xref: /dpdk/drivers/net/thunderx/nicvf_ethdev.c (revision 43362c6a76476f6807390ac51a6d87e6dedada6e)
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 
70 #include "nicvf_logs.h"
71 
72 static inline int
73 nicvf_atomic_write_link_status(struct rte_eth_dev *dev,
74 			       struct rte_eth_link *link)
75 {
76 	struct rte_eth_link *dst = &dev->data->dev_link;
77 	struct rte_eth_link *src = link;
78 
79 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
80 		*(uint64_t *)src) == 0)
81 		return -1;
82 
83 	return 0;
84 }
85 
86 static inline void
87 nicvf_set_eth_link_status(struct nicvf *nic, struct rte_eth_link *link)
88 {
89 	link->link_status = nic->link_up;
90 	link->link_duplex = ETH_LINK_AUTONEG;
91 	if (nic->duplex == NICVF_HALF_DUPLEX)
92 		link->link_duplex = ETH_LINK_HALF_DUPLEX;
93 	else if (nic->duplex == NICVF_FULL_DUPLEX)
94 		link->link_duplex = ETH_LINK_FULL_DUPLEX;
95 	link->link_speed = nic->speed;
96 	link->link_autoneg = ETH_LINK_SPEED_AUTONEG;
97 }
98 
99 static void
100 nicvf_interrupt(void *arg)
101 {
102 	struct nicvf *nic = arg;
103 
104 	if (nicvf_reg_poll_interrupts(nic) == NIC_MBOX_MSG_BGX_LINK_CHANGE) {
105 		if (nic->eth_dev->data->dev_conf.intr_conf.lsc)
106 			nicvf_set_eth_link_status(nic,
107 					&nic->eth_dev->data->dev_link);
108 		_rte_eth_dev_callback_process(nic->eth_dev,
109 				RTE_ETH_EVENT_INTR_LSC);
110 	}
111 
112 	rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
113 				nicvf_interrupt, nic);
114 }
115 
116 static int
117 nicvf_periodic_alarm_start(struct nicvf *nic)
118 {
119 	return rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
120 					nicvf_interrupt, nic);
121 }
122 
123 static int
124 nicvf_periodic_alarm_stop(struct nicvf *nic)
125 {
126 	return rte_eal_alarm_cancel(nicvf_interrupt, nic);
127 }
128 
129 /*
130  * Return 0 means link status changed, -1 means not changed
131  */
132 static int
133 nicvf_dev_link_update(struct rte_eth_dev *dev,
134 		      int wait_to_complete __rte_unused)
135 {
136 	struct rte_eth_link link;
137 	struct nicvf *nic = nicvf_pmd_priv(dev);
138 
139 	PMD_INIT_FUNC_TRACE();
140 
141 	memset(&link, 0, sizeof(link));
142 	nicvf_set_eth_link_status(nic, &link);
143 	return nicvf_atomic_write_link_status(dev, &link);
144 }
145 
146 static int
147 nicvf_dev_get_reg_length(struct rte_eth_dev *dev  __rte_unused)
148 {
149 	return nicvf_reg_get_count();
150 }
151 
152 static int
153 nicvf_dev_get_regs(struct rte_eth_dev *dev, struct rte_dev_reg_info *regs)
154 {
155 	uint64_t *data = regs->data;
156 	struct nicvf *nic = nicvf_pmd_priv(dev);
157 
158 	if (data == NULL)
159 		return -EINVAL;
160 
161 	/* Support only full register dump */
162 	if ((regs->length == 0) ||
163 		(regs->length == (uint32_t)nicvf_reg_get_count())) {
164 		regs->version = nic->vendor_id << 16 | nic->device_id;
165 		nicvf_reg_dump(nic, data);
166 		return 0;
167 	}
168 	return -ENOTSUP;
169 }
170 
171 static inline uint64_t
172 nicvf_rss_ethdev_to_nic(struct nicvf *nic, uint64_t ethdev_rss)
173 {
174 	uint64_t nic_rss = 0;
175 
176 	if (ethdev_rss & ETH_RSS_IPV4)
177 		nic_rss |= RSS_IP_ENA;
178 
179 	if (ethdev_rss & ETH_RSS_IPV6)
180 		nic_rss |= RSS_IP_ENA;
181 
182 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_UDP)
183 		nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
184 
