xref: /dpdk/drivers/net/thunderx/nicvf_ethdev.c (revision bbbe38a6d59ccdda25917712701e629d0b10af6f)
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 <ethdev_driver.h>
27 #include <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 static int nicvf_dev_stop(struct rte_eth_dev *dev);
48 static void nicvf_dev_stop_cleanup(struct rte_eth_dev *dev, bool cleanup);
49 static void nicvf_vf_stop(struct rte_eth_dev *dev, struct nicvf *nic,
50 			  bool cleanup);
51 static int nicvf_vlan_offload_config(struct rte_eth_dev *dev, int mask);
52 static int nicvf_vlan_offload_set(struct rte_eth_dev *dev, int mask);
53 
54 RTE_LOG_REGISTER_SUFFIX(nicvf_logtype_mbox, mbox, NOTICE);
55 RTE_LOG_REGISTER_SUFFIX(nicvf_logtype_init, init, NOTICE);
56 RTE_LOG_REGISTER_SUFFIX(nicvf_logtype_driver, driver, NOTICE);
57 
58 static void
59 nicvf_link_status_update(struct nicvf *nic,
60 			 struct rte_eth_link *link)
61 {
62 	memset(link, 0, sizeof(*link));
63 
64 	link->link_status = nic->link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
65 
66 	if (nic->duplex == NICVF_HALF_DUPLEX)
67 		link->link_duplex = ETH_LINK_HALF_DUPLEX;
68 	else if (nic->duplex == NICVF_FULL_DUPLEX)
69 		link->link_duplex = ETH_LINK_FULL_DUPLEX;
70 	link->link_speed = nic->speed;
71 	link->link_autoneg = ETH_LINK_AUTONEG;
72 }
73 
74 static void
75 nicvf_interrupt(void *arg)
76 {
77 	struct rte_eth_dev *dev = arg;
78 	struct nicvf *nic = nicvf_pmd_priv(dev);
79 	struct rte_eth_link link;
80 
81 	if (nicvf_reg_poll_interrupts(nic) == NIC_MBOX_MSG_BGX_LINK_CHANGE) {
82 		if (dev->data->dev_conf.intr_conf.lsc) {
83 			nicvf_link_status_update(nic, &link);
84 			rte_eth_linkstatus_set(dev, &link);
85 
86 			rte_eth_dev_callback_process(dev,
87 						     RTE_ETH_EVENT_INTR_LSC,
88 						     NULL);
89 		}
90 	}
91 
92 	rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
93 				nicvf_interrupt, dev);
94 }
95 
96 static void
97 nicvf_vf_interrupt(void *arg)
98 {
99 	struct nicvf *nic = arg;
100 
101 	nicvf_reg_poll_interrupts(nic);
102 
103 	rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
104 				nicvf_vf_interrupt, nic);
105 }
106 
107 static int
108 nicvf_periodic_alarm_start(void (fn)(void *), void *arg)
109 {
110 	return rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000, fn, arg);
111 }
112 
113 static int
114 nicvf_periodic_alarm_stop(void (fn)(void *), void *arg)
115 {
116 	return rte_eal_alarm_cancel(fn, arg);
117 }
118 
119 /*
120  * Return 0 means link status changed, -1 means not changed
121  */
122 static int
123 nicvf_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
124 {
125 #define CHECK_INTERVAL 100  /* 100ms */
126 #define MAX_CHECK_TIME 90   /* 9s (90 * 100ms) in total */
127 	struct rte_eth_link link;
128 	struct nicvf *nic = nicvf_pmd_priv(dev);
129 	int i;
130 
131 	PMD_INIT_FUNC_TRACE();
132 
133 	if (wait_to_complete) {
134 		/* rte_eth_link_get() might need to wait up to 9 seconds */
135 		for (i = 0; i < MAX_CHECK_TIME; i++) {
136 			nicvf_link_status_update(nic, &link);
137 			if (link.link_status == ETH_LINK_UP)
138 				break;
139 			rte_delay_ms(CHECK_INTERVAL);
140 		}
141 	} else {
142 		nicvf_link_status_update(nic, &link);
143 	}
144 
145 	return rte_eth_linkstatus_set(dev, &link);
146 }
147 
148 static int
149 nicvf_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
150 {
151 	struct nicvf *nic = nicvf_pmd_priv(dev);
152 	uint32_t buffsz, frame_size = mtu + NIC_HW_L2_OVERHEAD;
153 	size_t i;
154 	struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
155 
156 	PMD_INIT_FUNC_TRACE();
157 
158 	if (frame_size > NIC_HW_MAX_FRS)
159 		return -EINVAL;
160 
161 	if (frame_size < NIC_HW_MIN_FRS)
162 		return -EINVAL;
163 
164 	buffsz = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
165 
166 	/*
167 	 * Refuse mtu that requires the support of scattered packets
168 	 * when this feature has not been enabled before.
169 	 */
170 	if (dev->data->dev_started && !dev->data->scattered_rx &&
171 		(frame_size + 2 * VLAN_TAG_SIZE > buffsz))
172 		return -EINVAL;
173 
174 	/* check <seg size> * <max_seg>  >= max_frame */
175 	if (dev->data->scattered_rx &&
176 		(frame_size + 2 * VLAN_TAG_SIZE > buffsz * NIC_HW_MAX_SEGS))
177 		return -EINVAL;
178 
179 	if (frame_size > NIC_HW_L2_MAX_LEN)
180 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
181 	else
182 		rxmode->offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
183 
184 	if (nicvf_mbox_update_hw_max_frs(nic, mtu))
185 		return -EINVAL;
186 
187 	/* Update max_rx_pkt_len */
188 	rxmode->max_rx_pkt_len = mtu + RTE_ETHER_HDR_LEN;
189 	nic->mtu = mtu;
190 
191 	for (i = 0; i < nic->sqs_count; i++)
192 		nic->snicvf[i]->mtu = mtu;
193 
194 	return 0;
195 }
196 
197 static int
198 nicvf_dev_get_regs(struct rte_eth_dev *dev, struct rte_dev_reg_info *regs)
199 {
200 	uint64_t *data = regs->data;
201 	struct nicvf *nic = nicvf_pmd_priv(dev);
202 
203 	if (data == NULL) {
204 		regs->length = nicvf_reg_get_count();
205 		regs->width = THUNDERX_REG_BYTES;
206 		return 0;
207 	}
208 
209 	/* Support only full register dump */
210 	if ((regs->length == 0) ||
211 		(regs->length == (uint32_t)nicvf_reg_get_count())) {
212 		regs->version = nic->vendor_id << 16 | nic->device_id;
213 		nicvf_reg_dump(nic, data);
214 		return 0;
215 	}
216 	return -ENOTSUP;
217 }
218 
219 static int
220 nicvf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
221 {
222 	uint16_t qidx;
223 	struct nicvf_hw_rx_qstats rx_qstats;
224 	struct nicvf_hw_tx_qstats tx_qstats;
225 	struct nicvf_hw_stats port_stats;
226 	struct nicvf *nic = nicvf_pmd_priv(dev);
227 	uint16_t rx_start, rx_end;
228 	uint16_t tx_start, tx_end;
229 	size_t i;
230 
231 	/* RX queue indices for the first VF */
232 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
233 
234 	/* Reading per RX ring stats */
235 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
236 		if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
237 			break;
238 
239 		nicvf_hw_get_rx_qstats(nic, &rx_qstats, qidx);
240 		stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
241 		stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
242 	}
243 
244 	/* TX queue indices for the first VF */
245 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
246 
247 	/* Reading per TX ring stats */
248 	for (qidx = tx_start; qidx <= tx_end; qidx++) {
249 		if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
250 			break;
251 
252 		nicvf_hw_get_tx_qstats(nic, &tx_qstats, qidx);
253 		stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
254 		stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
255 	}
256 
257 	for (i = 0; i < nic->sqs_count; i++) {
258 		struct nicvf *snic = nic->snicvf[i];
259 
260 		if (snic == NULL)
261 			break;
262 
263 		/* RX queue indices for a secondary VF */
264 		nicvf_rx_range(dev, snic, &rx_start, &rx_end);
265 
266 		/* Reading per RX ring stats */
267 		for (qidx = rx_start; qidx <= rx_end; qidx++) {
268 			if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
269 				break;
270 
271 			nicvf_hw_get_rx_qstats(snic, &rx_qstats,
272 					       qidx % MAX_RCV_QUEUES_PER_QS);
273 			stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
274 			stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
275 		}
276 
277 		/* TX queue indices for a secondary VF */
278 		nicvf_tx_range(dev, snic, &tx_start, &tx_end);
279 		/* Reading per TX ring stats */
280 		for (qidx = tx_start; qidx <= tx_end; qidx++) {
281 			if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
282 				break;
283 
284 			nicvf_hw_get_tx_qstats(snic, &tx_qstats,
285 					       qidx % MAX_SND_QUEUES_PER_QS);
286 			stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
287 			stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
288 		}
289 	}
290 
291 	nicvf_hw_get_stats(nic, &port_stats);
292 	stats->ibytes = port_stats.rx_bytes;
293 	stats->ipackets = port_stats.rx_ucast_frames;
294 	stats->ipackets += port_stats.rx_bcast_frames;
295 	stats->ipackets += port_stats.rx_mcast_frames;
296 	stats->ierrors = port_stats.rx_l2_errors;
297 	stats->imissed = port_stats.rx_drop_red;
298 	stats->imissed += port_stats.rx_drop_overrun;
299 	stats->imissed += port_stats.rx_drop_bcast;
300 	stats->imissed += port_stats.rx_drop_mcast;
301 	stats->imissed += port_stats.rx_drop_l3_bcast;
302 	stats->imissed += port_stats.rx_drop_l3_mcast;
303 
304 	stats->obytes = port_stats.tx_bytes_ok;
305 	stats->opackets = port_stats.tx_ucast_frames_ok;
306 	stats->opackets += port_stats.tx_bcast_frames_ok;
307 	stats->opackets += port_stats.tx_mcast_frames_ok;
308 	stats->oerrors = port_stats.tx_drops;
309 
310 	return 0;
311 }
312 
313 static const uint32_t *
314 nicvf_dev_supported_ptypes_get(struct rte_eth_dev *dev)
315 {
316 	size_t copied;
317 	static uint32_t ptypes[32];
318 	struct nicvf *nic = nicvf_pmd_priv(dev);
319 	static const uint32_t ptypes_common[] = {
320 		RTE_PTYPE_L3_IPV4,
321 		RTE_PTYPE_L3_IPV4_EXT,
322 		RTE_PTYPE_L3_IPV6,
323 		RTE_PTYPE_L3_IPV6_EXT,
324 		RTE_PTYPE_L4_TCP,
325 		RTE_PTYPE_L4_UDP,
326 		RTE_PTYPE_L4_FRAG,
327 	};
328 	static const uint32_t ptypes_tunnel[] = {
329 		RTE_PTYPE_TUNNEL_GRE,
330 		RTE_PTYPE_TUNNEL_GENEVE,
331 		RTE_PTYPE_TUNNEL_VXLAN,
332 		RTE_PTYPE_TUNNEL_NVGRE,
333 	};
334 	static const uint32_t ptypes_end = RTE_PTYPE_UNKNOWN;
335 
336 	copied = sizeof(ptypes_common);
337 	memcpy(ptypes, ptypes_common, copied);
338 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
339 		memcpy((char *)ptypes + copied, ptypes_tunnel,
340 			sizeof(ptypes_tunnel));
341 		copied += sizeof(ptypes_tunnel);
342 	}
343 
344 	memcpy((char *)ptypes + copied, &ptypes_end, sizeof(ptypes_end));
345 
346 	/* All Ptypes are supported in all Rx functions. */
347 	return ptypes;
348 }
349 
350 static int
351 nicvf_dev_stats_reset(struct rte_eth_dev *dev)
352 {
353 	int i;
354 	uint16_t rxqs = 0, txqs = 0;
355 	struct nicvf *nic = nicvf_pmd_priv(dev);
356 	uint16_t rx_start, rx_end;
357 	uint16_t tx_start, tx_end;
