=%d)", (unsigned int)tx_free_thresh,
975 			(int)dev->data->port_id, (int)qidx);
976 		return -EINVAL;
977 	}
978 
979 	/* Free memory prior to re-allocation if needed. */
980 	if (dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
981 		PMD_TX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
982 				nicvf_netdev_qidx(nic, qidx));
983 		nicvf_dev_tx_queue_release(
984 			dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)]);
985 		dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
986 	}
987 
988 	/* Allocating tx queue data structure */
989 	txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct nicvf_txq),
990 					RTE_CACHE_LINE_SIZE, nic->node);
991 	if (txq == NULL) {
992 		PMD_INIT_LOG(ERR, "Failed to allocate txq=%d",
993 			     nicvf_netdev_qidx(nic, qidx));
994 		return -ENOMEM;
995 	}
996 
997 	txq->nic = nic;
998 	txq->queue_id = qidx;
999 	txq->tx_free_thresh = tx_free_thresh;
1000 	txq->sq_head = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_HEAD;
1001 	txq->sq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_DOOR;
1002 	offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1003 	txq->offloads = offloads;
1004 
1005 	is_single_pool = !!(offloads & DEV_TX_OFFLOAD_MBUF_FAST_FREE);
1006 
1007 	/* Choose optimum free threshold value for multipool case */
1008 	if (!is_single_pool) {
1009 		txq->tx_free_thresh = (uint16_t)
1010 		(tx_conf->tx_free_thresh == NICVF_DEFAULT_TX_FREE_THRESH ?
1011 				NICVF_TX_FREE_MPOOL_THRESH :
1012 				tx_conf->tx_free_thresh);
1013 		txq->pool_free = nicvf_multi_pool_free_xmited_buffers;
1014 	} else {
1015 		txq->pool_free = nicvf_single_pool_free_xmited_buffers;
1016 	}
1017 
1018 	/* Allocate software ring */
1019 	txq->txbuffs = rte_zmalloc_socket("txq->txbuffs",
1020 				nb_desc * sizeof(struct rte_mbuf *),
1021 				RTE_CACHE_LINE_SIZE, nic->node);
1022 
1023 	if (txq->txbuffs == NULL) {
1024 		nicvf_dev_tx_queue_release(txq);
1025 		return -ENOMEM;
1026 	}
1027 
1028 	if (nicvf_qset_sq_alloc(dev, nic, txq, qidx, nb_desc)) {
1029 		PMD_INIT_LOG(ERR, "Failed to allocate mem for sq %d", qidx);
1030 		nicvf_dev_tx_queue_release(txq);
1031 		return -ENOMEM;
1032 	}
1033 
1034 	nicvf_tx_queue_reset(txq);
1035 
1036 	PMD_INIT_LOG(DEBUG, "[%d] txq=%p nb_desc=%d desc=%p"
1037 			" phys=0x%" PRIx64 " offloads=0x%" PRIx64,
1038 			nicvf_netdev_qidx(nic, qidx), txq, nb_desc, txq->desc,
1039 			txq->phys, txq->offloads);
1040 
1041 	dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = txq;
1042 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1043 		RTE_ETH_QUEUE_STATE_STOPPED;
1044 	return 0;
1045 }
1046 
1047 static inline void
1048 nicvf_rx_queue_release_mbufs(struct rte_eth_dev *dev, struct nicvf_rxq *rxq)
1049 {
1050 	uint32_t rxq_cnt;
1051 	uint32_t nb_pkts, released_pkts = 0;
1052 	uint32_t refill_cnt = 0;
1053 	struct rte_mbuf *rx_pkts[NICVF_MAX_RX_FREE_THRESH];
1054 
1055 	if (dev->rx_pkt_burst == NULL)
1056 		return;
1057 
1058 	while ((rxq_cnt = nicvf_dev_rx_queue_count(dev,
1059 				nicvf_netdev_qidx(rxq->nic, rxq->queue_id)))) {
1060 		nb_pkts = dev->rx_pkt_burst(rxq, rx_pkts,
1061 					NICVF_MAX_RX_FREE_THRESH);
1062 		PMD_DRV_LOG(INFO, "nb_pkts=%d  rxq_cnt=%d", nb_pkts, rxq_cnt);
1063 		while (nb_pkts) {
1064 			rte_pktmbuf_free_seg(rx_pkts[--nb_pkts]);
1065 			released_pkts++;
1066 		}
1067 	}
1068 
1069 
1070 	refill_cnt += nicvf_dev_rbdr_refill(dev,
1071 			nicvf_netdev_qidx(rxq->nic, rxq->queue_id));
1072 
1073 	PMD_DRV_LOG(INFO, "free_cnt=%d  refill_cnt=%d",
1074 		    released_pkts, refill_cnt);
1075 }
1076 
1077 static void
1078 nicvf_rx_queue_reset(struct nicvf_rxq *rxq)
1079 {
1080 	rxq->head = 0;
1081 	rxq->available_space = 0;
1082 	rxq->recv_buffers = 0;
1083 }
1084 
1085 static inline int
1086 nicvf_vf_start_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
1087 			uint16_t qidx)
1088 {
1089 	struct nicvf_rxq *rxq;
1090 	int ret;
1091 
1092 	assert(qidx < MAX_RCV_QUEUES_PER_QS);
1093 
1094 	if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
1095 		RTE_ETH_QUEUE_STATE_STARTED)
1096 		return 0;
1097 
1098 	/* Update rbdr pointer to all rxq */
1099 	rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
1100 	rxq->shared_rbdr = nic->rbdr;
1101 
1102 	ret = nicvf_qset_rq_config(nic, qidx, rxq);
1103 	if (ret) {
1104 		PMD_INIT_LOG(ERR, "Failed to configure rq VF%d %d %d",
1105 			     nic->vf_id, qidx, ret);
1106 		goto config_rq_error;
1107 	}
1108 	ret = nicvf_qset_cq_config(nic, qidx, rxq);
1109 	if (ret) {
1110 		PMD_INIT_LOG(ERR, "Failed to configure cq VF%d %d %d",
1111 			     nic->vf_id, qidx, ret);
1112 		goto config_cq_error;
1113 	}
1114 
1115 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1116 		RTE_ETH_QUEUE_STATE_STARTED;
1117 	return 0;
1118 
1119 config_cq_error:
1120 	nicvf_qset_cq_reclaim(nic, qidx);
1121 config_rq_error:
1122 	nicvf_qset_rq_reclaim(nic, qidx);
1123 	return ret;
1124 }
1125 
1126 static inline int
1127 nicvf_vf_stop_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
1128 		       uint16_t qidx)
1129 {
1130 	struct nicvf_rxq *rxq;
1131 	int ret, other_error;
1132 
1133 	if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
1134 		RTE_ETH_QUEUE_STATE_STOPPED)
1135 		return 0;
1136 
1137 	ret = nicvf_qset_rq_reclaim(nic, qidx);
1138 	if (ret)
1139 		PMD_INIT_LOG(ERR, "Failed to reclaim rq VF%d %d %d",
1140 			     nic->vf_id, qidx, ret);
1141 
1142 	other_error = ret;
1143 	rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
1144 	nicvf_rx_queue_release_mbufs(dev, rxq);
