xref: /dpdk/drivers/net/thunderx/nicvf_ethdev.c (revision a997a33b2a0145ad3e6320ea1fc7df8d51a2fcdf)
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, NULL);
115 	}
116 
117 	rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
118 				nicvf_interrupt, dev);
119 }
120 
121 static void
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 	uint16_t rx_start, rx_end;
240 	uint16_t tx_start, tx_end;
241 	size_t i;
242 
243 	/* RX queue indices for the first VF */
244 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
245 
246 	/* Reading per RX ring stats */
247 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
248 		if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
249 			break;
250 
251 		nicvf_hw_get_rx_qstats(nic, &rx_qstats, qidx);
252 		stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
253 		stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
254 	}
255 
256 	/* TX queue indices for the first VF */
257 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
258 
259 	/* Reading per TX ring stats */
260 	for (qidx = tx_start; qidx <= tx_end; qidx++) {
261 		if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
262 			break;
263 
264 		nicvf_hw_get_tx_qstats(nic, &tx_qstats, qidx);
265 		stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
266 		stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
267 	}
268 
269 	for (i = 0; i < nic->sqs_count; i++) {
270 		struct nicvf *snic = nic->snicvf[i];
271 
272 		if (snic == NULL)
273 			break;
274 
275 		/* RX queue indices for a secondary VF */
276 		nicvf_rx_range(dev, snic, &rx_start, &rx_end);
277 
278 		/* Reading per RX ring stats */
279 		for (qidx = rx_start; qidx <= rx_end; qidx++) {
280 			if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
281 				break;
282 
283 			nicvf_hw_get_rx_qstats(snic, &rx_qstats,
284 					       qidx % MAX_RCV_QUEUES_PER_QS);
285 			stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
286 			stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
287 		}
288 
289 		/* TX queue indices for a secondary VF */
290 		nicvf_tx_range(dev, snic, &tx_start, &tx_end);
291 		/* Reading per TX ring stats */
292 		for (qidx = tx_start; qidx <= tx_end; qidx++) {
293 			if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
294 				break;
295 
296 			nicvf_hw_get_tx_qstats(snic, &tx_qstats,
297 					       qidx % MAX_SND_QUEUES_PER_QS);
298 			stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
299 			stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
300 		}
301 	}
302 
303 	nicvf_hw_get_stats(nic, &port_stats);
304 	stats->ibytes = port_stats.rx_bytes;
305 	stats->ipackets = port_stats.rx_ucast_frames;
306 	stats->ipackets += port_stats.rx_bcast_frames;
307 	stats->ipackets += port_stats.rx_mcast_frames;
308 	stats->ierrors = port_stats.rx_l2_errors;
309 	stats->imissed = port_stats.rx_drop_red;
310 	stats->imissed += port_stats.rx_drop_overrun;
311 	stats->imissed += port_stats.rx_drop_bcast;
312 	stats->imissed += port_stats.rx_drop_mcast;
313 	stats->imissed += port_stats.rx_drop_l3_bcast;
314 	stats->imissed += port_stats.rx_drop_l3_mcast;
315 
316 	stats->obytes = port_stats.tx_bytes_ok;
317 	stats->opackets = port_stats.tx_ucast_frames_ok;
318 	stats->opackets += port_stats.tx_bcast_frames_ok;
319 	stats->opackets += port_stats.tx_mcast_frames_ok;
320 	stats->oerrors = port_stats.tx_drops;
321 }
322 
323 static const uint32_t *
324 nicvf_dev_supported_ptypes_get(struct rte_eth_dev *dev)
325 {
326 	size_t copied;
327 	static uint32_t ptypes[32];
328 	struct nicvf *nic = nicvf_pmd_priv(dev);
329 	static const uint32_t ptypes_common[] = {
330 		RTE_PTYPE_L3_IPV4,
331 		RTE_PTYPE_L3_IPV4_EXT,
332 		RTE_PTYPE_L3_IPV6,
333 		RTE_PTYPE_L3_IPV6_EXT,
334 		RTE_PTYPE_L4_TCP,
335 		RTE_PTYPE_L4_UDP,
336 		RTE_PTYPE_L4_FRAG,
337 	};
338 	static const uint32_t ptypes_tunnel[] = {
339 		RTE_PTYPE_TUNNEL_GRE,
340 		RTE_PTYPE_TUNNEL_GENEVE,
341 		RTE_PTYPE_TUNNEL_VXLAN,
342 		RTE_PTYPE_TUNNEL_NVGRE,
343 	};
344 	static const uint32_t ptypes_end = RTE_PTYPE_UNKNOWN;
345 
346 	copied = sizeof(ptypes_common);
347 	memcpy(ptypes, ptypes_common, copied);
348 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
349 		memcpy((char *)ptypes + copied, ptypes_tunnel,
350 			sizeof(ptypes_tunnel));
351 		copied += sizeof(ptypes_tunnel);
352 	}
353 
354 	memcpy((char *)ptypes + copied, &ptypes_end, sizeof(ptypes_end));
355 	if (dev->rx_pkt_burst == nicvf_recv_pkts ||
356 		dev->rx_pkt_burst == nicvf_recv_pkts_multiseg)
357 		return ptypes;
358 
359 	return NULL;
360 }
361 
362 static void
363 nicvf_dev_stats_reset(struct rte_eth_dev *dev)
364 {
365 	int i;
366 	uint16_t rxqs = 0, txqs = 0;
367 	struct nicvf *nic = nicvf_pmd_priv(dev);
368 	uint16_t rx_start, rx_end;
369 	uint16_t tx_start, tx_end;
370 
371 	/* Reset all primary nic counters */
372 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
373 	for (i = rx_start; i <= rx_end; i++)
374 		rxqs |= (0x3 << (i * 2));
375 
376 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
377 	for (i = tx_start; i <= tx_end; i++)
378 		txqs |= (0x3 << (i * 2));
379 
380 	nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, rxqs, txqs);
381 
382 	/* Reset secondary nic queue counters */
383 	for (i = 0; i < nic->sqs_count; i++) {
384 		struct nicvf *snic = nic->snicvf[i];
385 		if (snic == NULL)
386 			break;
387 
388 		nicvf_rx_range(dev, snic, &rx_start, &rx_end);
389 		for (i = rx_start; i <= rx_end; i++)
390 			rxqs |= (0x3 << ((i % MAX_CMP_QUEUES_PER_QS) * 2));
391 
392 		nicvf_tx_range(dev, snic, &tx_start, &tx_end);
393 		for (i = tx_start; i <= tx_end; i++)
394 			txqs |= (0x3 << ((i % MAX_SND_QUEUES_PER_QS) * 2));
395 
396 		nicvf_mbox_reset_stat_counters(snic, 0, 0, rxqs, txqs);
397 	}
398 }
399 
400 /* Promiscuous mode enabled by default in LMAC to VF 1:1 map configuration */
401 static void
402 nicvf_dev_promisc_enable(struct rte_eth_dev *dev __rte_unused)
403 {
404 }
405 
406 static inline uint64_t
407 nicvf_rss_ethdev_to_nic(struct nicvf *nic, uint64_t ethdev_rss)
408 {
409 	uint64_t nic_rss = 0;
410 
411 	if (ethdev_rss & ETH_RSS_IPV4)
412 		nic_rss |= RSS_IP_ENA;
413 
414 	if (ethdev_rss & ETH_RSS_IPV6)
415 		nic_rss |= RSS_IP_ENA;
416 
417 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_UDP)
418 		nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
419 
420 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_TCP)
421 		nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
422 
423 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_UDP)
424 		nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
425 
426 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_TCP)
427 		nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
428 
429 	if (ethdev_rss & ETH_RSS_PORT)
430 		nic_rss |= RSS_L2_EXTENDED_HASH_ENA;
431 
432 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
433 		if (ethdev_rss & ETH_RSS_VXLAN)
434 			nic_rss |= RSS_TUN_VXLAN_ENA;
435 
436 		if (ethdev_rss & ETH_RSS_GENEVE)
437 			nic_rss |= RSS_TUN_GENEVE_ENA;
438 
439 		if (ethdev_rss & ETH_RSS_NVGRE)
440 			nic_rss |= RSS_TUN_NVGRE_ENA;
441 	}
442 
443 	return nic_rss;
444 }
445 
446 static inline uint64_t
447 nicvf_rss_nic_to_ethdev(struct nicvf *nic,  uint64_t nic_rss)
448 {
449 	uint64_t ethdev_rss = 0;
450 
451 	if (nic_rss & RSS_IP_ENA)
452 		ethdev_rss |= (ETH_RSS_IPV4 | ETH_RSS_IPV6);
453 
454 	if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_TCP_ENA))
455 		ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_TCP |
456 				ETH_RSS_NONFRAG_IPV6_TCP);
457 
458 	if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_UDP_ENA))
459 		ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_UDP |
460 				ETH_RSS_NONFRAG_IPV6_UDP);
461 
462 	if (nic_rss & RSS_L2_EXTENDED_HASH_ENA)
463 		ethdev_rss |= ETH_RSS_PORT;
464 
465 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
466 		if (nic_rss & RSS_TUN_VXLAN_ENA)
467 			ethdev_rss |= ETH_RSS_VXLAN;
468 
469 		if (nic_rss & RSS_TUN_GENEVE_ENA)
470 			ethdev_rss |= ETH_RSS_GENEVE;