358 	int ret;
359 
360 	/* Reset all primary nic counters */
361 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
362 	for (i = rx_start; i <= rx_end; i++)
363 		rxqs |= (0x3 << (i * 2));
364 
365 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
366 	for (i = tx_start; i <= tx_end; i++)
367 		txqs |= (0x3 << (i * 2));
368 
369 	ret = nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, rxqs, txqs);
370 	if (ret != 0)
371 		return ret;
372 
373 	/* Reset secondary nic queue counters */
374 	for (i = 0; i < nic->sqs_count; i++) {
375 		struct nicvf *snic = nic->snicvf[i];
376 		if (snic == NULL)
377 			break;
378 
379 		nicvf_rx_range(dev, snic, &rx_start, &rx_end);
380 		for (i = rx_start; i <= rx_end; i++)
381 			rxqs |= (0x3 << ((i % MAX_CMP_QUEUES_PER_QS) * 2));
382 
383 		nicvf_tx_range(dev, snic, &tx_start, &tx_end);
384 		for (i = tx_start; i <= tx_end; i++)
385 			txqs |= (0x3 << ((i % MAX_SND_QUEUES_PER_QS) * 2));
386 
387 		ret = nicvf_mbox_reset_stat_counters(snic, 0, 0, rxqs, txqs);
388 		if (ret != 0)
389 			return ret;
390 	}
391 
392 	return 0;
393 }
394 
395 /* Promiscuous mode enabled by default in LMAC to VF 1:1 map configuration */
396 static int
397 nicvf_dev_promisc_enable(struct rte_eth_dev *dev __rte_unused)
398 {
399 	return 0;
400 }
401 
402 static inline uint64_t
403 nicvf_rss_ethdev_to_nic(struct nicvf *nic, uint64_t ethdev_rss)
404 {
405 	uint64_t nic_rss = 0;
406 
407 	if (ethdev_rss & ETH_RSS_IPV4)
408 		nic_rss |= RSS_IP_ENA;
409 
410 	if (ethdev_rss & ETH_RSS_IPV6)
411 		nic_rss |= RSS_IP_ENA;
412 
413 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_UDP)
414 		nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
415 
416 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_TCP)
417 		nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
418 
419 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_UDP)
420 		nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
421 
422 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_TCP)
423 		nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
424 
425 	if (ethdev_rss & ETH_RSS_PORT)
426 		nic_rss |= RSS_L2_EXTENDED_HASH_ENA;
427 
428 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
429 		if (ethdev_rss & ETH_RSS_VXLAN)
430 			nic_rss |= RSS_TUN_VXLAN_ENA;
431 
432 		if (ethdev_rss & ETH_RSS_GENEVE)
433 			nic_rss |= RSS_TUN_GENEVE_ENA;
434 
435 		if (ethdev_rss & ETH_RSS_NVGRE)
436 			nic_rss |= RSS_TUN_NVGRE_ENA;
437 	}
438 
439 	return nic_rss;
440 }
441 
442 static inline uint64_t
443 nicvf_rss_nic_to_ethdev(struct nicvf *nic,  uint64_t nic_rss)
444 {
445 	uint64_t ethdev_rss = 0;
446 
447 	if (nic_rss & RSS_IP_ENA)
448 		ethdev_rss |= (ETH_RSS_IPV4 | ETH_RSS_IPV6);
449 
450 	if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_TCP_ENA))
451 		ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_TCP |
452 				ETH_RSS_NONFRAG_IPV6_TCP);
453 
454 	if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_UDP_ENA))
455 		ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_UDP |
456 				ETH_RSS_NONFRAG_IPV6_UDP);
457 
458 	if (nic_rss & RSS_L2_EXTENDED_HASH_ENA)
459 		ethdev_rss |= ETH_RSS_PORT;
460 
461 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
462 		if (nic_rss & RSS_TUN_VXLAN_ENA)
463 			ethdev_rss |= ETH_RSS_VXLAN;
464 
465 		if (nic_rss & RSS_TUN_GENEVE_ENA)
466 			ethdev_rss |= ETH_RSS_GENEVE;
467 
468 		if (nic_rss & RSS_TUN_NVGRE_ENA)
469 			ethdev_rss |= ETH_RSS_NVGRE;
470 	}
471 	return ethdev_rss;
472 }
473 
474 static int
475 nicvf_dev_reta_query(struct rte_eth_dev *dev,
476 		     struct rte_eth_rss_reta_entry64 *reta_conf,
477 		     uint16_t reta_size)
478 {
479 	struct nicvf *nic = nicvf_pmd_priv(dev);
480 	uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
481 	int ret, i, j;
482 
483 	if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
484 		PMD_DRV_LOG(ERR,
485 			    "The size of hash lookup table configured "
486 			    "(%u) doesn't match the number hardware can supported "
487 			    "(%u)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
488 		return -EINVAL;
489 	}
490 
491 	ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
492 	if (ret)
493 		return ret;
494 
495 	/* Copy RETA table */
496 	for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
497 		for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
498 			if ((reta_conf[i].mask >> j) & 0x01)
499 				reta_conf[i].reta[j] = tbl[j];
500 	}
501 
502 	return 0;
503 }
504 
505 static int
506 nicvf_dev_reta_update(struct rte_eth_dev *dev,
507 		      struct rte_eth_rss_reta_entry64 *reta_conf,
508 		      uint16_t reta_size)
509 {
510 	struct nicvf *nic = nicvf_pmd_priv(dev);
511 	uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
512 	int ret, i, j;
513 
514 	if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
515 		PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
516 			"(%u) doesn't match the number hardware can supported "
517 			"(%u)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
518 		return -EINVAL;
519 	}
520 
521 	ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
522 	if (ret)
523 		return ret;
524 
525 	/* Copy RETA table */
526 	for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
527 		for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
528 			if ((reta_conf[i].mask >> j) & 0x01)
529 				tbl[j] = reta_conf[i].reta[j];
530 	}
531 
532 	return nicvf_rss_reta_update(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
533 }
534 
535 static int
536 nicvf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
537 			    struct rte_eth_rss_conf *rss_conf)
538 {
539 	struct nicvf *nic = nicvf_pmd_priv(dev);
540 
541 	if (rss_conf->rss_key)
542 		nicvf_rss_get_key(nic, rss_conf->rss_key);
543 
544 	rss_conf->rss_key_len =  RSS_HASH_KEY_BYTE_SIZE;
545 	rss_conf->rss_hf = nicvf_rss_nic_to_ethdev(nic, nicvf_rss_get_cfg(nic));
546 	return 0;
547 }
548 
549 static int
550 nicvf_dev_rss_hash_update(struct rte_eth_dev *dev,
551 			  struct rte_eth_rss_conf *rss_conf)
552 {
553 	struct nicvf *nic = nicvf_pmd_priv(dev);
554 	uint64_t nic_rss;
555 
556 	if (rss_conf->rss_key &&
557 		rss_conf->rss_key_len != RSS_HASH_KEY_BYTE_SIZE) {
558 		PMD_DRV_LOG(ERR, "Hash key size mismatch %u",
559 			    rss_conf->rss_key_len);
560 		return -EINVAL;
561 	}
562 
563 	if (rss_conf->rss_key)
564 		nicvf_rss_set_key(nic, rss_conf->rss_key);
565 
566 	nic_rss = nicvf_rss_ethdev_to_nic(nic, rss_conf->rss_hf);
567 	nicvf_rss_set_cfg(nic, nic_rss);
568 	return 0;
569 }
570 
571 static int
572 nicvf_qset_cq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
573 		    struct nicvf_rxq *rxq, uint16_t qidx, uint32_t desc_cnt)
574 {
575 	const struct rte_memzone *rz;
576 	uint32_t ring_size = CMP_QUEUE_SZ_MAX * sizeof(union cq_entry_t);
577 
578 	rz = rte_eth_dma_zone_reserve(dev, "cq_ring",
579 				      nicvf_netdev_qidx(nic, qidx), ring_size,
580 				      NICVF_CQ_BASE_ALIGN_BYTES, nic->node);
581 	if (rz == NULL) {
582 		PMD_INIT_LOG(ERR, "Failed to allocate mem for cq hw ring");
583 		return -ENOMEM;
584 	}
585 
586 	memset(rz->addr, 0, ring_size);
587 
588 	rxq->phys = rz->iova;
589 	rxq->desc = rz->addr;
590 	rxq->qlen_mask = desc_cnt - 1;
591 
592 	return 0;
593 }
594 
595 static int
596 nicvf_qset_sq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
597 		    struct nicvf_txq *sq, uint16_t qidx, uint32_t desc_cnt)
598 {
599 	const struct rte_memzone *rz;
600 	uint32_t ring_size = SND_QUEUE_SZ_MAX * sizeof(union sq_entry_t);
601 
602 	rz = rte_eth_dma_zone_reserve(dev, "sq",
603 				      nicvf_netdev_qidx(nic, qidx), ring_size,
604 				      NICVF_SQ_BASE_ALIGN_BYTES, nic->node);
605 	if (rz == NULL) {
606 		PMD_INIT_LOG(ERR, "Failed allocate mem for sq hw ring");
607 		return -ENOMEM;
608 	}
609 
610 	memset(rz->addr, 0, ring_size);
611 
612 	sq->phys = rz->iova;
613 	sq->desc = rz->addr;
614 	sq->qlen_mask = desc_cnt - 1;
615 
616 	return 0;
617 }
618 
619 static int
620 nicvf_qset_rbdr_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
621 		      uint32_t desc_cnt, uint32_t buffsz)
622 {
623 	struct nicvf_rbdr *rbdr;
624 	const struct rte_memzone *rz;
625 	uint32_t ring_size;
626 
627 	assert(nic->rbdr == NULL);
628 	rbdr = rte_zmalloc_socket("rbdr", sizeof(struct nicvf_rbdr),
629 				  RTE_CACHE_LINE_SIZE, nic->node);
630 	if (rbdr == NULL) {
631 		PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr");
632 		return -ENOMEM;
633 	}
634 
635 	ring_size = sizeof(struct rbdr_entry_t) * RBDR_QUEUE_SZ_MAX;
636 	rz = rte_eth_dma_zone_reserve(dev, "rbdr",
637 				      nicvf_netdev_qidx(nic, 0), ring_size,
638 				      NICVF_RBDR_BASE_ALIGN_BYTES, nic->node);
639 	if (rz == NULL) {
640 		PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr desc ring");
641 		rte_free(rbdr);
642 		return -ENOMEM;
643 	}
644 
645 	memset(rz->addr, 0, ring_size);
646 
647 	rbdr->phys = rz->iova;
648 	rbdr->tail = 0;
649 	rbdr->next_tail = 0;
650 	rbdr->desc = rz->addr;
651 	rbdr->buffsz = buffsz;
652 	rbdr->qlen_mask = desc_cnt - 1;
653 	rbdr->rbdr_status =
654 		nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_STATUS0;
655 	rbdr->rbdr_door =
656 		nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_DOOR;
657 
658 	nic->rbdr = rbdr;
659 	return 0;
660 }
661 
662 static void
663 nicvf_rbdr_release_mbuf(struct rte_eth_dev *dev, struct nicvf *nic,
664 			nicvf_iova_addr_t phy)
665 {
666 	uint16_t qidx;
667 	void *obj;
668 	struct nicvf_rxq *rxq;
669 	uint16_t rx_start, rx_end;
670 
671 	/* Get queue ranges for this VF */
672 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
673 
674 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
675 		rxq = dev->data->rx_queues[qidx];
676 		if (rxq->precharge_cnt) {
677 			obj = (void *)nicvf_mbuff_phy2virt(phy,
678 							   rxq->mbuf_phys_off);
679 			rte_mempool_put(rxq->pool, obj);
680 			rxq->precharge_cnt--;
681 			break;
682 		}
683 	}
684 }
685 
686 static inline void
687 nicvf_rbdr_release_mbufs(struct rte_eth_dev *dev, struct nicvf *nic)