1145 	nicvf_rx_queue_reset(rxq);
1146 
1147 	ret = nicvf_qset_cq_reclaim(nic, qidx);
1148 	if (ret)
1149 		PMD_INIT_LOG(ERR, "Failed to reclaim cq VF%d %d %d",
1150 			     nic->vf_id, qidx, ret);
1151 
1152 	other_error |= ret;
1153 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1154 		RTE_ETH_QUEUE_STATE_STOPPED;
1155 	return other_error;
1156 }
1157 
1158 static void
1159 nicvf_dev_rx_queue_release(void *rx_queue)
1160 {
1161 	PMD_INIT_FUNC_TRACE();
1162 
1163 	rte_free(rx_queue);
1164 }
1165 
1166 static int
1167 nicvf_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
1168 {
1169 	struct nicvf *nic = nicvf_pmd_priv(dev);
1170 	int ret;
1171 
1172 	if (qidx >= MAX_RCV_QUEUES_PER_QS)
1173 		nic = nic->snicvf[(qidx / MAX_RCV_QUEUES_PER_QS - 1)];
1174 
1175 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1176 
1177 	ret = nicvf_vf_start_rx_queue(dev, nic, qidx);
1178 	if (ret)
1179 		return ret;
1180 
1181 	ret = nicvf_configure_cpi(dev);
1182 	if (ret)
1183 		return ret;
1184 
1185 	return nicvf_configure_rss_reta(dev);
1186 }
1187 
1188 static int
1189 nicvf_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
1190 {
1191 	int ret;
1192 	struct nicvf *nic = nicvf_pmd_priv(dev);
1193 
1194 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1195 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1196 
1197 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1198 
1199 	ret = nicvf_vf_stop_rx_queue(dev, nic, qidx);
1200 	ret |= nicvf_configure_cpi(dev);
1201 	ret |= nicvf_configure_rss_reta(dev);
1202 	return ret;
1203 }
1204 
1205 static int
1206 nicvf_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
1207 {
1208 	struct nicvf *nic = nicvf_pmd_priv(dev);
1209 
1210 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1211 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1212 
1213 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
1214 
1215 	return nicvf_vf_start_tx_queue(dev, nic, qidx);
1216 }
1217 
1218 static int
1219 nicvf_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
1220 {
1221 	struct nicvf *nic = nicvf_pmd_priv(dev);
1222 
1223 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1224 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1225 
1226 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
1227 
1228 	return nicvf_vf_stop_tx_queue(dev, nic, qidx);
1229 }
1230 
1231 static inline void
1232 nicvf_rxq_mbuf_setup(struct nicvf_rxq *rxq)
1233 {
1234 	uintptr_t p;
1235 	struct rte_mbuf mb_def;
1236 	struct nicvf *nic = rxq->nic;
1237 
1238 	RTE_BUILD_BUG_ON(sizeof(union mbuf_initializer) != 8);
1239 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, data_off) % 8 != 0);
1240 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, refcnt) -
1241 				offsetof(struct rte_mbuf, data_off) != 2);
1242 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, nb_segs) -
1243 				offsetof(struct rte_mbuf, data_off) != 4);
1244 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, port) -
1245 				offsetof(struct rte_mbuf, data_off) != 6);
1246 	mb_def.nb_segs = 1;
1247 	mb_def.data_off = RTE_PKTMBUF_HEADROOM + (nic->skip_bytes);
1248 	mb_def.port = rxq->port_id;
1249 	rte_mbuf_refcnt_set(&mb_def, 1);
1250 
1251 	/* Prevent compiler reordering: rearm_data covers previous fields */
1252 	rte_compiler_barrier();
1253 	p = (uintptr_t)&mb_def.rearm_data;
1254 	rxq->mbuf_initializer.value = *(uint64_t *)p;
1255 }
1256 
1257 static int
1258 nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
1259 			 uint16_t nb_desc, unsigned int socket_id,
1260 			 const struct rte_eth_rxconf *rx_conf,
1261 			 struct rte_mempool *mp)
1262 {
1263 	uint16_t rx_free_thresh;
1264 	struct nicvf_rxq *rxq;
1265 	struct nicvf *nic = nicvf_pmd_priv(dev);
1266 	uint64_t offloads;
1267 	uint32_t buffsz;
1268 	struct rte_pktmbuf_pool_private *mbp_priv;
1269 
1270 	PMD_INIT_FUNC_TRACE();
1271 
1272 	/* First skip check */
1273 	mbp_priv = rte_mempool_get_priv(mp);
1274 	buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
1275 	if (buffsz < (uint32_t)(nic->skip_bytes)) {
1276 		PMD_INIT_LOG(ERR, "First skip is more than configured buffer size");
1277 		return -EINVAL;
1278 	}
1279 
1280 	if (qidx >= MAX_RCV_QUEUES_PER_QS)
1281 		nic = nic->snicvf[qidx / MAX_RCV_QUEUES_PER_QS - 1];
1282 
1283 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1284 
1285 	/* Socket id check */
1286 	if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
1287 		PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
1288 		socket_id, nic->node);
1289 
1290 	/* Mempool memory must be contiguous, so must be one memory segment*/
1291 	if (mp->nb_mem_chunks != 1) {
1292 		PMD_INIT_LOG(ERR, "Non-contiguous mempool, add more huge pages");
1293 		return -EINVAL;
1294 	}
1295 
1296 	/* Mempool memory must be physically contiguous */
1297 	if (mp->flags & MEMPOOL_F_NO_IOVA_CONTIG) {
1298 		PMD_INIT_LOG(ERR, "Mempool memory must be physically contiguous");
1299 		return -EINVAL;
1300 	}
1301 
1302 	/* Rx deferred start is not supported */
1303 	if (rx_conf->rx_deferred_start) {
1304 		PMD_INIT_LOG(ERR, "Rx deferred start not supported");
1305 		return -EINVAL;
1306 	}
1307 
1308 	/* Roundup nb_desc to available qsize and validate max number of desc */
1309 	nb_desc = nicvf_qsize_cq_roundup(nb_desc);
1310 	if (nb_desc == 0) {
1311 		PMD_INIT_LOG(ERR, "Value nb_desc beyond available hw cq qsize");
1312 		return -EINVAL;
1313 	}
1314 
1315 
1316 	/* Check rx_free_thresh upper bound */
1317 	rx_free_thresh = (uint16_t)((rx_conf->rx_free_thresh) ?