185 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_TCP)
186 		nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
187 
188 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_UDP)
189 		nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
190 
191 	if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_TCP)
192 		nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
193 
194 	if (ethdev_rss & ETH_RSS_PORT)
195 		nic_rss |= RSS_L2_EXTENDED_HASH_ENA;
196 
197 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
198 		if (ethdev_rss & ETH_RSS_VXLAN)
199 			nic_rss |= RSS_TUN_VXLAN_ENA;
200 
201 		if (ethdev_rss & ETH_RSS_GENEVE)
202 			nic_rss |= RSS_TUN_GENEVE_ENA;
203 
204 		if (ethdev_rss & ETH_RSS_NVGRE)
205 			nic_rss |= RSS_TUN_NVGRE_ENA;
206 	}
207 
208 	return nic_rss;
209 }
210 
211 static inline uint64_t
212 nicvf_rss_nic_to_ethdev(struct nicvf *nic,  uint64_t nic_rss)
213 {
214 	uint64_t ethdev_rss = 0;
215 
216 	if (nic_rss & RSS_IP_ENA)
217 		ethdev_rss |= (ETH_RSS_IPV4 | ETH_RSS_IPV6);
218 
219 	if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_TCP_ENA))
220 		ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_TCP |
221 				ETH_RSS_NONFRAG_IPV6_TCP);
222 
223 	if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_UDP_ENA))
224 		ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_UDP |
225 				ETH_RSS_NONFRAG_IPV6_UDP);
226 
227 	if (nic_rss & RSS_L2_EXTENDED_HASH_ENA)
228 		ethdev_rss |= ETH_RSS_PORT;
229 
230 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
231 		if (nic_rss & RSS_TUN_VXLAN_ENA)
232 			ethdev_rss |= ETH_RSS_VXLAN;
233 
234 		if (nic_rss & RSS_TUN_GENEVE_ENA)
235 			ethdev_rss |= ETH_RSS_GENEVE;
236 
237 		if (nic_rss & RSS_TUN_NVGRE_ENA)
238 			ethdev_rss |= ETH_RSS_NVGRE;
239 	}
240 	return ethdev_rss;
241 }
242 
243 static int
244 nicvf_dev_reta_query(struct rte_eth_dev *dev,
245 		     struct rte_eth_rss_reta_entry64 *reta_conf,
246 		     uint16_t reta_size)
247 {
248 	struct nicvf *nic = nicvf_pmd_priv(dev);
249 	uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
250 	int ret, i, j;
251 
252 	if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
253 		RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
254 			"(%d) doesn't match the number hardware can supported "
255 			"(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
256 		return -EINVAL;
257 	}
258 
259 	ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
260 	if (ret)
261 		return ret;
262 
263 	/* Copy RETA table */
264 	for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
265 		for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
266 			if ((reta_conf[i].mask >> j) & 0x01)
267 				reta_conf[i].reta[j] = tbl[j];
268 	}
269 
270 	return 0;
271 }
272 
273 static int
274 nicvf_dev_reta_update(struct rte_eth_dev *dev,
275 		      struct rte_eth_rss_reta_entry64 *reta_conf,
276 		      uint16_t reta_size)
277 {
278 	struct nicvf *nic = nicvf_pmd_priv(dev);
279 	uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
280 	int ret, i, j;
281 
282 	if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
283 		RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
284 			"(%d) doesn't match the number hardware can supported "
285 			"(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
286 		return -EINVAL;
287 	}
288 
289 	ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
