xref: /dpdk/drivers/net/thunderx/nicvf_ethdev.c (revision b7004ab27edc2f36ea42fa79a4aa6b38d7fa31d6)
1 /*
2  *   BSD LICENSE
3  *
4  *   Copyright (C) Cavium networks Ltd. 2016.
5  *
6  *   Redistribution and use in source and binary forms, with or without
7  *   modification, are permitted provided that the following conditions
8  *   are met:
9  *
10  *     * Redistributions of source code must retain the above copyright
11  *       notice, this list of conditions and the following disclaimer.
12  *     * Redistributions in binary form must reproduce the above copyright
13  *       notice, this list of conditions and the following disclaimer in
14  *       the documentation and/or other materials provided with the
15  *       distribution.
16  *     * Neither the name of Cavium networks nor the names of its
17  *       contributors may be used to endorse or promote products derived
18  *       from this software without specific prior written permission.
19  *
20  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
23  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
24  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
25  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
26  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
30  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 #include <assert.h>
34 #include <stdio.h>
35 #include <stdbool.h>
36 #include <errno.h>
37 #include <stdint.h>
38 #include <string.h>
39 #include <unistd.h>
40 #include <stdarg.h>
41 #include <inttypes.h>
42 #include <netinet/in.h>
43 #include <sys/queue.h>
44 #include <sys/timerfd.h>
45 
46 #include <rte_alarm.h>
47 #include <rte_atomic.h>
48 #include <rte_branch_prediction.h>
49 #include <rte_byteorder.h>
50 #include <rte_common.h>
51 #include <rte_cycles.h>
52 #include <rte_debug.h>
53 #include <rte_dev.h>
54 #include <rte_eal.h>
55 #include <rte_ether.h>
56 #include <rte_ethdev.h>
57 #include <rte_interrupts.h>
58 #include <rte_log.h>
59 #include <rte_memory.h>
60 #include <rte_memzone.h>
61 #include <rte_malloc.h>
62 #include <rte_random.h>
63 #include <rte_pci.h>
64 #include <rte_tailq.h>
65 
66 #include "base/nicvf_plat.h"
67 
68 #include "nicvf_ethdev.h"
69 #include "nicvf_rxtx.h"
70 #include "nicvf_svf.h"
71 #include "nicvf_logs.h"
72 
73 static void nicvf_dev_stop(struct rte_eth_dev *dev);
74 static void nicvf_dev_stop_cleanup(struct rte_eth_dev *dev, bool cleanup);
75 static void nicvf_vf_stop(struct rte_eth_dev *dev, struct nicvf *nic,
76 			  bool cleanup);
77 
78 static inline int
79 nicvf_atomic_write_link_status(struct rte_eth_dev *dev,
80 			       struct rte_eth_link *link)
81 {
82 	struct rte_eth_link *dst = &dev->data->dev_link;
83 	struct rte_eth_link *src = link;
84 
85 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
86 		*(uint64_t *)src) == 0)
87 		return -1;
88 
89 	return 0;
90 }
91 
92 static inline void
93 nicvf_set_eth_link_status(struct nicvf *nic, struct rte_eth_link *link)
94 {
95 	link->link_status = nic->link_up;
96 	link->link_duplex = ETH_LINK_AUTONEG;
97 	if (nic->duplex == NICVF_HALF_DUPLEX)
98 		link->link_duplex = ETH_LINK_HALF_DUPLEX;
99 	else if (nic->duplex == NICVF_FULL_DUPLEX)
100 		link->link_duplex = ETH_LINK_FULL_DUPLEX;
101 	link->link_speed = nic->speed;
102 	link->link_autoneg = ETH_LINK_SPEED_AUTONEG;
103 }
104 
105 static void
106 nicvf_interrupt(void *arg)
107 {
108 	struct rte_eth_dev *dev = arg;
109 	struct nicvf *nic = nicvf_pmd_priv(dev);
110 
111 	if (nicvf_reg_poll_interrupts(nic) == NIC_MBOX_MSG_BGX_LINK_CHANGE) {
112 		if (dev->data->dev_conf.intr_conf.lsc)
113 			nicvf_set_eth_link_status(nic, &dev->data->dev_link);
114 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC);
115 	}
116 
117 	rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
118 				nicvf_interrupt, dev);
119 }
120 
121 static void __rte_unused
122 nicvf_vf_interrupt(void *arg)
123 {
124 	struct nicvf *nic = arg;
125 
126 	nicvf_reg_poll_interrupts(nic);
127 
128 	rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
129 				nicvf_vf_interrupt, nic);
130 }
131 
132 static int
133 nicvf_periodic_alarm_start(void (fn)(void *), void *arg)
134 {
135 	return rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000, fn, arg);
136 }
137 
138 static int
139 nicvf_periodic_alarm_stop(void (fn)(void *), void *arg)
140 {
141 	return rte_eal_alarm_cancel(fn, arg);
142 }
143 
144 /*
145  * Return 0 means link status changed, -1 means not changed
146  */
147 static int
148 nicvf_dev_link_update(struct rte_eth_dev *dev,
149 		      int wait_to_complete __rte_unused)
150 {
151 	struct rte_eth_link link;
152 	struct nicvf *nic = nicvf_pmd_priv(dev);
153 
154 	PMD_INIT_FUNC_TRACE();
155 
156 	memset(&link, 0, sizeof(link));
157 	nicvf_set_eth_link_status(nic, &link);
158 	return nicvf_atomic_write_link_status(dev, &link);
159 }
160 
161 static int
162 nicvf_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
163 {
164 	struct nicvf *nic = nicvf_pmd_priv(dev);
165 	uint32_t buffsz, frame_size = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
166 	size_t i;
167 
168 	PMD_INIT_FUNC_TRACE();
169 
170 	if (frame_size > NIC_HW_MAX_FRS)
171 		return -EINVAL;
172 
173 	if (frame_size < NIC_HW_MIN_FRS)
174 		return -EINVAL;
175 
176 	buffsz = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
177 
178 	/*
179 	 * Refuse mtu that requires the support of scattered packets
180 	 * when this feature has not been enabled before.
181 	 */
182 	if (!dev->data->scattered_rx &&
183 		(frame_size + 2 * VLAN_TAG_SIZE > buffsz))
184 		return -EINVAL;
185 
186 	/* check <seg size> * <max_seg>  >= max_frame */
187 	if (dev->data->scattered_rx &&
188 		(frame_size + 2 * VLAN_TAG_SIZE > buffsz * NIC_HW_MAX_SEGS))
189 		return -EINVAL;
190 
191 	if (frame_size > ETHER_MAX_LEN)
192 		dev->data->dev_conf.rxmode.jumbo_frame = 1;
193 	else
194 		dev->data->dev_conf.rxmode.jumbo_frame = 0;
195 
196 	if (nicvf_mbox_update_hw_max_frs(nic, frame_size))
197 		return -EINVAL;
198 
199 	/* Update max frame size */
200 	dev->data->dev_conf.rxmode.max_rx_pkt_len = (uint32_t)frame_size;
201 	nic->mtu = mtu;
202 
203 	for (i = 0; i < nic->sqs_count; i++)
204 		nic->snicvf[i]->mtu = mtu;
205 
206 	return 0;
207 }
208 
209 static int
210 nicvf_dev_get_regs(struct rte_eth_dev *dev, struct rte_dev_reg_info *regs)
211 {
212 	uint64_t *data = regs->data;
213 	struct nicvf *nic = nicvf_pmd_priv(dev);
214 
215 	if (data == NULL) {
216 		regs->length = nicvf_reg_get_count();
217 		regs->width = THUNDERX_REG_BYTES;
218 		return 0;
219 	}
220 
221 	/* Support only full register dump */
222 	if ((regs->length == 0) ||
223 		(regs->length == (uint32_t)nicvf_reg_get_count())) {
224 		regs->version = nic->vendor_id << 16 | nic->device_id;
225 		nicvf_reg_dump(nic, data);
226 		return 0;
227 	}
228 	return -ENOTSUP;
229 }
230 
231 static void
232 nicvf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
233 {
234 	uint16_t qidx;
235 	struct nicvf_hw_rx_qstats rx_qstats;
236 	struct nicvf_hw_tx_qstats tx_qstats;
237 	struct nicvf_hw_stats port_stats;
238 	struct nicvf *nic = nicvf_pmd_priv(dev);
239 
240 	/* Reading per RX ring stats */
241 	for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
242 		if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
243 			break;
244 
245 		nicvf_hw_get_rx_qstats(nic, &rx_qstats, qidx);
246 		stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
247 		stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
248 	}
249 
250 	/* Reading per TX ring stats */
251 	for (qidx = 0; qidx < dev->data->nb_tx_queues; qidx++) {
252 		if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
253 			break;
254 
255 		nicvf_hw_get_tx_qstats(nic, &tx_qstats, qidx);
256 		stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
257 		stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
258 	}
259 
260 	nicvf_hw_get_stats(nic, &port_stats);
261 	stats->ibytes = port_stats.rx_bytes;
262 	stats->ipackets = port_stats.rx_ucast_frames;
263 	stats->ipackets += port_stats.rx_bcast_frames;
264 	stats->ipackets += port_stats.rx_mcast_frames;
265 	stats->ierrors = port_stats.rx_l2_errors;
266 	stats->imissed = port_stats.rx_drop_red;
267 	stats->imissed += port_stats.rx_drop_overrun;
268 	stats->imissed += port_stats.rx_drop_bcast;