471 
472 		if (nic_rss & RSS_TUN_NVGRE_ENA)
473 			ethdev_rss |= ETH_RSS_NVGRE;
474 	}
475 	return ethdev_rss;
476 }
477 
478 static int
479 nicvf_dev_reta_query(struct rte_eth_dev *dev,
480 		     struct rte_eth_rss_reta_entry64 *reta_conf,
481 		     uint16_t reta_size)
482 {
483 	struct nicvf *nic = nicvf_pmd_priv(dev);
484 	uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
485 	int ret, i, j;
486 
487 	if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
488 		RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
489 			"(%d) doesn't match the number hardware can supported "
490 			"(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
491 		return -EINVAL;
492 	}
493 
494 	ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
495 	if (ret)
496 		return ret;
497 
498 	/* Copy RETA table */
499 	for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
500 		for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
501 			if ((reta_conf[i].mask >> j) & 0x01)
502 				reta_conf[i].reta[j] = tbl[j];
503 	}
504 
505 	return 0;
506 }
507 
508 static int
509 nicvf_dev_reta_update(struct rte_eth_dev *dev,
510 		      struct rte_eth_rss_reta_entry64 *reta_conf,
511 		      uint16_t reta_size)
512 {
513 	struct nicvf *nic = nicvf_pmd_priv(dev);
514 	uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
515 	int ret, i, j;
516 
517 	if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
518 		RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
519 			"(%d) doesn't match the number hardware can supported "
520 			"(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
521 		return -EINVAL;
522 	}
523 
524 	ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
525 	if (ret)
526 		return ret;
527 
528 	/* Copy RETA table */
529 	for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
530 		for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
531 			if ((reta_conf[i].mask >> j) & 0x01)
532 				tbl[j] = reta_conf[i].reta[j];
533 	}
534 
535 	return nicvf_rss_reta_update(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
536 }
537 
538 static int
539 nicvf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
540 			    struct rte_eth_rss_conf *rss_conf)
541 {
542 	struct nicvf *nic = nicvf_pmd_priv(dev);
543 
544 	if (rss_conf->rss_key)
545 		nicvf_rss_get_key(nic, rss_conf->rss_key);
546 
547 	rss_conf->rss_key_len =  RSS_HASH_KEY_BYTE_SIZE;
548 	rss_conf->rss_hf = nicvf_rss_nic_to_ethdev(nic, nicvf_rss_get_cfg(nic));
549 	return 0;
550 }
551 
552 static int
553 nicvf_dev_rss_hash_update(struct rte_eth_dev *dev,
554 			  struct rte_eth_rss_conf *rss_conf)
555 {
556 	struct nicvf *nic = nicvf_pmd_priv(dev);
557 	uint64_t nic_rss;
558 
559 	if (rss_conf->rss_key &&
560 		rss_conf->rss_key_len != RSS_HASH_KEY_BYTE_SIZE) {
561 		RTE_LOG(ERR, PMD, "Hash key size mismatch %d",
562 				rss_conf->rss_key_len);
563 		return -EINVAL;
564 	}
565 
566 	if (rss_conf->rss_key)
567 		nicvf_rss_set_key(nic, rss_conf->rss_key);
568 
569 	nic_rss = nicvf_rss_ethdev_to_nic(nic, rss_conf->rss_hf);
570 	nicvf_rss_set_cfg(nic, nic_rss);
571 	return 0;
572 }
573 
574 static int
575 nicvf_qset_cq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
576 		    struct nicvf_rxq *rxq, uint16_t qidx, uint32_t desc_cnt)
577 {
578 	const struct rte_memzone *rz;
579 	uint32_t ring_size = CMP_QUEUE_SZ_MAX * sizeof(union cq_entry_t);
580 
581 	rz = rte_eth_dma_zone_reserve(dev, "cq_ring",
582 				      nicvf_netdev_qidx(nic, qidx), ring_size,
583 				      NICVF_CQ_BASE_ALIGN_BYTES, nic->node);
584 	if (rz == NULL) {
585 		PMD_INIT_LOG(ERR, "Failed to allocate mem for cq hw ring");
586 		return -ENOMEM;
587 	}
588 
589 	memset(rz->addr, 0, ring_size);
590 
591 	rxq->phys = rz->phys_addr;
592 	rxq->desc = rz->addr;
593 	rxq->qlen_mask = desc_cnt - 1;
594 
595 	return 0;
596 }
597 
598 static int
599 nicvf_qset_sq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
600 		    struct nicvf_txq *sq, uint16_t qidx, uint32_t desc_cnt)
601 {
602 	const struct rte_memzone *rz;
603 	uint32_t ring_size = SND_QUEUE_SZ_MAX * sizeof(union sq_entry_t);
604 
605 	rz = rte_eth_dma_zone_reserve(dev, "sq",
606 				      nicvf_netdev_qidx(nic, qidx), ring_size,
607 				      NICVF_SQ_BASE_ALIGN_BYTES, nic->node);
608 	if (rz == NULL) {
609 		PMD_INIT_LOG(ERR, "Failed allocate mem for sq hw ring");
610 		return -ENOMEM;
611 	}
612 
613 	memset(rz->addr, 0, ring_size);
614 
615 	sq->phys = rz->phys_addr;
616 	sq->desc = rz->addr;
617 	sq->qlen_mask = desc_cnt - 1;
618 
619 	return 0;
620 }
621 
622 static int
623 nicvf_qset_rbdr_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
624 		      uint32_t desc_cnt, uint32_t buffsz)
625 {
626 	struct nicvf_rbdr *rbdr;
627 	const struct rte_memzone *rz;
628 	uint32_t ring_size;
629 
630 	assert(nic->rbdr == NULL);
631 	rbdr = rte_zmalloc_socket("rbdr", sizeof(struct nicvf_rbdr),
632 				  RTE_CACHE_LINE_SIZE, nic->node);
633 	if (rbdr == NULL) {
634 		PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr");
635 		return -ENOMEM;
636 	}
637 
638 	ring_size = sizeof(struct rbdr_entry_t) * RBDR_QUEUE_SZ_MAX;
639 	rz = rte_eth_dma_zone_reserve(dev, "rbdr",
640 				      nicvf_netdev_qidx(nic, 0), ring_size,
641 				      NICVF_RBDR_BASE_ALIGN_BYTES, nic->node);
642 	if (rz == NULL) {
643 		PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr desc ring");
644 		return -ENOMEM;
645 	}
646 
647 	memset(rz->addr, 0, ring_size);
648 
649 	rbdr->phys = rz->phys_addr;
650 	rbdr->tail = 0;
651 	rbdr->next_tail = 0;
652 	rbdr->desc = rz->addr;
653 	rbdr->buffsz = buffsz;
654 	rbdr->qlen_mask = desc_cnt - 1;
655 	rbdr->rbdr_status =
656 		nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_STATUS0;
657 	rbdr->rbdr_door =
658 		nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_DOOR;
659 
660 	nic->rbdr = rbdr;
661 	return 0;
662 }
663 
664 static void
665 nicvf_rbdr_release_mbuf(struct rte_eth_dev *dev, struct nicvf *nic,
666 			nicvf_phys_addr_t phy)
667 {
668 	uint16_t qidx;
669 	void *obj;
670 	struct nicvf_rxq *rxq;
671 	uint16_t rx_start, rx_end;
672 
673 	/* Get queue ranges for this VF */
674 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
675 
676 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
677 		rxq = dev->data->rx_queues[qidx];
678 		if (rxq->precharge_cnt) {
679 			obj = (void *)nicvf_mbuff_phy2virt(phy,
680 							   rxq->mbuf_phys_off);
681 			rte_mempool_put(rxq->pool, obj);
682 			rxq->precharge_cnt--;
683 			break;
684 		}
685 	}
686 }
687 
688 static inline void
689 nicvf_rbdr_release_mbufs(struct rte_eth_dev *dev, struct nicvf *nic)
690 {
691 	uint32_t qlen_mask, head;
692 	struct rbdr_entry_t *entry;
693 	struct nicvf_rbdr *rbdr = nic->rbdr;
694 
695 	qlen_mask = rbdr->qlen_mask;
696 	head = rbdr->head;
697 	while (head != rbdr->tail) {
698 		entry = rbdr->desc + head;
699 		nicvf_rbdr_release_mbuf(dev, nic, entry->full_addr);
700 		head++;
701 		head = head & qlen_mask;
702 	}
703 }
704 
705 static inline void
706 nicvf_tx_queue_release_mbufs(struct nicvf_txq *txq)
707 {
708 	uint32_t head;
709 
710 	head = txq->head;
711 	while (head != txq->tail) {
712 		if (txq->txbuffs[head]) {
713 			rte_pktmbuf_free_seg(txq->txbuffs[head]);
714 			txq->txbuffs[head] = NULL;
715 		}
716 		head++;
717 		head = head & txq->qlen_mask;
718 	}
719 }
720 
721 static void
722 nicvf_tx_queue_reset(struct nicvf_txq *txq)
723 {
724 	uint32_t txq_desc_cnt = txq->qlen_mask + 1;
725 
726 	memset(txq->desc, 0, sizeof(union sq_entry_t) * txq_desc_cnt);
727 	memset(txq->txbuffs, 0, sizeof(struct rte_mbuf *) * txq_desc_cnt);
728 	txq->tail = 0;
729 	txq->head = 0;
730 	txq->xmit_bufs = 0;
731 }
732 
733 static inline int
734 nicvf_vf_start_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
735 			uint16_t qidx)
736 {
737 	struct nicvf_txq *txq;
738 	int ret;
739 
740 	assert(qidx < MAX_SND_QUEUES_PER_QS);
741 
742 	if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
743 		RTE_ETH_QUEUE_STATE_STARTED)
744 		return 0;
745 