688 {
689 	uint32_t qlen_mask, head;
690 	struct rbdr_entry_t *entry;
691 	struct nicvf_rbdr *rbdr = nic->rbdr;
692 
693 	qlen_mask = rbdr->qlen_mask;
694 	head = rbdr->head;
695 	while (head != rbdr->tail) {
696 		entry = rbdr->desc + head;
697 		nicvf_rbdr_release_mbuf(dev, nic, entry->full_addr);
698 		head++;
699 		head = head & qlen_mask;
700 	}
701 }
702 
703 static inline void
704 nicvf_tx_queue_release_mbufs(struct nicvf_txq *txq)
705 {
706 	uint32_t head;
707 
708 	head = txq->head;
709 	while (head != txq->tail) {
710 		if (txq->txbuffs[head]) {
711 			rte_pktmbuf_free_seg(txq->txbuffs[head]);
712 			txq->txbuffs[head] = NULL;
713 		}
714 		head++;
715 		head = head & txq->qlen_mask;
716 	}
717 }
718 
719 static void
720 nicvf_tx_queue_reset(struct nicvf_txq *txq)
721 {
722 	uint32_t txq_desc_cnt = txq->qlen_mask + 1;
723 
724 	memset(txq->desc, 0, sizeof(union sq_entry_t) * txq_desc_cnt);
725 	memset(txq->txbuffs, 0, sizeof(struct rte_mbuf *) * txq_desc_cnt);
726 	txq->tail = 0;
727 	txq->head = 0;
728 	txq->xmit_bufs = 0;
729 }
730 
731 static inline int
732 nicvf_vf_start_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
733 			uint16_t qidx)
734 {
735 	struct nicvf_txq *txq;
736 	int ret;
737 
738 	assert(qidx < MAX_SND_QUEUES_PER_QS);
739 
740 	if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
741 		RTE_ETH_QUEUE_STATE_STARTED)
742 		return 0;
743 
744 	txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
745 	txq->pool = NULL;
746 	ret = nicvf_qset_sq_config(nic, qidx, txq);
747 	if (ret) {
748 		PMD_INIT_LOG(ERR, "Failed to configure sq VF%d %d %d",
749 			     nic->vf_id, qidx, ret);
750 		goto config_sq_error;
751 	}
752 
753 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
754 		RTE_ETH_QUEUE_STATE_STARTED;
755 	return ret;
756 
757 config_sq_error:
758 	nicvf_qset_sq_reclaim(nic, qidx);
759 	return ret;
760 }
761 
762 static inline int
763 nicvf_vf_stop_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
764 		       uint16_t qidx)
765 {
766 	struct nicvf_txq *txq;
767 	int ret;
768 
769 	assert(qidx < MAX_SND_QUEUES_PER_QS);
770 
771 	if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
772 		RTE_ETH_QUEUE_STATE_STOPPED)
773 		return 0;
774 
775 	ret = nicvf_qset_sq_reclaim(nic, qidx);
776 	if (ret)
777 		PMD_INIT_LOG(ERR, "Failed to reclaim sq VF%d %d %d",
778 			     nic->vf_id, qidx, ret);
779 
780 	txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
781 	nicvf_tx_queue_release_mbufs(txq);
782 	nicvf_tx_queue_reset(txq);
783 
784 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
785 		RTE_ETH_QUEUE_STATE_STOPPED;
786 	return ret;
787 }
788 
789 static inline int
790 nicvf_configure_cpi(struct rte_eth_dev *dev)
791 {
792 	struct nicvf *nic = nicvf_pmd_priv(dev);
793 	uint16_t qidx, qcnt;
794 	int ret;
795 
796 	/* Count started rx queues */
797 	for (qidx = qcnt = 0; qidx < dev->data->nb_rx_queues; qidx++)
798 		if (dev->data->rx_queue_state[qidx] ==
799 		    RTE_ETH_QUEUE_STATE_STARTED)
800 			qcnt++;
801 
802 	nic->cpi_alg = CPI_ALG_NONE;
803 	ret = nicvf_mbox_config_cpi(nic, qcnt);
804 	if (ret)
805 		PMD_INIT_LOG(ERR, "Failed to configure CPI %d", ret);
806 
807 	return ret;
808 }
809 
810 static inline int
811 nicvf_configure_rss(struct rte_eth_dev *dev)
812 {
813 	struct nicvf *nic = nicvf_pmd_priv(dev);
814 	uint64_t rsshf;
815 	int ret = -EINVAL;
816 
817 	rsshf = nicvf_rss_ethdev_to_nic(nic,
818 			dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf);
819 	PMD_DRV_LOG(INFO, "mode=%d rx_queues=%d loopback=%d rsshf=0x%" PRIx64,
820 		    dev->data->dev_conf.rxmode.mq_mode,
821 		    dev->data->nb_rx_queues,
822 		    dev->data->dev_conf.lpbk_mode, rsshf);
823 
824 	if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_NONE)
825 		ret = nicvf_rss_term(nic);
826 	else if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS)
827 		ret = nicvf_rss_config(nic, dev->data->nb_rx_queues, rsshf);
828 	if (ret)
829 		PMD_INIT_LOG(ERR, "Failed to configure RSS %d", ret);
830 
831 	return ret;
832 }
833 
834 static int
835 nicvf_configure_rss_reta(struct rte_eth_dev *dev)
836 {
837 	struct nicvf *nic = nicvf_pmd_priv(dev);
838 	unsigned int idx, qmap_size;
839 	uint8_t qmap[RTE_MAX_QUEUES_PER_PORT];
840 	uint8_t default_reta[NIC_MAX_RSS_IDR_TBL_SIZE];
841 
842 	if (nic->cpi_alg != CPI_ALG_NONE)
843 		return -EINVAL;
844 
845 	/* Prepare queue map */
846 	for (idx = 0, qmap_size = 0; idx < dev->data->nb_rx_queues; idx++) {
847 		if (dev->data->rx_queue_state[idx] ==
848 				RTE_ETH_QUEUE_STATE_STARTED)
849 			qmap[qmap_size++] = idx;
850 	}
851 
852 	/* Update default RSS RETA */
853 	for (idx = 0; idx < NIC_MAX_RSS_IDR_TBL_SIZE; idx++)
854 		default_reta[idx] = qmap[idx % qmap_size];
855 
856 	return nicvf_rss_reta_update(nic, default_reta,
857 				     NIC_MAX_RSS_IDR_TBL_SIZE);
858 }
859 
860 static void
861 nicvf_dev_tx_queue_release(void *sq)
862 {
863 	struct nicvf_txq *txq;
864 
865 	PMD_INIT_FUNC_TRACE();
866 
867 	txq = (struct nicvf_txq *)sq;
868 	if (txq) {
869 		if (txq->txbuffs != NULL) {
870 			nicvf_tx_queue_release_mbufs(txq);
871 			rte_free(txq->txbuffs);
872 			txq->txbuffs = NULL;
873 		}
874 		rte_free(txq);
875 	}
876 }
877 
878 static void
879 nicvf_set_tx_function(struct rte_eth_dev *dev)
880 {
881 	struct nicvf_txq *txq = NULL;
882 	size_t i;
883 	bool multiseg = false;
884 
885 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
886 		txq = dev->data->tx_queues[i];
887 		if (txq->offloads & DEV_TX_OFFLOAD_MULTI_SEGS) {
888 			multiseg = true;
889 			break;
890 		}
891 	}
892 
893 	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
894 	if (multiseg) {
895 		PMD_DRV_LOG(DEBUG, "Using multi-segment tx callback");
896 		dev->tx_pkt_burst = nicvf_xmit_pkts_multiseg;
897 	} else {
898 		PMD_DRV_LOG(DEBUG, "Using single-segment tx callback");
899 		dev->tx_pkt_burst = nicvf_xmit_pkts;
900 	}
901 
902 	if (!txq)
903 		return;
904 
905 	if (txq->pool_free == nicvf_single_pool_free_xmited_buffers)
906 		PMD_DRV_LOG(DEBUG, "Using single-mempool tx free method");
907 	else
908 		PMD_DRV_LOG(DEBUG, "Using multi-mempool tx free method");
909 }
910 
911 static void
912 nicvf_set_rx_function(struct rte_eth_dev *dev)
913 {
914 	struct nicvf *nic = nicvf_pmd_priv(dev);
915 
916 	const eth_rx_burst_t rx_burst_func[2][2][2] = {
917 	/* [NORMAL/SCATTER] [CKSUM/NO_CKSUM] [VLAN_STRIP/NO_VLAN_STRIP] */
918 		[0][0][0] = nicvf_recv_pkts_no_offload,
919 		[0][0][1] = nicvf_recv_pkts_vlan_strip,
920 		[0][1][0] = nicvf_recv_pkts_cksum,
921 		[0][1][1] = nicvf_recv_pkts_cksum_vlan_strip,
922 		[1][0][0] = nicvf_recv_pkts_multiseg_no_offload,
923 		[1][0][1] = nicvf_recv_pkts_multiseg_vlan_strip,
924 		[1][1][0] = nicvf_recv_pkts_multiseg_cksum,
925 		[1][1][1] = nicvf_recv_pkts_multiseg_cksum_vlan_strip,
926 	};
927 
928 	dev->rx_pkt_burst =
929 		rx_burst_func[dev->data->scattered_rx]
930 			[nic->offload_cksum][nic->vlan_strip];
931 }
932 
933 static int
934 nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
935 			 uint16_t nb_desc, unsigned int socket_id,
936 			 const struct rte_eth_txconf *tx_conf)
937 {
938 	uint16_t tx_free_thresh;
939 	bool is_single_pool;
940 	struct nicvf_txq *txq;
941 	struct nicvf *nic = nicvf_pmd_priv(dev);
942 	uint64_t offloads;
943 
944 	PMD_INIT_FUNC_TRACE();
945 
946 	if (qidx >= MAX_SND_QUEUES_PER_QS)
947 		nic = nic->snicvf[qidx / MAX_SND_QUEUES_PER_QS - 1];
948 
949 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
950 
951 	/* Socket id check */
952 	if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
953 		PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
954 		socket_id, nic->node);
955 
956 	/* Tx deferred start is not supported */
957 	if (tx_conf->tx_deferred_start) {
958 		PMD_INIT_LOG(ERR, "Tx deferred start not supported");
959 		return -EINVAL;
960 	}
961 
962 	/* Roundup nb_desc to available qsize and validate max number of desc */
963 	nb_desc = nicvf_qsize_sq_roundup(nb_desc);
964 	if (nb_desc == 0) {
965 		PMD_INIT_LOG(ERR, "Value of nb_desc beyond available sq qsize");
966 		return -EINVAL;
967 	}
968 
969 	/* Validate tx_free_thresh */
970 	tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
971 				tx_conf->tx_free_thresh :
972 				NICVF_DEFAULT_TX_FREE_THRESH);
973 
974 	if (tx_free_thresh > (nb_desc) ||
975 		tx_free_thresh > NICVF_MAX_TX_FREE_THRESH) {
976 		PMD_INIT_LOG(ERR,
977 			"tx_free_thresh must be less than the number of TX "
978 			"descriptors. (tx_free_thresh=%u port=%d "
979 			"queue=%d)", (unsigned int)tx_free_thresh,
980 			(int)dev->data->port_id, (int)qidx);
981 		return -EINVAL;
982 	}
983 
984 	/* Free memory prior to re-allocation if needed. */
985 	if (dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
986 		PMD_TX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
987 				nicvf_netdev_qidx(nic, qidx));
988 		nicvf_dev_tx_queue_release(
989 			dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)]);
990 		dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
991 	}
992 
993 	/* Allocating tx queue data structure */
994 	txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct nicvf_txq),
995 					RTE_CACHE_LINE_SIZE, nic->node);
996 	if (txq == NULL) {
997 		PMD_INIT_LOG(ERR, "Failed to allocate txq=%d",
998 			     nicvf_netdev_qidx(nic, qidx));
999 		return -ENOMEM;
1000 	}
1001 
1002 	txq->nic = nic;
1003 	txq->queue_id = qidx;
1004 	txq->tx_free_thresh = tx_free_thresh;
1005 	txq->sq_head = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_HEAD;
1006 	txq->sq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_DOOR;
1007 	offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1008 	txq->offloads = offloads;
1009 
1010 	is_single_pool = !!(offloads & DEV_TX_OFFLOAD_MBUF_FAST_FREE);
1011 
1012 	/* Choose optimum free threshold value for multipool case */
1013 	if (!is_single_pool) {
1014 		txq->tx_free_thresh = (uint16_t)
1015 		(tx_conf->tx_free_thresh == NICVF_DEFAULT_TX_FREE_THRESH ?