269 	stats->imissed += port_stats.rx_drop_mcast;
270 	stats->imissed += port_stats.rx_drop_l3_bcast;
271 	stats->imissed += port_stats.rx_drop_l3_mcast;
272 
273 	stats->obytes = port_stats.tx_bytes_ok;
274 	stats->opackets = port_stats.tx_ucast_frames_ok;
275 	stats->opackets += port_stats.tx_bcast_frames_ok;
276 	stats->opackets += port_stats.tx_mcast_frames_ok;
277 	stats->oerrors = port_stats.tx_drops;
278 }
279 
280 static const uint32_t *
281 nicvf_dev_supported_ptypes_get(struct rte_eth_dev *dev)
282 {
283 	size_t copied;
284 	static uint32_t ptypes[32];
285 	struct nicvf *nic = nicvf_pmd_priv(dev);
286 	static const uint32_t ptypes_common[] = {
287 		RTE_PTYPE_L3_IPV4,
288 		RTE_PTYPE_L3_IPV4_EXT,
289 		RTE_PTYPE_L3_IPV6,
290 		RTE_PTYPE_L3_IPV6_EXT,
291 		RTE_PTYPE_L4_TCP,
292 		RTE_PTYPE_L4_UDP,
293 		RTE_PTYPE_L4_FRAG,
294 	};
295 	static const uint32_t ptypes_tunnel[] = {
296 		RTE_PTYPE_TUNNEL_GRE,
297 		RTE_PTYPE_TUNNEL_GENEVE,
298 		RTE_PTYPE_TUNNEL_VXLAN,
299 		RTE_PTYPE_TUNNEL_NVGRE,
300 	};
301 	static const uint32_t ptypes_end = RTE_PTYPE_UNKNOWN;
302 
303 	copied = sizeof(ptypes_common);
304 	memcpy(ptypes, ptypes_common, copied);
305 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
306 		memcpy((char *)ptypes + copied, ptypes_tunnel,
307 			sizeof(ptypes_tunnel));
308 		copied += sizeof(ptypes_tunnel);
309 	}
310 
311 	memcpy((char *)ptypes + copied, &ptypes_end, sizeof(ptypes_end));
312 	if (dev->rx_pkt_burst == nicvf_recv_pkts ||
313 		dev->rx_pkt_burst == nicvf_recv_pkts_multiseg)
314 		return ptypes;
315 
316 	return NULL;
317 }
318 
319 static void
320 nicvf_dev_stats_reset(struct rte_eth_dev *dev)
321 {
322 	int i;
323 	uint16_t rxqs = 0, txqs = 0;
324 	struct nicvf *nic = nicvf_pmd_priv(dev);
325 
326 	for (i = 0; i < dev->data->nb_rx_queues; i++)
327 		rxqs |= (0x3 << (i * 2));
328 	for (i = 0; i < dev->data->nb_tx_queues; i++)
329 		txqs |= (0x3 << (i * 2));
330 
331 	nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, rxqs, txqs);
332 }
333 
334 /* Promiscuous mode enabled by default in LMAC to VF 1:1 map configuration */
335 static void
336 nicvf_dev_promisc_enable(struct rte_eth_dev *dev __rte_unused)
337 {
338 }
339 
340 static inline uint64_t
341 nicvf_rss_ethdev_to_nic(struct nicvf *nic, uint64_t ethdev_rss)
342 {
343 	uint64_t nic_rss = 0;
344 
345 	if (ethdev_rss & ETH_RSS_IPV4)
346 		nic_rss |= RSS_IP_ENA;
347 
348 	if (ethdev_rss & ETH_RSS_IPV6)
349 		nic_rss |= RSS_IP_ENA;
350 
351 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_UDP)
352 		nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
353 
354 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_TCP)
355 		nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
356 
357 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_UDP)
358 		nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
359 
360 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_TCP)
361 		nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
362 
363 	if (ethdev_rss & ETH_RSS_PORT)
364 		nic_rss |= RSS_L2_EXTENDED_HASH_ENA;
365 
366 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
367 		if (ethdev_rss & ETH_RSS_VXLAN)
368 			nic_rss |= RSS_TUN_VXLAN_ENA;
369 
370 		if (ethdev_rss & ETH_RSS_GENEVE)
371 			nic_rss |= RSS_TUN_GENEVE_ENA;
372 
373 		if (ethdev_rss & ETH_RSS_NVGRE)
374 			nic_rss |= RSS_TUN_NVGRE_ENA;
375 	}
376 
377 	return nic_rss;
378 }
379 
380 static inline uint64_t
381 nicvf_rss_nic_to_ethdev(struct nicvf *nic,  uint64_t nic_rss)
382 {
383 	uint64_t ethdev_rss = 0;
384 
385 	if (nic_rss & RSS_IP_ENA)
386 		ethdev_rss |= (ETH_RSS_IPV4 | ETH_RSS_IPV6);
387 
388 	if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_TCP_ENA))
389 		ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_TCP |
390 				ETH_RSS_NONFRAG_IPV6_TCP);
391 
392 	if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_UDP_ENA))
393 		ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_UDP |
394 				ETH_RSS_NONFRAG_IPV6_UDP);
395 
396 	if (nic_rss & RSS_L2_EXTENDED_HASH_ENA)
397 		ethdev_rss |= ETH_RSS_PORT;
398 
399 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
400 		if (nic_rss & RSS_TUN_VXLAN_ENA)
401 			ethdev_rss |= ETH_RSS_VXLAN;
402 
403 		if (nic_rss & RSS_TUN_GENEVE_ENA)
404 			ethdev_rss |= ETH_RSS_GENEVE;
405 
406 		if (nic_rss & RSS_TUN_NVGRE_ENA)
407 			ethdev_rss |= ETH_RSS_NVGRE;
408 	}
409 	return ethdev_rss;
410 }
411 
412 static int
413 nicvf_dev_reta_query(struct rte_eth_dev *dev,
414 		     struct rte_eth_rss_reta_entry64 *reta_conf,
415 		     uint16_t reta_size)
416 {
417 	struct nicvf *nic = nicvf_pmd_priv(dev);
418 	uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
419 	int ret, i, j;
420 
421 	if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
422 		RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
423 			"(%d) doesn't match the number hardware can supported "
424 			"(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
425 		return -EINVAL;
426 	}
427 
428 	ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
429 	if (ret)
430 		return ret;
431 
432 	/* Copy RETA table */
433 	for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
434 		for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
435 			if ((reta_conf[i].mask >> j) & 0x01)
436 				reta_conf[i].reta[j] = tbl[j];
437 	}
438 
439 	return 0;
440 }
441 
442 static int
443 nicvf_dev_reta_update(struct rte_eth_dev *dev,
444 		      struct rte_eth_rss_reta_entry64 *reta_conf,
445 		      uint16_t reta_size)
446 {
447 	struct nicvf *nic = nicvf_pmd_priv(dev);
448 	uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
449 	int ret, i, j;
450 
451 	if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
452 		RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
453 			"(%d) doesn't match the number hardware can supported "
454 			"(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
455 		return -EINVAL;
456 	}
457 
458 	ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
459 	if (ret)
460 		return ret;
461 
462 	/* Copy RETA table */
463 	for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
464 		for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
465 			if ((reta_conf[i].mask >> j) & 0x01)
466 				tbl[j] = reta_conf[i].reta[j];
467 	}
468 
469 	return nicvf_rss_reta_update(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
470 }
471 
472 static int
473 nicvf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
474 			    struct rte_eth_rss_conf *rss_conf)
475 {
476 	struct nicvf *nic = nicvf_pmd_priv(dev);
477 
478 	if (rss_conf->rss_key)
479 		nicvf_rss_get_key(nic, rss_conf->rss_key);
480 
481 	rss_conf->rss_key_len =  RSS_HASH_KEY_BYTE_SIZE;
482 	rss_conf->rss_hf = nicvf_rss_nic_to_ethdev(nic, nicvf_rss_get_cfg(nic));
483 	return 0;
484 }
485 
486 static int
487 nicvf_dev_rss_hash_update(struct rte_eth_dev *dev,
488 			  struct rte_eth_rss_conf *rss_conf)
489 {
490 	struct nicvf *nic = nicvf_pmd_priv(dev);
491 	uint64_t nic_rss;
492 
493 	if (rss_conf->rss_key &&
494 		rss_conf->rss_key_len != RSS_HASH_KEY_BYTE_SIZE) {
495 		RTE_LOG(ERR, PMD, "Hash key size mismatch %d",
496 				rss_conf->rss_key_len);
497 		return -EINVAL;
498 	}
499 
500 	if (rss_conf->rss_key)
501 		nicvf_rss_set_key(nic, rss_conf->rss_key);
502 
503 	nic_rss = nicvf_rss_ethdev_to_nic(nic, rss_conf->rss_hf);
504 	nicvf_rss_set_cfg(nic, nic_rss);
505 	return 0;
506 }
507 
508 static int
509 nicvf_qset_cq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
510 		    struct nicvf_rxq *rxq, uint16_t qidx, uint32_t desc_cnt)
511 {
512 	const struct rte_memzone *rz;
513 	uint32_t ring_size = CMP_QUEUE_SZ_MAX * sizeof(union cq_entry_t);
514 
515 	rz = rte_eth_dma_zone_reserve(dev, "cq_ring",
516 				      nicvf_netdev_qidx(nic, qidx), ring_size,
517 				      NICVF_CQ_BASE_ALIGN_BYTES, nic->node);
518 	if (rz == NULL) {
519 		PMD_INIT_LOG(ERR, "Failed to allocate mem for cq hw ring");
520 		return -ENOMEM;
521 	}
522 
523 	memset(rz->addr, 0, ring_size);
524 
525 	rxq->phys = rz->phys_addr;
526 	rxq->desc = rz->addr;
527 	rxq->qlen_mask = desc_cnt - 1;
528 
529 	return 0;
530 }
531 
532 static int
533 nicvf_qset_sq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
534 		    struct nicvf_txq *sq, uint16_t qidx, uint32_t desc_cnt)