746 	txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
747 	txq->pool = NULL;
748 	ret = nicvf_qset_sq_config(nic, qidx, txq);
749 	if (ret) {
750 		PMD_INIT_LOG(ERR, "Failed to configure sq VF%d %d %d",
751 			     nic->vf_id, qidx, ret);
752 		goto config_sq_error;
753 	}
754 
755 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
756 		RTE_ETH_QUEUE_STATE_STARTED;
757 	return ret;
758 
759 config_sq_error:
760 	nicvf_qset_sq_reclaim(nic, qidx);
761 	return ret;
762 }
763 
764 static inline int
765 nicvf_vf_stop_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
766 		       uint16_t qidx)
767 {
768 	struct nicvf_txq *txq;
769 	int ret;
770 
771 	assert(qidx < MAX_SND_QUEUES_PER_QS);
772 
773 	if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
774 		RTE_ETH_QUEUE_STATE_STOPPED)
775 		return 0;
776 
777 	ret = nicvf_qset_sq_reclaim(nic, qidx);
778 	if (ret)
779 		PMD_INIT_LOG(ERR, "Failed to reclaim sq VF%d %d %d",
780 			     nic->vf_id, qidx, ret);
781 
782 	txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
783 	nicvf_tx_queue_release_mbufs(txq);
784 	nicvf_tx_queue_reset(txq);
785 
786 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
787 		RTE_ETH_QUEUE_STATE_STOPPED;
788 	return ret;
789 }
790 
791 static inline int
792 nicvf_configure_cpi(struct rte_eth_dev *dev)
793 {
794 	struct nicvf *nic = nicvf_pmd_priv(dev);
795 	uint16_t qidx, qcnt;
796 	int ret;
797 
798 	/* Count started rx queues */
799 	for (qidx = qcnt = 0; qidx < dev->data->nb_rx_queues; qidx++)
800 		if (dev->data->rx_queue_state[qidx] ==
801 		    RTE_ETH_QUEUE_STATE_STARTED)
802 			qcnt++;
803 
804 	nic->cpi_alg = CPI_ALG_NONE;
805 	ret = nicvf_mbox_config_cpi(nic, qcnt);
806 	if (ret)
807 		PMD_INIT_LOG(ERR, "Failed to configure CPI %d", ret);
808 
809 	return ret;
810 }
811 
812 static inline int
813 nicvf_configure_rss(struct rte_eth_dev *dev)
814 {
815 	struct nicvf *nic = nicvf_pmd_priv(dev);
816 	uint64_t rsshf;
817 	int ret = -EINVAL;
818 
819 	rsshf = nicvf_rss_ethdev_to_nic(nic,
820 			dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf);
821 	PMD_DRV_LOG(INFO, "mode=%d rx_queues=%d loopback=%d rsshf=0x%" PRIx64,
822 		    dev->data->dev_conf.rxmode.mq_mode,
823 		    dev->data->nb_rx_queues,
824 		    dev->data->dev_conf.lpbk_mode, rsshf);
825 
826 	if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_NONE)
827 		ret = nicvf_rss_term(nic);
828 	else if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS)
829 		ret = nicvf_rss_config(nic, dev->data->nb_rx_queues, rsshf);
830 	if (ret)
831 		PMD_INIT_LOG(ERR, "Failed to configure RSS %d", ret);
832 
833 	return ret;
834 }
835 
836 static int
837 nicvf_configure_rss_reta(struct rte_eth_dev *dev)
838 {
839 	struct nicvf *nic = nicvf_pmd_priv(dev);
840 	unsigned int idx, qmap_size;
841 	uint8_t qmap[RTE_MAX_QUEUES_PER_PORT];
842 	uint8_t default_reta[NIC_MAX_RSS_IDR_TBL_SIZE];
843 
844 	if (nic->cpi_alg != CPI_ALG_NONE)
845 		return -EINVAL;
846 
847 	/* Prepare queue map */
848 	for (idx = 0, qmap_size = 0; idx < dev->data->nb_rx_queues; idx++) {
849 		if (dev->data->rx_queue_state[idx] ==
850 				RTE_ETH_QUEUE_STATE_STARTED)
851 			qmap[qmap_size++] = idx;
852 	}
853 
854 	/* Update default RSS RETA */
855 	for (idx = 0; idx < NIC_MAX_RSS_IDR_TBL_SIZE; idx++)
856 		default_reta[idx] = qmap[idx % qmap_size];
857 
858 	return nicvf_rss_reta_update(nic, default_reta,
859 				     NIC_MAX_RSS_IDR_TBL_SIZE);
860 }
861 
862 static void
863 nicvf_dev_tx_queue_release(void *sq)
864 {
865 	struct nicvf_txq *txq;
866 
867 	PMD_INIT_FUNC_TRACE();
868 
869 	txq = (struct nicvf_txq *)sq;
870 	if (txq) {
871 		if (txq->txbuffs != NULL) {
872 			nicvf_tx_queue_release_mbufs(txq);
873 			rte_free(txq->txbuffs);
874 			txq->txbuffs = NULL;
875 		}
876 		rte_free(txq);
877 	}
878 }
879 
880 static void
881 nicvf_set_tx_function(struct rte_eth_dev *dev)
882 {
883 	struct nicvf_txq *txq;
884 	size_t i;
885 	bool multiseg = false;
886 
887 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
888 		txq = dev->data->tx_queues[i];
889 		if ((txq->txq_flags & ETH_TXQ_FLAGS_NOMULTSEGS) == 0) {
890 			multiseg = true;
891 			break;
892 		}
893 	}
894 
895 	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
896 	if (multiseg) {
897 		PMD_DRV_LOG(DEBUG, "Using multi-segment tx callback");
898 		dev->tx_pkt_burst = nicvf_xmit_pkts_multiseg;
899 	} else {
900 		PMD_DRV_LOG(DEBUG, "Using single-segment tx callback");
901 		dev->tx_pkt_burst = nicvf_xmit_pkts;
902 	}
903 
904 	if (txq->pool_free == nicvf_single_pool_free_xmited_buffers)
905 		PMD_DRV_LOG(DEBUG, "Using single-mempool tx free method");
906 	else
907 		PMD_DRV_LOG(DEBUG, "Using multi-mempool tx free method");
908 }
909 
910 static void
911 nicvf_set_rx_function(struct rte_eth_dev *dev)
912 {
913 	if (dev->data->scattered_rx) {
914 		PMD_DRV_LOG(DEBUG, "Using multi-segment rx callback");
915 		dev->rx_pkt_burst = nicvf_recv_pkts_multiseg;
916 	} else {
917 		PMD_DRV_LOG(DEBUG, "Using single-segment rx callback");
918 		dev->rx_pkt_burst = nicvf_recv_pkts;
919 	}
920 }
921 
922 static int
923 nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
924 			 uint16_t nb_desc, unsigned int socket_id,
925 			 const struct rte_eth_txconf *tx_conf)
926 {
927 	uint16_t tx_free_thresh;
928 	uint8_t is_single_pool;
929 	struct nicvf_txq *txq;
930 	struct nicvf *nic = nicvf_pmd_priv(dev);
931 
932 	PMD_INIT_FUNC_TRACE();
933 
934 	if (qidx >= MAX_SND_QUEUES_PER_QS)
935 		nic = nic->snicvf[qidx / MAX_SND_QUEUES_PER_QS - 1];
936 
937 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
938 
939 	/* Socket id check */
940 	if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
941 		PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
942 		socket_id, nic->node);
943 
944 	/* Tx deferred start is not supported */
945 	if (tx_conf->tx_deferred_start) {
946 		PMD_INIT_LOG(ERR, "Tx deferred start not supported");
947 		return -EINVAL;
948 	}
949 
950 	/* Roundup nb_desc to available qsize and validate max number of desc */
951 	nb_desc = nicvf_qsize_sq_roundup(nb_desc);
952 	if (nb_desc == 0) {
953 		PMD_INIT_LOG(ERR, "Value of nb_desc beyond available sq qsize");
954 		return -EINVAL;
955 	}
956 
957 	/* Validate tx_free_thresh */
958 	tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
959 				tx_conf->tx_free_thresh :
960 				NICVF_DEFAULT_TX_FREE_THRESH);
961 
962 	if (tx_free_thresh > (nb_desc) ||
963 		tx_free_thresh > NICVF_MAX_TX_FREE_THRESH) {
964 		PMD_INIT_LOG(ERR,
965 			"tx_free_thresh must be less than the number of TX "
966 			"descriptors. (tx_free_thresh=%u port=%d "
967 			"queue=%d)", (unsigned int)tx_free_thresh,
968 			(int)dev->data->port_id, (int)qidx);
969 		return -EINVAL;
970 	}
971 
972 	/* Free memory prior to re-allocation if needed. */
973 	if (dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
974 		PMD_TX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
975 				nicvf_netdev_qidx(nic, qidx));
976 		nicvf_dev_tx_queue_release(
977 			dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)]);
978 		dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
979 	}
980 
981 	/* Allocating tx queue data structure */
982 	txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct nicvf_txq),
983 					RTE_CACHE_LINE_SIZE, nic->node);
984 	if (txq == NULL) {
985 		PMD_INIT_LOG(ERR, "Failed to allocate txq=%d",
986 			     nicvf_netdev_qidx(nic, qidx));
987 		return -ENOMEM;
988 	}
989 
990 	txq->nic = nic;
991 	txq->queue_id = qidx;
992 	txq->tx_free_thresh = tx_free_thresh;
993 	txq->txq_flags = tx_conf->txq_flags;
994 	txq->sq_head = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_HEAD;
995 	txq->sq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_DOOR;
996 	is_single_pool = (txq->txq_flags & ETH_TXQ_FLAGS_NOREFCOUNT &&
997 				txq->txq_flags & ETH_TXQ_FLAGS_NOMULTMEMP);
998 
999 	/* Choose optimum free threshold value for multipool case */
1000 	if (!is_single_pool) {
1001 		txq->tx_free_thresh = (uint16_t)
1002 		(tx_conf->tx_free_thresh == NICVF_DEFAULT_TX_FREE_THRESH ?