1016 				NICVF_TX_FREE_MPOOL_THRESH :
1017 				tx_conf->tx_free_thresh);
1018 		txq->pool_free = nicvf_multi_pool_free_xmited_buffers;
1019 	} else {
1020 		txq->pool_free = nicvf_single_pool_free_xmited_buffers;
1021 	}
1022 
1023 	/* Allocate software ring */
1024 	txq->txbuffs = rte_zmalloc_socket("txq->txbuffs",
1025 				nb_desc * sizeof(struct rte_mbuf *),
1026 				RTE_CACHE_LINE_SIZE, nic->node);
1027 
1028 	if (txq->txbuffs == NULL) {
1029 		nicvf_dev_tx_queue_release(txq);
1030 		return -ENOMEM;
1031 	}
1032 
1033 	if (nicvf_qset_sq_alloc(dev, nic, txq, qidx, nb_desc)) {
1034 		PMD_INIT_LOG(ERR, "Failed to allocate mem for sq %d", qidx);
1035 		nicvf_dev_tx_queue_release(txq);
1036 		return -ENOMEM;
1037 	}
1038 
1039 	nicvf_tx_queue_reset(txq);
1040 
1041 	PMD_INIT_LOG(DEBUG, "[%d] txq=%p nb_desc=%d desc=%p"
1042 			" phys=0x%" PRIx64 " offloads=0x%" PRIx64,
1043 			nicvf_netdev_qidx(nic, qidx), txq, nb_desc, txq->desc,
1044 			txq->phys, txq->offloads);
1045 
1046 	dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = txq;
1047 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1048 		RTE_ETH_QUEUE_STATE_STOPPED;
1049 	return 0;
1050 }
1051 
1052 static inline void
1053 nicvf_rx_queue_release_mbufs(struct rte_eth_dev *dev, struct nicvf_rxq *rxq)
1054 {
1055 	uint32_t rxq_cnt;
1056 	uint32_t nb_pkts, released_pkts = 0;
1057 	uint32_t refill_cnt = 0;
1058 	struct rte_mbuf *rx_pkts[NICVF_MAX_RX_FREE_THRESH];
1059 
1060 	if (dev->rx_pkt_burst == NULL)
1061 		return;
1062 
1063 	while ((rxq_cnt = nicvf_dev_rx_queue_count(dev,
1064 				nicvf_netdev_qidx(rxq->nic, rxq->queue_id)))) {
1065 		nb_pkts = dev->rx_pkt_burst(rxq, rx_pkts,
1066 					NICVF_MAX_RX_FREE_THRESH);
1067 		PMD_DRV_LOG(INFO, "nb_pkts=%d  rxq_cnt=%d", nb_pkts, rxq_cnt);
1068 		while (nb_pkts) {
1069 			rte_pktmbuf_free_seg(rx_pkts[--nb_pkts]);
1070 			released_pkts++;
1071 		}
1072 	}
1073 
1074 
1075 	refill_cnt += nicvf_dev_rbdr_refill(dev,
1076 			nicvf_netdev_qidx(rxq->nic, rxq->queue_id));
1077 
1078 	PMD_DRV_LOG(INFO, "free_cnt=%d  refill_cnt=%d",
1079 		    released_pkts, refill_cnt);
1080 }
1081 
1082 static void
1083 nicvf_rx_queue_reset(struct nicvf_rxq *rxq)
1084 {
1085 	rxq->head = 0;
1086 	rxq->available_space = 0;
1087 	rxq->recv_buffers = 0;
1088 }
1089 
1090 static inline int
1091 nicvf_vf_start_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
1092 			uint16_t qidx)
1093 {
1094 	struct nicvf_rxq *rxq;
1095 	int ret;
1096 
1097 	assert(qidx < MAX_RCV_QUEUES_PER_QS);
1098 
1099 	if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
1100 		RTE_ETH_QUEUE_STATE_STARTED)
1101 		return 0;
1102 
1103 	/* Update rbdr pointer to all rxq */
1104 	rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
1105 	rxq->shared_rbdr = nic->rbdr;
1106 
1107 	ret = nicvf_qset_rq_config(nic, qidx, rxq);
1108 	if (ret) {
1109 		PMD_INIT_LOG(ERR, "Failed to configure rq VF%d %d %d",
1110 			     nic->vf_id, qidx, ret);
1111 		goto config_rq_error;
1112 	}
1113 	ret = nicvf_qset_cq_config(nic, qidx, rxq);
1114 	if (ret) {
1115 		PMD_INIT_LOG(ERR, "Failed to configure cq VF%d %d %d",
1116 			     nic->vf_id, qidx, ret);
1117 		goto config_cq_error;
1118 	}
1119 
1120 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1121 		RTE_ETH_QUEUE_STATE_STARTED;
1122 	return 0;
1123 
1124 config_cq_error:
1125 	nicvf_qset_cq_reclaim(nic, qidx);
1126 config_rq_error:
1127 	nicvf_qset_rq_reclaim(nic, qidx);
1128 	return ret;
1129 }
1130 
1131 static inline int
1132 nicvf_vf_stop_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
1133 		       uint16_t qidx)
1134 {
1135 	struct nicvf_rxq *rxq;
1136 	int ret, other_error;
1137 
1138 	if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
1139 		RTE_ETH_QUEUE_STATE_STOPPED)
1140 		return 0;
1141 
1142 	ret = nicvf_qset_rq_reclaim(nic, qidx);
1143 	if (ret)
1144 		PMD_INIT_LOG(ERR, "Failed to reclaim rq VF%d %d %d",
1145 			     nic->vf_id, qidx, ret);
1146 
1147 	other_error = ret;
1148 	rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
1149 	nicvf_rx_queue_release_mbufs(dev, rxq);
1150 	nicvf_rx_queue_reset(rxq);
1151 
1152 	ret = nicvf_qset_cq_reclaim(nic, qidx);
1153 	if (ret)
1154 		PMD_INIT_LOG(ERR, "Failed to reclaim cq VF%d %d %d",
1155 			     nic->vf_id, qidx, ret);
1156 
1157 	other_error |= ret;
1158 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1159 		RTE_ETH_QUEUE_STATE_STOPPED;
1160 	return other_error;
1161 }
1162 
1163 static void
1164 nicvf_dev_rx_queue_release(void *rx_queue)
1165 {
1166 	PMD_INIT_FUNC_TRACE();
1167 
1168 	rte_free(rx_queue);
1169 }
1170 
1171 static int
1172 nicvf_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
1173 {
1174 	struct nicvf *nic = nicvf_pmd_priv(dev);
1175 	int ret;
1176 
1177 	if (qidx >= MAX_RCV_QUEUES_PER_QS)
1178 		nic = nic->snicvf[(qidx / MAX_RCV_QUEUES_PER_QS - 1)];
1179 
1180 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1181 
1182 	ret = nicvf_vf_start_rx_queue(dev, nic, qidx);
1183 	if (ret)
1184 		return ret;
1185 
1186 	ret = nicvf_configure_cpi(dev);
1187 	if (ret)
1188 		return ret;
1189 
1190 	return nicvf_configure_rss_reta(dev);
1191 }
1192 
1193 static int
1194 nicvf_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
1195 {
1196 	int ret;
1197 	struct nicvf *nic = nicvf_pmd_priv(dev);
1198 
1199 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1200 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1201 
1202 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1203 
1204 	ret = nicvf_vf_stop_rx_queue(dev, nic, qidx);
1205 	ret |= nicvf_configure_cpi(dev);
1206 	ret |= nicvf_configure_rss_reta(dev);
1207 	return ret;
1208 }
1209 
1210 static int
1211 nicvf_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
1212 {
1213 	struct nicvf *nic = nicvf_pmd_priv(dev);
1214 
1215 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1216 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1217 
1218 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
1219 
1220 	return nicvf_vf_start_tx_queue(dev, nic, qidx);
1221 }
1222 
1223 static int
1224 nicvf_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
1225 {
1226 	struct nicvf *nic = nicvf_pmd_priv(dev);
1227 
1228 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1229 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1230 
1231 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
1232 
1233 	return nicvf_vf_stop_tx_queue(dev, nic, qidx);
1234 }
1235 
1236 static inline void
1237 nicvf_rxq_mbuf_setup(struct nicvf_rxq *rxq)
1238 {
1239 	uintptr_t p;
1240 	struct rte_mbuf mb_def;
1241 	struct nicvf *nic = rxq->nic;
1242 
1243 	RTE_BUILD_BUG_ON(sizeof(union mbuf_initializer) != 8);
1244 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, data_off) % 8 != 0);
1245 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, refcnt) -
1246 				offsetof(struct rte_mbuf, data_off) != 2);
1247 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, nb_segs) -
1248 				offsetof(struct rte_mbuf, data_off) != 4);
1249 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, port) -
1250 				offsetof(struct rte_mbuf, data_off) != 6);
1251 	RTE_BUILD_BUG_ON(offsetof(struct nicvf_rxq, rxq_fastpath_data_end) -
1252 				offsetof(struct nicvf_rxq,
1253 					rxq_fastpath_data_start) > 128);
1254 	mb_def.nb_segs = 1;
1255 	mb_def.data_off = RTE_PKTMBUF_HEADROOM + (nic->skip_bytes);
1256 	mb_def.port = rxq->port_id;
1257 	rte_mbuf_refcnt_set(&mb_def, 1);
1258 
1259 	/* Prevent compiler reordering: rearm_data covers previous fields */
1260 	rte_compiler_barrier();
1261 	p = (uintptr_t)&mb_def.rearm_data;
1262 	rxq->mbuf_initializer.value = *(uint64_t *)p;
1263 }
1264 
1265 static int
1266 nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
1267 			 uint16_t nb_desc, unsigned int socket_id,
1268 			 const struct rte_eth_rxconf *rx_conf,
1269 			 struct rte_mempool *mp)
1270 {
1271 	uint16_t rx_free_thresh;
1272 	struct nicvf_rxq *rxq;
1273 	struct nicvf *nic = nicvf_pmd_priv(dev);
1274 	uint64_t offloads;
1275 	uint32_t buffsz;
1276 	struct rte_pktmbuf_pool_private *mbp_priv;
1277 
1278 	PMD_INIT_FUNC_TRACE();
1279 
1280 	/* First skip check */
1281 	mbp_priv = rte_mempool_get_priv(mp);
1282 	buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
1283 	if (buffsz < (uint32_t)(nic->skip_bytes)) {
1284 		PMD_INIT_LOG(ERR, "First skip is more than configured buffer size");
1285 		return -EINVAL;
1286 	}
1287 
1288 	if (qidx >= MAX_RCV_QUEUES_PER_QS)
1289 		nic = nic->snicvf[qidx / MAX_RCV_QUEUES_PER_QS - 1];
1290 
1291 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1292 
1293 	/* Socket id check */
1294 	if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
1295 		PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
1296 		socket_id, nic->node);
1297 
1298 	/* Mempool memory must be contiguous, so must be one memory segment*/
1299 	if (mp->nb_mem_chunks != 1) {
1300 		PMD_INIT_LOG(ERR, "Non-contiguous mempool, add more huge pages");
1301 		return -EINVAL;
1302 	}
1303 
1304 	/* Mempool memory must be physically contiguous */
1305 	if (mp->flags & MEMPOOL_F_NO_IOVA_CONTIG) {
1306 		PMD_INIT_LOG(ERR, "Mempool memory must be physically contiguous");
1307 		return -EINVAL;
1308 	}
1309 
1310 	/* Rx deferred start is not supported */
1311 	if (rx_conf->rx_deferred_start) {
1312 		PMD_INIT_LOG(ERR, "Rx deferred start not supported");
1313 		return -EINVAL;
1314 	}
1315 
1316 	/* Roundup nb_desc to available qsize and validate max number of desc */
1317 	nb_desc = nicvf_qsize_cq_roundup(nb_desc);
1318 	if (nb_desc == 0) {
1319 		PMD_INIT_LOG(ERR, "Value nb_desc beyond available hw cq qsize");
1320 		return -EINVAL;
1321 	}
1322 
1323 
1324 	/* Check rx_free_thresh upper bound */
1325 	rx_free_thresh = (uint16_t)((rx_conf->rx_free_thresh) ?