290 	if (ret)
291 		return ret;
292 
293 	/* Copy RETA table */
294 	for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
295 		for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
296 			if ((reta_conf[i].mask >> j) & 0x01)
297 				tbl[j] = reta_conf[i].reta[j];
298 	}
299 
300 	return nicvf_rss_reta_update(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
301 }
302 
303 static int
304 nicvf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
305 			    struct rte_eth_rss_conf *rss_conf)
306 {
307 	struct nicvf *nic = nicvf_pmd_priv(dev);
308 
309 	if (rss_conf->rss_key)
310 		nicvf_rss_get_key(nic, rss_conf->rss_key);
311 
312 	rss_conf->rss_key_len =  RSS_HASH_KEY_BYTE_SIZE;
313 	rss_conf->rss_hf = nicvf_rss_nic_to_ethdev(nic, nicvf_rss_get_cfg(nic));
314 	return 0;
315 }
316 
317 static int
318 nicvf_dev_rss_hash_update(struct rte_eth_dev *dev,
319 			  struct rte_eth_rss_conf *rss_conf)
320 {
321 	struct nicvf *nic = nicvf_pmd_priv(dev);
322 	uint64_t nic_rss;
323 
324 	if (rss_conf->rss_key &&
325 		rss_conf->rss_key_len != RSS_HASH_KEY_BYTE_SIZE) {
326 		RTE_LOG(ERR, PMD, "Hash key size mismatch %d",
327 				rss_conf->rss_key_len);
328 		return -EINVAL;
329 	}
330 
331 	if (rss_conf->rss_key)
332 		nicvf_rss_set_key(nic, rss_conf->rss_key);
333 
334 	nic_rss = nicvf_rss_ethdev_to_nic(nic, rss_conf->rss_hf);
335 	nicvf_rss_set_cfg(nic, nic_rss);
336 	return 0;
337 }
338 
339 static int
340 nicvf_qset_cq_alloc(struct nicvf *nic, struct nicvf_rxq *rxq, uint16_t qidx,
341 		    uint32_t desc_cnt)
342 {
343 	const struct rte_memzone *rz;
344 	uint32_t ring_size = desc_cnt * sizeof(union cq_entry_t);
345 
346 	rz = rte_eth_dma_zone_reserve(nic->eth_dev, "cq_ring", qidx, ring_size,
347 					NICVF_CQ_BASE_ALIGN_BYTES, nic->node);
348 	if (rz == NULL) {
349 		PMD_INIT_LOG(ERR, "Failed to allocate mem for cq hw ring");
350 		return -ENOMEM;
351 	}
352 
353 	memset(rz->addr, 0, ring_size);
354 
355 	rxq->phys = rz->phys_addr;
356 	rxq->desc = rz->addr;
357 	rxq->qlen_mask = desc_cnt - 1;
358 
359 	return 0;
360 }
361 
362 static int
363 nicvf_qset_sq_alloc(struct nicvf *nic,  struct nicvf_txq *sq, uint16_t qidx,
364 		    uint32_t desc_cnt)
365 {
366 	const struct rte_memzone *rz;
367 	uint32_t ring_size = desc_cnt * sizeof(union sq_entry_t);
368 
369 	rz = rte_eth_dma_zone_reserve(nic->eth_dev, "sq", qidx, ring_size,
370 				NICVF_SQ_BASE_ALIGN_BYTES, nic->node);
371 	if (rz == NULL) {
372 		PMD_INIT_LOG(ERR, "Failed allocate mem for sq hw ring");
373 		return -ENOMEM;
374 	}
375 
376 	memset(rz->addr, 0, ring_size);
377 
378 	sq->phys = rz->phys_addr;
379 	sq->desc = rz->addr;
380 	sq->qlen_mask = desc_cnt - 1;
381 
382 	return 0;
383 }
384 
385 static inline void
386 nicvf_tx_queue_release_mbufs(struct nicvf_txq *txq)
387 {
388 	uint32_t head;
389 
390 	head = txq->head;
391 	while (head != txq->tail) {
392 		if (txq->txbuffs[head]) {
393 			rte_pktmbuf_free_seg(txq->txbuffs[head]);
394 			txq->txbuffs[head] = NULL;
395 		}
396 		head++;
397 		head = head & txq->qlen_mask;
398 	}
399 }
400 
401 static void
402 nicvf_tx_queue_reset(struct nicvf_txq *txq)
403 {
404 	uint32_t txq_desc_cnt = txq->qlen_mask + 1;
405 
406 	memset(txq->desc, 0, sizeof(union sq_entry_t) * txq_desc_cnt);
407 	memset(txq->txbuffs, 0, sizeof(struct rte_mbuf *) * txq_desc_cnt);
