xref: /dpdk/drivers/net/cnxk/cnxk_ethdev.c (revision bdf4a3c6)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(C) 2021 Marvell.
3  */
4 #include <cnxk_ethdev.h>
5 
6 static inline uint64_t
7 nix_get_rx_offload_capa(struct cnxk_eth_dev *dev)
8 {
9 	uint64_t capa = CNXK_NIX_RX_OFFLOAD_CAPA;
10 
11 	if (roc_nix_is_vf_or_sdp(&dev->nix) ||
12 	    dev->npc.switch_header_type == ROC_PRIV_FLAGS_HIGIG)
13 		capa &= ~DEV_RX_OFFLOAD_TIMESTAMP;
14 
15 	return capa;
16 }
17 
18 static inline uint64_t
19 nix_get_tx_offload_capa(struct cnxk_eth_dev *dev)
20 {
21 	RTE_SET_USED(dev);
22 	return CNXK_NIX_TX_OFFLOAD_CAPA;
23 }
24 
25 static inline uint32_t
26 nix_get_speed_capa(struct cnxk_eth_dev *dev)
27 {
28 	uint32_t speed_capa;
29 
30 	/* Auto negotiation disabled */
31 	speed_capa = ETH_LINK_SPEED_FIXED;
32 	if (!roc_nix_is_vf_or_sdp(&dev->nix) && !roc_nix_is_lbk(&dev->nix)) {
33 		speed_capa |= ETH_LINK_SPEED_1G | ETH_LINK_SPEED_10G |
34 			      ETH_LINK_SPEED_25G | ETH_LINK_SPEED_40G |
35 			      ETH_LINK_SPEED_50G | ETH_LINK_SPEED_100G;
36 	}
37 
38 	return speed_capa;
39 }
40 
41 static void
42 nix_enable_mseg_on_jumbo(struct cnxk_eth_rxq_sp *rxq)
43 {
44 	struct rte_pktmbuf_pool_private *mbp_priv;
45 	struct rte_eth_dev *eth_dev;
46 	struct cnxk_eth_dev *dev;
47 	uint32_t buffsz;
48 
49 	dev = rxq->dev;
50 	eth_dev = dev->eth_dev;
51 
52 	/* Get rx buffer size */
53 	mbp_priv = rte_mempool_get_priv(rxq->qconf.mp);
54 	buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
55 
56 	if (eth_dev->data->dev_conf.rxmode.max_rx_pkt_len > buffsz) {
57 		dev->rx_offloads |= DEV_RX_OFFLOAD_SCATTER;
58 		dev->tx_offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
59 	}
60 }
61 
62 int
63 nix_recalc_mtu(struct rte_eth_dev *eth_dev)
64 {
65 	struct rte_eth_dev_data *data = eth_dev->data;
66 	struct cnxk_eth_rxq_sp *rxq;
67 	uint16_t mtu;
68 	int rc;
69 
70 	rxq = ((struct cnxk_eth_rxq_sp *)data->rx_queues[0]) - 1;
71 	/* Setup scatter mode if needed by jumbo */
72 	nix_enable_mseg_on_jumbo(rxq);
73 
74 	/* Setup MTU based on max_rx_pkt_len */
75 	mtu = data->dev_conf.rxmode.max_rx_pkt_len - CNXK_NIX_L2_OVERHEAD +
76 				CNXK_NIX_MAX_VTAG_ACT_SIZE;
77 
78 	rc = cnxk_nix_mtu_set(eth_dev, mtu);
79 	if (rc)
80 		plt_err("Failed to set default MTU size, rc=%d", rc);
81 
82 	return rc;
83 }
84 
85 static int
86 nix_init_flow_ctrl_config(struct rte_eth_dev *eth_dev)
87 {
88 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
89 	struct cnxk_fc_cfg *fc = &dev->fc_cfg;
90 	struct rte_eth_fc_conf fc_conf = {0};
91 	int rc;
92 
93 	/* Both Rx & Tx flow ctrl get enabled(RTE_FC_FULL) in HW
94 	 * by AF driver, update those info in PMD structure.