1326 				rx_conf->rx_free_thresh :
1327 				NICVF_DEFAULT_RX_FREE_THRESH);
1328 	if (rx_free_thresh > NICVF_MAX_RX_FREE_THRESH ||
1329 		rx_free_thresh >= nb_desc * .75) {
1330 		PMD_INIT_LOG(ERR, "rx_free_thresh greater than expected %d",
1331 				rx_free_thresh);
1332 		return -EINVAL;
1333 	}
1334 
1335 	/* Free memory prior to re-allocation if needed */
1336 	if (dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
1337 		PMD_RX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
1338 				nicvf_netdev_qidx(nic, qidx));
1339 		nicvf_dev_rx_queue_release(
1340 			dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)]);
1341 		dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
1342 	}
1343 
1344 	/* Allocate rxq memory */
1345 	rxq = rte_zmalloc_socket("ethdev rx queue", sizeof(struct nicvf_rxq),
1346 					RTE_CACHE_LINE_SIZE, nic->node);
1347 	if (rxq == NULL) {
1348 		PMD_INIT_LOG(ERR, "Failed to allocate rxq=%d",
1349 			     nicvf_netdev_qidx(nic, qidx));
1350 		return -ENOMEM;
1351 	}
1352 
1353 	rxq->nic = nic;
1354 	rxq->pool = mp;
1355 	rxq->queue_id = qidx;
1356 	rxq->port_id = dev->data->port_id;
1357 	rxq->rx_free_thresh = rx_free_thresh;
1358 	rxq->rx_drop_en = rx_conf->rx_drop_en;
1359 	rxq->cq_status = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_STATUS;
1360 	rxq->cq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_DOOR;
1361 	rxq->precharge_cnt = 0;
1362 
1363 	if (nicvf_hw_cap(nic) & NICVF_CAP_CQE_RX2)
1364 		rxq->rbptr_offset = NICVF_CQE_RX2_RBPTR_WORD;
1365 	else
1366 		rxq->rbptr_offset = NICVF_CQE_RBPTR_WORD;
1367 
1368 	nicvf_rxq_mbuf_setup(rxq);
1369 
1370 	/* Alloc completion queue */
1371 	if (nicvf_qset_cq_alloc(dev, nic, rxq, rxq->queue_id, nb_desc)) {
1372 		PMD_INIT_LOG(ERR, "failed to allocate cq %u", rxq->queue_id);
1373 		nicvf_dev_rx_queue_release(rxq);
1374 		return -ENOMEM;
1375 	}
1376 
1377 	nicvf_rx_queue_reset(rxq);
1378 
1379 	offloads = rx_conf->offloads | dev->data->dev_conf.rxmode.offloads;
1380 	PMD_INIT_LOG(DEBUG, "[%d] rxq=%p pool=%s nb_desc=(%d/%d)"
1381 			" phy=0x%" PRIx64 " offloads=0x%" PRIx64,
1382 			nicvf_netdev_qidx(nic, qidx), rxq, mp->name, nb_desc,
1383 			rte_mempool_avail_count(mp), rxq->phys, offloads);
1384 
1385 	dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = rxq;
1386 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1387 		RTE_ETH_QUEUE_STATE_STOPPED;
1388 	return 0;
1389 }
1390 
1391 static int
1392 nicvf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1393 {
1394 	struct nicvf *nic = nicvf_pmd_priv(dev);
1395 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1396 
1397 	PMD_INIT_FUNC_TRACE();
1398 
1399 	/* Autonegotiation may be disabled */
1400 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
1401 	dev_info->speed_capa |= ETH_LINK_SPEED_10M | ETH_LINK_SPEED_100M |
1402 				 ETH_LINK_SPEED_1G | ETH_LINK_SPEED_10G;
1403 	if (nicvf_hw_version(nic) != PCI_SUB_DEVICE_ID_CN81XX_NICVF)
1404 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
1405 
1406 	dev_info->min_rx_bufsize = RTE_ETHER_MIN_MTU;
1407 	dev_info->max_rx_pktlen = NIC_HW_MAX_MTU + RTE_ETHER_HDR_LEN;
1408 	dev_info->max_rx_queues =
1409 			(uint16_t)MAX_RCV_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
1410 	dev_info->max_tx_queues =
1411 			(uint16_t)MAX_SND_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
1412 	dev_info->max_mac_addrs = 1;
1413 	dev_info->max_vfs = pci_dev->max_vfs;
1414 
1415 	dev_info->rx_offload_capa = NICVF_RX_OFFLOAD_CAPA;
1416 	dev_info->tx_offload_capa = NICVF_TX_OFFLOAD_CAPA;
1417 	dev_info->rx_queue_offload_capa = NICVF_RX_OFFLOAD_CAPA;
1418 	dev_info->tx_queue_offload_capa = NICVF_TX_OFFLOAD_CAPA;
1419 
1420 	dev_info->reta_size = nic->rss_info.rss_size;
1421 	dev_info->hash_key_size = RSS_HASH_KEY_BYTE_SIZE;
1422 	dev_info->flow_type_rss_offloads = NICVF_RSS_OFFLOAD_PASS1;
1423 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING)
1424 		dev_info->flow_type_rss_offloads |= NICVF_RSS_OFFLOAD_TUNNEL;
1425 
1426 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
1427 		.rx_free_thresh = NICVF_DEFAULT_RX_FREE_THRESH,
1428 		.rx_drop_en = 0,
1429 	};
1430 
1431 	dev_info->default_txconf = (struct rte_eth_txconf) {
1432 		.tx_free_thresh = NICVF_DEFAULT_TX_FREE_THRESH,
1433 		.offloads = DEV_TX_OFFLOAD_MBUF_FAST_FREE |
1434 			DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM   |
1435 			DEV_TX_OFFLOAD_UDP_CKSUM          |
1436 			DEV_TX_OFFLOAD_TCP_CKSUM,
1437 	};
1438 
1439 	return 0;
1440 }
1441 
1442 static nicvf_iova_addr_t
1443 rbdr_rte_mempool_get(void *dev, void *opaque)
1444 {
1445 	uint16_t qidx;
1446 	uintptr_t mbuf;
1447 	struct nicvf_rxq *rxq;
1448 	struct rte_eth_dev *eth_dev = (struct rte_eth_dev *)dev;
1449 	struct nicvf *nic = (struct nicvf *)opaque;
1450 	uint16_t rx_start, rx_end;
1451 
1452 	/* Get queue ranges for this VF */
1453 	nicvf_rx_range(eth_dev, nic, &rx_start, &rx_end);
1454 
1455 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1456 		rxq = eth_dev->data->rx_queues[qidx];
1457 		/* Maintain equal buffer count across all pools */
1458 		if (rxq->precharge_cnt >= rxq->qlen_mask)
1459 			continue;
1460 		rxq->precharge_cnt++;
1461 		mbuf = (uintptr_t)rte_pktmbuf_alloc(rxq->pool);
1462 		if (mbuf)
1463 			return nicvf_mbuff_virt2phy(mbuf, rxq->mbuf_phys_off);
1464 	}
1465 	return 0;
1466 }
1467 
1468 static int
1469 nicvf_vf_start(struct rte_eth_dev *dev, struct nicvf *nic, uint32_t rbdrsz)
1470 {
1471 	int ret;
1472 	uint16_t qidx, data_off;
1473 	uint32_t total_rxq_desc, nb_rbdr_desc, exp_buffs;
1474 	uint64_t mbuf_phys_off = 0;
1475 	struct nicvf_rxq *rxq;
1476 	struct rte_mbuf *mbuf;
1477 	uint16_t rx_start, rx_end;
1478 	uint16_t tx_start, tx_end;
1479 	int mask;
1480 
1481 	PMD_INIT_FUNC_TRACE();
1482 
1483 	/* Userspace process exited without proper shutdown in last run */
1484 	if (nicvf_qset_rbdr_active(nic, 0))
1485 		nicvf_vf_stop(dev, nic, false);
1486 
1487 	/* Get queue ranges for this VF */
1488 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
1489 
1490 	/*
1491 	 * Thunderx nicvf PMD can support more than one pool per port only when
1492 	 * 1) Data payload size is same across all the pools in given port
1493 	 * AND
1494 	 * 2) All mbuffs in the pools are from the same hugepage
1495 	 * AND
1496 	 * 3) Mbuff metadata size is same across all the pools in given port
1497 	 *
1498 	 * This is to support existing application that uses multiple pool/port.
1499 	 * But, the purpose of using multipool for QoS will not be addressed.