408 	txq->tail = 0;
409 	txq->head = 0;
410 	txq->xmit_bufs = 0;
411 }
412 
413 static void
414 nicvf_dev_tx_queue_release(void *sq)
415 {
416 	struct nicvf_txq *txq;
417 
418 	PMD_INIT_FUNC_TRACE();
419 
420 	txq = (struct nicvf_txq *)sq;
421 	if (txq) {
422 		if (txq->txbuffs != NULL) {
423 			nicvf_tx_queue_release_mbufs(txq);
424 			rte_free(txq->txbuffs);
425 			txq->txbuffs = NULL;
426 		}
427 		rte_free(txq);
428 	}
429 }
430 
431 static int
432 nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
433 			 uint16_t nb_desc, unsigned int socket_id,
434 			 const struct rte_eth_txconf *tx_conf)
435 {
436 	uint16_t tx_free_thresh;
437 	uint8_t is_single_pool;
438 	struct nicvf_txq *txq;
439 	struct nicvf *nic = nicvf_pmd_priv(dev);
440 
441 	PMD_INIT_FUNC_TRACE();
442 
443 	/* Socket id check */
444 	if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
445 		PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
446 		socket_id, nic->node);
447 
448 	/* Tx deferred start is not supported */
449 	if (tx_conf->tx_deferred_start) {
450 		PMD_INIT_LOG(ERR, "Tx deferred start not supported");
451 		return -EINVAL;
452 	}
453 
454 	/* Roundup nb_desc to available qsize and validate max number of desc */
455 	nb_desc = nicvf_qsize_sq_roundup(nb_desc);
456 	if (nb_desc == 0) {
457 		PMD_INIT_LOG(ERR, "Value of nb_desc beyond available sq qsize");
458 		return -EINVAL;
459 	}
460 
461 	/* Validate tx_free_thresh */
462 	tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
463 				tx_conf->tx_free_thresh :
464 				NICVF_DEFAULT_TX_FREE_THRESH);
465 
466 	if (tx_free_thresh > (nb_desc) ||
467 		tx_free_thresh > NICVF_MAX_TX_FREE_THRESH) {
468 		PMD_INIT_LOG(ERR,
469 			"tx_free_thresh must be less than the number of TX "
470 			"descriptors. (tx_free_thresh=%u port=%d "
471 			"queue=%d)", (unsigned int)tx_free_thresh,
472 			(int)dev->data->port_id, (int)qidx);
473 		return -EINVAL;
474 	}
475 
476 	/* Free memory prior to re-allocation if needed. */
477 	if (dev->data->tx_queues[qidx] != NULL) {
478 		PMD_TX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
479 				qidx);
480 		nicvf_dev_tx_queue_release(dev->data->tx_queues[qidx]);
481 		dev->data->tx_queues[qidx] = NULL;
482 	}
483 
484 	/* Allocating tx queue data structure */
485 	txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct nicvf_txq),
486 					RTE_CACHE_LINE_SIZE, nic->node);
487 	if (txq == NULL) {
488 		PMD_INIT_LOG(ERR, "Failed to allocate txq=%d", qidx);
489 		return -ENOMEM;
490 	}
491 
492 	txq->nic = nic;
493 	txq->queue_id = qidx;
494 	txq->tx_free_thresh = tx_free_thresh;
495 	txq->txq_flags = tx_conf->txq_flags;
496 	txq->sq_head = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_HEAD;
497 	txq->sq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_DOOR;
498 	is_single_pool = (txq->txq_flags & ETH_TXQ_FLAGS_NOREFCOUNT &&
499 				txq->txq_flags & ETH_TXQ_FLAGS_NOMULTMEMP);
500 
501 	/* Choose optimum free threshold value for multipool case */
502 	if (!is_single_pool) {
503 		txq->tx_free_thresh = (uint16_t)
504 		(tx_conf->tx_free_thresh == NICVF_DEFAULT_TX_FREE_THRESH ?
505 				NICVF_TX_FREE_MPOOL_THRESH :
506 				tx_conf->tx_free_thresh);
507 	}
508 
509 	/* Allocate software ring */
510 	txq->txbuffs = rte_zmalloc_socket("txq->txbuffs",
511 				nb_desc * sizeof(struct rte_mbuf *),
