xref: /dpdk/drivers/net/thunderx/nicvf_ethdev.c (revision 250c9eb3ca895127f21a729caf4a928eb2f04d2c)
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 static void
58 nicvf_init_log(void)
59 {
60 	nicvf_logtype_mbox = rte_log_register("pmd.net.thunderx.mbox");
61 	if (nicvf_logtype_mbox >= 0)
62 		rte_log_set_level(nicvf_logtype_mbox, RTE_LOG_NOTICE);
63 
64 	nicvf_logtype_init = rte_log_register("pmd.net.thunderx.init");
65 	if (nicvf_logtype_init >= 0)
66 		rte_log_set_level(nicvf_logtype_init, RTE_LOG_NOTICE);
67 
68 	nicvf_logtype_driver = rte_log_register("pmd.net.thunderx.driver");
69 	if (nicvf_logtype_driver >= 0)
70 		rte_log_set_level(nicvf_logtype_driver, RTE_LOG_NOTICE);
71 }
72 
73 static void
74 nicvf_link_status_update(struct nicvf *nic,
75 			 struct rte_eth_link *link)
76 {
77 	memset(link, 0, sizeof(*link));
78 
79 	link->link_status = nic->link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
80 
81 	if (nic->duplex == NICVF_HALF_DUPLEX)
82 		link->link_duplex = ETH_LINK_HALF_DUPLEX;
83 	else if (nic->duplex == NICVF_FULL_DUPLEX)
84 		link->link_duplex = ETH_LINK_FULL_DUPLEX;
85 	link->link_speed = nic->speed;
86 	link->link_autoneg = ETH_LINK_AUTONEG;
87 }
88 
89 static void
90 nicvf_interrupt(void *arg)
91 {
92 	struct rte_eth_dev *dev = arg;
93 	struct nicvf *nic = nicvf_pmd_priv(dev);
94 	struct rte_eth_link link;
95 
96 	if (nicvf_reg_poll_interrupts(nic) == NIC_MBOX_MSG_BGX_LINK_CHANGE) {
97 		if (dev->data->dev_conf.intr_conf.lsc) {
98 			nicvf_link_status_update(nic, &link);
99 			rte_eth_linkstatus_set(dev, &link);
100 
101 			_rte_eth_dev_callback_process(dev,
102 						      RTE_ETH_EVENT_INTR_LSC,
103 						      NULL);
104 		}
105 	}
106 
107 	rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
108 				nicvf_interrupt, dev);
109 }
110 
111 static void
112 nicvf_vf_interrupt(void *arg)
113 {
114 	struct nicvf *nic = arg;
115 
116 	nicvf_reg_poll_interrupts(nic);
117 
118 	rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
119 				nicvf_vf_interrupt, nic);
120 }
121 
122 static int
123 nicvf_periodic_alarm_start(void (fn)(void *), void *arg)
124 {
125 	return rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000, fn, arg);
126 }
127 
128 static int
129 nicvf_periodic_alarm_stop(void (fn)(void *), void *arg)
130 {
131 	return rte_eal_alarm_cancel(fn, arg);
132 }
133 
134 /*
135  * Return 0 means link status changed, -1 means not changed
136  */
137 static int
138 nicvf_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
139 {
140 #define CHECK_INTERVAL 100  /* 100ms */
141 #define MAX_CHECK_TIME 90   /* 9s (90 * 100ms) in total */
142 	struct rte_eth_link link;
143 	struct nicvf *nic = nicvf_pmd_priv(dev);
144 	int i;
145 
146 	PMD_INIT_FUNC_TRACE();
147 
148 	if (wait_to_complete) {
149 		/* rte_eth_link_get() might need to wait up to 9 seconds */
150 		for (i = 0; i < MAX_CHECK_TIME; i++) {
151 			nicvf_link_status_update(nic, &link);
152 			if (link.link_status == ETH_LINK_UP)
153 				break;
154 			rte_delay_ms(CHECK_INTERVAL);
155 		}
156 	} else {
157 		nicvf_link_status_update(nic, &link);
158 	}
159 
160 	return rte_eth_linkstatus_set(dev, &link);
161 }
162 
163 static int
164 nicvf_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
165 {
166 	struct nicvf *nic = nicvf_pmd_priv(dev);
167 	uint32_t buffsz, frame_size = mtu + NIC_HW_L2_OVERHEAD;
168 	size_t i;
169 	struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
170 
171 	PMD_INIT_FUNC_TRACE();
172 
173 	if (frame_size > NIC_HW_MAX_FRS)
174 		return -EINVAL;
175 
176 	if (frame_size < NIC_HW_MIN_FRS)
177 		return -EINVAL;
178 
179 	buffsz = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
180 
181 	/*
182 	 * Refuse mtu that requires the support of scattered packets
183 	 * when this feature has not been enabled before.
184 	 */
185 	if (dev->data->dev_started && !dev->data->scattered_rx &&
186 		(frame_size + 2 * VLAN_TAG_SIZE > buffsz))
187 		return -EINVAL;
188 
189 	/* check <seg size> * <max_seg>  >= max_frame */
190 	if (dev->data->scattered_rx &&
191 		(frame_size + 2 * VLAN_TAG_SIZE > buffsz * NIC_HW_MAX_SEGS))
192 		return -EINVAL;
193 
194 	if (frame_size > ETHER_MAX_LEN)
195 		rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
196 	else
197 		rxmode->offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
198 
199 	if (nicvf_mbox_update_hw_max_frs(nic, mtu))
200 		return -EINVAL;
201 
202 	/* Update max_rx_pkt_len */
203 	rxmode->max_rx_pkt_len = mtu + ETHER_HDR_LEN;
204 	nic->mtu = mtu;
205 
206 	for (i = 0; i < nic->sqs_count; i++)
207 		nic->snicvf[i]->mtu = mtu;
208 
209 	return 0;
210 }
211 
212 static int
213 nicvf_dev_get_regs(struct rte_eth_dev *dev, struct rte_dev_reg_info *regs)
214 {
215 	uint64_t *data = regs->data;
216 	struct nicvf *nic = nicvf_pmd_priv(dev);
217 
218 	if (data == NULL) {
219 		regs->length = nicvf_reg_get_count();
220 		regs->width = THUNDERX_REG_BYTES;
221 		return 0;
222 	}
223 
224 	/* Support only full register dump */
225 	if ((regs->length == 0) ||
226 		(regs->length == (uint32_t)nicvf_reg_get_count())) {
227 		regs->version = nic->vendor_id << 16 | nic->device_id;
228 		nicvf_reg_dump(nic, data);
229 		return 0;
230 	}
231 	return -ENOTSUP;
232 }
233 
234 static int
235 nicvf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
236 {
237 	uint16_t qidx;
238 	struct nicvf_hw_rx_qstats rx_qstats;
239 	struct nicvf_hw_tx_qstats tx_qstats;
240 	struct nicvf_hw_stats port_stats;
241 	struct nicvf *nic = nicvf_pmd_priv(dev);
242 	uint16_t rx_start, rx_end;
243 	uint16_t tx_start, tx_end;
244 	size_t i;
245 
246 	/* RX queue indices for the first VF */
247 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
248 
249 	/* Reading per RX ring stats */
250 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
251 		if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
252 			break;
253 
254 		nicvf_hw_get_rx_qstats(nic, &rx_qstats, qidx);
255 		stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
256 		stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
257 	}
258 
259 	/* TX queue indices for the first VF */
260 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
261 
262 	/* Reading per TX ring stats */
263 	for (qidx = tx_start; qidx <= tx_end; qidx++) {
264 		if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
265 			break;
266 
267 		nicvf_hw_get_tx_qstats(nic, &tx_qstats, qidx);
268 		stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
269 		stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
270 	}
271 
272 	for (i = 0; i < nic->sqs_count; i++) {
273 		struct nicvf *snic = nic->snicvf[i];
274 
275 		if (snic == NULL)
276 			break;
277 
278 		/* RX queue indices for a secondary VF */
279 		nicvf_rx_range(dev, snic, &rx_start, &rx_end);
280 
281 		/* Reading per RX ring stats */
282 		for (qidx = rx_start; qidx <= rx_end; qidx++) {
283 			if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
284 				break;
285 
286 			nicvf_hw_get_rx_qstats(snic, &rx_qstats,
287 					       qidx % MAX_RCV_QUEUES_PER_QS);
288 			stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
289 			stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
290 		}
291 
292 		/* TX queue indices for a secondary VF */
293 		nicvf_tx_range(dev, snic, &tx_start, &tx_end);
294 		/* Reading per TX ring stats */
295 		for (qidx = tx_start; qidx <= tx_end; qidx++) {
296 			if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
297 				break;
298 
299 			nicvf_hw_get_tx_qstats(snic, &tx_qstats,
300 					       qidx % MAX_SND_QUEUES_PER_QS);
301 			stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
302 			stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
303 		}
304 	}
305 
306 	nicvf_hw_get_stats(nic, &port_stats);
307 	stats->ibytes = port_stats.rx_bytes;
308 	stats->ipackets = port_stats.rx_ucast_frames;
309 	stats->ipackets += port_stats.rx_bcast_frames;
310 	stats->ipackets += port_stats.rx_mcast_frames;
311 	stats->ierrors = port_stats.rx_l2_errors;
312 	stats->imissed = port_stats.rx_drop_red;
313 	stats->imissed += port_stats.rx_drop_overrun;
314 	stats->imissed += port_stats.rx_drop_bcast;
315 	stats->imissed += port_stats.rx_drop_mcast;
316 	stats->imissed += port_stats.rx_drop_l3_bcast;
317 	stats->imissed += port_stats.rx_drop_l3_mcast;
318 
319 	stats->obytes = port_stats.tx_bytes_ok;
320 	stats->opackets = port_stats.tx_ucast_frames_ok;
321 	stats->opackets += port_stats.tx_bcast_frames_ok;
322 	stats->opackets += port_stats.tx_mcast_frames_ok;
323 	stats->oerrors = port_stats.tx_drops;
324 
325 	return 0;
326 }
327 
328 static const uint32_t *
329 nicvf_dev_supported_ptypes_get(struct rte_eth_dev *dev)
330 {
331 	size_t copied;
332 	static uint32_t ptypes[32];
333 	struct nicvf *nic = nicvf_pmd_priv(dev);
334 	static const uint32_t ptypes_common[] = {
335 		RTE_PTYPE_L3_IPV4,
336 		RTE_PTYPE_L3_IPV4_EXT,
337 		RTE_PTYPE_L3_IPV6,
338 		RTE_PTYPE_L3_IPV6_EXT,
339 		RTE_PTYPE_L4_TCP,
340 		RTE_PTYPE_L4_UDP,
341 		RTE_PTYPE_L4_FRAG,
342 	};
343 	static const uint32_t ptypes_tunnel[] = {
344 		RTE_PTYPE_TUNNEL_GRE,
345 		RTE_PTYPE_TUNNEL_GENEVE,
346 		RTE_PTYPE_TUNNEL_VXLAN,
347 		RTE_PTYPE_TUNNEL_NVGRE,
348 	};
349 	static const uint32_t ptypes_end = RTE_PTYPE_UNKNOWN;
350 
351 	copied = sizeof(ptypes_common);
352 	memcpy(ptypes, ptypes_common, copied);
353 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
354 		memcpy((char *)ptypes + copied, ptypes_tunnel,
355 			sizeof(ptypes_tunnel));
356 		copied += sizeof(ptypes_tunnel);
357 	}
358 
359 	memcpy((char *)ptypes + copied, &ptypes_end, sizeof(ptypes_end));
360 	if (dev->rx_pkt_burst == nicvf_recv_pkts ||
361 		dev->rx_pkt_burst == nicvf_recv_pkts_multiseg)
362 		return ptypes;
363 
364 	return NULL;
365 }
366 
367 static void