1500 	 *
1501 	 */
1502 
1503 	/* Validate mempool attributes */
1504 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1505 		rxq = dev->data->rx_queues[qidx];
1506 		rxq->mbuf_phys_off = nicvf_mempool_phy_offset(rxq->pool);
1507 		mbuf = rte_pktmbuf_alloc(rxq->pool);
1508 		if (mbuf == NULL) {
1509 			PMD_INIT_LOG(ERR, "Failed allocate mbuf VF%d qid=%d "
1510 				     "pool=%s",
1511 				     nic->vf_id, qidx, rxq->pool->name);
1512 			return -ENOMEM;
1513 		}
1514 		data_off = nicvf_mbuff_meta_length(mbuf);
1515 		data_off += RTE_PKTMBUF_HEADROOM;
1516 		rte_pktmbuf_free(mbuf);
1517 
1518 		if (data_off % RTE_CACHE_LINE_SIZE) {
1519 			PMD_INIT_LOG(ERR, "%s: unaligned data_off=%d delta=%d",
1520 				rxq->pool->name, data_off,
1521 				data_off % RTE_CACHE_LINE_SIZE);
1522 			return -EINVAL;
1523 		}
1524 		rxq->mbuf_phys_off -= data_off;
1525 		rxq->mbuf_phys_off -= nic->skip_bytes;
1526 
1527 		if (mbuf_phys_off == 0)
1528 			mbuf_phys_off = rxq->mbuf_phys_off;
1529 		if (mbuf_phys_off != rxq->mbuf_phys_off) {
1530 			PMD_INIT_LOG(ERR, "pool params not same,%s VF%d %"
1531 				     PRIx64, rxq->pool->name, nic->vf_id,
1532 				     mbuf_phys_off);
1533 			return -EINVAL;
1534 		}
1535 	}
1536 
1537 	/* Check the level of buffers in the pool */
1538 	total_rxq_desc = 0;
1539 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1540 		rxq = dev->data->rx_queues[qidx];
1541 		/* Count total numbers of rxq descs */
1542 		total_rxq_desc += rxq->qlen_mask + 1;
1543 		exp_buffs = RTE_MEMPOOL_CACHE_MAX_SIZE + rxq->rx_free_thresh;
1544 		exp_buffs *= dev->data->nb_rx_queues;
1545 		if (rte_mempool_avail_count(rxq->pool) < exp_buffs) {
1546 			PMD_INIT_LOG(ERR, "Buff shortage in pool=%s (%d/%d)",
1547 				     rxq->pool->name,
1548 				     rte_mempool_avail_count(rxq->pool),
1549 				     exp_buffs);
1550 			return -ENOENT;
1551 		}
1552 	}
1553 
1554 	/* Check RBDR desc overflow */
1555 	ret = nicvf_qsize_rbdr_roundup(total_rxq_desc);
1556 	if (ret == 0) {
1557 		PMD_INIT_LOG(ERR, "Reached RBDR desc limit, reduce nr desc "
1558 			     "VF%d", nic->vf_id);
1559 		return -ENOMEM;
1560 	}
1561 
1562 	/* Enable qset */
1563 	ret = nicvf_qset_config(nic);
1564 	if (ret) {
1565 		PMD_INIT_LOG(ERR, "Failed to enable qset %d VF%d", ret,
1566 			     nic->vf_id);
1567 		return ret;
1568 	}
1569 
1570 	/* Allocate RBDR and RBDR ring desc */
1571 	nb_rbdr_desc = nicvf_qsize_rbdr_roundup(total_rxq_desc);
1572 	ret = nicvf_qset_rbdr_alloc(dev, nic, nb_rbdr_desc, rbdrsz);
1573 	if (ret) {
1574 		PMD_INIT_LOG(ERR, "Failed to allocate memory for rbdr alloc "
1575 			     "VF%d", nic->vf_id);
1576 		goto qset_reclaim;
1577 	}
1578 
1579 	/* Enable and configure RBDR registers */
1580 	ret = nicvf_qset_rbdr_config(nic, 0);
1581 	if (ret) {
1582 		PMD_INIT_LOG(ERR, "Failed to configure rbdr %d VF%d", ret,
1583 			     nic->vf_id);
1584 		goto qset_rbdr_free;
1585 	}
1586 
1587 	/* Fill rte_mempool buffers in RBDR pool and precharge it */
1588 	ret = nicvf_qset_rbdr_precharge(dev, nic, 0, rbdr_rte_mempool_get,
1589 					total_rxq_desc);
1590 	if (ret) {
1591 		PMD_INIT_LOG(ERR, "Failed to fill rbdr %d VF%d", ret,
1592 			     nic->vf_id);
1593 		goto qset_rbdr_reclaim;
1594 	}
1595 
1596 	PMD_DRV_LOG(INFO, "Filled %d out of %d entries in RBDR VF%d",
1597 		     nic->rbdr->tail, nb_rbdr_desc, nic->vf_id);
1598 
1599 	/* Configure VLAN Strip */
1600 	mask = ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK |
1601 		ETH_VLAN_EXTEND_MASK;
1602 	ret = nicvf_vlan_offload_config(dev, mask);
1603 
1604 	/* Based on the packet type(IPv4 or IPv6), the nicvf HW aligns L3 data
1605 	 * to the 64bit memory address.
1606 	 * The alignment creates a hole in mbuf(between the end of headroom and
1607 	 * packet data start). The new revision of the HW provides an option to
1608 	 * disable the L3 alignment feature and make mbuf layout looks
1609 	 * more like other NICs. For better application compatibility, disabling
1610 	 * l3 alignment feature on the hardware revisions it supports
1611 	 */
1612 	nicvf_apad_config(nic, false);
1613 
1614 	/* Get queue ranges for this VF */
1615 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
1616 
1617 	/* Configure TX queues */
1618 	for (qidx = tx_start; qidx <= tx_end; qidx++) {
1619 		ret = nicvf_vf_start_tx_queue(dev, nic,
1620 			qidx % MAX_SND_QUEUES_PER_QS);
1621 		if (ret)
1622 			goto start_txq_error;
1623 	}
1624 
1625 	/* Configure RX queues */
1626 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1627 		ret = nicvf_vf_start_rx_queue(dev, nic,
1628 			qidx % MAX_RCV_QUEUES_PER_QS);
1629 		if (ret)
1630 			goto start_rxq_error;
1631 	}
1632 
1633 	if (!nic->sqs_mode) {
1634 		/* Configure CPI algorithm */
1635 		ret = nicvf_configure_cpi(dev);
1636 		if (ret)
1637 			goto start_txq_error;
1638 
1639 		ret = nicvf_mbox_get_rss_size(nic);
1640 		if (ret) {
1641 			PMD_INIT_LOG(ERR, "Failed to get rss table size");
1642 			goto qset_rss_error;
1643 		}
1644 
1645 		/* Configure RSS */
1646 		ret = nicvf_configure_rss(dev);
1647 		if (ret)
1648 			goto qset_rss_error;
1649 	}
1650 
1651 	/* Done; Let PF make the BGX's RX and TX switches to ON position */
1652 	nicvf_mbox_cfg_done(nic);
1653 	return 0;
1654 
1655 qset_rss_error:
1656 	nicvf_rss_term(nic);
1657 start_rxq_error:
1658 	for (qidx = rx_start; qidx <= rx_end; qidx++)
1659 		nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
1660 start_txq_error:
1661 	for (qidx = tx_start; qidx <= tx_end; qidx++)
1662 		nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
1663 qset_rbdr_reclaim:
1664 	nicvf_qset_rbdr_reclaim(nic, 0);
1665 	nicvf_rbdr_release_mbufs(dev, nic);
1666 qset_rbdr_free:
1667 	if (nic->rbdr) {
1668 		rte_free(nic->rbdr);
1669 		nic->rbdr = NULL;
1670 	}
1671 qset_reclaim:
1672 	nicvf_qset_reclaim(nic);
1673 	return ret;
1674 }
1675 
1676 static int
1677 nicvf_dev_start(struct rte_eth_dev *dev)
1678 {
1679 	uint16_t qidx;
1680 	int ret;
1681 	size_t i;
1682 	struct nicvf *nic = nicvf_pmd_priv(dev);
1683 	struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
1684 	uint16_t mtu;
1685 	uint32_t buffsz = 0, rbdrsz = 0;
1686 	struct rte_pktmbuf_pool_private *mbp_priv;
1687 	struct nicvf_rxq *rxq;
1688 
1689 	PMD_INIT_FUNC_TRACE();
1690 
1691 	/* This function must be called for a primary device */
1692 	assert_primary(nic);
1693 
1694 	/* Validate RBDR buff size */
1695 	for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
1696 		rxq = dev->data->rx_queues[qidx];
1697 		mbp_priv = rte_mempool_get_priv(rxq->pool);
1698 		buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
1699 		if (buffsz % 128) {
1700 			PMD_INIT_LOG(ERR, "rxbuf size must be multiply of 128");
1701 			return -EINVAL;
1702 		}
1703 		if (rbdrsz == 0)
1704 			rbdrsz = buffsz;
1705 		if (rbdrsz != buffsz) {
1706 			PMD_INIT_LOG(ERR, "buffsz not same, qidx=%d (%d/%d)",
1707 				     qidx, rbdrsz, buffsz);
1708 			return -EINVAL;
1709 		}
1710 	}
1711 
1712 	/* Configure loopback */
1713 	ret = nicvf_loopback_config(nic, dev->data->dev_conf.lpbk_mode);
1714 	if (ret) {
1715 		PMD_INIT_LOG(ERR, "Failed to configure loopback %d", ret);
1716 		return ret;
1717 	}
1718 
1719 	/* Reset all statistics counters attached to this port */
1720 	ret = nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, 0xFFFF, 0xFFFF);
1721 	if (ret) {
1722 		PMD_INIT_LOG(ERR, "Failed to reset stat counters %d", ret);
1723 		return ret;
1724 	}
1725 
1726 	/* Setup scatter mode if needed by jumbo */
1727 	if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
1728 					    2 * VLAN_TAG_SIZE > buffsz)
1729 		dev->data->scattered_rx = 1;
1730 	if ((rx_conf->offloads & DEV_RX_OFFLOAD_SCATTER) != 0)
1731 		dev->data->scattered_rx = 1;
1732 
1733 	/* Setup MTU based on max_rx_pkt_len or default */
1734 	mtu = dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME ?