1318 				rx_conf->rx_free_thresh :
1319 				NICVF_DEFAULT_RX_FREE_THRESH);
1320 	if (rx_free_thresh > NICVF_MAX_RX_FREE_THRESH ||
1321 		rx_free_thresh >= nb_desc * .75) {
1322 		PMD_INIT_LOG(ERR, "rx_free_thresh greater than expected %d",
1323 				rx_free_thresh);
1324 		return -EINVAL;
1325 	}
1326 
1327 	/* Free memory prior to re-allocation if needed */
1328 	if (dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
1329 		PMD_RX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
1330 				nicvf_netdev_qidx(nic, qidx));
1331 		nicvf_dev_rx_queue_release(
1332 			dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)]);
1333 		dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
1334 	}
1335 
1336 	/* Allocate rxq memory */
1337 	rxq = rte_zmalloc_socket("ethdev rx queue", sizeof(struct nicvf_rxq),
1338 					RTE_CACHE_LINE_SIZE, nic->node);
1339 	if (rxq == NULL) {
1340 		PMD_INIT_LOG(ERR, "Failed to allocate rxq=%d",
1341 			     nicvf_netdev_qidx(nic, qidx));
1342 		return -ENOMEM;
1343 	}
1344 
1345 	rxq->nic = nic;
1346 	rxq->pool = mp;
1347 	rxq->queue_id = qidx;
1348 	rxq->port_id = dev->data->port_id;
1349 	rxq->rx_free_thresh = rx_free_thresh;
1350 	rxq->rx_drop_en = rx_conf->rx_drop_en;
1351 	rxq->cq_status = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_STATUS;
1352 	rxq->cq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_DOOR;
1353 	rxq->precharge_cnt = 0;
1354 
1355 	if (nicvf_hw_cap(nic) & NICVF_CAP_CQE_RX2)
1356 		rxq->rbptr_offset = NICVF_CQE_RX2_RBPTR_WORD;
1357 	else
1358 		rxq->rbptr_offset = NICVF_CQE_RBPTR_WORD;
1359 
1360 	nicvf_rxq_mbuf_setup(rxq);
1361 
1362 	/* Alloc completion queue */
1363 	if (nicvf_qset_cq_alloc(dev, nic, rxq, rxq->queue_id, nb_desc)) {
1364 		PMD_INIT_LOG(ERR, "failed to allocate cq %u", rxq->queue_id);
1365 		nicvf_dev_rx_queue_release(rxq);
1366 		return -ENOMEM;
1367 	}
1368 
1369 	nicvf_rx_queue_reset(rxq);
1370 
1371 	offloads = rx_conf->offloads | dev->data->dev_conf.rxmode.offloads;
1372 	PMD_INIT_LOG(DEBUG, "[%d] rxq=%p pool=%s nb_desc=(%d/%d)"
1373 			" phy=0x%" PRIx64 " offloads=0x%" PRIx64,
1374 			nicvf_netdev_qidx(nic, qidx), rxq, mp->name, nb_desc,
1375 			rte_mempool_avail_count(mp), rxq->phys, offloads);
1376 
1377 	dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = rxq;
1378 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1379 		RTE_ETH_QUEUE_STATE_STOPPED;
1380 	return 0;
1381 }
1382 
1383 static void
1384 nicvf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1385 {
1386 	struct nicvf *nic = nicvf_pmd_priv(dev);
1387 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1388 
1389 	PMD_INIT_FUNC_TRACE();
1390 
1391 	/* Autonegotiation may be disabled */
1392 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
1393 	dev_info->speed_capa |= ETH_LINK_SPEED_10M | ETH_LINK_SPEED_100M |
1394 				 ETH_LINK_SPEED_1G | ETH_LINK_SPEED_10G;
1395 	if (nicvf_hw_version(nic) != PCI_SUB_DEVICE_ID_CN81XX_NICVF)
1396 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
1397 
1398 	dev_info->min_rx_bufsize = ETHER_MIN_MTU;
1399 	dev_info->max_rx_pktlen = NIC_HW_MAX_MTU + ETHER_HDR_LEN;
1400 	dev_info->max_rx_queues =
1401 			(uint16_t)MAX_RCV_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
1402 	dev_info->max_tx_queues =
1403 			(uint16_t)MAX_SND_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
1404 	dev_info->max_mac_addrs = 1;
1405 	dev_info->max_vfs = pci_dev->max_vfs;
1406 
1407 	dev_info->rx_offload_capa = NICVF_RX_OFFLOAD_CAPA;
1408 	dev_info->tx_offload_capa = NICVF_TX_OFFLOAD_CAPA;
1409 	dev_info->rx_queue_offload_capa = NICVF_RX_OFFLOAD_CAPA;
1410 	dev_info->tx_queue_offload_capa = NICVF_TX_OFFLOAD_CAPA;
1411 
1412 	dev_info->reta_size = nic->rss_info.rss_size;
1413 	dev_info->hash_key_size = RSS_HASH_KEY_BYTE_SIZE;
1414 	dev_info->flow_type_rss_offloads = NICVF_RSS_OFFLOAD_PASS1;
1415 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING)
1416 		dev_info->flow_type_rss_offloads |= NICVF_RSS_OFFLOAD_TUNNEL;
1417 
1418 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
1419 		.rx_free_thresh = NICVF_DEFAULT_RX_FREE_THRESH,
1420 		.rx_drop_en = 0,
1421 		.offloads = DEV_RX_OFFLOAD_CRC_STRIP,
1422 	};
1423 
1424 	dev_info->default_txconf = (struct rte_eth_txconf) {
1425 		.tx_free_thresh = NICVF_DEFAULT_TX_FREE_THRESH,
1426 		.offloads = DEV_TX_OFFLOAD_MBUF_FAST_FREE |
1427 			DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM   |
1428 			DEV_TX_OFFLOAD_UDP_CKSUM          |
1429 			DEV_TX_OFFLOAD_TCP_CKSUM,
1430 	};
1431 }
1432 
1433 static nicvf_iova_addr_t
1434 rbdr_rte_mempool_get(void *dev, void *opaque)
1435 {
1436 	uint16_t qidx;
1437 	uintptr_t mbuf;
1438 	struct nicvf_rxq *rxq;
1439 	struct rte_eth_dev *eth_dev = (struct rte_eth_dev *)dev;
1440 	struct nicvf *nic = (struct nicvf *)opaque;
1441 	uint16_t rx_start, rx_end;
1442 
1443 	/* Get queue ranges for this VF */
1444 	nicvf_rx_range(eth_dev, nic, &rx_start, &rx_end);
1445 
1446 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1447 		rxq = eth_dev->data->rx_queues[qidx];
1448 		/* Maintain equal buffer count across all pools */
1449 		if (rxq->precharge_cnt >= rxq->qlen_mask)
1450 			continue;
1451 		rxq->precharge_cnt++;
1452 		mbuf = (uintptr_t)rte_pktmbuf_alloc(rxq->pool);
1453 		if (mbuf)
1454 			return nicvf_mbuff_virt2phy(mbuf, rxq->mbuf_phys_off);
1455 	}
1456 	return 0;
1457 }
1458 
1459 static int
1460 nicvf_vf_start(struct rte_eth_dev *dev, struct nicvf *nic, uint32_t rbdrsz)
1461 {
1462 	int ret;
1463 	uint16_t qidx, data_off;
1464 	uint32_t total_rxq_desc, nb_rbdr_desc, exp_buffs;
1465 	uint64_t mbuf_phys_off = 0;
1466 	struct nicvf_rxq *rxq;
1467 	struct rte_mbuf *mbuf;
1468 	uint16_t rx_start, rx_end;
1469 	uint16_t tx_start, tx_end;
1470 	bool vlan_strip;
1471 
1472 	PMD_INIT_FUNC_TRACE();
1473 
1474 	/* Userspace process exited without proper shutdown in last run */
1475 	if (nicvf_qset_rbdr_active(nic, 0))
1476 		nicvf_vf_stop(dev, nic, false);
1477 
1478 	/* Get queue ranges for this VF */
1479 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
1480 
1481 	/*
1482 	 * Thunderx nicvf PMD can support more than one pool per port only when
1483 	 * 1) Data payload size is same across all the pools in given port
1484 	 * AND
1485 	 * 2) All mbuffs in the pools are from the same hugepage
1486 	 * AND
1487 	 * 3) Mbuff metadata size is same across all the pools in given port
1488 	 *
1489 	 * This is to support existing application that uses multiple pool/port.