368 nicvf_dev_stats_reset(struct rte_eth_dev *dev)
369 {
370 	int i;
371 	uint16_t rxqs = 0, txqs = 0;
372 	struct nicvf *nic = nicvf_pmd_priv(dev);
373 	uint16_t rx_start, rx_end;
374 	uint16_t tx_start, tx_end;
375 
376 	/* Reset all primary nic counters */
377 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
378 	for (i = rx_start; i <= rx_end; i++)
379 		rxqs |= (0x3 << (i * 2));
380 
381 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
382 	for (i = tx_start; i <= tx_end; i++)
383 		txqs |= (0x3 << (i * 2));
384 
385 	nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, rxqs, txqs);
386 
387 	/* Reset secondary nic queue counters */
388 	for (i = 0; i < nic->sqs_count; i++) {
389 		struct nicvf *snic = nic->snicvf[i];
390 		if (snic == NULL)
391 			break;
392 
393 		nicvf_rx_range(dev, snic, &rx_start, &rx_end);
394 		for (i = rx_start; i <= rx_end; i++)
395 			rxqs |= (0x3 << ((i % MAX_CMP_QUEUES_PER_QS) * 2));
396 
397 		nicvf_tx_range(dev, snic, &tx_start, &tx_end);
398 		for (i = tx_start; i <= tx_end; i++)
399 			txqs |= (0x3 << ((i % MAX_SND_QUEUES_PER_QS) * 2));
400 
401 		nicvf_mbox_reset_stat_counters(snic, 0, 0, rxqs, txqs);
402 	}
403 }
404 
405 /* Promiscuous mode enabled by default in LMAC to VF 1:1 map configuration */
406 static void
407 nicvf_dev_promisc_enable(struct rte_eth_dev *dev __rte_unused)
408 {
409 }
410 
411 static inline uint64_t
412 nicvf_rss_ethdev_to_nic(struct nicvf *nic, uint64_t ethdev_rss)
413 {
414 	uint64_t nic_rss = 0;
415 
416 	if (ethdev_rss & ETH_RSS_IPV4)
417 		nic_rss |= RSS_IP_ENA;
418 
419 	if (ethdev_rss & ETH_RSS_IPV6)
420 		nic_rss |= RSS_IP_ENA;
421 
422 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_UDP)
423 		nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
424 
425 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_TCP)
426 		nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
427 
428 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_UDP)
429 		nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
430 
431 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_TCP)
432 		nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
433 
434 	if (ethdev_rss & ETH_RSS_PORT)
435 		nic_rss |= RSS_L2_EXTENDED_HASH_ENA;
436 
437 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
438 		if (ethdev_rss & ETH_RSS_VXLAN)
439 			nic_rss |= RSS_TUN_VXLAN_ENA;
440 
441 		if (ethdev_rss & ETH_RSS_GENEVE)
442 			nic_rss |= RSS_TUN_GENEVE_ENA;
443 
444 		if (ethdev_rss & ETH_RSS_NVGRE)
445 			nic_rss |= RSS_TUN_NVGRE_ENA;
446 	}
447 
448 	return nic_rss;
449 }
450 
451 static inline uint64_t
452 nicvf_rss_nic_to_ethdev(struct nicvf *nic,  uint64_t nic_rss)
453 {
454 	uint64_t ethdev_rss = 0;
455 
456 	if (nic_rss & RSS_IP_ENA)
457 		ethdev_rss |= (ETH_RSS_IPV4 | ETH_RSS_IPV6);
458 
459 	if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_TCP_ENA))
460 		ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_TCP |
461 				ETH_RSS_NONFRAG_IPV6_TCP);
462 
463 	if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_UDP_ENA))
464 		ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_UDP |
465 				ETH_RSS_NONFRAG_IPV6_UDP);
466 
467 	if (nic_rss & RSS_L2_EXTENDED_HASH_ENA)
468 		ethdev_rss |= ETH_RSS_PORT;
469 
470 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
471 		if (nic_rss & RSS_TUN_VXLAN_ENA)
472 			ethdev_rss |= ETH_RSS_VXLAN;
473 
474 		if (nic_rss & RSS_TUN_GENEVE_ENA)
475 			ethdev_rss |= ETH_RSS_GENEVE;
476 
477 		if (nic_rss & RSS_TUN_NVGRE_ENA)
478 			ethdev_rss |= ETH_RSS_NVGRE;
479 	}
480 	return ethdev_rss;
481 }
482 
483 static int
484 nicvf_dev_reta_query(struct rte_eth_dev *dev,
485 		     struct rte_eth_rss_reta_entry64 *reta_conf,
486 		     uint16_t reta_size)
487 {
488 	struct nicvf *nic = nicvf_pmd_priv(dev);
489 	uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
490 	int ret, i, j;
491 
492 	if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
493 		RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
494 			"(%d) doesn't match the number hardware can supported "
495 			"(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
496 		return -EINVAL;
497 	}
498 
499 	ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
500 	if (ret)
501 		return ret;
502 
503 	/* Copy RETA table */
504 	for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
505 		for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
506 			if ((reta_conf[i].mask >> j) & 0x01)
507 				reta_conf[i].reta[j] = tbl[j];
508 	}
509 
510 	return 0;
511 }
512 
513 static int
514 nicvf_dev_reta_update(struct rte_eth_dev *dev,
515 		      struct rte_eth_rss_reta_entry64 *reta_conf,
516 		      uint16_t reta_size)
517 {
518 	struct nicvf *nic = nicvf_pmd_priv(dev);
519 	uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
520 	int ret, i, j;
521 
522 	if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
523 		RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
524 			"(%d) doesn't match the number hardware can supported "
525 			"(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
526 		return -EINVAL;
527 	}
528 
529 	ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
530 	if (ret)
531 		return ret;
532 
533 	/* Copy RETA table */
534 	for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
535 		for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
536 			if ((reta_conf[i].mask >> j) & 0x01)
537 				tbl[j] = reta_conf[i].reta[j];
538 	}
539 
540 	return nicvf_rss_reta_update(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
541 }
542 
543 static int
544 nicvf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
545 			    struct rte_eth_rss_conf *rss_conf)
546 {
547 	struct nicvf *nic = nicvf_pmd_priv(dev);
548 
549 	if (rss_conf->rss_key)
550 		nicvf_rss_get_key(nic, rss_conf->rss_key);
551 
552 	rss_conf->rss_key_len =  RSS_HASH_KEY_BYTE_SIZE;
553 	rss_conf->rss_hf = nicvf_rss_nic_to_ethdev(nic, nicvf_rss_get_cfg(nic));
554 	return 0;
555 }
556 
557 static int
558 nicvf_dev_rss_hash_update(struct rte_eth_dev *dev,
559 			  struct rte_eth_rss_conf *rss_conf)
560 {
561 	struct nicvf *nic = nicvf_pmd_priv(dev);
562 	uint64_t nic_rss;
563 
564 	if (rss_conf->rss_key &&
565 		rss_conf->rss_key_len != RSS_HASH_KEY_BYTE_SIZE) {
566 		RTE_LOG(ERR, PMD, "Hash key size mismatch %d",
567 				rss_conf->rss_key_len);
568 		return -EINVAL;
569 	}
570 
571 	if (rss_conf->rss_key)
572 		nicvf_rss_set_key(nic, rss_conf->rss_key);
573 
574 	nic_rss = nicvf_rss_ethdev_to_nic(nic, rss_conf->rss_hf);
575 	nicvf_rss_set_cfg(nic, nic_rss);
576 	return 0;
577 }
578 
579 static int
580 nicvf_qset_cq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
581 		    struct nicvf_rxq *rxq, uint16_t qidx, uint32_t desc_cnt)
582 {
583 	const struct rte_memzone *rz;
584 	uint32_t ring_size = CMP_QUEUE_SZ_MAX * sizeof(union cq_entry_t);
585 
586 	rz = rte_eth_dma_zone_reserve(dev, "cq_ring",
587 				      nicvf_netdev_qidx(nic, qidx), ring_size,
588 				      NICVF_CQ_BASE_ALIGN_BYTES, nic->node);
589 	if (rz == NULL) {
590 		PMD_INIT_LOG(ERR, "Failed to allocate mem for cq hw ring");
591 		return -ENOMEM;
592 	}
593 
594 	memset(rz->addr, 0, ring_size);
595 
596 	rxq->phys = rz->iova;
597 	rxq->desc = rz->addr;
598 	rxq->qlen_mask = desc_cnt - 1;
599 
600 	return 0;
601 }
602 
603 static int
604 nicvf_qset_sq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
605 		    struct nicvf_txq *sq, uint16_t qidx, uint32_t desc_cnt)
606 {
607 	const struct rte_memzone *rz;
608 	uint32_t ring_size = SND_QUEUE_SZ_MAX * sizeof(union sq_entry_t);
609 
610 	rz = rte_eth_dma_zone_reserve(dev, "sq",
611 				      nicvf_netdev_qidx(nic, qidx), ring_size,
612 				      NICVF_SQ_BASE_ALIGN_BYTES, nic->node);
613 	if (rz == NULL) {
614 		PMD_INIT_LOG(ERR, "Failed allocate mem for sq hw ring");
615 		return -ENOMEM;
616 	}
617 
618 	memset(rz->addr, 0, ring_size);
619 
620 	sq->phys = rz->iova;
621 	sq->desc = rz->addr;
622 	sq->qlen_mask = desc_cnt - 1;
623 
624 	return 0;
625 }
626 
627 static int
628 nicvf_qset_rbdr_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
629 		      uint32_t desc_cnt, uint32_t buffsz)
630 {
631 	struct nicvf_rbdr *rbdr;
632 	const struct rte_memzone *rz;
633 	uint32_t ring_size;
634 
635 	assert(nic->rbdr == NULL);
636 	rbdr = rte_zmalloc_socket("rbdr", sizeof(struct nicvf_rbdr),
637 				  RTE_CACHE_LINE_SIZE, nic->node);
638 	if (rbdr == NULL) {
639 		PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr");
640 		return -ENOMEM;
641 	}
642 
643 	ring_size = sizeof(struct rbdr_entry_t) * RBDR_QUEUE_SZ_MAX;
644 	rz = rte_eth_dma_zone_reserve(dev, "rbdr",
645 				      nicvf_netdev_qidx(nic, 0), ring_size,
646 				      NICVF_RBDR_BASE_ALIGN_BYTES, nic->node);
647 	if (rz == NULL) {
648 		PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr desc ring");
649 		return -ENOMEM;
650 	}
651 
652 	memset(rz->addr, 0, ring_size);
653 
654 	rbdr->phys = rz->iova;
655 	rbdr->tail = 0;
656 	rbdr->next_tail = 0;
657 	rbdr->desc = rz->addr;
658 	rbdr->buffsz = buffsz;
659 	rbdr->qlen_mask = desc_cnt - 1;
660 	rbdr->rbdr_status =
661 		nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_STATUS0;
662 	rbdr->rbdr_door =
663 		nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_DOOR;
664 
665 	nic->rbdr = rbdr;
666 	return 0;
667 }
668 
669 static void
670 nicvf_rbdr_release_mbuf(struct rte_eth_dev *dev, struct nicvf *nic,
671 			nicvf_iova_addr_t phy)
672 {
673 	uint16_t qidx;
674 	void *obj;
675 	struct nicvf_rxq *rxq;
676 	uint16_t rx_start, rx_end;
677 
678 	/* Get queue ranges for this VF */
679 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
680 
681 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
682 		rxq = dev->data->rx_queues[qidx];
683 		if (rxq->precharge_cnt) {
684 			obj = (void *)nicvf_mbuff_phy2virt(phy,