512 				RTE_CACHE_LINE_SIZE, nic->node);
513 
514 	if (txq->txbuffs == NULL) {
515 		nicvf_dev_tx_queue_release(txq);
516 		return -ENOMEM;
517 	}
518 
519 	if (nicvf_qset_sq_alloc(nic, txq, qidx, nb_desc)) {
520 		PMD_INIT_LOG(ERR, "Failed to allocate mem for sq %d", qidx);
521 		nicvf_dev_tx_queue_release(txq);
522 		return -ENOMEM;
523 	}
524 
525 	nicvf_tx_queue_reset(txq);
526 
527 	PMD_TX_LOG(DEBUG, "[%d] txq=%p nb_desc=%d desc=%p phys=0x%" PRIx64,
528 			qidx, txq, nb_desc, txq->desc, txq->phys);
529 
530 	dev->data->tx_queues[qidx] = txq;
531 	dev->data->tx_queue_state[qidx] = RTE_ETH_QUEUE_STATE_STOPPED;
532 	return 0;
533 }
534 
535 static void
536 nicvf_rx_queue_reset(struct nicvf_rxq *rxq)
537 {
538 	rxq->head = 0;
539 	rxq->available_space = 0;
540 	rxq->recv_buffers = 0;
541 }
542 
543 static void
544 nicvf_dev_rx_queue_release(void *rx_queue)
545 {
546 	struct nicvf_rxq *rxq = rx_queue;
547 
548 	PMD_INIT_FUNC_TRACE();
549 
550 	if (rxq)
551 		rte_free(rxq);
552 }
553 
554 static int
555 nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
556 			 uint16_t nb_desc, unsigned int socket_id,
557 			 const struct rte_eth_rxconf *rx_conf,
558 			 struct rte_mempool *mp)
559 {
560 	uint16_t rx_free_thresh;
561 	struct nicvf_rxq *rxq;
562 	struct nicvf *nic = nicvf_pmd_priv(dev);
563 
564 	PMD_INIT_FUNC_TRACE();
565 
566 	/* Socket id check */
567 	if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
568 		PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
569 		socket_id, nic->node);
570 
571 	/* Mempool memory should be contiguous */
572 	if (mp->nb_mem_chunks != 1) {
573 		PMD_INIT_LOG(ERR, "Non contiguous mempool, check huge page sz");
574 		return -EINVAL;
575 	}
576 
577 	/* Rx deferred start is not supported */
578 	if (rx_conf->rx_deferred_start) {
579 		PMD_INIT_LOG(ERR, "Rx deferred start not supported");
580 		return -EINVAL;
581 	}
582 
583 	/* Roundup nb_desc to available qsize and validate max number of desc */
584 	nb_desc = nicvf_qsize_cq_roundup(nb_desc);
585 	if (nb_desc == 0) {
586 		PMD_INIT_LOG(ERR, "Value nb_desc beyond available hw cq qsize");
587 		return -EINVAL;
588 	}
589 
590 	/* Check rx_free_thresh upper bound */
591 	rx_free_thresh = (uint16_t)((rx_conf->rx_free_thresh) ?
592 				rx_conf->rx_free_thresh :
593 				NICVF_DEFAULT_RX_FREE_THRESH);
594 	if (rx_free_thresh > NICVF_MAX_RX_FREE_THRESH ||
595 		rx_free_thresh >= nb_desc * .75) {
596 		PMD_INIT_LOG(ERR, "rx_free_thresh greater than expected %d",
597 				rx_free_thresh);
598 		return -EINVAL;
599 	}
600 
601 	/* Free memory prior to re-allocation if needed */
602 	if (dev->data->rx_queues[qidx] != NULL) {
603 		PMD_RX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
604 				qidx);
605 		nicvf_dev_rx_queue_release(dev->data->rx_queues[qidx]);
606 		dev->data->rx_queues[qidx] = NULL;
607 	}
608 
609 	/* Allocate rxq memory */
610 	rxq = rte_zmalloc_socket("ethdev rx queue", sizeof(struct nicvf_rxq),
611 					RTE_CACHE_LINE_SIZE, nic->node);
612 	if (rxq == NULL) {
613 		PMD_INIT_LOG(ERR, "Failed to allocate rxq=%d", qidx);
614 		return -ENOMEM;
615 	}
616 
617 	rxq->nic = nic;
618 	rxq->pool = mp;
619 	rxq->queue_id = qidx;
620 	rxq->port_id = dev->data->port_id;
621 	rxq->rx_free_thresh = rx_free_thresh;
622 	rxq->rx_drop_en = rx_conf->rx_drop_en;