1003 				NICVF_TX_FREE_MPOOL_THRESH :
1004 				tx_conf->tx_free_thresh);
1005 		txq->pool_free = nicvf_multi_pool_free_xmited_buffers;
1006 	} else {
1007 		txq->pool_free = nicvf_single_pool_free_xmited_buffers;
1008 	}
1009 
1010 	/* Allocate software ring */
1011 	txq->txbuffs = rte_zmalloc_socket("txq->txbuffs",
1012 				nb_desc * sizeof(struct rte_mbuf *),
1013 				RTE_CACHE_LINE_SIZE, nic->node);
1014 
1015 	if (txq->txbuffs == NULL) {
1016 		nicvf_dev_tx_queue_release(txq);
1017 		return -ENOMEM;
1018 	}
1019 
1020 	if (nicvf_qset_sq_alloc(dev, nic, txq, qidx, nb_desc)) {
1021 		PMD_INIT_LOG(ERR, "Failed to allocate mem for sq %d", qidx);
1022 		nicvf_dev_tx_queue_release(txq);
1023 		return -ENOMEM;
1024 	}
1025 
1026 	nicvf_tx_queue_reset(txq);
1027 
1028 	PMD_TX_LOG(DEBUG, "[%d] txq=%p nb_desc=%d desc=%p phys=0x%" PRIx64,
1029 			nicvf_netdev_qidx(nic, qidx), txq, nb_desc, txq->desc,
1030 			txq->phys);
1031 
1032 	dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = txq;
1033 	dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1034 		RTE_ETH_QUEUE_STATE_STOPPED;
1035 	return 0;
1036 }
1037 
1038 static inline void
1039 nicvf_rx_queue_release_mbufs(struct rte_eth_dev *dev, struct nicvf_rxq *rxq)
1040 {
1041 	uint32_t rxq_cnt;
1042 	uint32_t nb_pkts, released_pkts = 0;
1043 	uint32_t refill_cnt = 0;
1044 	struct rte_mbuf *rx_pkts[NICVF_MAX_RX_FREE_THRESH];
1045 
1046 	if (dev->rx_pkt_burst == NULL)
1047 		return;
1048 
1049 	while ((rxq_cnt = nicvf_dev_rx_queue_count(dev,
1050 				nicvf_netdev_qidx(rxq->nic, rxq->queue_id)))) {
1051 		nb_pkts = dev->rx_pkt_burst(rxq, rx_pkts,
1052 					NICVF_MAX_RX_FREE_THRESH);
1053 		PMD_DRV_LOG(INFO, "nb_pkts=%d  rxq_cnt=%d", nb_pkts, rxq_cnt);
1054 		while (nb_pkts) {
1055 			rte_pktmbuf_free_seg(rx_pkts[--nb_pkts]);
1056 			released_pkts++;
1057 		}
1058 	}
1059 
1060 
1061 	refill_cnt += nicvf_dev_rbdr_refill(dev,
1062 			nicvf_netdev_qidx(rxq->nic, rxq->queue_id));
1063 
1064 	PMD_DRV_LOG(INFO, "free_cnt=%d  refill_cnt=%d",
1065 		    released_pkts, refill_cnt);
1066 }
1067 
1068 static void
1069 nicvf_rx_queue_reset(struct nicvf_rxq *rxq)
1070 {
1071 	rxq->head = 0;
1072 	rxq->available_space = 0;
1073 	rxq->recv_buffers = 0;
1074 }
1075 
1076 static inline int
1077 nicvf_vf_start_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
1078 			uint16_t qidx)
1079 {
1080 	struct nicvf_rxq *rxq;
1081 	int ret;
1082 
1083 	assert(qidx < MAX_RCV_QUEUES_PER_QS);
1084 
1085 	if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
1086 		RTE_ETH_QUEUE_STATE_STARTED)
1087 		return 0;
1088 
1089 	/* Update rbdr pointer to all rxq */
1090 	rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
1091 	rxq->shared_rbdr = nic->rbdr;
1092 
1093 	ret = nicvf_qset_rq_config(nic, qidx, rxq);
1094 	if (ret) {
1095 		PMD_INIT_LOG(ERR, "Failed to configure rq VF%d %d %d",
1096 			     nic->vf_id, qidx, ret);
1097 		goto config_rq_error;
1098 	}
1099 	ret = nicvf_qset_cq_config(nic, qidx, rxq);
1100 	if (ret) {
1101 		PMD_INIT_LOG(ERR, "Failed to configure cq VF%d %d %d",
1102 			     nic->vf_id, qidx, ret);
1103 		goto config_cq_error;
1104 	}
1105 
1106 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1107 		RTE_ETH_QUEUE_STATE_STARTED;
1108 	return 0;
1109 
1110 config_cq_error:
1111 	nicvf_qset_cq_reclaim(nic, qidx);
1112 config_rq_error:
1113 	nicvf_qset_rq_reclaim(nic, qidx);
1114 	return ret;
1115 }
1116 
1117 static inline int
1118 nicvf_vf_stop_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
1119 		       uint16_t qidx)
1120 {
1121 	struct nicvf_rxq *rxq;
1122 	int ret, other_error;
1123 
1124 	if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
1125 		RTE_ETH_QUEUE_STATE_STOPPED)
1126 		return 0;
1127 
1128 	ret = nicvf_qset_rq_reclaim(nic, qidx);
1129 	if (ret)
1130 		PMD_INIT_LOG(ERR, "Failed to reclaim rq VF%d %d %d",
1131 			     nic->vf_id, qidx, ret);
1132 
1133 	other_error = ret;
1134 	rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
1135 	nicvf_rx_queue_release_mbufs(dev, rxq);
1136 	nicvf_rx_queue_reset(rxq);
1137 
1138 	ret = nicvf_qset_cq_reclaim(nic, qidx);
1139 	if (ret)
1140 		PMD_INIT_LOG(ERR, "Failed to reclaim cq VF%d %d %d",
1141 			     nic->vf_id, qidx, ret);
1142 
1143 	other_error |= ret;
1144 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1145 		RTE_ETH_QUEUE_STATE_STOPPED;
1146 	return other_error;
1147 }
1148 
1149 static void
1150 nicvf_dev_rx_queue_release(void *rx_queue)
1151 {
1152 	PMD_INIT_FUNC_TRACE();
1153 
1154 	rte_free(rx_queue);
1155 }
1156 
1157 static int
1158 nicvf_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
1159 {
1160 	struct nicvf *nic = nicvf_pmd_priv(dev);
1161 	int ret;
1162 
1163 	if (qidx >= MAX_RCV_QUEUES_PER_QS)
1164 		nic = nic->snicvf[(qidx / MAX_RCV_QUEUES_PER_QS - 1)];
1165 
1166 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1167 
1168 	ret = nicvf_vf_start_rx_queue(dev, nic, qidx);
1169 	if (ret)
1170 		return ret;
1171 
1172 	ret = nicvf_configure_cpi(dev);
1173 	if (ret)
1174 		return ret;
1175 
1176 	return nicvf_configure_rss_reta(dev);
1177 }
1178 
1179 static int
1180 nicvf_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
1181 {
1182 	int ret;
1183 	struct nicvf *nic = nicvf_pmd_priv(dev);
1184 
1185 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1186 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1187 
1188 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1189 
1190 	ret = nicvf_vf_stop_rx_queue(dev, nic, qidx);
1191 	ret |= nicvf_configure_cpi(dev);
1192 	ret |= nicvf_configure_rss_reta(dev);
1193 	return ret;
1194 }
1195 
1196 static int
1197 nicvf_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
1198 {
1199 	struct nicvf *nic = nicvf_pmd_priv(dev);
1200 
1201 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1202 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1203 
1204 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
1205 
1206 	return nicvf_vf_start_tx_queue(dev, nic, qidx);
1207 }
1208 
1209 static int
1210 nicvf_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
1211 {
1212 	struct nicvf *nic = nicvf_pmd_priv(dev);
1213 
1214 	if (qidx >= MAX_SND_QUEUES_PER_QS)
1215 		nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1216 
1217 	qidx = qidx % MAX_SND_QUEUES_PER_QS;
1218 
1219 	return nicvf_vf_stop_tx_queue(dev, nic, qidx);
1220 }
1221 
1222 
1223 static int
1224 nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
1225 			 uint16_t nb_desc, unsigned int socket_id,
1226 			 const struct rte_eth_rxconf *rx_conf,
1227 			 struct rte_mempool *mp)
1228 {
1229 	uint16_t rx_free_thresh;
1230 	struct nicvf_rxq *rxq;
1231 	struct nicvf *nic = nicvf_pmd_priv(dev);
1232 
1233 	PMD_INIT_FUNC_TRACE();
1234 
1235 	if (qidx >= MAX_RCV_QUEUES_PER_QS)
1236 		nic = nic->snicvf[qidx / MAX_RCV_QUEUES_PER_QS - 1];
1237 
1238 	qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1239 
1240 	/* Socket id check */
1241 	if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
1242 		PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
1243 		socket_id, nic->node);
1244 
1245 	/* Mempool memory must be contiguous, so must be one memory segment*/
1246 	if (mp->nb_mem_chunks != 1) {
1247 		PMD_INIT_LOG(ERR, "Non-contiguous mempool, add more huge pages");
1248 		return -EINVAL;
1249 	}
1250 
1251 	/* Mempool memory must be physically contiguous */
1252 	if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG) {
1253 		PMD_INIT_LOG(ERR, "Mempool memory must be physically contiguous");
1254 		return -EINVAL;
1255 	}
1256 
1257 	/* Rx deferred start is not supported */
1258 	if (rx_conf->rx_deferred_start) {
1259 		PMD_INIT_LOG(ERR, "Rx deferred start not supported");
1260 		return -EINVAL;
1261 	}
1262 
1263 	/* Roundup nb_desc to available qsize and validate max number of desc */
1264 	nb_desc = nicvf_qsize_cq_roundup(nb_desc);
1265 	if (nb_desc == 0) {
1266 		PMD_INIT_LOG(ERR, "Value nb_desc beyond available hw cq qsize");
1267 		return -EINVAL;
1268 	}
1269 
1270 	/* Check rx_free_thresh upper bound */
1271 	rx_free_thresh = (uint16_t)((rx_conf->rx_free_thresh) ?