685 							   rxq->mbuf_phys_off);
686 			rte_mempool_put(rxq->pool, obj);
687 			rxq->precharge_cnt--;
688 			break;
689 		}
690 	}
691 }
692 
693 static inline void
694 nicvf_rbdr_release_mbufs(struct rte_eth_dev *dev, struct nicvf *nic)
695 {
696 	uint32_t qlen_mask, head;
697 	struct rbdr_entry_t *entry;
698 	struct nicvf_rbdr *rbdr = nic->rbdr;
699 
700 	qlen_mask = rbdr->qlen_mask;
701 	head = rbdr->head;
702 	while (head != rbdr->tail) {
703 		entry = rbdr->desc + head;
704 		nicvf_rbdr_release_mbuf(dev, nic, entry->full_addr);
705 		head++;
706 		head = head & qlen_mask;
707 	}
708 }
709 
710 static inline void
711 nicvf_tx_queue_release_mbufs(struct nicvf_txq *txq)
712 {
713 	uint32_t head;
714 
715 	head = txq->head;
716 	while (head != txq->tail) {
717 		if (txq->txbuffs[head]) {
718 			rte_pktmbuf_free_seg(txq->txbuffs[head]);
719 			txq->txbuffs[head] = NULL;
720 		}
721 		head++;
722 		head = head & txq->qlen_mask;
723 	}
724 }
725 
726 static void
727 nicvf_tx_queue_reset(struct nicvf_txq *txq)
728 {
729 	uint32_t txq_desc_cnt = txq->qlen_mask + 1;
730 
731 	memset(txq->desc, 0, sizeof(union sq_entry_t) * txq_desc_cnt);
732 	memset(txq->txbuffs, 0, sizeof(struct rte_mbuf *) * txq_desc_cnt);
733 	txq->tail = 0;
734 	txq->head = 0;
735 	txq->xmit_bufs = 0;
736 }
737 
738 static inline int
739 nicvf_vf_start_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
740 			uint16_t qidx)
741 {
742 	struct nicvf_txq *txq;
743 	int ret;
744 
745 	assert(qidx < MAX_SND_QUEUES_PER_QS);
746 
747 	if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
748 		RTE_ETH_QUEUE_STATE_STARTED)
749 		return 0;
750 
751 	txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
752 	txq->pool = NULL;
753 	ret = nicvf_qset_sq_config(nic, qidx, txq);
754 	if (ret) {
755 		PMD_INIT_LOG(ERR, "Failed to configure sq VF%d %d %d",
756 			     nic->vf_id, qidx, ret);
757 		goto config_sq_error;
758 	}
759 
760 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
761 		RTE_ETH_QUEUE_STATE_STARTED;
762 	return ret;
763 
764 config_sq_error:
765 	nicvf_qset_sq_reclaim(nic, qidx);
766 	return ret;
767 }
768 
769 static inline int
770 nicvf_vf_stop_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
771 		       uint16_t qidx)
772 {
773 	struct nicvf_txq *txq;
774 	int ret;
775 
776 	assert(qidx < MAX_SND_QUEUES_PER_QS);
777 
778 	if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
779 		RTE_ETH_QUEUE_STATE_STOPPED)
780 		return 0;
781 
782 	ret = nicvf_qset_sq_reclaim(nic, qidx);
783 	if (ret)
784 		PMD_INIT_LOG(ERR, "Failed to reclaim sq VF%d %d %d",
785 			     nic->vf_id, qidx, ret);
786 
787 	txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
788 	nicvf_tx_queue_release_mbufs(txq);
789 	nicvf_tx_queue_reset(txq);
790 
791 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
792 		RTE_ETH_QUEUE_STATE_STOPPED;
793 	return ret;
794 }
795 
796 static inline int
797 nicvf_configure_cpi(struct rte_eth_dev *dev)
798 {
799 	struct nicvf *nic = nicvf_pmd_priv(dev);
800 	uint16_t qidx, qcnt;
801 	int ret;
802 
803 	/* Count started rx queues */
804 	for (qidx = qcnt = 0; qidx < dev->data->nb_rx_queues; qidx++)
805 		if (dev->data->rx_queue_state[qidx] ==
806 		    RTE_ETH_QUEUE_STATE_STARTED)
807 			qcnt++;
808 
809 	nic->cpi_alg = CPI_ALG_NONE;
810 	ret = nicvf_mbox_config_cpi(nic, qcnt);
811 	if (ret)
812 		PMD_INIT_LOG(ERR, "Failed to configure CPI %d", ret);
813 
814 	return ret;
815 }
816 
817 static inline int
818 nicvf_configure_rss(struct rte_eth_dev *dev)
819 {
820 	struct nicvf *nic = nicvf_pmd_priv(dev);
821 	uint64_t rsshf;
822 	int ret = -EINVAL;
823 
824 	rsshf = nicvf_rss_ethdev_to_nic(nic,
825 			dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf);
826 	PMD_DRV_LOG(INFO, "mode=%d rx_queues=%d loopback=%d rsshf=0x%" PRIx64,
827 		    dev->data->dev_conf.rxmode.mq_mode,
828 		    dev->data->nb_rx_queues,
829 		    dev->data->dev_conf.lpbk_mode, rsshf);
830 
831 	if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_NONE)
832 		ret = nicvf_rss_term(nic);
833 	else if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS)
834 		ret = nicvf_rss_config(nic, dev->data->nb_rx_queues, rsshf);
835 	if (ret)
836 		PMD_INIT_LOG(ERR, "Failed to configure RSS %d", ret);
837 
838 	return ret;
839 }
840 
841 static int
842 nicvf_configure_rss_reta(struct rte_eth_dev *dev)
843 {
844 	struct nicvf *nic = nicvf_pmd_priv(dev);
845 	unsigned int idx, qmap_size;
846 	uint8_t qmap[RTE_MAX_QUEUES_PER_PORT];
847 	uint8_t default_reta[NIC_MAX_RSS_IDR_TBL_SIZE];
848 
849 	if (nic->cpi_alg != CPI_ALG_NONE)
850 		return -EINVAL;
851 
852 	/* Prepare queue map */
853 	for (idx = 0, qmap_size = 0; idx < dev->data->nb_rx_queues; idx++) {
854 		if (dev->data->rx_queue_state[idx] ==
855 				RTE_ETH_QUEUE_STATE_STARTED)
856 			qmap[qmap_size++] = idx;
857 	}
858 
859 	/* Update default RSS RETA */
860 	for (idx = 0; idx < NIC_MAX_RSS_IDR_TBL_SIZE; idx++)
861 		default_reta[idx] = qmap[idx % qmap_size];
862 
863 	return nicvf_rss_reta_update(nic, default_reta,
864 				     NIC_MAX_RSS_IDR_TBL_SIZE);
865 }
866 
867 static void
868 nicvf_dev_tx_queue_release(void *sq)
869 {
870 	struct nicvf_txq *txq;
871 
872 	PMD_INIT_FUNC_TRACE();
873 
874 	txq = (struct nicvf_txq *)sq;
875 	if (txq) {
876 		if (txq->txbuffs != NULL) {
877 			nicvf_tx_queue_release_mbufs(txq);
878 			rte_free(txq->txbuffs);
879 			txq->txbuffs = NULL;
880 		}
881 		rte_free(txq);
882 	}
883 }
884 
885 static void
886 nicvf_set_tx_function(struct rte_eth_dev *dev)
887 {
888 	struct nicvf_txq *txq = NULL;
889 	size_t i;
890 	bool multiseg = false;
891 
892 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
893 		txq = dev->data->tx_queues[i];
894 		if (txq->offloads & DEV_TX_OFFLOAD_MULTI_SEGS) {
895 			multiseg = true;
896 			break;
897 		}
898 	}
899 
900 	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
901 	if (multiseg) {
902 		PMD_DRV_LOG(DEBUG, "Using multi-segment tx callback");
903 		dev->tx_pkt_burst = nicvf_xmit_pkts_multiseg;
904 	} else {
905 		PMD_DRV_LOG(DEBUG, "Using single-segment tx callback");
906 		dev->tx_pkt_burst = nicvf_xmit_pkts;
907 	}
908 
909 	if (!txq)
910 		return;
911 
912 	if (txq->pool_free == nicvf_single_pool_free_xmited_buffers)
913 		PMD_DRV_LOG(DEBUG, "Using single-mempool tx free method");
914 	else
915 		PMD_DRV_LOG(DEBUG, "Using multi-mempool tx free method");
916 }
917 
918 static void
919 nicvf_set_rx_function(struct rte_eth_dev *dev)
920 {
921 	if (dev->data->scattered_rx) {
922 		PMD_DRV_LOG(DEBUG, "Using multi-segment rx callback");
923 		dev->rx_pkt_burst = nicvf_recv_pkts_multiseg;
924 	} else {
925 		PMD_DRV_LOG(DEBUG, "Using single-segment rx callback");
926 		dev->rx_pkt_burst = nicvf_recv_pkts;
927 	}
928 }
929 
930 static int
931 nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
932 			 uint16_t nb_desc, unsigned int socket_id,
933 			 const struct rte_eth_txconf *tx_conf)
934 {
935 	uint16_t tx_free_thresh;
936 	bool is_single_pool;
937 	struct nicvf_txq *txq;
938 	struct nicvf *nic = nicvf_pmd_priv(dev);
939 	uint64_t offloads;
940 
941 	PMD_INIT_FUNC_TRACE();
942 
943 	if (qidx >= MAX_SND_QUEUES_PER_QS)
944 		nic = nic->snicvf[qidx / MAX_SND_QUEUES_PER_QS - 1];
945 
946 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
947 
948 	/* Socket id check */
949 	if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
950 		PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
951 		socket_id, nic->node);
952 
953 	/* Tx deferred start is not supported */
954 	if (tx_conf->tx_deferred_start) {
955 		PMD_INIT_LOG(ERR, "Tx deferred start not supported");
956 		return -EINVAL;
957 	}
958 
959 	/* Roundup nb_desc to available qsize and validate max number of desc */
960 	nb_desc = nicvf_qsize_sq_roundup(nb_desc);
961 	if (nb_desc == 0) {
962 		PMD_INIT_LOG(ERR, "Value of nb_desc beyond available sq qsize");
963 		return -EINVAL;
964 	}
965 
966 	/* Validate tx_free_thresh */
967 	tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
968 				tx_conf->tx_free_thresh :
969 				NICVF_DEFAULT_TX_FREE_THRESH);
970 
971 	if (tx_free_thresh > (nb_desc) ||
972 		tx_free_thresh > NICVF_MAX_TX_FREE_THRESH) {
973 		PMD_INIT_LOG(ERR,
974 			"tx_free_thresh must be less than the number of TX "
975 			"descriptors. (tx_free_thresh=%u port=%d "
976 			"queue=%d)", (unsigned int)tx_free_thresh,
977 			(int)dev->data->port_id, (int)qidx);
978 		return -EINVAL;
979 	}
980 
981 	/* Free memory prior to re-allocation if needed. */
982 	if (dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
983 		PMD_TX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
984 				nicvf_netdev_qidx(nic, qidx));
985 		nicvf_dev_tx_queue_release(
986 			dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)]);
987 		dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
988 	}
989 
990 	/* Allocating tx queue data structure */
991 	txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct nicvf_txq),
992 					RTE_CACHE_LINE_SIZE, nic->node);
993 	if (txq == NULL) {
994 		PMD_INIT_LOG(ERR, "Failed to allocate txq=%d",
995 			     nicvf_netdev_qidx(nic, qidx));
996 		return -ENOMEM;
997 	}
998 
999 	txq->nic = nic;
1000 	txq->queue_id = qidx;
1001 	txq->tx_free_thresh = tx_free_thresh;
1002 	txq->sq_head = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_HEAD;
1003 	txq->sq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_DOOR;
1004 	offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1005 	txq->offloads = offloads;
1006 
1007 	is_single_pool = !!(offloads & DEV_TX_OFFLOAD_MBUF_FAST_FREE);
1008 
1009 	/* Choose optimum free threshold value for multipool case */
1010 	if (!is_single_pool) {
1011 		txq->tx_free_thresh = (uint16_t)
1012 		(tx_conf->tx_free_thresh == NICVF_DEFAULT_TX_FREE_THRESH ?