623 	rxq->cq_status = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_STATUS;
624 	rxq->cq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_DOOR;
625 	rxq->precharge_cnt = 0;
626 	rxq->rbptr_offset = NICVF_CQE_RBPTR_WORD;
627 
628 	/* Alloc completion queue */
629 	if (nicvf_qset_cq_alloc(nic, rxq, rxq->queue_id, nb_desc)) {
630 		PMD_INIT_LOG(ERR, "failed to allocate cq %u", rxq->queue_id);
631 		nicvf_dev_rx_queue_release(rxq);
632 		return -ENOMEM;
633 	}
634 
635 	nicvf_rx_queue_reset(rxq);
636 
637 	PMD_RX_LOG(DEBUG, "[%d] rxq=%p pool=%s nb_desc=(%d/%d) phy=%" PRIx64,
638 			qidx, rxq, mp->name, nb_desc,
639 			rte_mempool_count(mp), rxq->phys);
640 
641 	dev->data->rx_queues[qidx] = rxq;
642 	dev->data->rx_queue_state[qidx] = RTE_ETH_QUEUE_STATE_STOPPED;
643 	return 0;
644 }
645 
646 static void
647 nicvf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
648 {
649 	struct nicvf *nic = nicvf_pmd_priv(dev);
650 
651 	PMD_INIT_FUNC_TRACE();
652 
653 	dev_info->min_rx_bufsize = ETHER_MIN_MTU;
654 	dev_info->max_rx_pktlen = NIC_HW_MAX_FRS;
655 	dev_info->max_rx_queues = (uint16_t)MAX_RCV_QUEUES_PER_QS;
656 	dev_info->max_tx_queues = (uint16_t)MAX_SND_QUEUES_PER_QS;
657 	dev_info->max_mac_addrs = 1;
658 	dev_info->max_vfs = dev->pci_dev->max_vfs;
659 
660 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
661 	dev_info->tx_offload_capa =
662 		DEV_TX_OFFLOAD_IPV4_CKSUM  |
663 		DEV_TX_OFFLOAD_UDP_CKSUM   |
664 		DEV_TX_OFFLOAD_TCP_CKSUM   |
665 		DEV_TX_OFFLOAD_TCP_TSO     |
666 		DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;
667 
668 	dev_info->reta_size = nic->rss_info.rss_size;
669 	dev_info->hash_key_size = RSS_HASH_KEY_BYTE_SIZE;
670 	dev_info->flow_type_rss_offloads = NICVF_RSS_OFFLOAD_PASS1;
671 	if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING)
672 		dev_info->flow_type_rss_offloads |= NICVF_RSS_OFFLOAD_TUNNEL;
673 
674 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
675 		.rx_free_thresh = NICVF_DEFAULT_RX_FREE_THRESH,
676 		.rx_drop_en = 0,
677 	};
678 
679 	dev_info->default_txconf = (struct rte_eth_txconf) {
680 		.tx_free_thresh = NICVF_DEFAULT_TX_FREE_THRESH,
681 		.txq_flags =
682 			ETH_TXQ_FLAGS_NOMULTSEGS  |
683 			ETH_TXQ_FLAGS_NOREFCOUNT  |
684 			ETH_TXQ_FLAGS_NOMULTMEMP  |
685 			ETH_TXQ_FLAGS_NOVLANOFFL  |
686 			ETH_TXQ_FLAGS_NOXSUMSCTP,
687 	};
688 }
689 
690 static int
691 nicvf_dev_configure(struct rte_eth_dev *dev)
692 {
693 	struct rte_eth_conf *conf = &dev->data->dev_conf;
694 	struct rte_eth_rxmode *rxmode = &conf->rxmode;
695 	struct rte_eth_txmode *txmode = &conf->txmode;
696 	struct nicvf *nic = nicvf_pmd_priv(dev);
697 
698 	PMD_INIT_FUNC_TRACE();
699 
700 	if (!rte_eal_has_hugepages()) {
701 		PMD_INIT_LOG(INFO, "Huge page is not configured");
702 		return -EINVAL;
703 	}
704 
705 	if (txmode->mq_mode) {
706 		PMD_INIT_LOG(INFO, "Tx mq_mode DCB or VMDq not supported");
707 		return -EINVAL;
708 	}
709 
710 	if (rxmode->mq_mode != ETH_MQ_RX_NONE &&
711 		rxmode->mq_mode != ETH_MQ_RX_RSS) {
712 		PMD_INIT_LOG(INFO, "Unsupported rx qmode %d", rxmode->mq_mode);
713 		return -EINVAL;
714 	}
715 
716 	if (!rxmode->hw_strip_crc) {
717 		PMD_INIT_LOG(NOTICE, "Can't disable hw crc strip");
718 		rxmode->hw_strip_crc = 1;
719 	}
720 
721 	if (rxmode->hw_ip_checksum) {
722 		PMD_INIT_LOG(NOTICE, "Rxcksum not supported");