1272 				rx_conf->rx_free_thresh :
1273 				NICVF_DEFAULT_RX_FREE_THRESH);
1274 	if (rx_free_thresh > NICVF_MAX_RX_FREE_THRESH ||
1275 		rx_free_thresh >= nb_desc * .75) {
1276 		PMD_INIT_LOG(ERR, "rx_free_thresh greater than expected %d",
1277 				rx_free_thresh);
1278 		return -EINVAL;
1279 	}
1280 
1281 	/* Free memory prior to re-allocation if needed */
1282 	if (dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
1283 		PMD_RX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
1284 				nicvf_netdev_qidx(nic, qidx));
1285 		nicvf_dev_rx_queue_release(
1286 			dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)]);
1287 		dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
1288 	}
1289 
1290 	/* Allocate rxq memory */
1291 	rxq = rte_zmalloc_socket("ethdev rx queue", sizeof(struct nicvf_rxq),
1292 					RTE_CACHE_LINE_SIZE, nic->node);
1293 	if (rxq == NULL) {
1294 		PMD_INIT_LOG(ERR, "Failed to allocate rxq=%d",
1295 			     nicvf_netdev_qidx(nic, qidx));
1296 		return -ENOMEM;
1297 	}
1298 
1299 	rxq->nic = nic;
1300 	rxq->pool = mp;
1301 	rxq->queue_id = qidx;
1302 	rxq->port_id = dev->data->port_id;
1303 	rxq->rx_free_thresh = rx_free_thresh;
1304 	rxq->rx_drop_en = rx_conf->rx_drop_en;
1305 	rxq->cq_status = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_STATUS;
1306 	rxq->cq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_DOOR;
1307 	rxq->precharge_cnt = 0;
1308 
1309 	if (nicvf_hw_cap(nic) & NICVF_CAP_CQE_RX2)
1310 		rxq->rbptr_offset = NICVF_CQE_RX2_RBPTR_WORD;
1311 	else
1312 		rxq->rbptr_offset = NICVF_CQE_RBPTR_WORD;
1313 
1314 
1315 	/* Alloc completion queue */
1316 	if (nicvf_qset_cq_alloc(dev, nic, rxq, rxq->queue_id, nb_desc)) {
1317 		PMD_INIT_LOG(ERR, "failed to allocate cq %u", rxq->queue_id);
1318 		nicvf_dev_rx_queue_release(rxq);
1319 		return -ENOMEM;
1320 	}
1321 
1322 	nicvf_rx_queue_reset(rxq);
1323 
1324 	PMD_RX_LOG(DEBUG, "[%d] rxq=%p pool=%s nb_desc=(%d/%d) phy=%" PRIx64,
1325 			nicvf_netdev_qidx(nic, qidx), rxq, mp->name, nb_desc,
1326 			rte_mempool_avail_count(mp), rxq->phys);
1327 
1328 	dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = rxq;
1329 	dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1330 		RTE_ETH_QUEUE_STATE_STOPPED;
1331 	return 0;
1332 }
1333 
1334 static void
1335 nicvf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1336 {
1337 	struct nicvf *nic = nicvf_pmd_priv(dev);
1338 
1339 	PMD_INIT_FUNC_TRACE();
1340 
1341 	dev_info->min_rx_bufsize = ETHER_MIN_MTU;
1342 	dev_info->max_rx_pktlen = NIC_HW_MAX_FRS;
1343 	dev_info->max_rx_queues =
1344 			(uint16_t)MAX_RCV_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
1345 	dev_info->max_tx_queues =
1346 			(uint16_t)MAX_SND_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
1347 	dev_info->max_mac_addrs = 1;
1348 	dev_info->max_vfs = dev->pci_dev->max_vfs;
1349 
1350 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
1351 	dev_info->tx_offload_capa =
1352 		DEV_TX_OFFLOAD_IPV4_CKSUM  |
1353 		DEV_TX_OFFLOAD_UDP_CKSUM   |
1354 		DEV_TX_OFFLOAD_TCP_CKSUM   |
1355 		DEV_TX_OFFLOAD_TCP_TSO     |
1356 		DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;
1357 
1358 	dev_info->reta_size = nic->rss_info.rss_size;
1359 	dev_info->hash_key_size = RSS_HASH_KEY_BYTE_SIZE;
1360 	dev_info->flow_type_rss_offloads = NICVF_RSS_OFFLOAD_PASS1;
1361 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING)
1362 		dev_info->flow_type_rss_offloads |= NICVF_RSS_OFFLOAD_TUNNEL;
1363 
1364 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
1365 		.rx_free_thresh = NICVF_DEFAULT_RX_FREE_THRESH,
1366 		.rx_drop_en = 0,
1367 	};
1368 
1369 	dev_info->default_txconf = (struct rte_eth_txconf) {
1370 		.tx_free_thresh = NICVF_DEFAULT_TX_FREE_THRESH,
1371 		.txq_flags =
1372 			ETH_TXQ_FLAGS_NOMULTSEGS  |
1373 			ETH_TXQ_FLAGS_NOREFCOUNT  |
1374 			ETH_TXQ_FLAGS_NOMULTMEMP  |
1375 			ETH_TXQ_FLAGS_NOVLANOFFL  |
1376 			ETH_TXQ_FLAGS_NOXSUMSCTP,
1377 	};
1378 }
1379 
1380 static nicvf_phys_addr_t
1381 rbdr_rte_mempool_get(void *dev, void *opaque)
1382 {
1383 	uint16_t qidx;
1384 	uintptr_t mbuf;
1385 	struct nicvf_rxq *rxq;
1386 	struct rte_eth_dev *eth_dev = (struct rte_eth_dev *)dev;
1387 	struct nicvf *nic = (struct nicvf *)opaque;
1388 	uint16_t rx_start, rx_end;
1389 
1390 	/* Get queue ranges for this VF */
1391 	nicvf_rx_range(eth_dev, nic, &rx_start, &rx_end);
1392 
1393 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1394 		rxq = eth_dev->data->rx_queues[qidx];
1395 		/* Maintain equal buffer count across all pools */
1396 		if (rxq->precharge_cnt >= rxq->qlen_mask)
1397 			continue;
1398 		rxq->precharge_cnt++;
1399 		mbuf = (uintptr_t)rte_pktmbuf_alloc(rxq->pool);
1400 		if (mbuf)
1401 			return nicvf_mbuff_virt2phy(mbuf, rxq->mbuf_phys_off);
1402 	}
1403 	return 0;
1404 }
1405 
1406 static int
1407 nicvf_vf_start(struct rte_eth_dev *dev, struct nicvf *nic, uint32_t rbdrsz)
1408 {
1409 	int ret;
1410 	uint16_t qidx;
1411 	uint32_t total_rxq_desc, nb_rbdr_desc, exp_buffs;
1412 	uint64_t mbuf_phys_off = 0;
1413 	struct nicvf_rxq *rxq;
1414 	struct rte_mbuf *mbuf;
1415 	uint16_t rx_start, rx_end;
1416 	uint16_t tx_start, tx_end;
1417 
1418 	PMD_INIT_FUNC_TRACE();
1419 
1420 	/* Userspace process exited without proper shutdown in last run */
1421 	if (nicvf_qset_rbdr_active(nic, 0))
1422 		nicvf_vf_stop(dev, nic, false);
1423 
1424 	/* Get queue ranges for this VF */
1425 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
1426 
1427 	/*
1428 	 * Thunderx nicvf PMD can support more than one pool per port only when
1429 	 * 1) Data payload size is same across all the pools in given port
1430 	 * AND
1431 	 * 2) All mbuffs in the pools are from the same hugepage
1432 	 * AND
1433 	 * 3) Mbuff metadata size is same across all the pools in given port
1434 	 *
1435 	 * This is to support existing application that uses multiple pool/port.
1436 	 * But, the purpose of using multipool for QoS will not be addressed.