1013 				NICVF_TX_FREE_MPOOL_THRESH :
1014 				tx_conf->tx_free_thresh);
1015 		txq->pool_free = nicvf_multi_pool_free_xmited_buffers;
1016 	} else {
1017 		txq->pool_free = nicvf_single_pool_free_xmited_buffers;
1018 	}
1019 
1020 	/* Allocate software ring */
1021 	txq->txbuffs = rte_zmalloc_socket("txq->txbuffs",
1022 				nb_desc * sizeof(struct rte_mbuf *),
1023 				RTE_CACHE_LINE_SIZE, nic->node);
1024 
1025 	if (txq->txbuffs == NULL) {
1026 		nicvf_dev_tx_queue_release(txq);
1027 		return -ENOMEM;
1028 	}
1029 
1030 	if (nicvf_qset_sq_alloc(dev, nic, txq, qidx, nb_desc)) {
1031 		PMD_INIT_LOG(ERR, "Failed to allocate mem for sq %d", qidx);
1032 		nicvf_dev_tx_queue_release(txq);
1033 		return -ENOMEM;
1034 	}
1035 
1036 	nicvf_tx_queue_reset(txq);
1037 
1038 	PMD_INIT_LOG(DEBUG, "[%d] txq=%p nb_desc=%d desc=%p"
1039 			" phys=0x%" PRIx64 " offloads=0x%" PRIx64,
1040 			nicvf_netdev_qidx(nic, qidx), txq, nb_desc, txq->desc,
1041 			txq->phys, txq->offloads);
1042 
1043 	dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = txq;
1044 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1045 		RTE_ETH_QUEUE_STATE_STOPPED;
1046 	return 0;
1047 }
1048 
1049 static inline void
1050 nicvf_rx_queue_release_mbufs(struct rte_eth_dev *dev, struct nicvf_rxq *rxq)
1051 {
1052 	uint32_t rxq_cnt;
1053 	uint32_t nb_pkts, released_pkts = 0;
1054 	uint32_t refill_cnt = 0;
1055 	struct rte_mbuf *rx_pkts[NICVF_MAX_RX_FREE_THRESH];
1056 
1057 	if (dev->rx_pkt_burst == NULL)
1058 		return;
1059 
1060 	while ((rxq_cnt = nicvf_dev_rx_queue_count(dev,
1061 				nicvf_netdev_qidx(rxq->nic, rxq->queue_id)))) {
1062 		nb_pkts = dev->rx_pkt_burst(rxq, rx_pkts,
1063 					NICVF_MAX_RX_FREE_THRESH);
1064 		PMD_DRV_LOG(INFO, "nb_pkts=%d  rxq_cnt=%d", nb_pkts, rxq_cnt);
1065 		while (nb_pkts) {
1066 			rte_pktmbuf_free_seg(rx_pkts[--nb_pkts]);
1067 			released_pkts++;
1068 		}
1069 	}
1070 
1071 
1072 	refill_cnt += nicvf_dev_rbdr_refill(dev,
1073 			nicvf_netdev_qidx(rxq->nic, rxq->queue_id));
1074 
1075 	PMD_DRV_LOG(INFO, "free_cnt=%d  refill_cnt=%d",
1076 		    released_pkts, refill_cnt);
1077 }
1078 
1079 static void
1080 nicvf_rx_queue_reset(struct nicvf_rxq *rxq)
1081 {
1082 	rxq->head = 0;
1083 	rxq->available_space = 0;
1084 	rxq->recv_buffers = 0;
1085 }
1086 
1087 static inline int
1088 nicvf_vf_start_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
1089 			uint16_t qidx)
1090 {
1091 	struct nicvf_rxq *rxq;
1092 	int ret;
1093 
1094 	assert(qidx < MAX_RCV_QUEUES_PER_QS);
1095 
1096 	if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
1097 		RTE_ETH_QUEUE_STATE_STARTED)
1098 		return 0;
1099 
1100 	/* Update rbdr pointer to all rxq */
1101 	rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
1102 	rxq->shared_rbdr = nic->rbdr;
1103 
1104 	ret = nicvf_qset_rq_config(nic, qidx, rxq);
1105 	if (ret) {
1106 		PMD_INIT_LOG(ERR, "Failed to configure rq VF%d %d %d",
1107 			     nic->vf_id, qidx, ret);
1108 		goto config_rq_error;
1109 	}
1110 	ret = nicvf_qset_cq_config(nic, qidx, rxq);
1111 	if (ret) {
1112 		PMD_INIT_LOG(ERR, "Failed to configure cq VF%d %d %d",
1113 			     nic->vf_id, qidx, ret);
1114 		goto config_cq_error;
1115 	}
1116 
1117 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1118 		RTE_ETH_QUEUE_STATE_STARTED;
1119 	return 0;
1120 
1121 config_cq_error:
1122 	nicvf_qset_cq_reclaim(nic, qidx);
1123 config_rq_error:
1124 	nicvf_qset_rq_reclaim(nic, qidx);
1125 	return ret;
1126 }
1127 
1128 static inline int
1129 nicvf_vf_stop_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
1130 		       uint16_t qidx)
1131 {
1132 	struct nicvf_rxq *rxq;
1133 	int ret, other_error;
1134 
1135 	if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
1136 		RTE_ETH_QUEUE_STATE_STOPPED)
1137 		return 0;
1138 
1139 	ret = nicvf_qset_rq_reclaim(nic, qidx);
1140 	if (ret)
1141 		PMD_INIT_LOG(ERR, "Failed to reclaim rq VF%d %d %d",
1142 			     nic->vf_id, qidx, ret);
1143 
1144 	other_error = ret;
1145 	rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
1146 	nicvf_rx_queue_release_mbufs(dev, rxq);
1147 	nicvf_rx_queue_reset(rxq);
1148 
1149 	ret = nicvf_qset_cq_reclaim(nic, qidx);
1150 	if (ret)
1151 		PMD_INIT_LOG(ERR, "Failed to reclaim cq VF%d %d %d",
1152 			     nic->vf_id, qidx, ret);
1153 
1154 	other_error |= ret;
1155 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1156 		RTE_ETH_QUEUE_STATE_STOPPED;
1157 	return other_error;
1158 }
1159 
1160 static void
1161 nicvf_dev_rx_queue_release(void *rx_queue)
1162 {
1163 	PMD_INIT_FUNC_TRACE();
1164 
1165 	rte_free(rx_queue);
1166 }
1167 
1168 static int
1169 nicvf_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
1170 {
1171 	struct nicvf *nic = nicvf_pmd_priv(dev);
1172 	int ret;
1173 
1174 	if (qidx >= MAX_RCV_QUEUES_PER_QS)
1175 		nic = nic->snicvf[(qidx / MAX_RCV_QUEUES_PER_QS - 1)];
1176 
1177 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1178 
1179 	ret = nicvf_vf_start_rx_queue(dev, nic, qidx);
1180 	if (ret)
1181 		return ret;
1182 
1183 	ret = nicvf_configure_cpi(dev);
1184 	if (ret)
1185 		return ret;
1186 
1187 	return nicvf_configure_rss_reta(dev);
1188 }
1189 
1190 static int
1191 nicvf_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
1192 {
1193 	int ret;
1194 	struct nicvf *nic = nicvf_pmd_priv(dev);
1195 
1196 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1197 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1198 
1199 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1200 
1201 	ret = nicvf_vf_stop_rx_queue(dev, nic, qidx);
1202 	ret |= nicvf_configure_cpi(dev);
1203 	ret |= nicvf_configure_rss_reta(dev);
1204 	return ret;
1205 }
1206 
1207 static int
1208 nicvf_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
1209 {
1210 	struct nicvf *nic = nicvf_pmd_priv(dev);
1211 
1212 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1213 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1214 
1215 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
1216 
1217 	return nicvf_vf_start_tx_queue(dev, nic, qidx);
1218 }
1219 
1220 static int
1221 nicvf_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
1222 {
1223 	struct nicvf *nic = nicvf_pmd_priv(dev);
1224 
1225 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1226 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1227 
1228 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
1229 
1230 	return nicvf_vf_stop_tx_queue(dev, nic, qidx);
1231 }
1232 
1233 static inline void
1234 nicvf_rxq_mbuf_setup(struct nicvf_rxq *rxq)
1235 {
1236 	uintptr_t p;
1237 	struct rte_mbuf mb_def;
1238 	struct nicvf *nic = rxq->nic;
1239 
1240 	RTE_BUILD_BUG_ON(sizeof(union mbuf_initializer) != 8);
1241 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, data_off) % 8 != 0);
1242 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, refcnt) -
1243 				offsetof(struct rte_mbuf, data_off) != 2);
1244 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, nb_segs) -
1245 				offsetof(struct rte_mbuf, data_off) != 4);
1246 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, port) -
1247 				offsetof(struct rte_mbuf, data_off) != 6);
1248 	mb_def.nb_segs = 1;
1249 	mb_def.data_off = RTE_PKTMBUF_HEADROOM + (nic->skip_bytes);
1250 	mb_def.port = rxq->port_id;
1251 	rte_mbuf_refcnt_set(&mb_def, 1);
1252 
1253 	/* Prevent compiler reordering: rearm_data covers previous fields */
1254 	rte_compiler_barrier();
1255 	p = (uintptr_t)&mb_def.rearm_data;
1256 	rxq->mbuf_initializer.value = *(uint64_t *)p;
1257 }
1258 
1259 static int
1260 nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
1261 			 uint16_t nb_desc, unsigned int socket_id,
1262 			 const struct rte_eth_rxconf *rx_conf,
1263 			 struct rte_mempool *mp)
1264 {
1265 	uint16_t rx_free_thresh;
1266 	struct nicvf_rxq *rxq;
1267 	struct nicvf *nic = nicvf_pmd_priv(dev);
1268 	uint64_t offloads;
1269 	uint32_t buffsz;
1270 	struct rte_pktmbuf_pool_private *mbp_priv;
1271 
1272 	PMD_INIT_FUNC_TRACE();
1273 
1274 	/* First skip check */
1275 	mbp_priv = rte_mempool_get_priv(mp);