535 {
536 	const struct rte_memzone *rz;
537 	uint32_t ring_size = SND_QUEUE_SZ_MAX * sizeof(union sq_entry_t);
538 
539 	rz = rte_eth_dma_zone_reserve(dev, "sq",
540 				      nicvf_netdev_qidx(nic, qidx), ring_size,
541 				      NICVF_SQ_BASE_ALIGN_BYTES, nic->node);
542 	if (rz == NULL) {
543 		PMD_INIT_LOG(ERR, "Failed allocate mem for sq hw ring");
544 		return -ENOMEM;
545 	}
546 
547 	memset(rz->addr, 0, ring_size);
548 
549 	sq->phys = rz->phys_addr;
550 	sq->desc = rz->addr;
551 	sq->qlen_mask = desc_cnt - 1;
552 
553 	return 0;
554 }
555 
556 static int
557 nicvf_qset_rbdr_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
558 		      uint32_t desc_cnt, uint32_t buffsz)
559 {
560 	struct nicvf_rbdr *rbdr;
561 	const struct rte_memzone *rz;
562 	uint32_t ring_size;
563 
564 	assert(nic->rbdr == NULL);
565 	rbdr = rte_zmalloc_socket("rbdr", sizeof(struct nicvf_rbdr),
566 				  RTE_CACHE_LINE_SIZE, nic->node);
567 	if (rbdr == NULL) {
568 		PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr");
569 		return -ENOMEM;
570 	}
571 
572 	ring_size = sizeof(struct rbdr_entry_t) * RBDR_QUEUE_SZ_MAX;
573 	rz = rte_eth_dma_zone_reserve(dev, "rbdr",
574 				      nicvf_netdev_qidx(nic, 0), ring_size,
575 				      NICVF_RBDR_BASE_ALIGN_BYTES, nic->node);
576 	if (rz == NULL) {
577 		PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr desc ring");
578 		return -ENOMEM;
579 	}
580 
581 	memset(rz->addr, 0, ring_size);
582 
583 	rbdr->phys = rz->phys_addr;
584 	rbdr->tail = 0;
585 	rbdr->next_tail = 0;
586 	rbdr->desc = rz->addr;
587 	rbdr->buffsz = buffsz;
588 	rbdr->qlen_mask = desc_cnt - 1;
589 	rbdr->rbdr_status =
590 		nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_STATUS0;
591 	rbdr->rbdr_door =
592 		nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_DOOR;
593 
594 	nic->rbdr = rbdr;
595 	return 0;
596 }
597 
598 static void
599 nicvf_rbdr_release_mbuf(struct rte_eth_dev *dev, struct nicvf *nic __rte_unused,
600 			nicvf_phys_addr_t phy)
601 {
602 	uint16_t qidx;
603 	void *obj;
604 	struct nicvf_rxq *rxq;
605 
606 	for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
607 		rxq = dev->data->rx_queues[qidx];
608 		if (rxq->precharge_cnt) {
609 			obj = (void *)nicvf_mbuff_phy2virt(phy,
610 							   rxq->mbuf_phys_off);
611 			rte_mempool_put(rxq->pool, obj);
612 			rxq->precharge_cnt--;
613 			break;
614 		}
615 	}
616 }
617 
618 static inline void
619 nicvf_rbdr_release_mbufs(struct rte_eth_dev *dev, struct nicvf *nic)
620 {
621 	uint32_t qlen_mask, head;
622 	struct rbdr_entry_t *entry;
623 	struct nicvf_rbdr *rbdr = nic->rbdr;
624 
625 	qlen_mask = rbdr->qlen_mask;
626 	head = rbdr->head;
627 	while (head != rbdr->tail) {
628 		entry = rbdr->desc + head;
629 		nicvf_rbdr_release_mbuf(dev, nic, entry->full_addr);
630 		head++;
631 		head = head & qlen_mask;
632 	}
633 }
634 
635 static inline void
636 nicvf_tx_queue_release_mbufs(struct nicvf_txq *txq)
637 {
638 	uint32_t head;
639 
640 	head = txq->head;
641 	while (head != txq->tail) {
642 		if (txq->txbuffs[head]) {
643 			rte_pktmbuf_free_seg(txq->txbuffs[head]);
644 			txq->txbuffs[head] = NULL;
645 		}
646 		head++;
647 		head = head & txq->qlen_mask;
648 	}
649 }
650 
651 static void
652 nicvf_tx_queue_reset(struct nicvf_txq *txq)
653 {
654 	uint32_t txq_desc_cnt = txq->qlen_mask + 1;
655 
656 	memset(txq->desc, 0, sizeof(union sq_entry_t) * txq_desc_cnt);
657 	memset(txq->txbuffs, 0, sizeof(struct rte_mbuf *) * txq_desc_cnt);
658 	txq->tail = 0;
659 	txq->head = 0;
660 	txq->xmit_bufs = 0;
661 }
662 
663 static inline int
664 nicvf_vf_start_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
665 			uint16_t qidx)
666 {
667 	struct nicvf_txq *txq;
668 	int ret;
669 
670 	assert(qidx < MAX_SND_QUEUES_PER_QS);
671 
672 	if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
673 		RTE_ETH_QUEUE_STATE_STARTED)
674 		return 0;
675 
676 	txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
677 	txq->pool = NULL;
678 	ret = nicvf_qset_sq_config(nic, qidx, txq);
679 	if (ret) {
680 		PMD_INIT_LOG(ERR, "Failed to configure sq VF%d %d %d",
681 			     nic->vf_id, qidx, ret);
682 		goto config_sq_error;
683 	}
684 
685 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
686 		RTE_ETH_QUEUE_STATE_STARTED;
687 	return ret;
688 
689 config_sq_error:
690 	nicvf_qset_sq_reclaim(nic, qidx);
691 	return ret;
692 }
693 
694 static inline int
695 nicvf_vf_stop_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
696 		       uint16_t qidx)
697 {
698 	struct nicvf_txq *txq;
699 	int ret;
700 
701 	assert(qidx < MAX_SND_QUEUES_PER_QS);
702 
703 	if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
704 		RTE_ETH_QUEUE_STATE_STOPPED)
705 		return 0;
706 
707 	ret = nicvf_qset_sq_reclaim(nic, qidx);
708 	if (ret)
709 		PMD_INIT_LOG(ERR, "Failed to reclaim sq VF%d %d %d",
710 			     nic->vf_id, qidx, ret);
711 
712 	txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
713 	nicvf_tx_queue_release_mbufs(txq);
714 	nicvf_tx_queue_reset(txq);
715 
716 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
717 		RTE_ETH_QUEUE_STATE_STOPPED;
718 	return ret;
719 }
720 
721 static inline int
722 nicvf_configure_cpi(struct rte_eth_dev *dev)
723 {
724 	struct nicvf *nic = nicvf_pmd_priv(dev);
725 	uint16_t qidx, qcnt;
726 	int ret;
727 
728 	/* Count started rx queues */
729 	for (qidx = qcnt = 0; qidx < dev->data->nb_rx_queues; qidx++)
730 		if (dev->data->rx_queue_state[qidx] ==
731 		    RTE_ETH_QUEUE_STATE_STARTED)
732 			qcnt++;
733 
734 	nic->cpi_alg = CPI_ALG_NONE;
735 	ret = nicvf_mbox_config_cpi(nic, qcnt);
736 	if (ret)
737 		PMD_INIT_LOG(ERR, "Failed to configure CPI %d", ret);
738 
739 	return ret;
740 }
741 
742 static inline int
743 nicvf_configure_rss(struct rte_eth_dev *dev)
744 {
745 	struct nicvf *nic = nicvf_pmd_priv(dev);
746 	uint64_t rsshf;
747 	int ret = -EINVAL;
748 
749 	rsshf = nicvf_rss_ethdev_to_nic(nic,
750 			dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf);
751 	PMD_DRV_LOG(INFO, "mode=%d rx_queues=%d loopback=%d rsshf=0x%" PRIx64,
752 		    dev->data->dev_conf.rxmode.mq_mode,
753 		    dev->data->nb_rx_queues,
754 		    dev->data->dev_conf.lpbk_mode, rsshf);
755 
756 	if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_NONE)
757 		ret = nicvf_rss_term(nic);
758 	else if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS)
759 		ret = nicvf_rss_config(nic, dev->data->nb_rx_queues, rsshf);
760 	if (ret)
761 		PMD_INIT_LOG(ERR, "Failed to configure RSS %d", ret);
762 
763 	return ret;
764 }
765 
766 static int
767 nicvf_configure_rss_reta(struct rte_eth_dev *dev)
768 {
769 	struct nicvf *nic = nicvf_pmd_priv(dev);
770 	unsigned int idx, qmap_size;
771 	uint8_t qmap[RTE_MAX_QUEUES_PER_PORT];
772 	uint8_t default_reta[NIC_MAX_RSS_IDR_TBL_SIZE];
773 
774 	if (nic->cpi_alg != CPI_ALG_NONE)
775 		return -EINVAL;
776 
777 	/* Prepare queue map */
778 	for (idx = 0, qmap_size = 0; idx < dev->data->nb_rx_queues; idx++) {
779 		if (dev->data->rx_queue_state[idx] ==
780 				RTE_ETH_QUEUE_STATE_STARTED)
781 			qmap[qmap_size++] = idx;
782 	}
783 
784 	/* Update default RSS RETA */
785 	for (idx = 0; idx < NIC_MAX_RSS_IDR_TBL_SIZE; idx++)
786 		default_reta[idx] = qmap[idx % qmap_size];
787 
788 	return nicvf_rss_reta_update(nic, default_reta,
789 				     NIC_MAX_RSS_IDR_TBL_SIZE);
790 }
791 
792 static void
793 nicvf_dev_tx_queue_release(void *sq)
794 {
795 	struct nicvf_txq *txq;
796 
797 	PMD_INIT_FUNC_TRACE();
798 
799 	txq = (struct nicvf_txq *)sq;
800 	if (txq) {
801 		if (txq->txbuffs != NULL) {
802 			nicvf_tx_queue_release_mbufs(txq);
803 			rte_free(txq->txbuffs);
804 			txq->txbuffs = NULL;
805 		}
806 		rte_free(txq);
807 	}
808 }
809 
810 static void
811 nicvf_set_tx_function(struct rte_eth_dev *dev)
812 {
813 	struct nicvf_txq *txq;
814 	size_t i;
815 	bool multiseg = false;
816 
817 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
818 		txq = dev->data->tx_queues[i];
819 		if ((txq->txq_flags & ETH_TXQ_FLAGS_NOMULTSEGS) == 0) {
820 			multiseg = true;
821 			break;
822 		}
823 	}
824 
825 	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
826 	if (multiseg) {
827 		PMD_DRV_LOG(DEBUG, "Using multi-segment tx callback");
828 		dev->tx_pkt_burst = nicvf_xmit_pkts_multiseg;