1490 	 * But, the purpose of using multipool for QoS will not be addressed.
1491 	 *
1492 	 */
1493 
1494 	/* Validate mempool attributes */
1495 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1496 		rxq = dev->data->rx_queues[qidx];
1497 		rxq->mbuf_phys_off = nicvf_mempool_phy_offset(rxq->pool);
1498 		mbuf = rte_pktmbuf_alloc(rxq->pool);
1499 		if (mbuf == NULL) {
1500 			PMD_INIT_LOG(ERR, "Failed allocate mbuf VF%d qid=%d "
1501 				     "pool=%s",
1502 				     nic->vf_id, qidx, rxq->pool->name);
1503 			return -ENOMEM;
1504 		}
1505 		data_off = nicvf_mbuff_meta_length(mbuf);
1506 		data_off += RTE_PKTMBUF_HEADROOM;
1507 		rte_pktmbuf_free(mbuf);
1508 
1509 		if (data_off % RTE_CACHE_LINE_SIZE) {
1510 			PMD_INIT_LOG(ERR, "%s: unaligned data_off=%d delta=%d",
1511 				rxq->pool->name, data_off,
1512 				data_off % RTE_CACHE_LINE_SIZE);
1513 			return -EINVAL;
1514 		}
1515 		rxq->mbuf_phys_off -= data_off;
1516 		rxq->mbuf_phys_off -= nic->skip_bytes;
1517 
1518 		if (mbuf_phys_off == 0)
1519 			mbuf_phys_off = rxq->mbuf_phys_off;
1520 		if (mbuf_phys_off != rxq->mbuf_phys_off) {
1521 			PMD_INIT_LOG(ERR, "pool params not same,%s VF%d %"
1522 				     PRIx64, rxq->pool->name, nic->vf_id,
1523 				     mbuf_phys_off);
1524 			return -EINVAL;
1525 		}
1526 	}
1527 
1528 	/* Check the level of buffers in the pool */
1529 	total_rxq_desc = 0;
1530 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1531 		rxq = dev->data->rx_queues[qidx];
1532 		/* Count total numbers of rxq descs */
1533 		total_rxq_desc += rxq->qlen_mask + 1;
1534 		exp_buffs = RTE_MEMPOOL_CACHE_MAX_SIZE + rxq->rx_free_thresh;
1535 		exp_buffs *= dev->data->nb_rx_queues;
1536 		if (rte_mempool_avail_count(rxq->pool) < exp_buffs) {
1537 			PMD_INIT_LOG(ERR, "Buff shortage in pool=%s (%d/%d)",
1538 				     rxq->pool->name,
1539 				     rte_mempool_avail_count(rxq->pool),
1540 				     exp_buffs);
1541 			return -ENOENT;
1542 		}
1543 	}
1544 
1545 	/* Check RBDR desc overflow */
1546 	ret = nicvf_qsize_rbdr_roundup(total_rxq_desc);
1547 	if (ret == 0) {
1548 		PMD_INIT_LOG(ERR, "Reached RBDR desc limit, reduce nr desc "
1549 			     "VF%d", nic->vf_id);
1550 		return -ENOMEM;
1551 	}
1552 
1553 	/* Enable qset */
1554 	ret = nicvf_qset_config(nic);
1555 	if (ret) {
1556 		PMD_INIT_LOG(ERR, "Failed to enable qset %d VF%d", ret,
1557 			     nic->vf_id);
1558 		return ret;
1559 	}
1560 
1561 	/* Allocate RBDR and RBDR ring desc */
1562 	nb_rbdr_desc = nicvf_qsize_rbdr_roundup(total_rxq_desc);
1563 	ret = nicvf_qset_rbdr_alloc(dev, nic, nb_rbdr_desc, rbdrsz);
1564 	if (ret) {
1565 		PMD_INIT_LOG(ERR, "Failed to allocate memory for rbdr alloc "
1566 			     "VF%d", nic->vf_id);
1567 		goto qset_reclaim;
1568 	}
1569 
1570 	/* Enable and configure RBDR registers */
1571 	ret = nicvf_qset_rbdr_config(nic, 0);
1572 	if (ret) {
1573 		PMD_INIT_LOG(ERR, "Failed to configure rbdr %d VF%d", ret,
1574 			     nic->vf_id);
1575 		goto qset_rbdr_free;
1576 	}
1577 
1578 	/* Fill rte_mempool buffers in RBDR pool and precharge it */
1579 	ret = nicvf_qset_rbdr_precharge(dev, nic, 0, rbdr_rte_mempool_get,
1580 					total_rxq_desc);
1581 	if (ret) {
1582 		PMD_INIT_LOG(ERR, "Failed to fill rbdr %d VF%d", ret,
1583 			     nic->vf_id);
1584 		goto qset_rbdr_reclaim;
1585 	}
1586 
1587 	PMD_DRV_LOG(INFO, "Filled %d out of %d entries in RBDR VF%d",
1588 		     nic->rbdr->tail, nb_rbdr_desc, nic->vf_id);
1589 
1590 	/* Configure VLAN Strip */
1591 	vlan_strip = !!(dev->data->dev_conf.rxmode.offloads &
1592 			DEV_RX_OFFLOAD_VLAN_STRIP);
1593 	nicvf_vlan_hw_strip(nic, vlan_strip);
1594 
1595 	/* Based on the packet type(IPv4 or IPv6), the nicvf HW aligns L3 data
1596 	 * to the 64bit memory address.