1437 	 *
1438 	 */
1439 
1440 	/* Validate mempool attributes */
1441 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1442 		rxq = dev->data->rx_queues[qidx];
1443 		rxq->mbuf_phys_off = nicvf_mempool_phy_offset(rxq->pool);
1444 		mbuf = rte_pktmbuf_alloc(rxq->pool);
1445 		if (mbuf == NULL) {
1446 			PMD_INIT_LOG(ERR, "Failed allocate mbuf VF%d qid=%d "
1447 				     "pool=%s",
1448 				     nic->vf_id, qidx, rxq->pool->name);
1449 			return -ENOMEM;
1450 		}
1451 		rxq->mbuf_phys_off -= nicvf_mbuff_meta_length(mbuf);
1452 		rxq->mbuf_phys_off -= RTE_PKTMBUF_HEADROOM;
1453 		rte_pktmbuf_free(mbuf);
1454 
1455 		if (mbuf_phys_off == 0)
1456 			mbuf_phys_off = rxq->mbuf_phys_off;
1457 		if (mbuf_phys_off != rxq->mbuf_phys_off) {
1458 			PMD_INIT_LOG(ERR, "pool params not same,%s VF%d %"
1459 				     PRIx64, rxq->pool->name, nic->vf_id,
1460 				     mbuf_phys_off);
1461 			return -EINVAL;
1462 		}
1463 	}
1464 
1465 	/* Check the level of buffers in the pool */
1466 	total_rxq_desc = 0;
1467 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1468 		rxq = dev->data->rx_queues[qidx];
1469 		/* Count total numbers of rxq descs */
1470 		total_rxq_desc += rxq->qlen_mask + 1;
1471 		exp_buffs = RTE_MEMPOOL_CACHE_MAX_SIZE + rxq->rx_free_thresh;
1472 		exp_buffs *= dev->data->nb_rx_queues;
1473 		if (rte_mempool_avail_count(rxq->pool) < exp_buffs) {
1474 			PMD_INIT_LOG(ERR, "Buff shortage in pool=%s (%d/%d)",
1475 				     rxq->pool->name,
1476 				     rte_mempool_avail_count(rxq->pool),
1477 				     exp_buffs);
1478 			return -ENOENT;
1479 		}
1480 	}
1481 
1482 	/* Check RBDR desc overflow */
1483 	ret = nicvf_qsize_rbdr_roundup(total_rxq_desc);
1484 	if (ret == 0) {
1485 		PMD_INIT_LOG(ERR, "Reached RBDR desc limit, reduce nr desc "
1486 			     "VF%d", nic->vf_id);
1487 		return -ENOMEM;
1488 	}
1489 
1490 	/* Enable qset */
1491 	ret = nicvf_qset_config(nic);
1492 	if (ret) {
1493 		PMD_INIT_LOG(ERR, "Failed to enable qset %d VF%d", ret,
1494 			     nic->vf_id);
1495 		return ret;
1496 	}
1497 
1498 	/* Allocate RBDR and RBDR ring desc */
1499 	nb_rbdr_desc = nicvf_qsize_rbdr_roundup(total_rxq_desc);
1500 	ret = nicvf_qset_rbdr_alloc(dev, nic, nb_rbdr_desc, rbdrsz);
1501 	if (ret) {
1502 		PMD_INIT_LOG(ERR, "Failed to allocate memory for rbdr alloc "
1503 			     "VF%d", nic->vf_id);
1504 		goto qset_reclaim;
1505 	}
1506 
1507 	/* Enable and configure RBDR registers */
1508 	ret = nicvf_qset_rbdr_config(nic, 0);
1509 	if (ret) {
1510 		PMD_INIT_LOG(ERR, "Failed to configure rbdr %d VF%d", ret,
1511 			     nic->vf_id);
1512 		goto qset_rbdr_free;
1513 	}
1514 
1515 	/* Fill rte_mempool buffers in RBDR pool and precharge it */
1516 	ret = nicvf_qset_rbdr_precharge(dev, nic, 0, rbdr_rte_mempool_get,
1517 					total_rxq_desc);
1518 	if (ret) {
1519 		PMD_INIT_LOG(ERR, "Failed to fill rbdr %d VF%d", ret,
1520 			     nic->vf_id);
1521 		goto qset_rbdr_reclaim;
1522 	}
1523 
1524 	PMD_DRV_LOG(INFO, "Filled %d out of %d entries in RBDR VF%d",
1525 		     nic->rbdr->tail, nb_rbdr_desc, nic->vf_id);
1526 
1527 	/* Configure VLAN Strip */
1528 	nicvf_vlan_hw_strip(nic, dev->data->dev_conf.rxmode.hw_vlan_strip);
1529 
1530 	/* Based on the packet type(IPv4 or IPv6), the nicvf HW aligns L3 data
1531 	 * to the 64bit memory address.
1532 	 * The alignment creates a hole in mbuf(between the end of headroom and
1533 	 * packet data start). The new revision of the HW provides an option to
1534 	 * disable the L3 alignment feature and make mbuf layout looks
1535 	 * more like other NICs. For better application compatibility, disabling
1536 	 * l3 alignment feature on the hardware revisions it supports
1537 	 */
1538 	nicvf_apad_config(nic, false);
1539 
1540 	/* Get queue ranges for this VF */
1541 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
1542 
1543 	/* Configure TX queues */
1544 	for (qidx = tx_start; qidx <= tx_end; qidx++) {
1545 		ret = nicvf_vf_start_tx_queue(dev, nic,
1546 			qidx % MAX_SND_QUEUES_PER_QS);
1547 		if (ret)
1548 			goto start_txq_error;
1549 	}
1550 
1551 	/* Configure RX queues */
1552 	for (qidx = rx_start; qidx <= rx_end; qidx++) {
1553 		ret = nicvf_vf_start_rx_queue(dev, nic,
1554 			qidx % MAX_RCV_QUEUES_PER_QS);
1555 		if (ret)
1556 			goto start_rxq_error;
1557 	}
1558 
1559 	if (!nic->sqs_mode) {
1560 		/* Configure CPI algorithm */
1561 		ret = nicvf_configure_cpi(dev);
1562 		if (ret)
1563 			goto start_txq_error;
1564 
1565 		ret = nicvf_mbox_get_rss_size(nic);
1566 		if (ret) {
1567 			PMD_INIT_LOG(ERR, "Failed to get rss table size");
1568 			goto qset_rss_error;
1569 		}
1570 
1571 		/* Configure RSS */
1572 		ret = nicvf_configure_rss(dev);
1573 		if (ret)
1574 			goto qset_rss_error;
1575 	}
1576 
1577 	/* Done; Let PF make the BGX's RX and TX switches to ON position */
1578 	nicvf_mbox_cfg_done(nic);
1579 	return 0;
1580 
1581 qset_rss_error:
1582 	nicvf_rss_term(nic);
1583 start_rxq_error:
1584 	for (qidx = rx_start; qidx <= rx_end; qidx++)
1585 		nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
1586 start_txq_error:
1587 	for (qidx = tx_start; qidx <= tx_end; qidx++)
1588 		nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
1589 qset_rbdr_reclaim:
1590 	nicvf_qset_rbdr_reclaim(nic, 0);
1591 	nicvf_rbdr_release_mbufs(dev, nic);
1592 qset_rbdr_free:
1593 	if (nic->rbdr) {
1594 		rte_free(nic->rbdr);
1595 		nic->rbdr = NULL;
1596 	}
1597 qset_reclaim:
1598 	nicvf_qset_reclaim(nic);
1599 	return ret;
1600 }
1601 
1602 static int
1603 nicvf_dev_start(struct rte_eth_dev *dev)
1604 {
1605 	uint16_t qidx;
1606 	int ret;
1607 	size_t i;
1608 	struct nicvf *nic = nicvf_pmd_priv(dev);
1609 	struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
1610 	uint16_t mtu;
1611 	uint32_t buffsz = 0, rbdrsz = 0;
1612 	struct rte_pktmbuf_pool_private *mbp_priv;
1613 	struct nicvf_rxq *rxq;
1614 
1615 	PMD_INIT_FUNC_TRACE();
1616 
1617 	/* This function must be called for a primary device */
1618 	assert_primary(nic);
1619 
1620 	/* Validate RBDR buff size */
1621 	for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
1622 		rxq = dev->data->rx_queues[qidx];
1623 		mbp_priv = rte_mempool_get_priv(rxq->pool);
1624 		buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
1625 		if (buffsz % 128) {
1626 			PMD_INIT_LOG(ERR, "rxbuf size must be multiply of 128");
1627 			return -EINVAL;
1628 		}
1629 		if (rbdrsz == 0)
1630 			rbdrsz = buffsz;
1631 		if (rbdrsz != buffsz) {
1632 			PMD_INIT_LOG(ERR, "buffsz not same, qidx=%d (%d/%d)",
1633 				     qidx, rbdrsz, buffsz);
1634 			return -EINVAL;
1635 		}
1636 	}
1637 
1638 	/* Configure loopback */
1639 	ret = nicvf_loopback_config(nic, dev->data->dev_conf.lpbk_mode);
1640 	if (ret) {
1641 		PMD_INIT_LOG(ERR, "Failed to configure loopback %d", ret);
1642 		return ret;
1643 	}
1644 
1645 	/* Reset all statistics counters attached to this port */
1646 	ret = nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, 0xFFFF, 0xFFFF);
1647 	if (ret) {
1648 		PMD_INIT_LOG(ERR, "Failed to reset stat counters %d", ret);
1649 		return ret;
1650 	}
1651 
1652 	/* Setup scatter mode if needed by jumbo */
1653 	if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
1654 					    2 * VLAN_TAG_SIZE > buffsz)
1655 		dev->data->scattered_rx = 1;
1656 	if (rx_conf->enable_scatter)
1657 		dev->data->scattered_rx = 1;
1658 
1659 	/* Setup MTU based on max_rx_pkt_len or default */
1660 	mtu = dev->data->dev_conf.rxmode.jumbo_frame ?