95 	 */
96 	rc = cnxk_nix_flow_ctrl_get(eth_dev, &fc_conf);
97 	if (rc)
98 		goto exit;
99 
100 	fc->mode = fc_conf.mode;
101 	fc->rx_pause = (fc_conf.mode == RTE_FC_FULL) ||
102 			(fc_conf.mode == RTE_FC_RX_PAUSE);
103 	fc->tx_pause = (fc_conf.mode == RTE_FC_FULL) ||
104 			(fc_conf.mode == RTE_FC_TX_PAUSE);
105 
106 exit:
107 	return rc;
108 }
109 
110 static int
111 nix_update_flow_ctrl_config(struct rte_eth_dev *eth_dev)
112 {
113 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
114 	struct cnxk_fc_cfg *fc = &dev->fc_cfg;
115 	struct rte_eth_fc_conf fc_cfg = {0};
116 
117 	if (roc_nix_is_vf_or_sdp(&dev->nix))
118 		return 0;
119 
120 	fc_cfg.mode = fc->mode;
121 
122 	/* To avoid Link credit deadlock on Ax, disable Tx FC if it's enabled */
123 	if (roc_model_is_cn96_ax() &&
124 	    dev->npc.switch_header_type != ROC_PRIV_FLAGS_HIGIG &&
125 	    (fc_cfg.mode == RTE_FC_FULL || fc_cfg.mode == RTE_FC_RX_PAUSE)) {
126 		fc_cfg.mode =
127 				(fc_cfg.mode == RTE_FC_FULL ||
128 				fc_cfg.mode == RTE_FC_TX_PAUSE) ?
129 				RTE_FC_TX_PAUSE : RTE_FC_NONE;
130 	}
131 
132 	return cnxk_nix_flow_ctrl_set(eth_dev, &fc_cfg);
133 }
134 
135 uint64_t
136 cnxk_nix_rxq_mbuf_setup(struct cnxk_eth_dev *dev)
137 {
138 	uint16_t port_id = dev->eth_dev->data->port_id;
139 	struct rte_mbuf mb_def;
140 	uint64_t *tmp;
141 
142 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, data_off) % 8 != 0);
143 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, refcnt) -
144 				 offsetof(struct rte_mbuf, data_off) !=
145 			 2);
146 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, nb_segs) -
147 				 offsetof(struct rte_mbuf, data_off) !=
148 			 4);
149 	RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, port) -
150 				 offsetof(struct rte_mbuf, data_off) !=
151 			 6);
152 	mb_def.nb_segs = 1;
153 	mb_def.data_off = RTE_PKTMBUF_HEADROOM +
154 			  (dev->ptp_en * CNXK_NIX_TIMESYNC_RX_OFFSET);
155 	mb_def.port = port_id;
156 	rte_mbuf_refcnt_set(&mb_def, 1);
157 
158 	/* Prevent compiler reordering: rearm_data covers previous fields */
159 	rte_compiler_barrier();
160 	tmp = (uint64_t *)&mb_def.rearm_data;
161 
162 	return *tmp;
163 }
164 
165 static inline uint8_t
166 nix_sq_max_sqe_sz(struct cnxk_eth_dev *dev)
167 {
168 	/*
169 	 * Maximum three segments can be supported with W8, Choose
170 	 * NIX_MAXSQESZ_W16 for multi segment offload.
171 	 */
172 	if (dev->tx_offloads & DEV_TX_OFFLOAD_MULTI_SEGS)
173 		return NIX_MAXSQESZ_W16;
174 	else
175 		return NIX_MAXSQESZ_W8;
176 }
177 
178 int
179 cnxk_nix_tx_queue_setup(struct rte_eth_dev *eth_dev, uint16_t qid,
180 			uint16_t nb_desc, uint16_t fp_tx_q_sz,
181 			const struct rte_eth_txconf *tx_conf)
182 {
183 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
184 	const struct eth_dev_ops *dev_ops = eth_dev->dev_ops;
185 	struct cnxk_eth_txq_sp *txq_sp;
186 	struct roc_nix_sq *sq;
187 	size_t txq_sz;
188 	int rc;
189 
190 	/* Free memory prior to re-allocation if needed. */
191 	if (eth_dev->data->tx_queues[qid] != NULL) {
192 		plt_nix_dbg("Freeing memory prior to re-allocation %d", qid);
193 		dev_ops->tx_queue_release(eth_dev->data->tx_queues[qid]);
194 		eth_dev->data->tx_queues[qid] = NULL;