1276 	buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
1277 	if (buffsz < (uint32_t)(nic->skip_bytes)) {
1278 		PMD_INIT_LOG(ERR, "First skip is more than configured buffer size");
1279 		return -EINVAL;
1280 	}
1281 
1282 	if (qidx >= MAX_RCV_QUEUES_PER_QS)
1283 		nic = nic->snicvf[qidx / MAX_RCV_QUEUES_PER_QS - 1];
1284 
1285 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1286 
1287 	/* Socket id check */
1288 	if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
1289 		PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
1290 		socket_id, nic->node);
1291 
1292 	/* Mempool memory must be contiguous, so must be one memory segment*/
1293 	if (mp->nb_mem_chunks != 1) {
1294 		PMD_INIT_LOG(ERR, "Non-contiguous mempool, add more huge pages");
1295 		return -EINVAL;
1296 	}
1297 
1298 	/* Mempool memory must be physically contiguous */
1299 	if (mp->flags & MEMPOOL_F_NO_IOVA_CONTIG) {
1300 		PMD_INIT_LOG(ERR, "Mempool memory must be physically contiguous");
1301 		return -EINVAL;
1302 	}
1303 
1304 	/* Rx deferred start is not supported */
1305 	if (rx_conf->rx_deferred_start) {
1306 		PMD_INIT_LOG(ERR, "Rx deferred start not supported");
1307 		return -EINVAL;
1308 	}
1309 
1310 	/* Roundup nb_desc to available qsize and validate max number of desc */
1311 	nb_desc = nicvf_qsize_cq_roundup(nb_desc);
1312 	if (nb_desc == 0) {
1313 		PMD_INIT_LOG(ERR, "Value nb_desc beyond available hw cq qsize");
1314 		return -EINVAL;
1315 	}
1316 
1317 
1318 	/* Check rx_free_thresh upper bound */
1319 	rx_free_thresh = (uint16_t)((rx_conf->rx_free_thresh) ?
1320 				rx_conf->rx_free_thresh :
1321 				NICVF_DEFAULT_RX_FREE_THRESH);
1322 	if (rx_free_thresh > NICVF_MAX_RX_FREE_THRESH ||
1323 		rx_free_thresh >= nb_desc * .75) {
1324 		PMD_INIT_LOG(ERR, "rx_free_thresh greater than expected %d",
1325 				rx_free_thresh);
1326 		return -EINVAL;
1327 	}
1328 
1329 	/* Free memory prior to re-allocation if needed */
1330 	if (dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
1331 		PMD_RX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
1332 				nicvf_netdev_qidx(nic, qidx));
1333 		nicvf_dev_rx_queue_release(
1334 			dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)]);
1335 		dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
1336 	}
1337 
1338 	/* Allocate rxq memory */
1339 	rxq = rte_zmalloc_socket("ethdev rx queue", sizeof(struct nicvf_rxq),
1340 					RTE_CACHE_LINE_SIZE, nic->node);
1341 	if (rxq == NULL) {
1342 		PMD_INIT_LOG(ERR, "Failed to allocate rxq=%d",
1343 			     nicvf_netdev_qidx(nic, qidx));
1344 		return -ENOMEM;
1345 	}
1346 
1347 	rxq->nic = nic;
1348 	rxq->pool = mp;
1349 	rxq->queue_id = qidx;
1350 	rxq->port_id = dev->data->port_id;
1351 	rxq->rx_free_thresh = rx_free_thresh;
1352 	rxq->rx_drop_en = rx_conf->rx_drop_en;
1353 	rxq->cq_status = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_STATUS;
1354 	rxq->cq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_DOOR;
1355 	rxq->precharge_cnt = 0;
1356 
1357 	if (nicvf_hw_cap(nic) & NICVF_CAP_CQE_RX2)
1358 		rxq->rbptr_offset = NICVF_CQE_RX2_RBPTR_WORD;
1359 	else
1360 		rxq->rbptr_offset = NICVF_CQE_RBPTR_WORD;
1361 
1362 	nicvf_rxq_mbuf_setup(rxq);
1363 
1364 	/* Alloc completion queue */
1365 	if (nicvf_qset_cq_alloc(dev, nic, rxq, rxq->queue_id, nb_desc)) {
1366 		PMD_INIT_LOG(ERR, "failed to allocate cq %u", rxq->queue_id);
1367 		nicvf_dev_rx_queue_release(rxq);
1368 		return -ENOMEM;
1369 	}
1370 
1371 	nicvf_rx_queue_reset(rxq);
1372 
1373 	offloads = rx_conf->offloads | dev->data->dev_conf.rxmode.offloads;
1374 	PMD_INIT_LOG(DEBUG, "[%d] rxq=%p pool=%s nb_desc=(%d/%d)"
1375 			" phy=0x%" PRIx64 " offloads=0x%" PRIx64,
1376 			nicvf_netdev_qidx(nic, qidx), rxq, mp->name, nb_desc,
1377 			rte_mempool_avail_count(mp), rxq->phys, offloads);
1378 
1379 	dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = rxq;
1380 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1381 		RTE_ETH_QUEUE_STATE_STOPPED;
1382 	return 0;
1383 }
1384 
1385 static void
1386 nicvf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1387 {
1388 	struct nicvf *nic = nicvf_pmd_priv(dev);
1389 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1390 
1391 	PMD_INIT_FUNC_TRACE();
1392 
1393 	/* Autonegotiation may be disabled */
1394 	dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
1395 	dev_info->speed_capa |= ETH_LINK_SPEED_10M | ETH_LINK_SPEED_100M |
1396 				 ETH_LINK_SPEED_1G | ETH_LINK_SPEED_10G;
1397 	if (nicvf_hw_version(nic) != PCI_SUB_DEVICE_ID_CN81XX_NICVF)
1398 		dev_info->speed_capa |= ETH_LINK_SPEED_40G;
1399 
1400 	dev_info->min_rx_bufsize = ETHER_MIN_MTU;
1401 	dev_info->max_rx_pktlen = NIC_HW_MAX_MTU + ETHER_HDR_LEN;
1402 	dev_info->max_rx_queues =
1403 			(uint16_t)MAX_RCV_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
1404 	dev_info->max_tx_queues =
1405 			(uint16_t)MAX_SND_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
1406 	dev_info->max_mac_addrs = 1;
1407 	dev_info->max_vfs = pci_dev->max_vfs;
1408 
1409 	dev_info->rx_offload_capa = NICVF_RX_OFFLOAD_CAPA;
1410 	dev_info->tx_offload_capa = NICVF_TX_OFFLOAD_CAPA;
1411 	dev_info->rx_queue_offload_capa = NICVF_RX_OFFLOAD_CAPA;
1412 	dev_info->tx_queue_offload_capa = NICVF_TX_OFFLOAD_CAPA;
1413 
1414 	dev_info->reta_size = nic->rss_info.rss_size;
1415 	dev_info->hash_key_size = RSS_HASH_KEY_BYTE_SIZE;
1416 	dev_info->flow_type_rss_offloads = NICVF_RSS_OFFLOAD_PASS1;
1417 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING)
1418 		dev_info->flow_type_rss_offloads |= NICVF_RSS_OFFLOAD_TUNNEL;
1419 
1420 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
1421 		.rx_free_thresh = NICVF_DEFAULT_RX_FREE_THRESH,
1422 		.rx_drop_en = 0,
1423 		.offloads = DEV_RX_OFFLOAD_CRC_STRIP,
1424 	};
1425 
1426 	dev_info->default_txconf = (struct rte_eth_txconf) {
1427 		.tx_free_thresh = NICVF_DEFAULT_TX_FREE_THRESH,
1428 		.offloads = DEV_TX_OFFLOAD_MBUF_FAST_FREE |
1429 			DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM   |
1430 			DEV_TX_OFFLOAD_UDP_CKSUM          |
1431 			DEV_TX_OFFLOAD_TCP_CKSUM,
1432 	};
1433 }
1434 
1435 static nicvf_iova_addr_t
1436 rbdr_rte_mempool_get(void *dev, void *opaque)
1437 {
1438 	uint16_t qidx;
1439 	uintptr_t mbuf;
1440 	struct nicvf_rxq *rxq;
1441 	struct rte_eth_dev *eth_dev = (struct rte_eth_dev *)dev;
1442 	struct nicvf *nic = (struct nicvf *)opaque;
1443 	uint16_t rx_start, rx_end;
1444 
1445 	/* Get queue ranges for this VF */
1446 	nicvf_rx_range(eth_dev, nic, &rx_start, &rx_end);
1447 
1448 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1449 		rxq = eth_dev->data->rx_queues[qidx];
1450 		/* Maintain equal buffer count across all pools */
1451 		if (rxq->precharge_cnt >= rxq->qlen_mask)
1452 			continue;
1453 		rxq->precharge_cnt++;
1454 		mbuf = (uintptr_t)rte_pktmbuf_alloc(rxq->pool);
1455 		if (mbuf)
1456 			return nicvf_mbuff_virt2phy(mbuf, rxq->mbuf_phys_off);
1457 	}
1458 	return 0;
1459 }
1460 
1461 static int
1462 nicvf_vf_start(struct rte_eth_dev *dev, struct nicvf *nic, uint32_t rbdrsz)
1463 {
1464 	int ret;
1465 	uint16_t qidx, data_off;
1466 	uint32_t total_rxq_desc, nb_rbdr_desc, exp_buffs;
1467 	uint64_t mbuf_phys_off = 0;
1468 	struct nicvf_rxq *rxq;
1469 	struct rte_mbuf *mbuf;
1470 	uint16_t rx_start, rx_end;
1471 	uint16_t tx_start, tx_end;
1472 	bool vlan_strip;
1473 
1474 	PMD_INIT_FUNC_TRACE();
1475 
1476 	/* Userspace process exited without proper shutdown in last run */
1477 	if (nicvf_qset_rbdr_active(nic, 0))
1478 		nicvf_vf_stop(dev, nic, false);
1479 
1480 	/* Get queue ranges for this VF */
1481 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
1482 
1483 	/*
1484 	 * Thunderx nicvf PMD can support more than one pool per port only when
1485 	 * 1) Data payload size is same across all the pools in given port
1486 	 * AND
1487 	 * 2) All mbuffs in the pools are from the same hugepage
1488 	 * AND
1489 	 * 3) Mbuff metadata size is same across all the pools in given port
1490 	 *
1491 	 * This is to support existing application that uses multiple pool/port.