1661 		dev->data->dev_conf.rxmode.max_rx_pkt_len
1662 			-  ETHER_HDR_LEN - ETHER_CRC_LEN
1663 		: ETHER_MTU;
1664 
1665 	if (nicvf_dev_set_mtu(dev, mtu)) {
1666 		PMD_INIT_LOG(ERR, "Failed to set default mtu size");
1667 		return -EBUSY;
1668 	}
1669 
1670 	ret = nicvf_vf_start(dev, nic, rbdrsz);
1671 	if (ret != 0)
1672 		return ret;
1673 
1674 	for (i = 0; i < nic->sqs_count; i++) {
1675 		assert(nic->snicvf[i]);
1676 
1677 		ret = nicvf_vf_start(dev, nic->snicvf[i], rbdrsz);
1678 		if (ret != 0)
1679 			return ret;
1680 	}
1681 
1682 	/* Configure callbacks based on scatter mode */
1683 	nicvf_set_tx_function(dev);
1684 	nicvf_set_rx_function(dev);
1685 
1686 	return 0;
1687 }
1688 
1689 static void
1690 nicvf_dev_stop_cleanup(struct rte_eth_dev *dev, bool cleanup)
1691 {
1692 	size_t i;
1693 	int ret;
1694 	struct nicvf *nic = nicvf_pmd_priv(dev);
1695 
1696 	PMD_INIT_FUNC_TRACE();
1697 
1698 	/* Teardown secondary vf first */
1699 	for (i = 0; i < nic->sqs_count; i++) {
1700 		if (!nic->snicvf[i])
1701 			continue;
1702 
1703 		nicvf_vf_stop(dev, nic->snicvf[i], cleanup);
1704 	}
1705 
1706 	/* Stop the primary VF now */
1707 	nicvf_vf_stop(dev, nic, cleanup);
1708 
1709 	/* Disable loopback */
1710 	ret = nicvf_loopback_config(nic, 0);
1711 	if (ret)
1712 		PMD_INIT_LOG(ERR, "Failed to disable loopback %d", ret);
1713 
1714 	/* Reclaim CPI configuration */
1715 	ret = nicvf_mbox_config_cpi(nic, 0);
1716 	if (ret)
1717 		PMD_INIT_LOG(ERR, "Failed to reclaim CPI config %d", ret);
1718 }
1719 
1720 static void
1721 nicvf_dev_stop(struct rte_eth_dev *dev)
1722 {
1723 	PMD_INIT_FUNC_TRACE();
1724 
1725 	nicvf_dev_stop_cleanup(dev, false);
1726 }
1727 
1728 static void
1729 nicvf_vf_stop(struct rte_eth_dev *dev, struct nicvf *nic, bool cleanup)
1730 {
1731 	int ret;
1732 	uint16_t qidx;
1733 	uint16_t tx_start, tx_end;
1734 	uint16_t rx_start, rx_end;
1735 
1736 	PMD_INIT_FUNC_TRACE();
1737 
1738 	if (cleanup) {
1739 		/* Let PF make the BGX's RX and TX switches to OFF position */
1740 		nicvf_mbox_shutdown(nic);
1741 	}
1742 
1743 	/* Disable VLAN Strip */
1744 	nicvf_vlan_hw_strip(nic, 0);
1745 
1746 	/* Get queue ranges for this VF */
1747 	nicvf_tx_range(dev, nic, &tx_start, &tx_end);
1748 
1749 	for (qidx = tx_start; qidx <= tx_end; qidx++)
1750 		nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
1751 
1752 	/* Get queue ranges for this VF */
1753 	nicvf_rx_range(dev, nic, &rx_start, &rx_end);
1754 
1755 	/* Reclaim rq */
1756 	for (qidx = rx_start; qidx <= rx_end; qidx++)
1757 		nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
1758 
1759 	/* Reclaim RBDR */
1760 	ret = nicvf_qset_rbdr_reclaim(nic, 0);
1761 	if (ret)
1762 		PMD_INIT_LOG(ERR, "Failed to reclaim RBDR %d", ret);
1763 
1764 	/* Move all charged buffers in RBDR back to pool */
1765 	if (nic->rbdr != NULL)
1766 		nicvf_rbdr_release_mbufs(dev, nic);
1767 
1768 	/* Disable qset */
1769 	ret = nicvf_qset_reclaim(nic);
1770 	if (ret)
1771 		PMD_INIT_LOG(ERR, "Failed to disable qset %d", ret);
1772 
1773 	/* Disable all interrupts */
1774 	nicvf_disable_all_interrupts(nic);
1775 
1776 	/* Free RBDR SW structure */
1777 	if (nic->rbdr) {
1778 		rte_free(nic->rbdr);
1779 		nic->rbdr = NULL;
1780 	}
1781 }
1782 
1783 static void
1784 nicvf_dev_close(struct rte_eth_dev *dev)
1785 {
1786 	size_t i;
1787 	struct nicvf *nic = nicvf_pmd_priv(dev);
1788 
1789 	PMD_INIT_FUNC_TRACE();
1790 
1791 	nicvf_dev_stop_cleanup(dev, true);
1792 	nicvf_periodic_alarm_stop(nicvf_interrupt, dev);
1793 
1794 	for (i = 0; i < nic->sqs_count; i++) {
1795 		if (!nic->snicvf[i])
1796 			continue;
1797 
1798 		nicvf_periodic_alarm_stop(nicvf_vf_interrupt, nic->snicvf[i]);
1799 	}
1800 }
1801 
1802 static int
1803 nicvf_request_sqs(struct nicvf *nic)
1804 {
1805 	size_t i;
1806 
1807 	assert_primary(nic);
1808 	assert(nic->sqs_count > 0);
1809 	assert(nic->sqs_count <= MAX_SQS_PER_VF);
1810 
1811 	/* Set no of Rx/Tx queues in each of the SQsets */
1812 	for (i = 0; i < nic->sqs_count; i++) {
1813 		if (nicvf_svf_empty())
1814 			rte_panic("Cannot assign sufficient number of "
1815 				  "secondary queues to primary VF%" PRIu8 "\n",
1816 				  nic->vf_id);
1817 
1818 		nic->snicvf[i] = nicvf_svf_pop();
1819 		nic->snicvf[i]->sqs_id = i;
1820 	}
1821 
1822 	return nicvf_mbox_request_sqs(nic);
1823 }
1824 
1825 static int
1826 nicvf_dev_configure(struct rte_eth_dev *dev)
1827 {
1828 	struct rte_eth_dev_data *data = dev->data;
1829 	struct rte_eth_conf *conf = &data->dev_conf;
1830 	struct rte_eth_rxmode *rxmode = &conf->rxmode;
1831 	struct rte_eth_txmode *txmode = &conf->txmode;
1832 	struct nicvf *nic = nicvf_pmd_priv(dev);
1833 	uint8_t cqcount;
1834 
1835 	PMD_INIT_FUNC_TRACE();
1836 
1837 	if (!rte_eal_has_hugepages()) {
1838 		PMD_INIT_LOG(INFO, "Huge page is not configured");
1839 		return -EINVAL;
1840 	}
1841 
1842 	if (txmode->mq_mode) {
1843 		PMD_INIT_LOG(INFO, "Tx mq_mode DCB or VMDq not supported");
1844 		return -EINVAL;
1845 	}
1846 
1847 	if (rxmode->mq_mode != ETH_MQ_RX_NONE &&
1848 		rxmode->mq_mode != ETH_MQ_RX_RSS) {
1849 		PMD_INIT_LOG(INFO, "Unsupported rx qmode %d", rxmode->mq_mode);
1850 		return -EINVAL;
1851 	}
1852 
1853 	if (!rxmode->hw_strip_crc) {
1854 		PMD_INIT_LOG(NOTICE, "Can't disable hw crc strip");
1855 		rxmode->hw_strip_crc = 1;
1856 	}
1857 
1858 	if (rxmode->hw_ip_checksum) {
1859 		PMD_INIT_LOG(NOTICE, "Rxcksum not supported");
1860 		rxmode->hw_ip_checksum = 0;
1861 	}
1862 
1863 	if (rxmode->split_hdr_size) {
1864 		PMD_INIT_LOG(INFO, "Rxmode does not support split header");
1865 		return -EINVAL;
1866 	}
1867 
1868 	if (rxmode->hw_vlan_filter) {
1869 		PMD_INIT_LOG(INFO, "VLAN filter not supported");
1870 		return -EINVAL;
1871 	}
1872 
1873 	if (rxmode->hw_vlan_extend) {
1874 		PMD_INIT_LOG(INFO, "VLAN extended not supported");
1875 		return -EINVAL;
1876 	}
1877 
1878 	if (rxmode->enable_lro) {
1879 		PMD_INIT_LOG(INFO, "LRO not supported");
1880 		return -EINVAL;
1881 	}
1882 
1883 	if (conf->link_speeds & ETH_LINK_SPEED_FIXED) {
1884 		PMD_INIT_LOG(INFO, "Setting link speed/duplex not supported");
1885 		return -EINVAL;
1886 	}
1887 
1888 	if (conf->dcb_capability_en) {
1889 		PMD_INIT_LOG(INFO, "DCB enable not supported");
1890 		return -EINVAL;
1891 	}
1892 
1893 	if (conf->fdir_conf.mode != RTE_FDIR_MODE_NONE) {
1894 		PMD_INIT_LOG(INFO, "Flow director not supported");
1895 		return -EINVAL;
1896 	}
1897 
1898 	assert_primary(nic);
1899 	NICVF_STATIC_ASSERT(MAX_RCV_QUEUES_PER_QS == MAX_SND_QUEUES_PER_QS);
1900 	cqcount = RTE_MAX(data->nb_tx_queues, data->nb_rx_queues);
1901 	if (cqcount > MAX_RCV_QUEUES_PER_QS) {
1902 		nic->sqs_count = RTE_ALIGN_CEIL(cqcount, MAX_RCV_QUEUES_PER_QS);
1903 		nic->sqs_count = (nic->sqs_count / MAX_RCV_QUEUES_PER_QS) - 1;
1904 	} else {
1905 		nic->sqs_count = 0;
1906 	}
1907 
1908 	assert(nic->sqs_count <= MAX_SQS_PER_VF);
1909 
1910 	if (nic->sqs_count > 0) {
1911 		if (nicvf_request_sqs(nic)) {
1912 			rte_panic("Cannot assign sufficient number of "
1913 				  "secondary queues to PORT%d VF%" PRIu8 "\n",
1914 				  dev->data->port_id, nic->vf_id);
1915 		}
1916 	}
1917 
1918 	PMD_INIT_LOG(DEBUG, "Configured ethdev port%d hwcap=0x%" PRIx64,
1919 		dev->data->port_id, nicvf_hw_cap(nic));
1920 
1921 	return 0;
1922 }
1923 
1924 /* Initialize and register driver with DPDK Application */