723 		rxmode->hw_ip_checksum = 0;
724 	}
725 
726 	if (rxmode->split_hdr_size) {
727 		PMD_INIT_LOG(INFO, "Rxmode does not support split header");
728 		return -EINVAL;
729 	}
730 
731 	if (rxmode->hw_vlan_filter) {
732 		PMD_INIT_LOG(INFO, "VLAN filter not supported");
733 		return -EINVAL;
734 	}
735 
736 	if (rxmode->hw_vlan_extend) {
737 		PMD_INIT_LOG(INFO, "VLAN extended not supported");
738 		return -EINVAL;
739 	}
740 
741 	if (rxmode->enable_lro) {
742 		PMD_INIT_LOG(INFO, "LRO not supported");
743 		return -EINVAL;
744 	}
745 
746 	if (conf->link_speeds & ETH_LINK_SPEED_FIXED) {
747 		PMD_INIT_LOG(INFO, "Setting link speed/duplex not supported");
748 		return -EINVAL;
749 	}
750 
751 	if (conf->dcb_capability_en) {
752 		PMD_INIT_LOG(INFO, "DCB enable not supported");
753 		return -EINVAL;
754 	}
755 
756 	if (conf->fdir_conf.mode != RTE_FDIR_MODE_NONE) {
757 		PMD_INIT_LOG(INFO, "Flow director not supported");
758 		return -EINVAL;
759 	}
760 
761 	PMD_INIT_LOG(DEBUG, "Configured ethdev port%d hwcap=0x%" PRIx64,
762 		dev->data->port_id, nicvf_hw_cap(nic));
763 
764 	return 0;
765 }
766 
767 /* Initialize and register driver with DPDK Application */
768 static const struct eth_dev_ops nicvf_eth_dev_ops = {
769 	.dev_configure            = nicvf_dev_configure,
770 	.link_update              = nicvf_dev_link_update,
771 	.dev_infos_get            = nicvf_dev_info_get,
772 	.reta_update              = nicvf_dev_reta_update,
773 	.reta_query               = nicvf_dev_reta_query,
774 	.rss_hash_update          = nicvf_dev_rss_hash_update,
775 	.rss_hash_conf_get        = nicvf_dev_rss_hash_conf_get,
776 	.rx_queue_setup           = nicvf_dev_rx_queue_setup,
777 	.rx_queue_release         = nicvf_dev_rx_queue_release,
778 	.tx_queue_setup           = nicvf_dev_tx_queue_setup,
779 	.tx_queue_release         = nicvf_dev_tx_queue_release,
780 	.get_reg_length           = nicvf_dev_get_reg_length,
781 	.get_reg                  = nicvf_dev_get_regs,
782 };
783 
784 static int
785 nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
786 {
787 	int ret;
788 	struct rte_pci_device *pci_dev;
789 	struct nicvf *nic = nicvf_pmd_priv(eth_dev);
790 
791 	PMD_INIT_FUNC_TRACE();
792 
793 	eth_dev->dev_ops = &nicvf_eth_dev_ops;
794 
795 	pci_dev = eth_dev->pci_dev;
796 	rte_eth_copy_pci_info(eth_dev, pci_dev);
797 
798 	nic->device_id = pci_dev->id.device_id;
799 	nic->vendor_id = pci_dev->id.vendor_id;
800 	nic->subsystem_device_id = pci_dev->id.subsystem_device_id;
801 	nic->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
802 	nic->eth_dev = eth_dev;
803 
804 	PMD_INIT_LOG(DEBUG, "nicvf: device (%x:%x) %u:%u:%u:%u",
805 			pci_dev->id.vendor_id, pci_dev->id.device_id,
806 			pci_dev->addr.domain, pci_dev->addr.bus,
807 			pci_dev->addr.devid, pci_dev->addr.function);
808 
809 	nic->reg_base = (uintptr_t)pci_dev->mem_resource[0].addr;
810 	if (!nic->reg_base) {
811 		PMD_INIT_LOG(ERR, "Failed to map BAR0");
812 		ret = -ENODEV;
813 		goto fail;
814 	}
815 
816 	nicvf_disable_all_interrupts(nic);
817 
818 	ret = nicvf_periodic_alarm_start(nic);
819 	if (ret) {
820 		PMD_INIT_LOG(ERR, "Failed to start period alarm");
821 		goto fail;
822 	}
823 
824 	ret = nicvf_mbox_check_pf_ready(nic);
825 	if (ret) {
826 		PMD_INIT_LOG(ERR, "Failed to get ready message from PF");
827 		goto alarm_fail;