1597 	 * The alignment creates a hole in mbuf(between the end of headroom and
1598 	 * packet data start). The new revision of the HW provides an option to
1599 	 * disable the L3 alignment feature and make mbuf layout looks
1600 	 * more like other NICs. For better application compatibility, disabling
1601 	 * l3 alignment feature on the hardware revisions it supports
1602 	 */
1603 	nicvf_apad_config(nic, false);
1604 
1605 	/* Get queue ranges for this VF */
1606 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
1607 
1608 	/* Configure TX queues */
1609 	for (qidx = tx_start; qidx <= tx_end; qidx++) {
1610 		ret = nicvf_vf_start_tx_queue(dev, nic,
1611 			qidx % MAX_SND_QUEUES_PER_QS);
1612 		if (ret)
1613 			goto start_txq_error;
1614 	}
1615 
1616 	/* Configure RX queues */
1617 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1618 		ret = nicvf_vf_start_rx_queue(dev, nic,
1619 			qidx % MAX_RCV_QUEUES_PER_QS);
1620 		if (ret)
1621 			goto start_rxq_error;
1622 	}
1623 
1624 	if (!nic->sqs_mode) {
1625 		/* Configure CPI algorithm */
1626 		ret = nicvf_configure_cpi(dev);
1627 		if (ret)
1628 			goto start_txq_error;
1629 
1630 		ret = nicvf_mbox_get_rss_size(nic);
1631 		if (ret) {
1632 			PMD_INIT_LOG(ERR, "Failed to get rss table size");
1633 			goto qset_rss_error;
1634 		}
1635 
1636 		/* Configure RSS */
1637 		ret = nicvf_configure_rss(dev);
1638 		if (ret)
1639 			goto qset_rss_error;
1640 	}
1641 
1642 	/* Done; Let PF make the BGX's RX and TX switches to ON position */
1643 	nicvf_mbox_cfg_done(nic);
1644 	return 0;
1645 
1646 qset_rss_error:
1647 	nicvf_rss_term(nic);
1648 start_rxq_error:
1649 	for (qidx = rx_start; qidx <= rx_end; qidx++)
1650 		nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
1651 start_txq_error:
1652 	for (qidx = tx_start; qidx <= tx_end; qidx++)
1653 		nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
1654 qset_rbdr_reclaim:
1655 	nicvf_qset_rbdr_reclaim(nic, 0);
1656 	nicvf_rbdr_release_mbufs(dev, nic);
1657 qset_rbdr_free:
1658 	if (nic->rbdr) {
1659 		rte_free(nic->rbdr);
1660 		nic->rbdr = NULL;
1661 	}
1662 qset_reclaim:
1663 	nicvf_qset_reclaim(nic);
1664 	return ret;
1665 }
1666 
1667 static int
1668 nicvf_dev_start(struct rte_eth_dev *dev)
1669 {
1670 	uint16_t qidx;
1671 	int ret;
1672 	size_t i;
1673 	struct nicvf *nic = nicvf_pmd_priv(dev);
1674 	struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
1675 	uint16_t mtu;
1676 	uint32_t buffsz = 0, rbdrsz = 0;
1677 	struct rte_pktmbuf_pool_private *mbp_priv;
1678 	struct nicvf_rxq *rxq;
1679 
1680 	PMD_INIT_FUNC_TRACE();
1681 
1682 	/* This function must be called for a primary device */
1683 	assert_primary(nic);
1684 
1685 	/* Validate RBDR buff size */
1686 	for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
1687 		rxq = dev->data->rx_queues[qidx];
1688 		mbp_priv = rte_mempool_get_priv(rxq->pool);
1689 		buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
1690 		if (buffsz % 128) {
1691 			PMD_INIT_LOG(ERR, "rxbuf size must be multiply of 128");
1692 			return -EINVAL;
1693 		}
1694 		if (rbdrsz == 0)
1695 			rbdrsz = buffsz;
1696 		if (rbdrsz != buffsz) {
1697 			PMD_INIT_LOG(ERR, "buffsz not same, qidx=%d (%d/%d)",
1698 				     qidx, rbdrsz, buffsz);
1699 			return -EINVAL;
1700 		}
1701 	}
1702 
1703 	/* Configure loopback */
1704 	ret = nicvf_loopback_config(nic, dev->data->dev_conf.lpbk_mode);
1705 	if (ret) {
1706 		PMD_INIT_LOG(ERR, "Failed to configure loopback %d", ret);
1707 		return ret;
1708 	}
1709 
1710 	/* Reset all statistics counters attached to this port */
1711 	ret = nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, 0xFFFF, 0xFFFF);
1712 	if (ret) {
1713 		PMD_INIT_LOG(ERR, "Failed to reset stat counters %d", ret);
1714 		return ret;
1715 	}
1716 
1717 	/* Setup scatter mode if needed by jumbo */
1718 	if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
1719 					    2 * VLAN_TAG_SIZE > buffsz)
1720 		dev->data->scattered_rx = 1;
1721 	if ((rx_conf->offloads & DEV_RX_OFFLOAD_SCATTER) != 0)
1722 		dev->data->scattered_rx = 1;
1723 
1724 	/* Setup MTU based on max_rx_pkt_len or default */
1725 	mtu = dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME ?