1925 static const struct eth_dev_ops nicvf_eth_dev_ops = {
1926 	.dev_configure            = nicvf_dev_configure,
1927 	.dev_start                = nicvf_dev_start,
1928 	.dev_stop                 = nicvf_dev_stop,
1929 	.link_update              = nicvf_dev_link_update,
1930 	.dev_close                = nicvf_dev_close,
1931 	.stats_get                = nicvf_dev_stats_get,
1932 	.stats_reset              = nicvf_dev_stats_reset,
1933 	.promiscuous_enable       = nicvf_dev_promisc_enable,
1934 	.dev_infos_get            = nicvf_dev_info_get,
1935 	.dev_supported_ptypes_get = nicvf_dev_supported_ptypes_get,
1936 	.mtu_set                  = nicvf_dev_set_mtu,
1937 	.reta_update              = nicvf_dev_reta_update,
1938 	.reta_query               = nicvf_dev_reta_query,
1939 	.rss_hash_update          = nicvf_dev_rss_hash_update,
1940 	.rss_hash_conf_get        = nicvf_dev_rss_hash_conf_get,
1941 	.rx_queue_start           = nicvf_dev_rx_queue_start,
1942 	.rx_queue_stop            = nicvf_dev_rx_queue_stop,
1943 	.tx_queue_start           = nicvf_dev_tx_queue_start,
1944 	.tx_queue_stop            = nicvf_dev_tx_queue_stop,
1945 	.rx_queue_setup           = nicvf_dev_rx_queue_setup,
1946 	.rx_queue_release         = nicvf_dev_rx_queue_release,
1947 	.rx_queue_count           = nicvf_dev_rx_queue_count,
1948 	.tx_queue_setup           = nicvf_dev_tx_queue_setup,
1949 	.tx_queue_release         = nicvf_dev_tx_queue_release,
1950 	.get_reg                  = nicvf_dev_get_regs,
1951 };
1952 
1953 static int
1954 nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
1955 {
1956 	int ret;
1957 	struct rte_pci_device *pci_dev;
1958 	struct nicvf *nic = nicvf_pmd_priv(eth_dev);
1959 
1960 	PMD_INIT_FUNC_TRACE();
1961 
1962 	eth_dev->dev_ops = &nicvf_eth_dev_ops;
1963 
1964 	/* For secondary processes, the primary has done all the work */
1965 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1966 		if (nic) {
1967 			/* Setup callbacks for secondary process */
1968 			nicvf_set_tx_function(eth_dev);
1969 			nicvf_set_rx_function(eth_dev);
1970 			return 0;
1971 		} else {
1972 			/* If nic == NULL than it is secondary function
1973 			 * so ethdev need to be released by caller */
1974 			return ENOTSUP;
1975 		}
1976 	}
1977 
1978 	pci_dev = eth_dev->pci_dev;
1979 	rte_eth_copy_pci_info(eth_dev, pci_dev);
1980 
1981 	nic->device_id = pci_dev->id.device_id;
1982 	nic->vendor_id = pci_dev->id.vendor_id;
1983 	nic->subsystem_device_id = pci_dev->id.subsystem_device_id;
1984 	nic->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
1985 
1986 	PMD_INIT_LOG(DEBUG, "nicvf: device (%x:%x) %u:%u:%u:%u",
1987 			pci_dev->id.vendor_id, pci_dev->id.device_id,
1988 			pci_dev->addr.domain, pci_dev->addr.bus,
1989 			pci_dev->addr.devid, pci_dev->addr.function);
1990 
1991 	nic->reg_base = (uintptr_t)pci_dev->mem_resource[0].addr;
1992 	if (!nic->reg_base) {
1993 		PMD_INIT_LOG(ERR, "Failed to map BAR0");
1994 		ret = -ENODEV;
1995 		goto fail;
1996 	}
1997 
1998 	nicvf_disable_all_interrupts(nic);
1999 
2000 	ret = nicvf_periodic_alarm_start(nicvf_interrupt, eth_dev);
2001 	if (ret) {
2002 		PMD_INIT_LOG(ERR, "Failed to start period alarm");
2003 		goto fail;
2004 	}
2005 
2006 	ret = nicvf_mbox_check_pf_ready(nic);
2007 	if (ret) {
2008 		PMD_INIT_LOG(ERR, "Failed to get ready message from PF");
2009 		goto alarm_fail;
2010 	} else {
2011 		PMD_INIT_LOG(INFO,
2012 			"node=%d vf=%d mode=%s sqs=%s loopback_supported=%s",
2013 			nic->node, nic->vf_id,
2014 			nic->tns_mode == NIC_TNS_MODE ? "tns" : "tns-bypass",
2015 			nic->sqs_mode ? "true" : "false",
2016 			nic->loopback_supported ? "true" : "false"
2017 			);
2018 	}
2019 
2020 	ret = nicvf_base_init(nic);
2021 	if (ret) {
2022 		PMD_INIT_LOG(ERR, "Failed to execute nicvf_base_init");
2023 		goto malloc_fail;
2024 	}
2025 
2026 	if (nic->sqs_mode) {
2027 		/* Push nic to stack of secondary vfs */
2028 		nicvf_svf_push(nic);
2029 
2030 		/* Steal nic pointer from the device for further reuse */
2031 		eth_dev->data->dev_private = NULL;
2032 
2033 		nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
2034 		ret = nicvf_periodic_alarm_start(nicvf_vf_interrupt, nic);
2035 		if (ret) {
2036 			PMD_INIT_LOG(ERR, "Failed to start period alarm");
2037 			goto fail;
2038 		}
2039 
2040 		/* Detach port by returning postive error number */
2041 		return ENOTSUP;
2042 	}
2043 
2044 	eth_dev->data->mac_addrs = rte_zmalloc("mac_addr", ETHER_ADDR_LEN, 0);
2045 	if (eth_dev->data->mac_addrs == NULL) {
2046 		PMD_INIT_LOG(ERR, "Failed to allocate memory for mac addr");
2047 		ret = -ENOMEM;
2048 		goto alarm_fail;
2049 	}
2050 	if (is_zero_ether_addr((struct ether_addr *)nic->mac_addr))
2051 		eth_random_addr(&nic->mac_addr[0]);
2052 
2053 	ether_addr_copy((struct ether_addr *)nic->mac_addr,
2054 			&eth_dev->data->mac_addrs[0]);
2055 
2056 	ret = nicvf_mbox_set_mac_addr(nic, nic->mac_addr);
2057 	if (ret) {
2058 		PMD_INIT_LOG(ERR, "Failed to set mac addr");
2059 		goto malloc_fail;
2060 	}
2061 
2062 	PMD_INIT_LOG(INFO, "Port %d (%x:%x) mac=%02x:%02x:%02x:%02x:%02x:%02x",
2063 		eth_dev->data->port_id, nic->vendor_id, nic->device_id,
2064 		nic->mac_addr[0], nic->mac_addr[1], nic->mac_addr[2],
2065 		nic->mac_addr[3], nic->mac_addr[4], nic->mac_addr[5]);
2066 
2067 	return 0;
2068 
2069 malloc_fail:
2070 	rte_free(eth_dev->data->mac_addrs);
2071 alarm_fail:
2072 	nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
2073 fail:
2074 	return ret;
2075 }
2076 
2077 static const struct rte_pci_id pci_id_nicvf_map[] = {
2078 	{
2079 		.class_id = RTE_CLASS_ANY_ID,
2080 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2081 		.device_id = PCI_DEVICE_ID_THUNDERX_CN88XX_PASS1_NICVF,
2082 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2083 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS1_NICVF,
2084 	},
2085 	{
2086 		.class_id = RTE_CLASS_ANY_ID,
2087 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2088 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2089 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2090 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS2_NICVF,
2091 	},
2092 	{
2093 		.class_id = RTE_CLASS_ANY_ID,
2094 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2095 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2096 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2097 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN81XX_NICVF,
2098 	},
2099 	{
2100 		.class_id = RTE_CLASS_ANY_ID,
2101 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
2102 		.device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2103 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2104 		.subsystem_device_id = PCI_SUB_DEVICE_ID_CN83XX_NICVF,
2105 	},
2106 	{
2107 		.vendor_id = 0,
2108 	},
2109 };
2110 
2111 static struct eth_driver rte_nicvf_pmd = {
2112 	.pci_drv = {
2113 		.id_table = pci_id_nicvf_map,
2114 		.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
2115 		.probe = rte_eth_dev_pci_probe,
2116 		.remove = rte_eth_dev_pci_remove,
2117 	},
2118 	.eth_dev_init = nicvf_eth_dev_init,
2119 	.dev_private_size = sizeof(struct nicvf),
2120 };
2121 
2122 RTE_PMD_REGISTER_PCI(net_thunderx, rte_nicvf_pmd.pci_drv);
2123 RTE_PMD_REGISTER_PCI_TABLE(net_thunderx, pci_id_nicvf_map);
2124 RTE_PMD_REGISTER_KMOD_DEP(net_thunderx, "* igb_uio | uio_pci_generic | vfio");
2125