828 	} else {
829 		PMD_INIT_LOG(INFO,
830 			"node=%d vf=%d mode=%s sqs=%s loopback_supported=%s",
831 			nic->node, nic->vf_id,
832 			nic->tns_mode == NIC_TNS_MODE ? "tns" : "tns-bypass",
833 			nic->sqs_mode ? "true" : "false",
834 			nic->loopback_supported ? "true" : "false"
835 			);
836 	}
837 
838 	if (nic->sqs_mode) {
839 		PMD_INIT_LOG(INFO, "Unsupported SQS VF detected, Detaching...");
840 		/* Detach port by returning Positive error number */
841 		ret = ENOTSUP;
842 		goto alarm_fail;
843 	}
844 
845 	eth_dev->data->mac_addrs = rte_zmalloc("mac_addr", ETHER_ADDR_LEN, 0);
846 	if (eth_dev->data->mac_addrs == NULL) {
847 		PMD_INIT_LOG(ERR, "Failed to allocate memory for mac addr");
848 		ret = -ENOMEM;
849 		goto alarm_fail;
850 	}
851 	if (is_zero_ether_addr((struct ether_addr *)nic->mac_addr))
852 		eth_random_addr(&nic->mac_addr[0]);
853 
854 	ether_addr_copy((struct ether_addr *)nic->mac_addr,
855 			&eth_dev->data->mac_addrs[0]);
856 
857 	ret = nicvf_mbox_set_mac_addr(nic, nic->mac_addr);
858 	if (ret) {
859 		PMD_INIT_LOG(ERR, "Failed to set mac addr");
860 		goto malloc_fail;
861 	}
862 
863 	ret = nicvf_base_init(nic);
864 	if (ret) {
865 		PMD_INIT_LOG(ERR, "Failed to execute nicvf_base_init");
866 		goto malloc_fail;
867 	}
868 
869 	ret = nicvf_mbox_get_rss_size(nic);
870 	if (ret) {
871 		PMD_INIT_LOG(ERR, "Failed to get rss table size");
872 		goto malloc_fail;
873 	}
874 
875 	PMD_INIT_LOG(INFO, "Port %d (%x:%x) mac=%02x:%02x:%02x:%02x:%02x:%02x",
876 		eth_dev->data->port_id, nic->vendor_id, nic->device_id,
877 		nic->mac_addr[0], nic->mac_addr[1], nic->mac_addr[2],
878 		nic->mac_addr[3], nic->mac_addr[4], nic->mac_addr[5]);
879 
880 	return 0;
881 
882 malloc_fail:
883 	rte_free(eth_dev->data->mac_addrs);
884 alarm_fail:
885 	nicvf_periodic_alarm_stop(nic);
886 fail:
887 	return ret;
888 }
889 
890 static const struct rte_pci_id pci_id_nicvf_map[] = {
891 	{
892 		.class_id = RTE_CLASS_ANY_ID,
893 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
894 		.device_id = PCI_DEVICE_ID_THUNDERX_PASS1_NICVF,
895 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
896 		.subsystem_device_id = PCI_SUB_DEVICE_ID_THUNDERX_PASS1_NICVF,
897 	},
898 	{
899 		.class_id = RTE_CLASS_ANY_ID,
900 		.vendor_id = PCI_VENDOR_ID_CAVIUM,
901 		.device_id = PCI_DEVICE_ID_THUNDERX_PASS2_NICVF,
902 		.subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
903 		.subsystem_device_id = PCI_SUB_DEVICE_ID_THUNDERX_PASS2_NICVF,
904 	},
905 	{
906 		.vendor_id = 0,
907 	},
908 };
909 
910 static struct eth_driver rte_nicvf_pmd = {
911 	.pci_drv = {
912 		.name = "rte_nicvf_pmd",
913 		.id_table = pci_id_nicvf_map,
914 		.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
915 	},
916 	.eth_dev_init = nicvf_eth_dev_init,
917 	.dev_private_size = sizeof(struct nicvf),
918 };
919 
920 static int
921 rte_nicvf_pmd_init(const char *name __rte_unused, const char *para __rte_unused)
922 {
923 	PMD_INIT_FUNC_TRACE();
924 	PMD_INIT_LOG(INFO, "librte_pmd_thunderx nicvf version %s",
925 			THUNDERX_NICVF_PMD_VERSION);
926 
927 	rte_eth_driver_register(&rte_nicvf_pmd);
928 	return 0;
929 }
930 
931 static struct rte_driver rte_nicvf_driver = {
932 	.name = "nicvf_driver",
933 	.type = PMD_PDEV,
934 	.init = rte_nicvf_pmd_init,
935 };
936 
937 PMD_REGISTER_DRIVER(rte_nicvf_driver);
938