1492 	 * But, the purpose of using multipool for QoS will not be addressed.
1493 	 *
1494 	 */
1495 
1496 	/* Validate mempool attributes */
1497 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1498 		rxq = dev->data->rx_queues[qidx];
1499 		rxq->mbuf_phys_off = nicvf_mempool_phy_offset(rxq->pool);
1500 		mbuf = rte_pktmbuf_alloc(rxq->pool);
1501 		if (mbuf == NULL) {
1502 			PMD_INIT_LOG(ERR, "Failed allocate mbuf VF%d qid=%d "
1503 				     "pool=%s",
1504 				     nic->vf_id, qidx, rxq->pool->name);
1505 			return -ENOMEM;
1506 		}
1507 		data_off = nicvf_mbuff_meta_length(mbuf);
1508 		data_off += RTE_PKTMBUF_HEADROOM;
1509 		rte_pktmbuf_free(mbuf);
1510 
1511 		if (data_off % RTE_CACHE_LINE_SIZE) {
1512 			PMD_INIT_LOG(ERR, "%s: unaligned data_off=%d delta=%d",
1513 				rxq->pool->name, data_off,
1514 				data_off % RTE_CACHE_LINE_SIZE);
1515 			return -EINVAL;
1516 		}
1517 		rxq->mbuf_phys_off -= data_off;
1518 		rxq->mbuf_phys_off -= nic->skip_bytes;
1519 
1520 		if (mbuf_phys_off == 0)
1521 			mbuf_phys_off = rxq->mbuf_phys_off;
1522 		if (mbuf_phys_off != rxq->mbuf_phys_off) {
1523 			PMD_INIT_LOG(ERR, "pool params not same,%s VF%d %"
1524 				     PRIx64, rxq->pool->name, nic->vf_id,
1525 				     mbuf_phys_off);
1526 			return -EINVAL;
1527 		}
1528 	}
1529 
1530 	/* Check the level of buffers in the pool */
1531 	total_rxq_desc = 0;
1532 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1533 		rxq = dev->data->rx_queues[qidx];
1534 		/* Count total numbers of rxq descs */
1535 		total_rxq_desc += rxq->qlen_mask + 1;
1536 		exp_buffs = RTE_MEMPOOL_CACHE_MAX_SIZE + rxq->rx_free_thresh;
1537 		exp_buffs *= dev->data->nb_rx_queues;
1538 		if (rte_mempool_avail_count(rxq->pool) < exp_buffs) {
1539 			PMD_INIT_LOG(ERR, "Buff shortage in pool=%s (%d/%d)",
1540 				     rxq->pool->name,
1541 				     rte_mempool_avail_count(rxq->pool),
1542 				     exp_buffs);
1543 			return -ENOENT;
1544 		}
1545 	}
1546 
1547 	/* Check RBDR desc overflow */
1548 	ret = nicvf_qsize_rbdr_roundup(total_rxq_desc);
1549 	if (ret == 0) {
1550 		PMD_INIT_LOG(ERR, "Reached RBDR desc limit, reduce nr desc "
1551 			     "VF%d", nic->vf_id);
1552 		return -ENOMEM;
1553 	}
1554 
1555 	/* Enable qset */
1556 	ret = nicvf_qset_config(nic);
1557 	if (ret) {
1558 		PMD_INIT_LOG(ERR, "Failed to enable qset %d VF%d", ret,
1559 			     nic->vf_id);
1560 		return ret;
1561 	}
1562 
1563 	/* Allocate RBDR and RBDR ring desc */
1564 	nb_rbdr_desc = nicvf_qsize_rbdr_roundup(total_rxq_desc);
1565 	ret = nicvf_qset_rbdr_alloc(dev, nic, nb_rbdr_desc, rbdrsz);
1566 	if (ret) {
1567 		PMD_INIT_LOG(ERR, "Failed to allocate memory for rbdr alloc "
1568 			     "VF%d", nic->vf_id);
1569 		goto qset_reclaim;
1570 	}
1571 
1572 	/* Enable and configure RBDR registers */
1573 	ret = nicvf_qset_rbdr_config(nic, 0);
1574 	if (ret) {
1575 		PMD_INIT_LOG(ERR, "Failed to configure rbdr %d VF%d", ret,
1576 			     nic->vf_id);
1577 		goto qset_rbdr_free;
1578 	}
1579 
1580 	/* Fill rte_mempool buffers in RBDR pool and precharge it */
1581 	ret = nicvf_qset_rbdr_precharge(dev, nic, 0, rbdr_rte_mempool_get,
1582 					total_rxq_desc);
1583 	if (ret) {
1584 		PMD_INIT_LOG(ERR, "Failed to fill rbdr %d VF%d", ret,
1585 			     nic->vf_id);
1586 		goto qset_rbdr_reclaim;
1587 	}
1588 
1589 	PMD_DRV_LOG(INFO, "Filled %d out of %d entries in RBDR VF%d",
1590 		     nic->rbdr->tail, nb_rbdr_desc, nic->vf_id);
1591 
1592 	/* Configure VLAN Strip */
1593 	vlan_strip = !!(dev->data->dev_conf.rxmode.offloads &
1594 			DEV_RX_OFFLOAD_VLAN_STRIP);
1595 	nicvf_vlan_hw_strip(nic, vlan_strip);
1596 
1597 	/* Based on the packet type(IPv4 or IPv6), the nicvf HW aligns L3 data
1598 	 * to the 64bit memory address.
1599 	 * The alignment creates a hole in mbuf(between the end of headroom and
1600 	 * packet data start). The new revision of the HW provides an option to
1601 	 * disable the L3 alignment feature and make mbuf layout looks
1602 	 * more like other NICs. For better application compatibility, disabling
1603 	 * l3 alignment feature on the hardware revisions it supports
1604 	 */
1605 	nicvf_apad_config(nic, false);
1606 
1607 	/* Get queue ranges for this VF */
1608 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
1609 
1610 	/* Configure TX queues */
1611 	for (qidx = tx_start; qidx <= tx_end; qidx++) {
1612 		ret = nicvf_vf_start_tx_queue(dev, nic,
1613 			qidx % MAX_SND_QUEUES_PER_QS);
1614 		if (ret)
1615 			goto start_txq_error;
1616 	}
1617 
1618 	/* Configure RX queues */
1619 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1620 		ret = nicvf_vf_start_rx_queue(dev, nic,
1621 			qidx % MAX_RCV_QUEUES_PER_QS);
1622 		if (ret)
1623 			goto start_rxq_error;
1624 	}
1625 
1626 	if (!nic->sqs_mode) {
1627 		/* Configure CPI algorithm */
1628 		ret = nicvf_configure_cpi(dev);
1629 		if (ret)
1630 			goto start_txq_error;
1631 
1632 		ret = nicvf_mbox_get_rss_size(nic);
1633 		if (ret) {
1634 			PMD_INIT_LOG(ERR, "Failed to get rss table size");
1635 			goto qset_rss_error;
1636 		}
1637 
1638 		/* Configure RSS */
1639 		ret = nicvf_configure_rss(dev);
1640 		if (ret)
1641 			goto qset_rss_error;
1642 	}
1643 
1644 	/* Done; Let PF make the BGX's RX and TX switches to ON position */
1645 	nicvf_mbox_cfg_done(nic);
1646 	return 0;
1647 
1648 qset_rss_error:
1649 	nicvf_rss_term(nic);
1650 start_rxq_error:
1651 	for (qidx = rx_start; qidx <= rx_end; qidx++)
1652 		nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
1653 start_txq_error:
1654 	for (qidx = tx_start; qidx <= tx_end; qidx++)
1655 		nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
1656 qset_rbdr_reclaim:
1657 	nicvf_qset_rbdr_reclaim(nic, 0);
1658 	nicvf_rbdr_release_mbufs(dev, nic);
1659 qset_rbdr_free:
1660 	if (nic->rbdr) {
1661 		rte_free(nic->rbdr);
1662 		nic->rbdr = NULL;
1663 	}
1664 qset_reclaim:
1665 	nicvf_qset_reclaim(nic);
1666 	return ret;
1667 }
1668 
1669 static int
1670 nicvf_dev_start(struct rte_eth_dev *dev)
1671 {
1672 	uint16_t qidx;
1673 	int ret;
1674 	size_t i;
1675 	struct nicvf *nic = nicvf_pmd_priv(dev);
1676 	struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
1677 	uint16_t mtu;
1678 	uint32_t buffsz = 0, rbdrsz = 0;
1679 	struct rte_pktmbuf_pool_private *mbp_priv;
1680 	struct nicvf_rxq *rxq;
1681 
1682 	PMD_INIT_FUNC_TRACE();
1683 
1684 	/* This function must be called for a primary device */
1685 	assert_primary(nic);
1686 
1687 	/* Validate RBDR buff size */
1688 	for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
1689 		rxq = dev->data->rx_queues[qidx];
1690 		mbp_priv = rte_mempool_get_priv(rxq->pool);
1691 		buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
1692 		if (buffsz % 128) {
1693 			PMD_INIT_LOG(ERR, "rxbuf size must be multiply of 128");
1694 			return -EINVAL;
1695 		}
1696 		if (rbdrsz == 0)
1697 			rbdrsz = buffsz;
1698 		if (rbdrsz != buffsz) {
1699 			PMD_INIT_LOG(ERR, "buffsz not same, qidx=%d (%d/%d)",
1700 				     qidx, rbdrsz, buffsz);
1701 			return -EINVAL;
1702 		}
1703 	}
1704 
1705 	/* Configure loopback */
1706 	ret = nicvf_loopback_config(nic, dev->data->dev_conf.lpbk_mode);
1707 	if (ret) {
1708 		PMD_INIT_LOG(ERR, "Failed to configure loopback %d", ret);
1709 		return ret;
1710 	}
1711 
1712 	/* Reset all statistics counters attached to this port */
1713 	ret = nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, 0xFFFF, 0xFFFF);
1714 	if (ret) {
1715 		PMD_INIT_LOG(ERR, "Failed to reset stat counters %d", ret);
1716 		return ret;
1717 	}
1718 
1719 	/* Setup scatter mode if needed by jumbo */
1720 	if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
1721 					    2 * VLAN_TAG_SIZE > buffsz)
1722 		dev->data->scattered_rx = 1;
1723 	if ((rx_conf->offloads & DEV_RX_OFFLOAD_SCATTER) != 0)
1724 		dev->data->scattered_rx = 1;
1725 
1726 	/* Setup MTU based on max_rx_pkt_len or default */
1727 	mtu = dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME ?