829 	} else {
830 		PMD_DRV_LOG(DEBUG, "Using single-segment tx callback");
831 		dev->tx_pkt_burst = nicvf_xmit_pkts;
832 	}
833 
834 	if (txq->pool_free == nicvf_single_pool_free_xmited_buffers)
835 		PMD_DRV_LOG(DEBUG, "Using single-mempool tx free method");
836 	else
837 		PMD_DRV_LOG(DEBUG, "Using multi-mempool tx free method");
838 }
839 
840 static void
841 nicvf_set_rx_function(struct rte_eth_dev *dev)
842 {
843 	if (dev->data->scattered_rx) {
844 		PMD_DRV_LOG(DEBUG, "Using multi-segment rx callback");
845 		dev->rx_pkt_burst = nicvf_recv_pkts_multiseg;
846 	} else {
847 		PMD_DRV_LOG(DEBUG, "Using single-segment rx callback");
848 		dev->rx_pkt_burst = nicvf_recv_pkts;
849 	}
850 }
851 
852 static int
853 nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
854 			 uint16_t nb_desc, unsigned int socket_id,
855 			 const struct rte_eth_txconf *tx_conf)
856 {
857 	uint16_t tx_free_thresh;
858 	uint8_t is_single_pool;
859 	struct nicvf_txq *txq;
860 	struct nicvf *nic = nicvf_pmd_priv(dev);
861 
862 	PMD_INIT_FUNC_TRACE();
863 
864 	/* Socket id check */
865 	if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
866 		PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
867 		socket_id, nic->node);
868 
869 	/* Tx deferred start is not supported */
870 	if (tx_conf->tx_deferred_start) {
871 		PMD_INIT_LOG(ERR, "Tx deferred start not supported");
872 		return -EINVAL;
873 	}
874 
875 	/* Roundup nb_desc to available qsize and validate max number of desc */
876 	nb_desc = nicvf_qsize_sq_roundup(nb_desc);
877 	if (nb_desc == 0) {
878 		PMD_INIT_LOG(ERR, "Value of nb_desc beyond available sq qsize");
879 		return -EINVAL;
880 	}
881 
882 	/* Validate tx_free_thresh */
883 	tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
884 				tx_conf->tx_free_thresh :
885 				NICVF_DEFAULT_TX_FREE_THRESH);
886 
887 	if (tx_free_thresh > (nb_desc) ||
888 		tx_free_thresh > NICVF_MAX_TX_FREE_THRESH) {
889 		PMD_INIT_LOG(ERR,
890 			"tx_free_thresh must be less than the number of TX "
891 			"descriptors. (tx_free_thresh=%u port=%d "
892 			"queue=%d)", (unsigned int)tx_free_thresh,
893 			(int)dev->data->port_id, (int)qidx);
894 		return -EINVAL;
895 	}
896 
897 	/* Free memory prior to re-allocation if needed. */
898 	if (dev->data->tx_queues[qidx] != NULL) {
899 		PMD_TX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
900 				qidx);
901 		nicvf_dev_tx_queue_release(dev->data->tx_queues[qidx]);
902 		dev->data->tx_queues[qidx] = NULL;
903 	}
904 
905 	/* Allocating tx queue data structure */
906 	txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct nicvf_txq),
907 					RTE_CACHE_LINE_SIZE, nic->node);
908 	if (txq == NULL) {
909 		PMD_INIT_LOG(ERR, "Failed to allocate txq=%d", qidx);
910 		return -ENOMEM;
911 	}
912 
913 	txq->nic = nic;
914 	txq->queue_id = qidx;
915 	txq->tx_free_thresh = tx_free_thresh;
916 	txq->txq_flags = tx_conf->txq_flags;
917 	txq->sq_head = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_HEAD;
918 	txq->sq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_DOOR;
919 	is_single_pool = (txq->txq_flags & ETH_TXQ_FLAGS_NOREFCOUNT &&
920 				txq->txq_flags & ETH_TXQ_FLAGS_NOMULTMEMP);
921 
922 	/* Choose optimum free threshold value for multipool case */
923 	if (!is_single_pool) {
924 		txq->tx_free_thresh = (uint16_t)
925 		(tx_conf->tx_free_thresh == NICVF_DEFAULT_TX_FREE_THRESH ?
926 				NICVF_TX_FREE_MPOOL_THRESH :
927 				tx_conf->tx_free_thresh);
928 		txq->pool_free = nicvf_multi_pool_free_xmited_buffers;
929 	} else {
930 		txq->pool_free = nicvf_single_pool_free_xmited_buffers;
931 	}
932 
933 	/* Allocate software ring */
934 	txq->txbuffs = rte_zmalloc_socket("txq->txbuffs",
935 				nb_desc * sizeof(struct rte_mbuf *),
936 				RTE_CACHE_LINE_SIZE, nic->node);
937 
938 	if (txq->txbuffs == NULL) {
939 		nicvf_dev_tx_queue_release(txq);
940 		return -ENOMEM;
941 	}
942 
943 	if (nicvf_qset_sq_alloc(dev, nic, txq, qidx, nb_desc)) {
944 		PMD_INIT_LOG(ERR, "Failed to allocate mem for sq %d", qidx);
945 		nicvf_dev_tx_queue_release(txq);
946 		return -ENOMEM;
947 	}
948 
949 	nicvf_tx_queue_reset(txq);
950 
951 	PMD_TX_LOG(DEBUG, "[%d] txq=%p nb_desc=%d desc=%p phys=0x%" PRIx64,
952 			qidx, txq, nb_desc, txq->desc, txq->phys);
953 
954 	dev->data->tx_queues[qidx] = txq;
955 	dev->data->tx_queue_state[qidx] = RTE_ETH_QUEUE_STATE_STOPPED;
956 	return 0;
957 }
958 
959 static inline void
960 nicvf_rx_queue_release_mbufs(struct rte_eth_dev *dev, struct nicvf_rxq *rxq)
961 {
962 	uint32_t rxq_cnt;
963 	uint32_t nb_pkts, released_pkts = 0;
964 	uint32_t refill_cnt = 0;
965 	struct rte_mbuf *rx_pkts[NICVF_MAX_RX_FREE_THRESH];
966 
967 	if (dev->rx_pkt_burst == NULL)
968 		return;
969 
970 	while ((rxq_cnt = nicvf_dev_rx_queue_count(dev, rxq->queue_id))) {
971 		nb_pkts = dev->rx_pkt_burst(rxq, rx_pkts,
972 					NICVF_MAX_RX_FREE_THRESH);
973 		PMD_DRV_LOG(INFO, "nb_pkts=%d  rxq_cnt=%d", nb_pkts, rxq_cnt);
974 		while (nb_pkts) {
975 			rte_pktmbuf_free_seg(rx_pkts[--nb_pkts]);
976 			released_pkts++;
977 		}
978 	}
979 
980 	refill_cnt += nicvf_dev_rbdr_refill(dev, rxq->queue_id);
981 	PMD_DRV_LOG(INFO, "free_cnt=%d  refill_cnt=%d",
982 		    released_pkts, refill_cnt);
983 }
984 
985 static void
986 nicvf_rx_queue_reset(struct nicvf_rxq *rxq)
987 {
988 	rxq->head = 0;
989 	rxq->available_space = 0;
990 	rxq->recv_buffers = 0;
991 }
992 
993 static inline int
994 nicvf_vf_start_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
995 			uint16_t qidx)
996 {
997 	struct nicvf_rxq *rxq;
998 	int ret;
999 
1000 	assert(qidx < MAX_RCV_QUEUES_PER_QS);
1001 
1002 	if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
1003 		RTE_ETH_QUEUE_STATE_STARTED)
1004 		return 0;
1005 
1006 	/* Update rbdr pointer to all rxq */
1007 	rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
1008 	rxq->shared_rbdr = nic->rbdr;
1009 
1010 	ret = nicvf_qset_rq_config(nic, qidx, rxq);
1011 	if (ret) {
1012 		PMD_INIT_LOG(ERR, "Failed to configure rq VF%d %d %d",
1013 			     nic->vf_id, qidx, ret);
1014 		goto config_rq_error;
1015 	}
1016 	ret = nicvf_qset_cq_config(nic, qidx, rxq);
1017 	if (ret) {
1018 		PMD_INIT_LOG(ERR, "Failed to configure cq VF%d %d %d",
1019 			     nic->vf_id, qidx, ret);
1020 		goto config_cq_error;
1021 	}
1022 
1023 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1024 		RTE_ETH_QUEUE_STATE_STARTED;
1025 	return 0;
1026 
1027 config_cq_error:
1028 	nicvf_qset_cq_reclaim(nic, qidx);
1029 config_rq_error:
1030 	nicvf_qset_rq_reclaim(nic, qidx);
1031 	return ret;
1032 }
1033 
1034 static inline int
1035 nicvf_vf_stop_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
1036 		       uint16_t qidx)
1037 {
1038 	struct nicvf_rxq *rxq;
1039 	int ret, other_error;
1040 
1041 	if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
1042 		RTE_ETH_QUEUE_STATE_STOPPED)
1043 		return 0;
1044 
1045 	ret = nicvf_qset_rq_reclaim(nic, qidx);
1046 	if (ret)
1047 		PMD_INIT_LOG(ERR, "Failed to reclaim rq VF%d %d %d",
1048 			     nic->vf_id, qidx, ret);
1049 
1050 	other_error = ret;
1051 	rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
1052 	nicvf_rx_queue_release_mbufs(dev, rxq);
1053 	nicvf_rx_queue_reset(rxq);
1054 
1055 	ret = nicvf_qset_cq_reclaim(nic, qidx);
1056 	if (ret)
1057 		PMD_INIT_LOG(ERR, "Failed to reclaim cq VF%d %d %d",
1058 			     nic->vf_id, qidx, ret);
1059 
1060 	other_error |= ret;
1061 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1062 		RTE_ETH_QUEUE_STATE_STOPPED;
1063 	return other_error;
1064 }
1065 
1066 static void
1067 nicvf_dev_rx_queue_release(void *rx_queue)
1068 {
1069 	PMD_INIT_FUNC_TRACE();
1070 
1071 	rte_free(rx_queue);
1072 }
1073 
1074 static int
1075 nicvf_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
1076 {
1077 	struct nicvf *nic = nicvf_pmd_priv(dev);
1078 	int ret;
1079 
1080 	if (qidx >= MAX_RCV_QUEUES_PER_QS)
1081 		nic = nic->snicvf[(qidx / MAX_RCV_QUEUES_PER_QS - 1)];
1082 
1083 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1084 
1085 	ret = nicvf_vf_start_rx_queue(dev, nic, qidx);
1086 	if (ret)
1087 		return ret;
1088 
1089 	ret = nicvf_configure_cpi(dev);
1090 	if (ret)
1091 		return ret;
1092 
1093 	return nicvf_configure_rss_reta(dev);
1094 }
1095 
1096 static int