1726 		dev->data->dev_conf.rxmode.max_rx_pkt_len
1727 			-  ETHER_HDR_LEN : ETHER_MTU;
1728 
1729 	if (nicvf_dev_set_mtu(dev, mtu)) {
1730 		PMD_INIT_LOG(ERR, "Failed to set default mtu size");
1731 		return -EBUSY;
1732 	}
1733 
1734 	ret = nicvf_vf_start(dev, nic, rbdrsz);
1735 	if (ret != 0)
1736 		return ret;
1737 
1738 	for (i = 0; i < nic->sqs_count; i++) {
1739 		assert(nic->snicvf[i]);
1740 
1741 		ret = nicvf_vf_start(dev, nic->snicvf[i], rbdrsz);
1742 		if (ret != 0)
1743 			return ret;
1744 	}
1745 
1746 	/* Configure callbacks based on scatter mode */
1747 	nicvf_set_tx_function(dev);
1748 	nicvf_set_rx_function(dev);
1749 
1750 	return 0;
1751 }
1752 
1753 static void
1754 nicvf_dev_stop_cleanup(struct rte_eth_dev *dev, bool cleanup)
1755 {
1756 	size_t i;
1757 	int ret;
1758 	struct nicvf *nic = nicvf_pmd_priv(dev);
1759 
1760 	PMD_INIT_FUNC_TRACE();
1761 
1762 	/* Teardown secondary vf first */
1763 	for (i = 0; i < nic->sqs_count; i++) {
1764 		if (!nic->snicvf[i])
1765 			continue;
1766 
1767 		nicvf_vf_stop(dev, nic->snicvf[i], cleanup);
1768 	}
1769 
1770 	/* Stop the primary VF now */
1771 	nicvf_vf_stop(dev, nic, cleanup);
1772 
1773 	/* Disable loopback */
1774 	ret = nicvf_loopback_config(nic, 0);
1775 	if (ret)
1776 		PMD_INIT_LOG(ERR, "Failed to disable loopback %d", ret);
1777 
1778 	/* Reclaim CPI configuration */
1779 	ret = nicvf_mbox_config_cpi(nic, 0);
1780 	if (ret)
1781 		PMD_INIT_LOG(ERR, "Failed to reclaim CPI config %d", ret);
1782 }
1783 
1784 static void
1785 nicvf_dev_stop(struct rte_eth_dev *dev)
1786 {
1787 	PMD_INIT_FUNC_TRACE();
1788 
1789 	nicvf_dev_stop_cleanup(dev, false);
1790 }
1791 
1792 static void
1793 nicvf_vf_stop(struct rte_eth_dev *dev, struct nicvf *nic, bool cleanup)
1794 {
1795 	int ret;
1796 	uint16_t qidx;
1797 	uint16_t tx_start, tx_end;
1798 	uint16_t rx_start, rx_end;
1799 
1800 	PMD_INIT_FUNC_TRACE();
1801 
1802 	if (cleanup) {
1803 		/* Let PF make the BGX's RX and TX switches to OFF position */
1804 		nicvf_mbox_shutdown(nic);
1805 	}
1806 
1807 	/* Disable VLAN Strip */
1808 	nicvf_vlan_hw_strip(nic, 0);
1809 
1810 	/* Get queue ranges for this VF */
1811 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
1812 
1813 	for (qidx = tx_start; qidx <= tx_end; qidx++)
1814 		nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
1815 
1816 	/* Get queue ranges for this VF */
1817 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
1818 
1819 	/* Reclaim rq */
1820 	for (qidx = rx_start; qidx <= rx_end; qidx++)
1821 		nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
1822 
1823 	/* Reclaim RBDR */
1824 	ret = nicvf_qset_rbdr_reclaim(nic, 0);
1825 	if (ret)
1826 		PMD_INIT_LOG(ERR, "Failed to reclaim RBDR %d", ret);
1827 
1828 	/* Move all charged buffers in RBDR back to pool */
1829 	if (nic->rbdr != NULL)
1830 		nicvf_rbdr_release_mbufs(dev, nic);
1831 
1832 	/* Disable qset */
1833 	ret = nicvf_qset_reclaim(nic);
1834 	if (ret)
1835 		PMD_INIT_LOG(ERR, "Failed to disable qset %d", ret);
1836 
1837 	/* Disable all interrupts */
1838 	nicvf_disable_all_interrupts(nic);
1839 
1840 	/* Free RBDR SW structure */
1841 	if (nic->rbdr) {
1842 		rte_free(nic->rbdr);
1843 		nic->rbdr = NULL;
1844 	}
1845 }
1846 
1847 static void
1848 nicvf_dev_close(struct rte_eth_dev *dev)
1849 {
1850 	size_t i;
1851 	struct nicvf *nic = nicvf_pmd_priv(dev);
1852 
1853 	PMD_INIT_FUNC_TRACE();
1854 
1855 	nicvf_dev_stop_cleanup(dev, true);
1856 	nicvf_periodic_alarm_stop(nicvf_interrupt, dev);
1857 
1858 	for (i = 0; i < nic->sqs_count; i++) {
1859 		if (!nic->snicvf[i])
1860 			continue;
1861 
1862 		nicvf_periodic_alarm_stop(nicvf_vf_interrupt, nic->snicvf[i]);
1863 	}
1864 }
1865 
1866 static int
1867 nicvf_request_sqs(struct nicvf *nic)
1868 {
1869 	size_t i;
1870 
1871 	assert_primary(nic);
1872 	assert(nic->sqs_count > 0);
1873 	assert(nic->sqs_count <= MAX_SQS_PER_VF);
1874 
1875 	/* Set no of Rx/Tx queues in each of the SQsets */
1876 	for (i = 0; i < nic->sqs_count; i++) {
1877 		if (nicvf_svf_empty())
1878 			rte_panic("Cannot assign sufficient number of "
1879 				  "secondary queues to primary VF%" PRIu8 "\n",
1880 				  nic->vf_id);
1881 
1882 		nic->snicvf[i] = nicvf_svf_pop();
1883 		nic->snicvf[i]->sqs_id = i;
1884 	}
1885 
1886 	return nicvf_mbox_request_sqs(nic);
1887 }
1888 
1889 static int
1890 nicvf_dev_configure(struct rte_eth_dev *dev)
1891 {
1892 	struct rte_eth_dev_data *data = dev->data;
1893 	struct rte_eth_conf *conf = &data->dev_conf;
1894 	struct rte_eth_rxmode *rxmode = &conf->rxmode;
1895 	struct rte_eth_txmode *txmode = &conf->txmode;
1896 	struct nicvf *nic = nicvf_pmd_priv(dev);
1897 	uint8_t cqcount;
1898 
1899 	PMD_INIT_FUNC_TRACE();
1900 
1901 	if (!rte_eal_has_hugepages()) {
1902 		PMD_INIT_LOG(INFO, "Huge page is not configured");
1903 		return -EINVAL;
1904 	}
1905 
1906 	/* KEEP_CRC offload flag is not supported by PMD
1907 	 * can remove the below block when DEV_RX_OFFLOAD_CRC_STRIP removed
1908 	 */
1909 	if (rte_eth_dev_must_keep_crc(rxmode->offloads)) {
1910 		PMD_INIT_LOG(NOTICE, "Can't disable hw crc strip");
1911 		rxmode->offloads |= DEV_RX_OFFLOAD_CRC_STRIP;
1912 	}
1913 
1914 	if (txmode->mq_mode) {
1915 		PMD_INIT_LOG(INFO, "Tx mq_mode DCB or VMDq not supported");
1916 		return -EINVAL;
1917 	}
1918 
1919 	if (rxmode->mq_mode != ETH_MQ_RX_NONE &&
1920 		rxmode->mq_mode != ETH_MQ_RX_RSS) {
1921 		PMD_INIT_LOG(INFO, "Unsupported rx qmode %d", rxmode->mq_mode);
1922 		return -EINVAL;