1735 		dev->data->dev_conf.rxmode.max_rx_pkt_len
1736 			-  RTE_ETHER_HDR_LEN : RTE_ETHER_MTU;
1737 
1738 	if (nicvf_dev_set_mtu(dev, mtu)) {
1739 		PMD_INIT_LOG(ERR, "Failed to set default mtu size");
1740 		return -EBUSY;
1741 	}
1742 
1743 	ret = nicvf_vf_start(dev, nic, rbdrsz);
1744 	if (ret != 0)
1745 		return ret;
1746 
1747 	for (i = 0; i < nic->sqs_count; i++) {
1748 		assert(nic->snicvf[i]);
1749 
1750 		ret = nicvf_vf_start(dev, nic->snicvf[i], rbdrsz);
1751 		if (ret != 0)
1752 			return ret;
1753 	}
1754 
1755 	/* Configure callbacks based on offloads */
1756 	nicvf_set_tx_function(dev);
1757 	nicvf_set_rx_function(dev);
1758 
1759 	return 0;
1760 }
1761 
1762 static void
1763 nicvf_dev_stop_cleanup(struct rte_eth_dev *dev, bool cleanup)
1764 {
1765 	size_t i;
1766 	int ret;
1767 	struct nicvf *nic = nicvf_pmd_priv(dev);
1768 
1769 	PMD_INIT_FUNC_TRACE();
1770 	dev->data->dev_started = 0;
1771 
1772 	/* Teardown secondary vf first */
1773 	for (i = 0; i < nic->sqs_count; i++) {
1774 		if (!nic->snicvf[i])
1775 			continue;
1776 
1777 		nicvf_vf_stop(dev, nic->snicvf[i], cleanup);
1778 	}
1779 
1780 	/* Stop the primary VF now */
1781 	nicvf_vf_stop(dev, nic, cleanup);
1782 
1783 	/* Disable loopback */
1784 	ret = nicvf_loopback_config(nic, 0);
1785 	if (ret)
1786 		PMD_INIT_LOG(ERR, "Failed to disable loopback %d", ret);
1787 
1788 	/* Reclaim CPI configuration */
1789 	ret = nicvf_mbox_config_cpi(nic, 0);
1790 	if (ret)
1791 		PMD_INIT_LOG(ERR, "Failed to reclaim CPI config %d", ret);
1792 }
1793 
1794 static int
1795 nicvf_dev_stop(struct rte_eth_dev *dev)
1796 {
1797 	PMD_INIT_FUNC_TRACE();
1798 
1799 	nicvf_dev_stop_cleanup(dev, false);
1800 
1801 	return 0;
1802 }
1803 
1804 static void
1805 nicvf_vf_stop(struct rte_eth_dev *dev, struct nicvf *nic, bool cleanup)
1806 {
1807 	int ret;
1808 	uint16_t qidx;
1809 	uint16_t tx_start, tx_end;
1810 	uint16_t rx_start, rx_end;
1811 
1812 	PMD_INIT_FUNC_TRACE();
1813 
1814 	if (cleanup) {
1815 		/* Let PF make the BGX's RX and TX switches to OFF position */
1816 		nicvf_mbox_shutdown(nic);
1817 	}
1818 
1819 	/* Disable VLAN Strip */
1820 	nicvf_vlan_hw_strip(nic, 0);
1821 
1822 	/* Get queue ranges for this VF */
1823 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
1824 
1825 	for (qidx = tx_start; qidx <= tx_end; qidx++)
1826 		nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
1827 
1828 	/* Get queue ranges for this VF */
1829 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
1830 
1831 	/* Reclaim rq */
1832 	for (qidx = rx_start; qidx <= rx_end; qidx++)
1833 		nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
1834 
1835 	/* Reclaim RBDR */
1836 	ret = nicvf_qset_rbdr_reclaim(nic, 0);
1837 	if (ret)
1838 		PMD_INIT_LOG(ERR, "Failed to reclaim RBDR %d", ret);
1839 
1840 	/* Move all charged buffers in RBDR back to pool */
1841 	if (nic->rbdr != NULL)
1842 		nicvf_rbdr_release_mbufs(dev, nic);
1843 
1844 	/* Disable qset */
1845 	ret = nicvf_qset_reclaim(nic);
1846 	if (ret)
1847 		PMD_INIT_LOG(ERR, "Failed to disable qset %d", ret);
1848 
1849 	/* Disable all interrupts */
1850 	nicvf_disable_all_interrupts(nic);
1851 
1852 	/* Free RBDR SW structure */
1853 	if (nic->rbdr) {
1854 		rte_free(nic->rbdr);
1855 		nic->rbdr = NULL;
1856 	}
1857 }
1858 
1859 static int
1860 nicvf_dev_close(struct rte_eth_dev *dev)
1861 {
1862 	size_t i;
1863 	struct nicvf *nic = nicvf_pmd_priv(dev);
1864 
1865 	PMD_INIT_FUNC_TRACE();
1866 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1867 		return 0;
1868 
1869 	nicvf_dev_stop_cleanup(dev, true);
1870 	nicvf_periodic_alarm_stop(nicvf_interrupt, dev);
1871 
1872 	for (i = 0; i < nic->sqs_count; i++) {
1873 		if (!nic->snicvf[i])
1874 			continue;
1875 
1876 		nicvf_periodic_alarm_stop(nicvf_vf_interrupt, nic->snicvf[i]);
1877 	}
1878 
1879 	return 0;
1880 }
1881 
1882 static int
1883 nicvf_request_sqs(struct nicvf *nic)
1884 {
1885 	size_t i;
1886 
1887 	assert_primary(nic);
1888 	assert(nic->sqs_count > 0);
1889 	assert(nic->sqs_count <= MAX_SQS_PER_VF);
1890 
1891 	/* Set no of Rx/Tx queues in each of the SQsets */
1892 	for (i = 0; i < nic->sqs_count; i++) {
1893 		if (nicvf_svf_empty())
1894 			rte_panic("Cannot assign sufficient number of "
1895 				  "secondary queues to primary VF%" PRIu8 "\n",
1896 				  nic->vf_id);
1897 
1898 		nic->snicvf[i] = nicvf_svf_pop();
1899 		nic->snicvf[i]->sqs_id = i;
1900 	}
1901 
1902 	return nicvf_mbox_request_sqs(nic);
1903 }
1904 
1905 static int
1906 nicvf_dev_configure(struct rte_eth_dev *dev)
1907 {
1908 	struct rte_eth_dev_data *data = dev->data;
1909 	struct rte_eth_conf *conf = &data->dev_conf;
1910 	struct rte_eth_rxmode *rxmode = &conf->rxmode;
1911 	struct rte_eth_txmode *txmode = &conf->txmode;
1912 	struct nicvf *nic = nicvf_pmd_priv(dev);
1913 	uint8_t cqcount;
1914 
1915 	PMD_INIT_FUNC_TRACE();
1916 
1917 	if (rxmode->mq_mode & ETH_MQ_RX_RSS_FLAG)
1918 		rxmode->offloads |= DEV_RX_OFFLOAD_RSS_HASH;
1919 
1920 	if (!rte_eal_has_hugepages()) {
1921 		PMD_INIT_LOG(INFO, "Huge page is not configured");
1922 		return -EINVAL;
1923 	}
1924 
1925 	if (txmode->mq_mode) {
1926 		PMD_INIT_LOG(INFO, "Tx mq_mode DCB or VMDq not supported");
1927 		return -EINVAL;
1928 	}
1929 
1930 	if (rxmode->mq_mode != ETH_MQ_RX_NONE &&
1931 		rxmode->mq_mode != ETH_MQ_RX_RSS) {
1932 		PMD_INIT_LOG(INFO, "Unsupported rx qmode %d", rxmode->mq_mode);
1933 		return -EINVAL;
1934 	}
1935 
1936 	if (rxmode->split_hdr_size) {
1937 		PMD_INIT_LOG(INFO, "Rxmode does not support split header");
1938 		return -EINVAL;
1939 	}
1940 
1941 	if (conf->link_speeds & ETH_LINK_SPEED_FIXED) {
1942 		PMD_INIT_LOG(INFO, "Setting link speed/duplex not supported");
1943 		return -EINVAL;
1944 	}
1945 
1946 	if (conf->dcb_capability_en) {
1947 		PMD_INIT_LOG(INFO, "DCB enable not supported");
1948 		return -EINVAL;
1949 	}
1950 
1951 	if (conf->fdir_conf.mode != RTE_FDIR_MODE_NONE) {
1952 		PMD_INIT_LOG(INFO, "Flow director not supported");
1953 		return -EINVAL;
1954 	}
1955 
1956 	assert_primary(nic);
1957 	NICVF_STATIC_ASSERT(MAX_RCV_QUEUES_PER_QS == MAX_SND_QUEUES_PER_QS);
1958 	cqcount = RTE_MAX(data->nb_tx_queues, data->nb_rx_queues);
1959 	if (cqcount > MAX_RCV_QUEUES_PER_QS) {
1960 		nic->sqs_count = RTE_ALIGN_CEIL(cqcount, MAX_RCV_QUEUES_PER_QS);
1961 		nic->sqs_count = (nic->sqs_count / MAX_RCV_QUEUES_PER_QS) - 1;
1962 	} else {
1963 		nic->sqs_count = 0;
1964 	}
1965 
1966 	assert(nic->sqs_count <= MAX_SQS_PER_VF);
1967 
1968 	if (nic->sqs_count > 0) {
1969 		if (nicvf_request_sqs(nic)) {
1970 			rte_panic("Cannot assign sufficient number of "
1971 				  "secondary queues to PORT%d VF%" PRIu8 "\n",
1972 				  dev->data->port_id, nic->vf_id);
1973 		}
1974 	}
1975 
1976 	if (rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
1977 		nic->offload_cksum = 1;
1978 
1979 	PMD_INIT_LOG(DEBUG, "Configured ethdev port%d hwcap=0x%" PRIx64,
1980 		dev->data->port_id, nicvf_hw_cap(nic));
1981 
1982 	return 0;
1983 }
1984 
1985 static int
1986 nicvf_dev_set_link_up(struct rte_eth_dev *dev)
1987 {
1988 	struct nicvf *nic = nicvf_pmd_priv(dev);
1989 	int rc, i;
1990 
1991 	rc = nicvf_mbox_set_link_up_down(nic, true);
1992 	if (rc)
1993 		goto done;
1994 
1995 	/* Start tx queues  */
1996 	for (i = 0; i < dev->data->nb_tx_queues; i++)
1997 		nicvf_dev_tx_queue_start(dev, i);
1998 
1999 done:
2000 	return rc;
2001 }
2002 
2003 static int
2004 nicvf_dev_set_link_down(struct rte_eth_dev *dev)
2005 {
2006 	struct nicvf *nic = nicvf_pmd_priv(dev);
2007 	int i;
2008 
2009 	/* Stop tx queues  */
2010 	for (i = 0; i < dev->data->nb_tx_queues; i++)
2011 		nicvf_dev_tx_queue_stop(dev, i);
2012 
2013 	return nicvf_mbox_set_link_up_down(nic, false);
2014 }
2015 
2016 /* Initialize and register driver with DPDK Application */
2017 static const struct eth_dev_ops nicvf_eth_dev_ops = {
2018 	.dev_configure            = nicvf_dev_configure,
2019 	.dev_start                = nicvf_dev_start,
2020 	.dev_stop                 = nicvf_dev_stop,
2021 	.link_update              = nicvf_dev_link_update,
2022 	.dev_close                = nicvf_dev_close,
2023 	.stats_get                = nicvf_dev_stats_get,
2024 	.stats_reset              = nicvf_dev_stats_reset,
2025 	.promiscuous_enable       = nicvf_dev_promisc_enable,
2026 	.dev_infos_get            = nicvf_dev_info_get,
2027 	.dev_supported_ptypes_get = nicvf_dev_supported_ptypes_get,
2028 	.mtu_set                  = nicvf_dev_set_mtu,
2029 	.vlan_offload_set         = nicvf_vlan_offload_set,
2030 	.reta_update              = nicvf_dev_reta_update,
2031 	.reta_query               = nicvf_dev_reta_query,
2032 	.rss_hash_update          = nicvf_dev_rss_hash_update,
2033 	.rss_hash_conf_get        = nicvf_dev_rss_hash_conf_get,
2034 	.rx_queue_start           = nicvf_dev_rx_queue_start,
2035 	.rx_queue_stop            = nicvf_dev_rx_queue_stop,
2036 	.tx_queue_start           = nicvf_dev_tx_queue_start,
2037 	.tx_queue_stop            = nicvf_dev_tx_queue_stop,
2038 	.rx_queue_setup           = nicvf_dev_rx_queue_setup,
2039 	.rx_queue_release         = nicvf_dev_rx_queue_release,
2040 	.tx_queue_setup           = nicvf_dev_tx_queue_setup,