1097 nicvf_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
1098 {
1099 	int ret;
1100 	struct nicvf *nic = nicvf_pmd_priv(dev);
1101 
1102 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1103 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1104 
1105 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1106 
1107 	ret = nicvf_vf_stop_rx_queue(dev, nic, qidx);
1108 	ret |= nicvf_configure_cpi(dev);
1109 	ret |= nicvf_configure_rss_reta(dev);
1110 	return ret;
1111 }
1112 
1113 static int
1114 nicvf_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
1115 {
1116 	struct nicvf *nic = nicvf_pmd_priv(dev);
1117 
1118 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1119 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1120 
1121 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
1122 
1123 	return nicvf_vf_start_tx_queue(dev, nic, qidx);
1124 }
1125 
1126 static int
1127 nicvf_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
1128 {
1129 	struct nicvf *nic = nicvf_pmd_priv(dev);
1130 
1131 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1132 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1133 
1134 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
1135 
1136 	return nicvf_vf_stop_tx_queue(dev, nic, qidx);
1137 }
1138 
1139 
1140 static int
1141 nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
1142 			 uint16_t nb_desc, unsigned int socket_id,
1143 			 const struct rte_eth_rxconf *rx_conf,
1144 			 struct rte_mempool *mp)
1145 {
1146 	uint16_t rx_free_thresh;
1147 	struct nicvf_rxq *rxq;
1148 	struct nicvf *nic = nicvf_pmd_priv(dev);
1149 
1150 	PMD_INIT_FUNC_TRACE();
1151 
1152 	/* Socket id check */
1153 	if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
1154 		PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
1155 		socket_id, nic->node);
1156 
1157 	/* Mempool memory must be contiguous, so must be one memory segment*/
1158 	if (mp->nb_mem_chunks != 1) {
1159 		PMD_INIT_LOG(ERR, "Non-contiguous mempool, add more huge pages");
1160 		return -EINVAL;
1161 	}
1162 
1163 	/* Mempool memory must be physically contiguous */
1164 	if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG) {
1165 		PMD_INIT_LOG(ERR, "Mempool memory must be physically contiguous");
1166 		return -EINVAL;
1167 	}
1168 
1169 	/* Rx deferred start is not supported */
1170 	if (rx_conf->rx_deferred_start) {
1171 		PMD_INIT_LOG(ERR, "Rx deferred start not supported");
1172 		return -EINVAL;
1173 	}
1174 
1175 	/* Roundup nb_desc to available qsize and validate max number of desc */
1176 	nb_desc = nicvf_qsize_cq_roundup(nb_desc);
1177 	if (nb_desc == 0) {
1178 		PMD_INIT_LOG(ERR, "Value nb_desc beyond available hw cq qsize");
1179 		return -EINVAL;
1180 	}
1181 
1182 	/* Check rx_free_thresh upper bound */
1183 	rx_free_thresh = (uint16_t)((rx_conf->rx_free_thresh) ?
1184 				rx_conf->rx_free_thresh :
1185 				NICVF_DEFAULT_RX_FREE_THRESH);
1186 	if (rx_free_thresh > NICVF_MAX_RX_FREE_THRESH ||
1187 		rx_free_thresh >= nb_desc * .75) {
1188 		PMD_INIT_LOG(ERR, "rx_free_thresh greater than expected %d",
1189 				rx_free_thresh);
1190 		return -EINVAL;
1191 	}
1192 
1193 	/* Free memory prior to re-allocation if needed */
1194 	if (dev->data->rx_queues[qidx] != NULL) {
1195 		PMD_RX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
1196 				qidx);
1197 		nicvf_dev_rx_queue_release(dev->data->rx_queues[qidx]);
1198 		dev->data->rx_queues[qidx] = NULL;
1199 	}
1200 
1201 	/* Allocate rxq memory */
1202 	rxq = rte_zmalloc_socket("ethdev rx queue", sizeof(struct nicvf_rxq),
1203 					RTE_CACHE_LINE_SIZE, nic->node);
1204 	if (rxq == NULL) {
1205 		PMD_INIT_LOG(ERR, "Failed to allocate rxq=%d", qidx);
1206 		return -ENOMEM;
1207 	}
1208 
1209 	rxq->nic = nic;
1210 	rxq->pool = mp;
1211 	rxq->queue_id = qidx;
1212 	rxq->port_id = dev->data->port_id;
1213 	rxq->rx_free_thresh = rx_free_thresh;
1214 	rxq->rx_drop_en = rx_conf->rx_drop_en;
1215 	rxq->cq_status = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_STATUS;
1216 	rxq->cq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_DOOR;
1217 	rxq->precharge_cnt = 0;
1218 
1219 	if (nicvf_hw_cap(nic) & NICVF_CAP_CQE_RX2)
1220 		rxq->rbptr_offset = NICVF_CQE_RX2_RBPTR_WORD;
1221 	else
1222 		rxq->rbptr_offset = NICVF_CQE_RBPTR_WORD;
1223 
1224 
1225 	/* Alloc completion queue */
1226 	if (nicvf_qset_cq_alloc(dev, nic, rxq, rxq->queue_id, nb_desc)) {
1227 		PMD_INIT_LOG(ERR, "failed to allocate cq %u", rxq->queue_id);
1228 		nicvf_dev_rx_queue_release(rxq);
1229 		return -ENOMEM;
1230 	}
1231 
1232 	nicvf_rx_queue_reset(rxq);
1233 
1234 	PMD_RX_LOG(DEBUG, "[%d] rxq=%p pool=%s nb_desc=(%d/%d) phy=%" PRIx64,
1235 			qidx, rxq, mp->name, nb_desc,
1236 			rte_mempool_avail_count(mp), rxq->phys);
1237 
1238 	dev->data->rx_queues[qidx] = rxq;
1239 	dev->data->rx_queue_state[qidx] = RTE_ETH_QUEUE_STATE_STOPPED;
1240 	return 0;
1241 }
1242 
1243 static void
1244 nicvf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1245 {
1246 	struct nicvf *nic = nicvf_pmd_priv(dev);
1247 
1248 	PMD_INIT_FUNC_TRACE();
1249 
1250 	dev_info->min_rx_bufsize = ETHER_MIN_MTU;
1251 	dev_info->max_rx_pktlen = NIC_HW_MAX_FRS;
1252 	dev_info->max_rx_queues = (uint16_t)MAX_RCV_QUEUES_PER_QS;
1253 	dev_info->max_tx_queues = (uint16_t)MAX_SND_QUEUES_PER_QS;
1254 	dev_info->max_mac_addrs = 1;
1255 	dev_info->max_vfs = dev->pci_dev->max_vfs;
1256 
1257 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
1258 	dev_info->tx_offload_capa =
1259 		DEV_TX_OFFLOAD_IPV4_CKSUM  |
1260 		DEV_TX_OFFLOAD_UDP_CKSUM   |
1261 		DEV_TX_OFFLOAD_TCP_CKSUM   |
1262 		DEV_TX_OFFLOAD_TCP_TSO     |
1263 		DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;
1264 
1265 	dev_info->reta_size = nic->rss_info.rss_size;
1266 	dev_info->hash_key_size = RSS_HASH_KEY_BYTE_SIZE;
1267 	dev_info->flow_type_rss_offloads = NICVF_RSS_OFFLOAD_PASS1;
1268 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING)
1269 		dev_info->flow_type_rss_offloads |= NICVF_RSS_OFFLOAD_TUNNEL;
1270 
1271 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
1272 		.rx_free_thresh = NICVF_DEFAULT_RX_FREE_THRESH,
1273 		.rx_drop_en = 0,
1274 	};
1275 
1276 	dev_info->default_txconf = (struct rte_eth_txconf) {
1277 		.tx_free_thresh = NICVF_DEFAULT_TX_FREE_THRESH,
1278 		.txq_flags =
1279 			ETH_TXQ_FLAGS_NOMULTSEGS  |
1280 			ETH_TXQ_FLAGS_NOREFCOUNT  |
1281 			ETH_TXQ_FLAGS_NOMULTMEMP  |
1282 			ETH_TXQ_FLAGS_NOVLANOFFL  |
1283 			ETH_TXQ_FLAGS_NOXSUMSCTP,
1284 	};
1285 }
1286 
1287 static nicvf_phys_addr_t
1288 rbdr_rte_mempool_get(void *dev, void *opaque)
1289 {
1290 	uint16_t qidx;
1291 	uintptr_t mbuf;
1292 	struct nicvf_rxq *rxq;
1293 	struct rte_eth_dev *eth_dev = (struct rte_eth_dev *)dev;
1294 	struct nicvf *nic __rte_unused = (struct nicvf *)opaque;
1295 
1296 	for (qidx = 0; qidx < eth_dev->data->nb_rx_queues; qidx++) {
1297 		rxq = eth_dev->data->rx_queues[qidx];
1298 		/* Maintain equal buffer count across all pools */
1299 		if (rxq->precharge_cnt >= rxq->qlen_mask)
1300 			continue;
1301 		rxq->precharge_cnt++;
1302 		mbuf = (uintptr_t)rte_pktmbuf_alloc(rxq->pool);
1303 		if (mbuf)
1304 			return nicvf_mbuff_virt2phy(mbuf, rxq->mbuf_phys_off);
1305 	}
1306 	return 0;
1307 }
1308 
1309 static int
1310 nicvf_vf_start(struct rte_eth_dev *dev, struct nicvf *nic, uint32_t rbdrsz)
1311 {
1312 	int ret;
1313 	uint16_t qidx;
1314 	uint32_t total_rxq_desc, nb_rbdr_desc, exp_buffs;
1315 	uint64_t mbuf_phys_off = 0;
1316 	struct nicvf_rxq *rxq;
1317 	struct rte_mbuf *mbuf;
1318 	uint16_t rx_start, rx_end;
1319 	uint16_t tx_start, tx_end;
1320 
1321 	PMD_INIT_FUNC_TRACE();
1322 
1323 	/* Userspace process exited without proper shutdown in last run */
1324 	if (nicvf_qset_rbdr_active(nic, 0))
1325 		nicvf_vf_stop(dev, nic, false);
1326 
1327 	/* Get queue ranges for this VF */
1328 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
1329 
1330 	/*
1331 	 * Thunderx nicvf PMD can support more than one pool per port only when
1332 	 * 1) Data payload size is same across all the pools in given port
1333 	 * AND
1334 	 * 2) All mbuffs in the pools are from the same hugepage
1335 	 * AND
1336 	 * 3) Mbuff metadata size is same across all the pools in given port
1337 	 *
1338 	 * This is to support existing application that uses multiple pool/port.