1728 		dev->data->dev_conf.rxmode.max_rx_pkt_len
1729 			-  ETHER_HDR_LEN : ETHER_MTU;
1730 
1731 	if (nicvf_dev_set_mtu(dev, mtu)) {
1732 		PMD_INIT_LOG(ERR, "Failed to set default mtu size");
1733 		return -EBUSY;
1734 	}
1735 
1736 	ret = nicvf_vf_start(dev, nic, rbdrsz);
1737 	if (ret != 0)
1738 		return ret;
1739 
1740 	for (i = 0; i < nic->sqs_count; i++) {
1741 		assert(nic->snicvf[i]);
1742 
1743 		ret = nicvf_vf_start(dev, nic->snicvf[i], rbdrsz);
1744 		if (ret != 0)
1745 			return ret;
1746 	}
1747 
1748 	/* Configure callbacks based on scatter mode */
1749 	nicvf_set_tx_function(dev);
1750 	nicvf_set_rx_function(dev);
1751 
1752 	return 0;
1753 }
1754 
1755 static void
1756 nicvf_dev_stop_cleanup(struct rte_eth_dev *dev, bool cleanup)
1757 {
1758 	size_t i;
1759 	int ret;
1760 	struct nicvf *nic = nicvf_pmd_priv(dev);
1761 
1762 	PMD_INIT_FUNC_TRACE();
1763 
1764 	/* Teardown secondary vf first */
1765 	for (i = 0; i < nic->sqs_count; i++) {
1766 		if (!nic->snicvf[i])
1767 			continue;
1768 
1769 		nicvf_vf_stop(dev, nic->snicvf[i], cleanup);
1770 	}
1771 
1772 	/* Stop the primary VF now */
1773 	nicvf_vf_stop(dev, nic, cleanup);
1774 
1775 	/* Disable loopback */
1776 	ret = nicvf_loopback_config(nic, 0);
1777 	if (ret)
1778 		PMD_INIT_LOG(ERR, "Failed to disable loopback %d", ret);
1779 
1780 	/* Reclaim CPI configuration */
1781 	ret = nicvf_mbox_config_cpi(nic, 0);
1782 	if (ret)
1783 		PMD_INIT_LOG(ERR, "Failed to reclaim CPI config %d", ret);
1784 }
1785 
1786 static void
1787 nicvf_dev_stop(struct rte_eth_dev *dev)
1788 {
1789 	PMD_INIT_FUNC_TRACE();
1790 
1791 	nicvf_dev_stop_cleanup(dev, false);
1792 }
1793 
1794 static void
1795 nicvf_vf_stop(struct rte_eth_dev *dev, struct nicvf *nic, bool cleanup)
1796 {
1797 	int ret;
1798 	uint16_t qidx;
1799 	uint16_t tx_start, tx_end;
1800 	uint16_t rx_start, rx_end;
1801 
1802 	PMD_INIT_FUNC_TRACE();
1803 
1804 	if (cleanup) {
1805 		/* Let PF make the BGX's RX and TX switches to OFF position */
1806 		nicvf_mbox_shutdown(nic);
1807 	}
1808 
1809 	/* Disable VLAN Strip */
1810 	nicvf_vlan_hw_strip(nic, 0);
1811 
1812 	/* Get queue ranges for this VF */
1813 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
1814 
1815 	for (qidx = tx_start; qidx <= tx_end; qidx++)
1816 		nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
1817 
1818 	/* Get queue ranges for this VF */
1819 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
1820 
1821 	/* Reclaim rq */
1822 	for (qidx = rx_start; qidx <= rx_end; qidx++)
1823 		nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
1824 
1825 	/* Reclaim RBDR */
1826 	ret = nicvf_qset_rbdr_reclaim(nic, 0);
1827 	if (ret)
1828 		PMD_INIT_LOG(ERR, "Failed to reclaim RBDR %d", ret);
1829 
1830 	/* Move all charged buffers in RBDR back to pool */
1831 	if (nic->rbdr != NULL)
1832 		nicvf_rbdr_release_mbufs(dev, nic);
1833 
1834 	/* Disable qset */
1835 	ret = nicvf_qset_reclaim(nic);
1836 	if (ret)
1837 		PMD_INIT_LOG(ERR, "Failed to disable qset %d", ret);
1838 
1839 	/* Disable all interrupts */
1840 	nicvf_disable_all_interrupts(nic);
1841 
1842 	/* Free RBDR SW structure */
1843 	if (nic->rbdr) {
1844 		rte_free(nic->rbdr);
1845 		nic->rbdr = NULL;
1846 	}
1847 }
1848 
1849 static void
1850 nicvf_dev_close(struct rte_eth_dev *dev)
1851 {
1852 	size_t i;
1853 	struct nicvf *nic = nicvf_pmd_priv(dev);
1854 
1855 	PMD_INIT_FUNC_TRACE();
1856 
1857 	nicvf_dev_stop_cleanup(dev, true);
1858 	nicvf_periodic_alarm_stop(nicvf_interrupt, dev);
1859 
1860 	for (i = 0; i < nic->sqs_count; i++) {
1861 		if (!nic->snicvf[i])
1862 			continue;
1863 
1864 		nicvf_periodic_alarm_stop(nicvf_vf_interrupt, nic->snicvf[i]);
1865 	}
1866 }
1867 
1868 static int
1869 nicvf_request_sqs(struct nicvf *nic)
1870 {
1871 	size_t i;
1872 
1873 	assert_primary(nic);
1874 	assert(nic->sqs_count > 0);
1875 	assert(nic->sqs_count <= MAX_SQS_PER_VF);
1876 
1877 	/* Set no of Rx/Tx queues in each of the SQsets */
1878 	for (i = 0; i < nic->sqs_count; i++) {
1879 		if (nicvf_svf_empty())
1880 			rte_panic("Cannot assign sufficient number of "
1881 				  "secondary queues to primary VF%" PRIu8 "\n",
1882 				  nic->vf_id);
1883 
1884 		nic->snicvf[i] = nicvf_svf_pop();
1885 		nic->snicvf[i]->sqs_id = i;
1886 	}
1887 
1888 	return nicvf_mbox_request_sqs(nic);
1889 }
1890 
1891 static int
1892 nicvf_dev_configure(struct rte_eth_dev *dev)
1893 {
1894 	struct rte_eth_dev_data *data = dev->data;
1895 	struct rte_eth_conf *conf = &data->dev_conf;
1896 	struct rte_eth_rxmode *rxmode = &conf->rxmode;
1897 	struct rte_eth_txmode *txmode = &conf->txmode;
1898 	struct nicvf *nic = nicvf_pmd_priv(dev);
1899 	uint8_t cqcount;
1900 
1901 	PMD_INIT_FUNC_TRACE();
1902 
1903 	if (!rte_eal_has_hugepages()) {
1904 		PMD_INIT_LOG(INFO, "Huge page is not configured");
1905 		return -EINVAL;
1906 	}
1907 
1908 	/* KEEP_CRC offload flag is not supported by PMD
1909 	 * can remove the below block when DEV_RX_OFFLOAD_CRC_STRIP removed
1910 	 */
1911 	if (rte_eth_dev_must_keep_crc(rxmode->offloads)) {
1912 		PMD_INIT_LOG(NOTICE, "Can't disable hw crc strip");
1913 		rxmode->offloads |= DEV_RX_OFFLOAD_CRC_STRIP;
1914 	}
1915 
1916 	if (txmode->mq_mode) {
1917 		PMD_INIT_LOG(INFO, "Tx mq_mode DCB or VMDq not supported");
1918 		return -EINVAL;
1919 	}
1920 
1921 	if (rxmode->mq_mode != ETH_MQ_RX_NONE &&
1922 		rxmode->mq_mode != ETH_MQ_RX_RSS) {
1923 		PMD_INIT_LOG(INFO, "Unsupported rx qmode %d", rxmode->mq_mode);
1924 		return -EINVAL;
1925 	}
1926 
1927 	if (rxmode->split_hdr_size) {
1928 		PMD_INIT_LOG(INFO, "Rxmode does not support split header");
1929 		return -EINVAL;
1930 	}
1931 
1932 	if (conf->link_speeds & ETH_LINK_SPEED_FIXED) {
1933 		PMD_INIT_LOG(INFO, "Setting link speed/duplex not supported");
1934 		return -EINVAL;
1935 	}
1936 
1937 	if (conf->dcb_capability_en) {
1938 		PMD_INIT_LOG(INFO, "DCB enable not supported");
1939 		return -EINVAL;
1940 	}
1941 
1942 	if (conf->fdir_conf.mode != RTE_FDIR_MODE_NONE) {
1943 		PMD_INIT_LOG(INFO, "Flow director not supported");
1944 		return -EINVAL;
1945 	}
1946 
1947 	assert_primary(nic);
1948 	NICVF_STATIC_ASSERT(MAX_RCV_QUEUES_PER_QS == MAX_SND_QUEUES_PER_QS);
1949 	cqcount = RTE_MAX(data->nb_tx_queues, data->nb_rx_queues);
1950 	if (cqcount > MAX_RCV_QUEUES_PER_QS) {
1951 		nic->sqs_count = RTE_ALIGN_CEIL(cqcount, MAX_RCV_QUEUES_PER_QS);
1952 		nic->sqs_count = (nic->sqs_count / MAX_RCV_QUEUES_PER_QS) - 1;
1953 	} else {
1954 		nic->sqs_count = 0;
1955 	}
1956 
1957 	assert(nic->sqs_count <= MAX_SQS_PER_VF);
1958 
1959 	if (nic->sqs_count > 0) {
1960 		if (nicvf_request_sqs(nic)) {
1961 			rte_panic("Cannot assign sufficient number of "
1962 				  "secondary queues to PORT%d VF%" PRIu8 "\n",
1963 				  dev->data->port_id, nic->vf_id);
1964 		}
1965 	}
1966 
1967 	PMD_INIT_LOG(DEBUG, "Configured ethdev port%d hwcap=0x%" PRIx64,
1968 		dev->data->port_id, nicvf_hw_cap(nic));
1969 
1970 	return 0;
1971 }
1972 
1973 /* Initialize and register driver with DPDK Application */
1974 static const struct eth_dev_ops nicvf_eth_dev_ops = {
1975 	.dev_configure            = nicvf_dev_configure,
1976 	.dev_start                = nicvf_dev_start,
1977 	.dev_stop                 = nicvf_dev_stop,
1978 	.link_update              = nicvf_dev_link_update,
1979 	.dev_close                = nicvf_dev_close,
1980 	.stats_get                = nicvf_dev_stats_get,
1981 	.stats_reset              = nicvf_dev_stats_reset,
1982 	.promiscuous_enable       = nicvf_dev_promisc_enable,
1983 	.dev_infos_get            = nicvf_dev_info_get,
1984 	.dev_supported_ptypes_get = nicvf_dev_supported_ptypes_get,
1985 	.mtu_set                  = nicvf_dev_set_mtu,
1986 	.reta_update              = nicvf_dev_reta_update,
1987 	.reta_query               = nicvf_dev_reta_query,
1988 	.rss_hash_update          = nicvf_dev_rss_hash_update,
1989 	.rss_hash_conf_get        = nicvf_dev_rss_hash_conf_get,
1990 	.rx_queue_start           = nicvf_dev_rx_queue_start,
1991 	.rx_queue_stop            = nicvf_dev_rx_queue_stop,
1992 	.tx_queue_start           = nicvf_dev_tx_queue_start,
1993 	.tx_queue_stop            = nicvf_dev_tx_queue_stop,
1994 	.rx_queue_setup           = nicvf_dev_rx_queue_setup,