1923 	}
1924 
1925 	if (rxmode->split_hdr_size) {
1926 		PMD_INIT_LOG(INFO, "Rxmode does not support split header");
1927 		return -EINVAL;
1928 	}
1929 
1930 	if (conf->link_speeds & ETH_LINK_SPEED_FIXED) {
1931 		PMD_INIT_LOG(INFO, "Setting link speed/duplex not supported");
1932 		return -EINVAL;
1933 	}
1934 
1935 	if (conf->dcb_capability_en) {
1936 		PMD_INIT_LOG(INFO, "DCB enable not supported");
1937 		return -EINVAL;
1938 	}
1939 
1940 	if (conf->fdir_conf.mode != RTE_FDIR_MODE_NONE) {
1941 		PMD_INIT_LOG(INFO, "Flow director not supported");
1942 		return -EINVAL;
1943 	}
1944 
1945 	assert_primary(nic);
1946 	NICVF_STATIC_ASSERT(MAX_RCV_QUEUES_PER_QS == MAX_SND_QUEUES_PER_QS);
1947 	cqcount = RTE_MAX(data->nb_tx_queues, data->nb_rx_queues);
1948 	if (cqcount > MAX_RCV_QUEUES_PER_QS) {
1949 		nic->sqs_count = RTE_ALIGN_CEIL(cqcount, MAX_RCV_QUEUES_PER_QS);
1950 		nic->sqs_count = (nic->sqs_count / MAX_RCV_QUEUES_PER_QS) - 1;
1951 	} else {
1952 		nic->sqs_count = 0;
1953 	}
1954 
1955 	assert(nic->sqs_count <= MAX_SQS_PER_VF);
1956 
1957 	if (nic->sqs_count > 0) {
1958 		if (nicvf_request_sqs(nic)) {
1959 			rte_panic("Cannot assign sufficient number of "
1960 				  "secondary queues to PORT%d VF%" PRIu8 "\n",
1961 				  dev->data->port_id, nic->vf_id);
1962 		}
1963 	}
1964 
1965 	PMD_INIT_LOG(DEBUG, "Configured ethdev port%d hwcap=0x%" PRIx64,
1966 		dev->data->port_id, nicvf_hw_cap(nic));
1967 
1968 	return 0;
1969 }
1970 
1971 /* Initialize and register driver with DPDK Application */
1972 static const struct eth_dev_ops nicvf_eth_dev_ops = {
1973 	.dev_configure            = nicvf_dev_configure,
1974 	.dev_start                = nicvf_dev_start,
1975 	.dev_stop                 = nicvf_dev_stop,
1976 	.link_update              = nicvf_dev_link_update,
1977 	.dev_close                = nicvf_dev_close,
1978 	.stats_get                = nicvf_dev_stats_get,
1979 	.stats_reset              = nicvf_dev_stats_reset,
1980 	.promiscuous_enable       = nicvf_dev_promisc_enable,
1981 	.dev_infos_get            = nicvf_dev_info_get,
1982 	.dev_supported_ptypes_get = nicvf_dev_supported_ptypes_get,
1983 	.mtu_set                  = nicvf_dev_set_mtu,
1984 	.reta_update              = nicvf_dev_reta_update,
1985 	.reta_query               = nicvf_dev_reta_query,
1986 	.rss_hash_update          = nicvf_dev_rss_hash_update,
1987 	.rss_hash_conf_get        = nicvf_dev_rss_hash_conf_get,
1988 	.rx_queue_start           = nicvf_dev_rx_queue_start,
1989 	.rx_queue_stop            = nicvf_dev_rx_queue_stop,
1990 	.tx_queue_start           = nicvf_dev_tx_queue_start,
1991 	.tx_queue_stop            = nicvf_dev_tx_queue_stop,
1992 	.rx_queue_setup           = nicvf_dev_rx_queue_setup,
1993 	.rx_queue_release         = nicvf_dev_rx_queue_release,
1994 	.rx_queue_count           = nicvf_dev_rx_queue_count,
1995 	.tx_queue_setup           = nicvf_dev_tx_queue_setup,
1996 	.tx_queue_release         = nicvf_dev_tx_queue_release,
1997 	.get_reg                  = nicvf_dev_get_regs,
1998 };
1999 
2000 static inline int
2001 nicvf_set_first_skip(struct rte_eth_dev *dev)
2002 {
2003 	int bytes_to_skip = 0;
2004 	int ret = 0;
2005 	unsigned int i;
2006 	struct rte_kvargs *kvlist;
2007 	static const char *const skip[] = {
2008 		SKIP_DATA_BYTES,
2009 		NULL};
2010 	struct nicvf *nic = nicvf_pmd_priv(dev);
2011 
2012 	if (!dev->device->devargs) {
2013 		nicvf_first_skip_config(nic, 0);
2014 		return ret;
2015 	}
2016 
2017 	kvlist = rte_kvargs_parse(dev->device->devargs->args, skip);
2018 	if (!kvlist)
2019 		return -EINVAL;
2020 
2021 	if (kvlist->count == 0)
2022 		goto exit;
2023 
2024 	for (i = 0; i != kvlist->count; ++i) {
2025 		const struct rte_kvargs_pair *pair = &kvlist->pairs[i];
2026 
2027 		if (!strcmp(pair->key, SKIP_DATA_BYTES))
2028 			bytes_to_skip = atoi(pair->value);
2029 	}
2030 
2031 	/*128 bytes amounts to one cache line*/
2032 	if (bytes_to_skip >= 0 && bytes_to_skip < 128) {
2033 		if (!(bytes_to_skip % 8)) {
2034 			nicvf_first_skip_config(nic, (bytes_to_skip / 8));
2035 			nic->skip_bytes = bytes_to_skip;
2036 			goto kvlist_free;
2037 		} else {
2038 			PMD_INIT_LOG(ERR, "skip_data_bytes should be multiple of 8");
2039 			ret = -EINVAL;
2040 			goto exit;
2041 		}
2042 	} else {
2043 		PMD_INIT_LOG(ERR, "skip_data_bytes should be less than 128");
2044 		ret = -EINVAL;
2045 		goto exit;
2046 	}
2047 exit:
2048 	nicvf_first_skip_config(nic, 0);
2049 kvlist_free:
2050 	rte_kvargs_free(kvlist);
2051 	return ret;
2052 }
2053 static int
2054 nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
2055 {
2056 	int ret;
2057 	struct rte_pci_device *pci_dev;
2058 	struct nicvf *nic = nicvf_pmd_priv(eth_dev);
2059 
2060 	PMD_INIT_FUNC_TRACE();
2061 
2062 	eth_dev->dev_ops = &nicvf_eth_dev_ops;
2063 
2064 	/* For secondary processes, the primary has done all the work */
2065 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2066 		if (nic) {
2067 			/* Setup callbacks for secondary process */
2068 			nicvf_set_tx_function(eth_dev);
2069 			nicvf_set_rx_function(eth_dev);
2070 			return 0;
2071 		} else {
2072 			/* If nic == NULL than it is secondary function
2073 			 * so ethdev need to be released by caller */
2074 			return ENOTSUP;
2075 		}
2076 	}
2077 
2078 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
2079 	rte_eth_copy_pci_info(eth_dev, pci_dev);
2080 
2081 	nic->device_id = pci_dev->id.device_id;
2082 	nic->vendor_id = pci_dev->id.vendor_id;