1339 	 * But, the purpose of using multipool for QoS will not be addressed.
1340 	 *
1341 	 */
1342 
1343 	/* Validate mempool attributes */
1344 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1345 		rxq = dev->data->rx_queues[qidx];
1346 		rxq->mbuf_phys_off = nicvf_mempool_phy_offset(rxq->pool);
1347 		mbuf = rte_pktmbuf_alloc(rxq->pool);
1348 		if (mbuf == NULL) {
1349 			PMD_INIT_LOG(ERR, "Failed allocate mbuf VF%d qid=%d "
1350 				     "pool=%s",
1351 				     nic->vf_id, qidx, rxq->pool->name);
1352 			return -ENOMEM;
1353 		}
1354 		rxq->mbuf_phys_off -= nicvf_mbuff_meta_length(mbuf);
1355 		rxq->mbuf_phys_off -= RTE_PKTMBUF_HEADROOM;
1356 		rte_pktmbuf_free(mbuf);
1357 
1358 		if (mbuf_phys_off == 0)
1359 			mbuf_phys_off = rxq->mbuf_phys_off;
1360 		if (mbuf_phys_off != rxq->mbuf_phys_off) {
1361 			PMD_INIT_LOG(ERR, "pool params not same,%s VF%d %"
1362 				     PRIx64, rxq->pool->name, nic->vf_id,
1363 				     mbuf_phys_off);
1364 			return -EINVAL;
1365 		}
1366 	}
1367 
1368 	/* Check the level of buffers in the pool */
1369 	total_rxq_desc = 0;
1370 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1371 		rxq = dev->data->rx_queues[qidx];
1372 		/* Count total numbers of rxq descs */
1373 		total_rxq_desc += rxq->qlen_mask + 1;
1374 		exp_buffs = RTE_MEMPOOL_CACHE_MAX_SIZE + rxq->rx_free_thresh;
1375 		exp_buffs *= dev->data->nb_rx_queues;
1376 		if (rte_mempool_avail_count(rxq->pool) < exp_buffs) {
1377 			PMD_INIT_LOG(ERR, "Buff shortage in pool=%s (%d/%d)",
1378 				     rxq->pool->name,
1379 				     rte_mempool_avail_count(rxq->pool),
1380 				     exp_buffs);
1381 			return -ENOENT;
1382 		}
1383 	}
1384 
1385 	/* Check RBDR desc overflow */
1386 	ret = nicvf_qsize_rbdr_roundup(total_rxq_desc);
1387 	if (ret == 0) {
1388 		PMD_INIT_LOG(ERR, "Reached RBDR desc limit, reduce nr desc "
1389 			     "VF%d", nic->vf_id);
1390 		return -ENOMEM;
1391 	}
1392 
1393 	/* Enable qset */
1394 	ret = nicvf_qset_config(nic);
1395 	if (ret) {
1396 		PMD_INIT_LOG(ERR, "Failed to enable qset %d VF%d", ret,
1397 			     nic->vf_id);
1398 		return ret;
1399 	}
1400 
1401 	/* Allocate RBDR and RBDR ring desc */
1402 	nb_rbdr_desc = nicvf_qsize_rbdr_roundup(total_rxq_desc);
1403 	ret = nicvf_qset_rbdr_alloc(dev, nic, nb_rbdr_desc, rbdrsz);
1404 	if (ret) {
1405 		PMD_INIT_LOG(ERR, "Failed to allocate memory for rbdr alloc "
1406 			     "VF%d", nic->vf_id);
1407 		goto qset_reclaim;
1408 	}
1409 
1410 	/* Enable and configure RBDR registers */
1411 	ret = nicvf_qset_rbdr_config(nic, 0);
1412 	if (ret) {
1413 		PMD_INIT_LOG(ERR, "Failed to configure rbdr %d VF%d", ret,
1414 			     nic->vf_id);
1415 		goto qset_rbdr_free;
1416 	}
1417 
1418 	/* Fill rte_mempool buffers in RBDR pool and precharge it */
1419 	ret = nicvf_qset_rbdr_precharge(dev, nic, 0, rbdr_rte_mempool_get,
1420 					total_rxq_desc);
1421 	if (ret) {
1422 		PMD_INIT_LOG(ERR, "Failed to fill rbdr %d VF%d", ret,
1423 			     nic->vf_id);
1424 		goto qset_rbdr_reclaim;
1425 	}
1426 
1427 	PMD_DRV_LOG(INFO, "Filled %d out of %d entries in RBDR VF%d",
1428 		     nic->rbdr->tail, nb_rbdr_desc, nic->vf_id);
1429 
1430 	/* Configure VLAN Strip */
1431 	nicvf_vlan_hw_strip(nic, dev->data->dev_conf.rxmode.hw_vlan_strip);
1432 
1433 	/* Get queue ranges for this VF */
1434 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
1435 
1436 	/* Configure TX queues */
1437 	for (qidx = tx_start; qidx <= tx_end; qidx++) {
1438 		ret = nicvf_vf_start_tx_queue(dev, nic,
1439 			qidx % MAX_SND_QUEUES_PER_QS);
1440 		if (ret)
1441 			goto start_txq_error;
1442 	}
1443 
1444 	/* Configure RX queues */
1445 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1446 		ret = nicvf_vf_start_rx_queue(dev, nic,
1447 			qidx % MAX_RCV_QUEUES_PER_QS);
1448 		if (ret)
1449 			goto start_rxq_error;
1450 	}
1451 
1452 	if (!nic->sqs_mode) {
1453 		/* Configure CPI algorithm */
1454 		ret = nicvf_configure_cpi(dev);
1455 		if (ret)
1456 			goto start_txq_error;
1457 
1458 		ret = nicvf_mbox_get_rss_size(nic);
1459 		if (ret) {
1460 			PMD_INIT_LOG(ERR, "Failed to get rss table size");
1461 			goto qset_rss_error;
1462 		}
1463 
1464 		/* Configure RSS */
1465 		ret = nicvf_configure_rss(dev);
1466 		if (ret)
1467 			goto qset_rss_error;
1468 	}
1469 
1470 	/* Done; Let PF make the BGX's RX and TX switches to ON position */
1471 	nicvf_mbox_cfg_done(nic);
1472 	return 0;
1473 
1474 qset_rss_error:
1475 	nicvf_rss_term(nic);
1476 start_rxq_error:
1477 	for (qidx = rx_start; qidx <= rx_end; qidx++)
1478 		nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
1479 start_txq_error:
1480 	for (qidx = tx_start; qidx <= tx_end; qidx++)
1481 		nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
1482 qset_rbdr_reclaim:
1483 	nicvf_qset_rbdr_reclaim(nic, 0);
1484 	nicvf_rbdr_release_mbufs(dev, nic);
1485 qset_rbdr_free:
1486 	if (nic->rbdr) {
1487 		rte_free(nic->rbdr);
1488 		nic->rbdr = NULL;
1489 	}
1490 qset_reclaim:
1491 	nicvf_qset_reclaim(nic);
1492 	return ret;
1493 }
1494 
1495 static int
1496 nicvf_dev_start(struct rte_eth_dev *dev)
1497 {
1498 	uint16_t qidx;
1499 	int ret;
1500 	size_t i;
1501 	struct nicvf *nic = nicvf_pmd_priv(dev);
1502 	struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
1503 	uint16_t mtu;
1504 	uint32_t buffsz = 0, rbdrsz = 0;
1505 	struct rte_pktmbuf_pool_private *mbp_priv;
1506 	struct nicvf_rxq *rxq;
1507 
1508 	PMD_INIT_FUNC_TRACE();
1509 
1510 	/* This function must be called for a primary device */
1511 	assert_primary(nic);
1512 
1513 	/* Validate RBDR buff size */
1514 	for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
1515 		rxq = dev->data->rx_queues[qidx];
1516 		mbp_priv = rte_mempool_get_priv(rxq->pool);
1517 		buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
1518 		if (buffsz % 128) {
1519 			PMD_INIT_LOG(ERR, "rxbuf size must be multiply of 128");
1520 			return -EINVAL;
1521 		}
1522 		if (rbdrsz == 0)
1523 			rbdrsz = buffsz;
1524 		if (rbdrsz != buffsz) {
1525 			PMD_INIT_LOG(ERR, "buffsz not same, qidx=%d (%d/%d)",
1526 				     qidx, rbdrsz, buffsz);
1527 			return -EINVAL;
1528 		}
1529 	}
1530 
1531 	/* Configure loopback */
1532 	ret = nicvf_loopback_config(nic, dev->data->dev_conf.lpbk_mode);
1533 	if (ret) {
1534 		PMD_INIT_LOG(ERR, "Failed to configure loopback %d", ret);
1535 		return ret;
1536 	}
1537 
1538 	/* Reset all statistics counters attached to this port */
1539 	ret = nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, 0xFFFF, 0xFFFF);
1540 	if (ret) {
1541 		PMD_INIT_LOG(ERR, "Failed to reset stat counters %d", ret);
1542 		return ret;
1543 	}
1544 
1545 	/* Setup scatter mode if needed by jumbo */
1546 	if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
1547 					    2 * VLAN_TAG_SIZE > buffsz)
1548 		dev->data->scattered_rx = 1;
1549 	if (rx_conf->enable_scatter)
1550 		dev->data->scattered_rx = 1;
1551 
1552 	/* Setup MTU based on max_rx_pkt_len or default */
1553 	mtu = dev->data->dev_conf.rxmode.jumbo_frame ?