2041 	.tx_queue_release         = nicvf_dev_tx_queue_release,
2042 	.dev_set_link_up          = nicvf_dev_set_link_up,
2043 	.dev_set_link_down        = nicvf_dev_set_link_down,
2044 	.get_reg                  = nicvf_dev_get_regs,
2045 };
2046 
2047 static int
2048 nicvf_vlan_offload_config(struct rte_eth_dev *dev, int mask)
2049 {
2050 	struct rte_eth_rxmode *rxmode;
2051 	struct nicvf *nic = nicvf_pmd_priv(dev);
2052 	rxmode = &dev->data->dev_conf.rxmode;
2053 	if (mask & ETH_VLAN_STRIP_MASK) {
2054 		if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2055 			nicvf_vlan_hw_strip(nic, true);
2056 		else
2057 			nicvf_vlan_hw_strip(nic, false);
2058 	}
2059 
2060 	return 0;
2061 }
2062 
2063 static int
2064 nicvf_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2065 {
2066 	nicvf_vlan_offload_config(dev, mask);
2067 
2068 	return 0;
2069 }
2070 
2071 static inline int
2072 nicvf_set_first_skip(struct rte_eth_dev *dev)
2073 {
2074 	int bytes_to_skip = 0;
2075 	int ret = 0;
2076 	unsigned int i;
2077 	struct rte_kvargs *kvlist;
2078 	static const char *const skip[] = {
2079 		SKIP_DATA_BYTES,
2080 		NULL};
2081 	struct nicvf *nic = nicvf_pmd_priv(dev);
2082 
2083 	if (!dev->device->devargs) {
2084 		nicvf_first_skip_config(nic, 0);
2085 		return ret;
2086 	}
2087 
2088 	kvlist = rte_kvargs_parse(dev->device->devargs->args, skip);
2089 	if (!kvlist)
2090 		return -EINVAL;
2091 
2092 	if (kvlist->count == 0)
2093 		goto exit;
2094 
2095 	for (i = 0; i != kvlist->count; ++i) {
2096 		const struct rte_kvargs_pair *pair = &kvlist->pairs[i];
2097 
2098 		if (!strcmp(pair->key, SKIP_DATA_BYTES))
2099 			bytes_to_skip = atoi(pair->value);
2100 	}
2101 
2102 	/*128 bytes amounts to one cache line*/
2103 	if (bytes_to_skip >= 0 && bytes_to_skip < 128) {
2104 		if (!(bytes_to_skip % 8)) {
2105 			nicvf_first_skip_config(nic, (bytes_to_skip / 8));
2106 			nic->skip_bytes = bytes_to_skip;
2107 			goto kvlist_free;
2108 		} else {
2109 			PMD_INIT_LOG(ERR, "skip_data_bytes should be multiple of 8");
2110 			ret = -EINVAL;
2111 			goto exit;
2112 		}
2113 	} else {
2114 		PMD_INIT_LOG(ERR, "skip_data_bytes should be less than 128");
2115 		ret = -EINVAL;
2116 		goto exit;
2117 	}
2118 exit:
2119 	nicvf_first_skip_config(nic, 0);
2120 kvlist_free:
2121 	rte_kvargs_free(kvlist);
2122 	return ret;
2123 }
2124 static int
2125 nicvf_eth_dev_uninit(struct rte_eth_dev *dev)
2126 {
2127 	PMD_INIT_FUNC_TRACE();
2128 	nicvf_dev_close(dev);
2129 	return 0;
2130 }
2131 static int
2132 nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
2133 {
2134 	int ret;
2135 	struct rte_pci_device *pci_dev;
2136 	struct nicvf *nic = nicvf_pmd_priv(eth_dev);
2137 
2138 	PMD_INIT_FUNC_TRACE();
2139 
2140 	eth_dev->dev_ops = &nicvf_eth_dev_ops;
2141 	eth_dev->rx_queue_count = nicvf_dev_rx_queue_count;
2142 
2143 	/* For secondary processes, the primary has done all the work */
2144 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2145 		if (nic) {
2146 			/* Setup callbacks for secondary process */
2147 			nicvf_set_tx_function(eth_dev);
2148 			nicvf_set_rx_function(eth_dev);
2149 			return 0;
2150 		} else {
2151 			/* If nic == NULL than it is secondary function
2152 			 * so ethdev need to be released by caller */
2153 			return ENOTSUP;
2154 		}
2155 	}
2156 
2157 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
2158 	rte_eth_copy_pci_info(eth_dev, pci_dev);
2159 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2160 
2161 	nic->device_id = pci_dev->id.device_id;
2162 	nic->vendor_id = pci_dev->id.vendor_id;
2163 	nic->subsystem_device_id = pci_dev->id.subsystem_device_id;
2164 	nic->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
2165 
2166 	PMD_INIT_LOG(DEBUG, "nicvf: device (%x:%x) %u:%u:%u:%u",
2167 			pci_dev->id.vendor_id, pci_dev->id.device_id,
2168 			pci_dev->addr.domain, pci_dev->addr.bus,
2169 			pci_dev->addr.devid, pci_dev->addr.function);
2170 
2171 	nic->reg_base = (uintptr_t)pci_dev->mem_resource[0].addr;
2172 	if (!nic->reg_base) {
2173 		PMD_INIT_LOG(ERR, "Failed to map BAR0");
2174 		ret = -ENODEV;
2175 		goto fail;
2176 	}
2177 
2178 	nicvf_disable_all_interrupts(nic);
2179 
2180 	ret = nicvf_periodic_alarm_start(nicvf_interrupt, eth_dev);
2181 	if (ret) {
2182 		PMD_INIT_LOG(ERR, "Failed to start period alarm");
2183 		goto fail;
2184 	}
2185 
2186 	ret = nicvf_mbox_check_pf_ready(nic);
2187 	if (ret) {
2188 		PMD_INIT_LOG(ERR, "Failed to get ready message from PF");
2189 		goto alarm_fail;
2190 	} else {
2191 		PMD_INIT_LOG(INFO,
2192 			"node=%d vf=%d mode=%s sqs=%s loopback_supported=%s",
2193 			nic->node, nic->vf_id,
2194 			nic->tns_mode == NIC_TNS_MODE ? "tns" : "tns-bypass",
2195 			nic->sqs_mode ? "true" : "false",
2196 			nic->loopback_supported ? "true" : "false"
2197 			);
2198 	}
2199 
2200 	ret = nicvf_base_init(nic);
2201 	if (ret) {
2202 		PMD_INIT_LOG(ERR, "Failed to execute nicvf_base_init");
2203 		goto malloc_fail;
2204 	}
2205 
2206 	if (nic->sqs_mode) {
2207 		/* Push nic to stack of secondary vfs */
2208 		nicvf_svf_push(nic);
2209 
2210 		/* Steal nic pointer from the device for further reuse */
2211 		eth_dev->data->dev_private = NULL;
2212 
2213 		nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
2214 		ret = nicvf_periodic_alarm_start(nicvf_vf_interrupt, nic);
2215 		if (ret) {
2216 			PMD_INIT_LOG(ERR, "Failed to start period alarm");
2217 			goto fail;
2218 		}
2219 
2220 		/* Detach port by returning positive error number */
2221 		return ENOTSUP;
2222 	}
2223 
2224 	eth_dev->data->mac_addrs = rte_zmalloc("mac_addr",
2225 					RTE_ETHER_ADDR_LEN, 0);
2226 	if (eth_dev->data->mac_addrs == NULL) {
2227 		PMD_INIT_LOG(ERR, "Failed to allocate memory for mac addr");
2228 		ret = -ENOMEM;
2229 		goto alarm_fail;
2230 	}
2231 	if (rte_is_zero_ether_addr((struct rte_ether_addr *)nic->mac_addr))
2232 		rte_eth_random_addr(&nic->mac_addr[0]);
2233 
2234 	rte_ether_addr_copy((struct rte_ether_addr *)nic->mac_addr,
2235 			&eth_dev->data->mac_addrs[0]);
2236 
2237 	ret = nicvf_mbox_set_mac_addr(nic, nic->mac_addr);
2238 	if (ret) {
2239 		PMD_INIT_LOG(ERR, "Failed to set mac addr");
2240 		goto malloc_fail;
2241 	}
2242 
2243 	ret = nicvf_set_first_skip(eth_dev);
2244 	if (ret) {
2245 		PMD_INIT_LOG(ERR, "Failed to configure first skip");
2246 		goto malloc_fail;
2247 	}
2248 	PMD_INIT_LOG(INFO, "Port %d (%x:%x) mac=%02x:%02x:%02x:%02x:%02x:%02x",
2249 		eth_dev->data->port_id, nic->vendor_id, nic->device_id,
2250 		nic->mac_addr[0], nic->mac_addr[1], nic->mac_addr[2],
2251 		nic->mac_addr[3], nic->mac_addr[4], nic->mac_addr[5]);
2252 
2253 	return 0;
2254 
2255 malloc_fail:
2256 	rte_free(eth_dev->data->mac_addrs);
2257 	eth_dev->data->mac_addrs = NULL;
2258 alarm_fail:
2259 	nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
2260 fail:
2261 	return ret;
2262 }
2263 
2264 static const struct rte_pci_id pci_id_nicvf_map[] = {
2265 	{
2266 		.class_id = RTE_CLASS_ANY_ID,
2267 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2268 		.device_id = PCI_DEVICE_ID_THUNDERX_CN88XX_PASS1_NICVF,
2269 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2270 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS1_NICVF,
2271 	},
2272 	{
2273 		.class_id = RTE_CLASS_ANY_ID,
2274 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2275 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2276 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2277 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS2_NICVF,
2278 	},
2279 	{
2280 		.class_id = RTE_CLASS_ANY_ID,
2281 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2282 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2283 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2284 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN81XX_NICVF,
2285 	},
2286 	{
2287 		.class_id = RTE_CLASS_ANY_ID,
2288 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2289 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2290 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2291 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN83XX_NICVF,
2292 	},
2293 	{
2294 		.vendor_id = 0,
2295 	},
2296 };
2297 
2298 static int nicvf_eth_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2299 	struct rte_pci_device *pci_dev)
2300 {
2301 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct nicvf),
2302 		nicvf_eth_dev_init);
2303 }
2304 
2305 static int nicvf_eth_pci_remove(struct rte_pci_device *pci_dev)
2306 {
2307 	return rte_eth_dev_pci_generic_remove(pci_dev, nicvf_eth_dev_uninit);
2308 }
2309 
2310 static struct rte_pci_driver rte_nicvf_pmd = {
2311 	.id_table = pci_id_nicvf_map,
2312 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_KEEP_MAPPED_RES |
2313 			RTE_PCI_DRV_INTR_LSC,
2314 	.probe = nicvf_eth_pci_probe,
2315 	.remove = nicvf_eth_pci_remove,
2316 };
2317 
2318 RTE_PMD_REGISTER_PCI(net_thunderx, rte_nicvf_pmd);
2319 RTE_PMD_REGISTER_PCI_TABLE(net_thunderx, pci_id_nicvf_map);
2320 RTE_PMD_REGISTER_KMOD_DEP(net_thunderx, "* igb_uio | uio_pci_generic | vfio-pci");
2321 RTE_PMD_REGISTER_PARAM_STRING(net_thunderx, SKIP_DATA_BYTES "=<int>");
2322