2083 	nic->subsystem_device_id = pci_dev->id.subsystem_device_id;
2084 	nic->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
2085 
2086 	PMD_INIT_LOG(DEBUG, "nicvf: device (%x:%x) %u:%u:%u:%u",
2087 			pci_dev->id.vendor_id, pci_dev->id.device_id,
2088 			pci_dev->addr.domain, pci_dev->addr.bus,
2089 			pci_dev->addr.devid, pci_dev->addr.function);
2090 
2091 	nic->reg_base = (uintptr_t)pci_dev->mem_resource[0].addr;
2092 	if (!nic->reg_base) {
2093 		PMD_INIT_LOG(ERR, "Failed to map BAR0");
2094 		ret = -ENODEV;
2095 		goto fail;
2096 	}
2097 
2098 	nicvf_disable_all_interrupts(nic);
2099 
2100 	ret = nicvf_periodic_alarm_start(nicvf_interrupt, eth_dev);
2101 	if (ret) {
2102 		PMD_INIT_LOG(ERR, "Failed to start period alarm");
2103 		goto fail;
2104 	}
2105 
2106 	ret = nicvf_mbox_check_pf_ready(nic);
2107 	if (ret) {
2108 		PMD_INIT_LOG(ERR, "Failed to get ready message from PF");
2109 		goto alarm_fail;
2110 	} else {
2111 		PMD_INIT_LOG(INFO,
2112 			"node=%d vf=%d mode=%s sqs=%s loopback_supported=%s",
2113 			nic->node, nic->vf_id,
2114 			nic->tns_mode == NIC_TNS_MODE ? "tns" : "tns-bypass",
2115 			nic->sqs_mode ? "true" : "false",
2116 			nic->loopback_supported ? "true" : "false"
2117 			);
2118 	}
2119 
2120 	ret = nicvf_base_init(nic);
2121 	if (ret) {
2122 		PMD_INIT_LOG(ERR, "Failed to execute nicvf_base_init");
2123 		goto malloc_fail;
2124 	}
2125 
2126 	if (nic->sqs_mode) {
2127 		/* Push nic to stack of secondary vfs */
2128 		nicvf_svf_push(nic);
2129 
2130 		/* Steal nic pointer from the device for further reuse */
2131 		eth_dev->data->dev_private = NULL;
2132 
2133 		nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
2134 		ret = nicvf_periodic_alarm_start(nicvf_vf_interrupt, nic);
2135 		if (ret) {
2136 			PMD_INIT_LOG(ERR, "Failed to start period alarm");
2137 			goto fail;
2138 		}
2139 
2140 		/* Detach port by returning positive error number */
2141 		return ENOTSUP;
2142 	}
2143 
2144 	eth_dev->data->mac_addrs = rte_zmalloc("mac_addr", ETHER_ADDR_LEN, 0);
2145 	if (eth_dev->data->mac_addrs == NULL) {
2146 		PMD_INIT_LOG(ERR, "Failed to allocate memory for mac addr");
2147 		ret = -ENOMEM;
2148 		goto alarm_fail;
2149 	}
2150 	if (is_zero_ether_addr((struct ether_addr *)nic->mac_addr))
2151 		eth_random_addr(&nic->mac_addr[0]);
2152 
2153 	ether_addr_copy((struct ether_addr *)nic->mac_addr,
2154 			&eth_dev->data->mac_addrs[0]);
2155 
2156 	ret = nicvf_mbox_set_mac_addr(nic, nic->mac_addr);
2157 	if (ret) {
2158 		PMD_INIT_LOG(ERR, "Failed to set mac addr");
2159 		goto malloc_fail;
2160 	}
2161 
2162 	ret = nicvf_set_first_skip(eth_dev);
2163 	if (ret) {
2164 		PMD_INIT_LOG(ERR, "Failed to configure first skip");
2165 		goto malloc_fail;
2166 	}
2167 	PMD_INIT_LOG(INFO, "Port %d (%x:%x) mac=%02x:%02x:%02x:%02x:%02x:%02x",
2168 		eth_dev->data->port_id, nic->vendor_id, nic->device_id,
2169 		nic->mac_addr[0], nic->mac_addr[1], nic->mac_addr[2],
2170 		nic->mac_addr[3], nic->mac_addr[4], nic->mac_addr[5]);
2171 
2172 	return 0;
2173 
2174 malloc_fail:
2175 	rte_free(eth_dev->data->mac_addrs);
2176 alarm_fail:
2177 	nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
2178 fail:
2179 	return ret;
2180 }
2181 
2182 static const struct rte_pci_id pci_id_nicvf_map[] = {
2183 	{
2184 		.class_id = RTE_CLASS_ANY_ID,
2185 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2186 		.device_id = PCI_DEVICE_ID_THUNDERX_CN88XX_PASS1_NICVF,
2187 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2188 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS1_NICVF,
2189 	},
2190 	{
2191 		.class_id = RTE_CLASS_ANY_ID,
2192 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2193 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2194 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2195 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS2_NICVF,
2196 	},
2197 	{
2198 		.class_id = RTE_CLASS_ANY_ID,
2199 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2200 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2201 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2202 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN81XX_NICVF,
2203 	},
2204 	{
2205 		.class_id = RTE_CLASS_ANY_ID,
2206 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2207 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2208 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2209 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN83XX_NICVF,
2210 	},
2211 	{
2212 		.vendor_id = 0,
2213 	},
2214 };
2215 
2216 static int nicvf_eth_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2217 	struct rte_pci_device *pci_dev)
2218 {
2219 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct nicvf),
2220 		nicvf_eth_dev_init);
2221 }
2222 
2223 static int nicvf_eth_pci_remove(struct rte_pci_device *pci_dev)
2224 {
2225 	return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
2226 }
2227 
2228 static struct rte_pci_driver rte_nicvf_pmd = {
2229 	.id_table = pci_id_nicvf_map,
2230 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_KEEP_MAPPED_RES |
2231 			RTE_PCI_DRV_INTR_LSC,
2232 	.probe = nicvf_eth_pci_probe,
2233 	.remove = nicvf_eth_pci_remove,
2234 };
2235 
2236 RTE_PMD_REGISTER_PCI(net_thunderx, rte_nicvf_pmd);
2237 RTE_PMD_REGISTER_PCI_TABLE(net_thunderx, pci_id_nicvf_map);
2238 RTE_PMD_REGISTER_KMOD_DEP(net_thunderx, "* igb_uio | uio_pci_generic | vfio-pci");
2239 RTE_PMD_REGISTER_PARAM_STRING(net_thunderx, SKIP_DATA_BYTES "=<int>");
2240