1554 		dev->data->dev_conf.rxmode.max_rx_pkt_len
1555 			-  ETHER_HDR_LEN - ETHER_CRC_LEN
1556 		: ETHER_MTU;
1557 
1558 	if (nicvf_dev_set_mtu(dev, mtu)) {
1559 		PMD_INIT_LOG(ERR, "Failed to set default mtu size");
1560 		return -EBUSY;
1561 	}
1562 
1563 	ret = nicvf_vf_start(dev, nic, rbdrsz);
1564 	if (ret != 0)
1565 		return ret;
1566 
1567 	for (i = 0; i < nic->sqs_count; i++) {
1568 		assert(nic->snicvf[i]);
1569 
1570 		ret = nicvf_vf_start(dev, nic->snicvf[i], rbdrsz);
1571 		if (ret != 0)
1572 			return ret;
1573 	}
1574 
1575 	/* Configure callbacks based on scatter mode */
1576 	nicvf_set_tx_function(dev);
1577 	nicvf_set_rx_function(dev);
1578 
1579 	return 0;
1580 }
1581 
1582 static void
1583 nicvf_dev_stop_cleanup(struct rte_eth_dev *dev, bool cleanup)
1584 {
1585 	size_t i;
1586 	int ret;
1587 	struct nicvf *nic = nicvf_pmd_priv(dev);
1588 
1589 	PMD_INIT_FUNC_TRACE();
1590 
1591 	/* Teardown secondary vf first */
1592 	for (i = 0; i < nic->sqs_count; i++) {
1593 		if (!nic->snicvf[i])
1594 			continue;
1595 
1596 		nicvf_vf_stop(dev, nic->snicvf[i], cleanup);
1597 	}
1598 
1599 	/* Stop the primary VF now */
1600 	nicvf_vf_stop(dev, nic, cleanup);
1601 
1602 	/* Disable loopback */
1603 	ret = nicvf_loopback_config(nic, 0);
1604 	if (ret)
1605 		PMD_INIT_LOG(ERR, "Failed to disable loopback %d", ret);
1606 
1607 	/* Reclaim CPI configuration */
1608 	ret = nicvf_mbox_config_cpi(nic, 0);
1609 	if (ret)
1610 		PMD_INIT_LOG(ERR, "Failed to reclaim CPI config %d", ret);
1611 }
1612 
1613 static void
1614 nicvf_dev_stop(struct rte_eth_dev *dev)
1615 {
1616 	PMD_INIT_FUNC_TRACE();
1617 
1618 	nicvf_dev_stop_cleanup(dev, false);
1619 }
1620 
1621 static void
1622 nicvf_vf_stop(struct rte_eth_dev *dev, struct nicvf *nic, bool cleanup)
1623 {
1624 	int ret;
1625 	uint16_t qidx;
1626 	uint16_t tx_start, tx_end;
1627 	uint16_t rx_start, rx_end;
1628 
1629 	PMD_INIT_FUNC_TRACE();
1630 
1631 	if (cleanup) {
1632 		/* Let PF make the BGX's RX and TX switches to OFF position */
1633 		nicvf_mbox_shutdown(nic);
1634 	}
1635 
1636 	/* Disable VLAN Strip */
1637 	nicvf_vlan_hw_strip(nic, 0);
1638 
1639 	/* Get queue ranges for this VF */
1640 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
1641 
1642 	for (qidx = tx_start; qidx <= tx_end; qidx++)
1643 		nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
1644 
1645 	/* Get queue ranges for this VF */
1646 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
1647 
1648 	/* Reclaim rq */
1649 	for (qidx = rx_start; qidx <= rx_end; qidx++)
1650 		nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
1651 
1652 	/* Reclaim RBDR */
1653 	ret = nicvf_qset_rbdr_reclaim(nic, 0);
1654 	if (ret)
1655 		PMD_INIT_LOG(ERR, "Failed to reclaim RBDR %d", ret);
1656 
1657 	/* Move all charged buffers in RBDR back to pool */
1658 	if (nic->rbdr != NULL)
1659 		nicvf_rbdr_release_mbufs(dev, nic);
1660 
1661 	/* Disable qset */
1662 	ret = nicvf_qset_reclaim(nic);
1663 	if (ret)
1664 		PMD_INIT_LOG(ERR, "Failed to disable qset %d", ret);
1665 
1666 	/* Disable all interrupts */
1667 	nicvf_disable_all_interrupts(nic);
1668 
1669 	/* Free RBDR SW structure */
1670 	if (nic->rbdr) {
1671 		rte_free(nic->rbdr);
1672 		nic->rbdr = NULL;
1673 	}
1674 }
1675 
1676 static void
1677 nicvf_dev_close(struct rte_eth_dev *dev)
1678 {
1679 	size_t i;
1680 	struct nicvf *nic = nicvf_pmd_priv(dev);
1681 
1682 	PMD_INIT_FUNC_TRACE();
1683 
1684 	nicvf_dev_stop_cleanup(dev, true);
1685 	nicvf_periodic_alarm_stop(nicvf_interrupt, dev);
1686 
1687 	for (i = 0; i < nic->sqs_count; i++) {
1688 		if (!nic->snicvf[i])
1689 			continue;
1690 
1691 		nicvf_periodic_alarm_stop(nicvf_vf_interrupt, nic->snicvf[i]);
1692 	}
1693 }
1694 
1695 static int
1696 nicvf_request_sqs(struct nicvf *nic)
1697 {
1698 	size_t i;
1699 
1700 	assert_primary(nic);
1701 	assert(nic->sqs_count > 0);
1702 	assert(nic->sqs_count <= MAX_SQS_PER_VF);
1703 
1704 	/* Set no of Rx/Tx queues in each of the SQsets */
1705 	for (i = 0; i < nic->sqs_count; i++) {
1706 		if (nicvf_svf_empty())
1707 			rte_panic("Cannot assign sufficient number of "
1708 				  "secondary queues to primary VF%" PRIu8 "\n",
1709 				  nic->vf_id);
1710 
1711 		nic->snicvf[i] = nicvf_svf_pop();
1712 		nic->snicvf[i]->sqs_id = i;
1713 	}
1714 
1715 	return nicvf_mbox_request_sqs(nic);
1716 }
1717 
1718 static int
1719 nicvf_dev_configure(struct rte_eth_dev *dev)
1720 {
1721 	struct rte_eth_dev_data *data = dev->data;
1722 	struct rte_eth_conf *conf = &data->dev_conf;
1723 	struct rte_eth_rxmode *rxmode = &conf->rxmode;
1724 	struct rte_eth_txmode *txmode = &conf->txmode;
1725 	struct nicvf *nic = nicvf_pmd_priv(dev);
1726 	uint8_t cqcount;
1727 
1728 	PMD_INIT_FUNC_TRACE();
1729 
1730 	if (!rte_eal_has_hugepages()) {
1731 		PMD_INIT_LOG(INFO, "Huge page is not configured");
1732 		return -EINVAL;
1733 	}
1734 
1735 	if (txmode->mq_mode) {
1736 		PMD_INIT_LOG(INFO, "Tx mq_mode DCB or VMDq not supported");
1737 		return -EINVAL;
1738 	}
1739 
1740 	if (rxmode->mq_mode != ETH_MQ_RX_NONE &&
1741 		rxmode->mq_mode != ETH_MQ_RX_RSS) {
1742 		PMD_INIT_LOG(INFO, "Unsupported rx qmode %d", rxmode->mq_mode);
1743 		return -EINVAL;
1744 	}
1745 
1746 	if (!rxmode->hw_strip_crc) {
1747 		PMD_INIT_LOG(NOTICE, "Can't disable hw crc strip");
1748 		rxmode->hw_strip_crc = 1;
1749 	}
1750 
1751 	if (rxmode->hw_ip_checksum) {
1752 		PMD_INIT_LOG(NOTICE, "Rxcksum not supported");
1753 		rxmode->hw_ip_checksum = 0;
1754 	}
1755 
1756 	if (rxmode->split_hdr_size) {
1757 		PMD_INIT_LOG(INFO, "Rxmode does not support split header");
1758 		return -EINVAL;
1759 	}
1760 
1761 	if (rxmode->hw_vlan_filter) {
1762 		PMD_INIT_LOG(INFO, "VLAN filter not supported");
1763 		return -EINVAL;
1764 	}
1765 
1766 	if (rxmode->hw_vlan_extend) {
1767 		PMD_INIT_LOG(INFO, "VLAN extended not supported");
1768 		return -EINVAL;
1769 	}
1770 
1771 	if (rxmode->enable_lro) {
1772 		PMD_INIT_LOG(INFO, "LRO not supported");
1773 		return -EINVAL;
1774 	}
1775 
1776 	if (conf->link_speeds & ETH_LINK_SPEED_FIXED) {
1777 		PMD_INIT_LOG(INFO, "Setting link speed/duplex not supported");
1778 		return -EINVAL;
1779 	}
1780 
1781 	if (conf->dcb_capability_en) {
1782 		PMD_INIT_LOG(INFO, "DCB enable not supported");
1783 		return -EINVAL;
1784 	}
1785 
1786 	if (conf->fdir_conf.mode != RTE_FDIR_MODE_NONE) {
1787 		PMD_INIT_LOG(INFO, "Flow director not supported");
1788 		return -EINVAL;
1789 	}
1790 
1791 	assert_primary(nic);
1792 	NICVF_STATIC_ASSERT(MAX_RCV_QUEUES_PER_QS == MAX_SND_QUEUES_PER_QS);
1793 	cqcount = RTE_MAX(data->nb_tx_queues, data->nb_rx_queues);
1794 	if (cqcount > MAX_RCV_QUEUES_PER_QS) {
1795 		nic->sqs_count = RTE_ALIGN_CEIL(cqcount, MAX_RCV_QUEUES_PER_QS);
1796 		nic->sqs_count = (nic->sqs_count / MAX_RCV_QUEUES_PER_QS) - 1;
1797 	} else {
1798 		nic->sqs_count = 0;
1799 	}
1800 
1801 	assert(nic->sqs_count <= MAX_SQS_PER_VF);
1802 
1803 	if (nic->sqs_count > 0) {