1995 	.rx_queue_release         = nicvf_dev_rx_queue_release,
1996 	.rx_queue_count           = nicvf_dev_rx_queue_count,
1997 	.tx_queue_setup           = nicvf_dev_tx_queue_setup,
1998 	.tx_queue_release         = nicvf_dev_tx_queue_release,
1999 	.get_reg                  = nicvf_dev_get_regs,
2000 };
2001 
2002 static inline int
2003 nicvf_set_first_skip(struct rte_eth_dev *dev)
2004 {
2005 	int bytes_to_skip = 0;
2006 	int ret = 0;
2007 	unsigned int i;
2008 	struct rte_kvargs *kvlist;
2009 	static const char *const skip[] = {
2010 		SKIP_DATA_BYTES,
2011 		NULL};
2012 	struct nicvf *nic = nicvf_pmd_priv(dev);
2013 
2014 	if (!dev->device->devargs) {
2015 		nicvf_first_skip_config(nic, 0);
2016 		return ret;
2017 	}
2018 
2019 	kvlist = rte_kvargs_parse(dev->device->devargs->args, skip);
2020 	if (!kvlist)
2021 		return -EINVAL;
2022 
2023 	if (kvlist->count == 0)
2024 		goto exit;
2025 
2026 	for (i = 0; i != kvlist->count; ++i) {
2027 		const struct rte_kvargs_pair *pair = &kvlist->pairs[i];
2028 
2029 		if (!strcmp(pair->key, SKIP_DATA_BYTES))
2030 			bytes_to_skip = atoi(pair->value);
2031 	}
2032 
2033 	/*128 bytes amounts to one cache line*/
2034 	if (bytes_to_skip >= 0 && bytes_to_skip < 128) {
2035 		if (!(bytes_to_skip % 8)) {
2036 			nicvf_first_skip_config(nic, (bytes_to_skip / 8));
2037 			nic->skip_bytes = bytes_to_skip;
2038 			goto kvlist_free;
2039 		} else {
2040 			PMD_INIT_LOG(ERR, "skip_data_bytes should be multiple of 8");
2041 			ret = -EINVAL;
2042 			goto exit;
2043 		}
2044 	} else {
2045 		PMD_INIT_LOG(ERR, "skip_data_bytes should be less than 128");
2046 		ret = -EINVAL;
2047 		goto exit;
2048 	}
2049 exit:
2050 	nicvf_first_skip_config(nic, 0);
2051 kvlist_free:
2052 	rte_kvargs_free(kvlist);
2053 	return ret;
2054 }
2055 static int
2056 nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
2057 {
2058 	int ret;
2059 	struct rte_pci_device *pci_dev;
2060 	struct nicvf *nic = nicvf_pmd_priv(eth_dev);
2061 
2062 	PMD_INIT_FUNC_TRACE();
2063 
2064 	eth_dev->dev_ops = &nicvf_eth_dev_ops;
2065 
2066 	/* For secondary processes, the primary has done all the work */
2067 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2068 		if (nic) {
2069 			/* Setup callbacks for secondary process */
2070 			nicvf_set_tx_function(eth_dev);
2071 			nicvf_set_rx_function(eth_dev);
2072 			return 0;
2073 		} else {
2074 			/* If nic == NULL than it is secondary function
2075 			 * so ethdev need to be released by caller */
2076 			return ENOTSUP;
2077 		}
2078 	}
2079 
2080 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
2081 	rte_eth_copy_pci_info(eth_dev, pci_dev);
2082 
2083 	nic->device_id = pci_dev->id.device_id;
2084 	nic->vendor_id = pci_dev->id.vendor_id;
2085 	nic->subsystem_device_id = pci_dev->id.subsystem_device_id;
2086 	nic->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
2087 
2088 	PMD_INIT_LOG(DEBUG, "nicvf: device (%x:%x) %u:%u:%u:%u",
2089 			pci_dev->id.vendor_id, pci_dev->id.device_id,
2090 			pci_dev->addr.domain, pci_dev->addr.bus,
2091 			pci_dev->addr.devid, pci_dev->addr.function);
2092 
2093 	nic->reg_base = (uintptr_t)pci_dev->mem_resource[0].addr;
2094 	if (!nic->reg_base) {
2095 		PMD_INIT_LOG(ERR, "Failed to map BAR0");
2096 		ret = -ENODEV;
2097 		goto fail;
2098 	}
2099 
2100 	nicvf_disable_all_interrupts(nic);
2101 
2102 	ret = nicvf_periodic_alarm_start(nicvf_interrupt, eth_dev);
2103 	if (ret) {
2104 		PMD_INIT_LOG(ERR, "Failed to start period alarm");
2105 		goto fail;
2106 	}
2107 
2108 	ret = nicvf_mbox_check_pf_ready(nic);
2109 	if (ret) {
2110 		PMD_INIT_LOG(ERR, "Failed to get ready message from PF");
2111 		goto alarm_fail;
2112 	} else {
2113 		PMD_INIT_LOG(INFO,
2114 			"node=%d vf=%d mode=%s sqs=%s loopback_supported=%s",
2115 			nic->node, nic->vf_id,
2116 			nic->tns_mode == NIC_TNS_MODE ? "tns" : "tns-bypass",
2117 			nic->sqs_mode ? "true" : "false",
2118 			nic->loopback_supported ? "true" : "false"
2119 			);
2120 	}
2121 
2122 	ret = nicvf_base_init(nic);
2123 	if (ret) {
2124 		PMD_INIT_LOG(ERR, "Failed to execute nicvf_base_init");
2125 		goto malloc_fail;
2126 	}
2127 
2128 	if (nic->sqs_mode) {
2129 		/* Push nic to stack of secondary vfs */
2130 		nicvf_svf_push(nic);
2131 
2132 		/* Steal nic pointer from the device for further reuse */
2133 		eth_dev->data->dev_private = NULL;
2134 
2135 		nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
2136 		ret = nicvf_periodic_alarm_start(nicvf_vf_interrupt, nic);
2137 		if (ret) {
2138 			PMD_INIT_LOG(ERR, "Failed to start period alarm");
2139 			goto fail;
2140 		}
2141 
2142 		/* Detach port by returning positive error number */
2143 		return ENOTSUP;
2144 	}
2145 
2146 	eth_dev->data->mac_addrs = rte_zmalloc("mac_addr", ETHER_ADDR_LEN, 0);
2147 	if (eth_dev->data->mac_addrs == NULL) {
2148 		PMD_INIT_LOG(ERR, "Failed to allocate memory for mac addr");
2149 		ret = -ENOMEM;
2150 		goto alarm_fail;
2151 	}
2152 	if (is_zero_ether_addr((struct ether_addr *)nic->mac_addr))
2153 		eth_random_addr(&nic->mac_addr[0]);
2154 
2155 	ether_addr_copy((struct ether_addr *)nic->mac_addr,
2156 			&eth_dev->data->mac_addrs[0]);
2157 
2158 	ret = nicvf_mbox_set_mac_addr(nic, nic->mac_addr);
2159 	if (ret) {
2160 		PMD_INIT_LOG(ERR, "Failed to set mac addr");
2161 		goto malloc_fail;
2162 	}
2163 
2164 	ret = nicvf_set_first_skip(eth_dev);
2165 	if (ret) {
2166 		PMD_INIT_LOG(ERR, "Failed to configure first skip");
2167 		goto malloc_fail;
2168 	}
2169 	PMD_INIT_LOG(INFO, "Port %d (%x:%x) mac=%02x:%02x:%02x:%02x:%02x:%02x",
2170 		eth_dev->data->port_id, nic->vendor_id, nic->device_id,
2171 		nic->mac_addr[0], nic->mac_addr[1], nic->mac_addr[2],
2172 		nic->mac_addr[3], nic->mac_addr[4], nic->mac_addr[5]);
2173 
2174 	return 0;
2175 
2176 malloc_fail:
2177 	rte_free(eth_dev->data->mac_addrs);
2178 alarm_fail:
2179 	nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
2180 fail:
2181 	return ret;
2182 }
2183 
2184 static const struct rte_pci_id pci_id_nicvf_map[] = {
2185 	{
2186 		.class_id = RTE_CLASS_ANY_ID,
2187 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2188 		.device_id = PCI_DEVICE_ID_THUNDERX_CN88XX_PASS1_NICVF,
2189 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2190 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS1_NICVF,
2191 	},
2192 	{
2193 		.class_id = RTE_CLASS_ANY_ID,
2194 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2195 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2196 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2197 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS2_NICVF,
2198 	},
2199 	{
2200 		.class_id = RTE_CLASS_ANY_ID,
2201 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2202 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2203 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2204 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN81XX_NICVF,
2205 	},
2206 	{
2207 		.class_id = RTE_CLASS_ANY_ID,
2208 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2209 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2210 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2211 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN83XX_NICVF,
2212 	},
2213 	{
2214 		.vendor_id = 0,
2215 	},
2216 };
2217 
2218 static int nicvf_eth_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2219 	struct rte_pci_device *pci_dev)
2220 {
2221 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct nicvf),
2222 		nicvf_eth_dev_init);
2223 }
2224 
2225 static int nicvf_eth_pci_remove(struct rte_pci_device *pci_dev)
2226 {
2227 	return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
2228 }
2229 
2230 static struct rte_pci_driver rte_nicvf_pmd = {
2231 	.id_table = pci_id_nicvf_map,
2232 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_KEEP_MAPPED_RES |
2233 			RTE_PCI_DRV_INTR_LSC,
2234 	.probe = nicvf_eth_pci_probe,
2235 	.remove = nicvf_eth_pci_remove,
2236 };
2237 
2238 RTE_PMD_REGISTER_PCI(net_thunderx, rte_nicvf_pmd);
2239 RTE_PMD_REGISTER_PCI_TABLE(net_thunderx, pci_id_nicvf_map);
2240 RTE_PMD_REGISTER_KMOD_DEP(net_thunderx, "* igb_uio | uio_pci_generic | vfio-pci");
2241 RTE_PMD_REGISTER_PARAM_STRING(net_thunderx, SKIP_DATA_BYTES "=<int>");
2242