1804 		if (nicvf_request_sqs(nic)) {
1805 			rte_panic("Cannot assign sufficient number of "
1806 				  "secondary queues to PORT%d VF%" PRIu8 "\n",
1807 				  dev->data->port_id, nic->vf_id);
1808 		}
1809 	}
1810 
1811 	PMD_INIT_LOG(DEBUG, "Configured ethdev port%d hwcap=0x%" PRIx64,
1812 		dev->data->port_id, nicvf_hw_cap(nic));
1813 
1814 	return 0;
1815 }
1816 
1817 /* Initialize and register driver with DPDK Application */
1818 static const struct eth_dev_ops nicvf_eth_dev_ops = {
1819 	.dev_configure            = nicvf_dev_configure,
1820 	.dev_start                = nicvf_dev_start,
1821 	.dev_stop                 = nicvf_dev_stop,
1822 	.link_update              = nicvf_dev_link_update,
1823 	.dev_close                = nicvf_dev_close,
1824 	.stats_get                = nicvf_dev_stats_get,
1825 	.stats_reset              = nicvf_dev_stats_reset,
1826 	.promiscuous_enable       = nicvf_dev_promisc_enable,
1827 	.dev_infos_get            = nicvf_dev_info_get,
1828 	.dev_supported_ptypes_get = nicvf_dev_supported_ptypes_get,
1829 	.mtu_set                  = nicvf_dev_set_mtu,
1830 	.reta_update              = nicvf_dev_reta_update,
1831 	.reta_query               = nicvf_dev_reta_query,
1832 	.rss_hash_update          = nicvf_dev_rss_hash_update,
1833 	.rss_hash_conf_get        = nicvf_dev_rss_hash_conf_get,
1834 	.rx_queue_start           = nicvf_dev_rx_queue_start,
1835 	.rx_queue_stop            = nicvf_dev_rx_queue_stop,
1836 	.tx_queue_start           = nicvf_dev_tx_queue_start,
1837 	.tx_queue_stop            = nicvf_dev_tx_queue_stop,
1838 	.rx_queue_setup           = nicvf_dev_rx_queue_setup,
1839 	.rx_queue_release         = nicvf_dev_rx_queue_release,
1840 	.rx_queue_count           = nicvf_dev_rx_queue_count,
1841 	.tx_queue_setup           = nicvf_dev_tx_queue_setup,
1842 	.tx_queue_release         = nicvf_dev_tx_queue_release,
1843 	.get_reg                  = nicvf_dev_get_regs,
1844 };
1845 
1846 static int
1847 nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
1848 {
1849 	int ret;
1850 	struct rte_pci_device *pci_dev;
1851 	struct nicvf *nic = nicvf_pmd_priv(eth_dev);
1852 
1853 	PMD_INIT_FUNC_TRACE();
1854 
1855 	eth_dev->dev_ops = &nicvf_eth_dev_ops;
1856 
1857 	/* For secondary processes, the primary has done all the work */
1858 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1859 		/* Setup callbacks for secondary process */
1860 		nicvf_set_tx_function(eth_dev);
1861 		nicvf_set_rx_function(eth_dev);
1862 		return 0;
1863 	}
1864 
1865 	pci_dev = eth_dev->pci_dev;
1866 	rte_eth_copy_pci_info(eth_dev, pci_dev);
1867 
1868 	nic->device_id = pci_dev->id.device_id;
1869 	nic->vendor_id = pci_dev->id.vendor_id;
1870 	nic->subsystem_device_id = pci_dev->id.subsystem_device_id;
1871 	nic->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
1872 
1873 	PMD_INIT_LOG(DEBUG, "nicvf: device (%x:%x) %u:%u:%u:%u",
1874 			pci_dev->id.vendor_id, pci_dev->id.device_id,
1875 			pci_dev->addr.domain, pci_dev->addr.bus,
1876 			pci_dev->addr.devid, pci_dev->addr.function);
1877 
1878 	nic->reg_base = (uintptr_t)pci_dev->mem_resource[0].addr;
1879 	if (!nic->reg_base) {
1880 		PMD_INIT_LOG(ERR, "Failed to map BAR0");
1881 		ret = -ENODEV;
1882 		goto fail;
1883 	}
1884 
1885 	nicvf_disable_all_interrupts(nic);
1886 
1887 	ret = nicvf_periodic_alarm_start(nicvf_interrupt, eth_dev);
1888 	if (ret) {
1889 		PMD_INIT_LOG(ERR, "Failed to start period alarm");
1890 		goto fail;
1891 	}
1892 
1893 	ret = nicvf_mbox_check_pf_ready(nic);
1894 	if (ret) {
1895 		PMD_INIT_LOG(ERR, "Failed to get ready message from PF");
1896 		goto alarm_fail;
1897 	} else {
1898 		PMD_INIT_LOG(INFO,
1899 			"node=%d vf=%d mode=%s sqs=%s loopback_supported=%s",
1900 			nic->node, nic->vf_id,
1901 			nic->tns_mode == NIC_TNS_MODE ? "tns" : "tns-bypass",
1902 			nic->sqs_mode ? "true" : "false",
1903 			nic->loopback_supported ? "true" : "false"
1904 			);
1905 	}
1906 
1907 	if (nic->sqs_mode) {
1908 		PMD_INIT_LOG(INFO, "Unsupported SQS VF detected, Detaching...");
1909 		/* Detach port by returning Positive error number */
1910 		ret = ENOTSUP;
1911 		goto alarm_fail;
1912 	}
1913 
1914 	eth_dev->data->mac_addrs = rte_zmalloc("mac_addr", ETHER_ADDR_LEN, 0);
1915 	if (eth_dev->data->mac_addrs == NULL) {
1916 		PMD_INIT_LOG(ERR, "Failed to allocate memory for mac addr");
1917 		ret = -ENOMEM;
1918 		goto alarm_fail;
1919 	}
1920 	if (is_zero_ether_addr((struct ether_addr *)nic->mac_addr))
1921 		eth_random_addr(&nic->mac_addr[0]);
1922 
1923 	ether_addr_copy((struct ether_addr *)nic->mac_addr,
1924 			&eth_dev->data->mac_addrs[0]);
1925 
1926 	ret = nicvf_mbox_set_mac_addr(nic, nic->mac_addr);
1927 	if (ret) {
1928 		PMD_INIT_LOG(ERR, "Failed to set mac addr");
1929 		goto malloc_fail;
1930 	}
1931 
1932 	ret = nicvf_base_init(nic);
1933 	if (ret) {
1934 		PMD_INIT_LOG(ERR, "Failed to execute nicvf_base_init");
1935 		goto malloc_fail;
1936 	}
1937 
1938 	PMD_INIT_LOG(INFO, "Port %d (%x:%x) mac=%02x:%02x:%02x:%02x:%02x:%02x",
1939 		eth_dev->data->port_id, nic->vendor_id, nic->device_id,
1940 		nic->mac_addr[0], nic->mac_addr[1], nic->mac_addr[2],
1941 		nic->mac_addr[3], nic->mac_addr[4], nic->mac_addr[5]);
1942 
1943 	return 0;
1944 
1945 malloc_fail:
1946 	rte_free(eth_dev->data->mac_addrs);
1947 alarm_fail:
1948 	nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
1949 fail:
1950 	return ret;
1951 }
1952 
1953 static const struct rte_pci_id pci_id_nicvf_map[] = {
1954 	{
1955 		.class_id = RTE_CLASS_ANY_ID,
1956 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
1957 		.device_id = PCI_DEVICE_ID_THUNDERX_CN88XX_PASS1_NICVF,
1958 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
1959 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS1_NICVF,
1960 	},
1961 	{
1962 		.class_id = RTE_CLASS_ANY_ID,
1963 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
1964 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
1965 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
1966 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS2_NICVF,
1967 	},
1968 	{
1969 		.class_id = RTE_CLASS_ANY_ID,
1970 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
1971 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
1972 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
1973 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN81XX_NICVF,
1974 	},
1975 	{
1976 		.vendor_id = 0,
1977 	},
1978 };
1979 
1980 static struct eth_driver rte_nicvf_pmd = {
1981 	.pci_drv = {
1982 		.id_table = pci_id_nicvf_map,
1983 		.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1984 		.probe = rte_eth_dev_pci_probe,
1985 		.remove = rte_eth_dev_pci_remove,
1986 	},
1987 	.eth_dev_init = nicvf_eth_dev_init,
1988 	.dev_private_size = sizeof(struct nicvf),
1989 };
1990 
1991 DRIVER_REGISTER_PCI(net_thunderx, rte_nicvf_pmd.pci_drv);
1992 DRIVER_REGISTER_PCI_TABLE(net_thunderx, pci_id_nicvf_map);
1993