195 	}
196 
197 	/* Setup ROC SQ */
198 	sq = &dev->sqs[qid];
199 	sq->qid = qid;
200 	sq->nb_desc = nb_desc;
201 	sq->max_sqe_sz = nix_sq_max_sqe_sz(dev);
202 
203 	rc = roc_nix_sq_init(&dev->nix, sq);
204 	if (rc) {
205 		plt_err("Failed to init sq=%d, rc=%d", qid, rc);
206 		return rc;
207 	}
208 
209 	rc = -ENOMEM;
210 	txq_sz = sizeof(struct cnxk_eth_txq_sp) + fp_tx_q_sz;
211 	txq_sp = plt_zmalloc(txq_sz, PLT_CACHE_LINE_SIZE);
212 	if (!txq_sp) {
213 		plt_err("Failed to alloc tx queue mem");
214 		rc |= roc_nix_sq_fini(sq);
215 		return rc;
216 	}
217 
218 	txq_sp->dev = dev;
219 	txq_sp->qid = qid;
220 	txq_sp->qconf.conf.tx = *tx_conf;
221 	txq_sp->qconf.nb_desc = nb_desc;
222 
223 	plt_nix_dbg("sq=%d fc=%p offload=0x%" PRIx64 " lmt_addr=%p"
224 		    " nb_sqb_bufs=%d sqes_per_sqb_log2=%d",
225 		    qid, sq->fc, dev->tx_offloads, sq->lmt_addr,
226 		    sq->nb_sqb_bufs, sq->sqes_per_sqb_log2);
227 
228 	/* Store start of fast path area */
229 	eth_dev->data->tx_queues[qid] = txq_sp + 1;
230 	eth_dev->data->tx_queue_state[qid] = RTE_ETH_QUEUE_STATE_STOPPED;
231 	return 0;
232 }
233 
234 static void
235 cnxk_nix_tx_queue_release(void *txq)
236 {
237 	struct cnxk_eth_txq_sp *txq_sp;
238 	struct cnxk_eth_dev *dev;
239 	struct roc_nix_sq *sq;
240 	uint16_t qid;
241 	int rc;
242 
243 	if (!txq)
244 		return;
245 
246 	txq_sp = cnxk_eth_txq_to_sp(txq);
247 	dev = txq_sp->dev;
248 	qid = txq_sp->qid;
249 
250 	plt_nix_dbg("Releasing txq %u", qid);
251 
252 	/* Cleanup ROC SQ */
253 	sq = &dev->sqs[qid];
254 	rc = roc_nix_sq_fini(sq);
255 	if (rc)
256 		plt_err("Failed to cleanup sq, rc=%d", rc);
257 
258 	/* Finally free */
259 	plt_free(txq_sp);
260 }
261 
262 int
263 cnxk_nix_rx_queue_setup(struct rte_eth_dev *eth_dev, uint16_t qid,
264 			uint16_t nb_desc, uint16_t fp_rx_q_sz,
265 			const struct rte_eth_rxconf *rx_conf,
266 			struct rte_mempool *mp)
267 {
268 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
269 	struct cnxk_eth_rxq_sp *rxq_sp;
270 	struct rte_mempool_ops *ops;
271 	const char *platform_ops;
272 	struct roc_nix_rq *rq;
273 	struct roc_nix_cq *cq;
274 	uint16_t first_skip;
275 	int rc = -EINVAL;
276 	size_t rxq_sz;
277 
278 	/* Sanity checks */
279 	if (rx_conf->rx_deferred_start == 1) {
280 		plt_err("Deferred Rx start is not supported");
281 		goto fail;
282 	}
283 
284 	platform_ops = rte_mbuf_platform_mempool_ops();
285 	/* This driver needs cnxk_npa mempool ops to work */
286 	ops = rte_mempool_get_ops(mp->ops_index);
287 	if (strncmp(ops->name, platform_ops, RTE_MEMPOOL_OPS_NAMESIZE)) {
288 		plt_err("mempool ops should be of cnxk_npa type");
289 		goto fail;
290 	}
291 
292 	if (mp->pool_id == 0) {
293 		plt_err("Invalid pool_id");
294 		goto fail;
295 	}
296 
297 	/* Free memory prior to re-allocation if needed */
298 	if (eth_dev->data->rx_queues[qid] != NULL) {
299 		const struct eth_dev_ops *dev_ops = eth_dev->dev_ops;
300 
301 		plt_nix_dbg("Freeing memory prior to re-allocation %d", qid);
302 		dev_ops->rx_queue_release(eth_dev->data->rx_queues[qid]);
303 		eth_dev->data->rx_queues[qid] = NULL;
304 	}
305 
306 	/* Setup ROC CQ */
307 	cq = &dev->cqs[qid];
308 	cq->qid = qid;
309 	cq->nb_desc = nb_desc;
310 	rc = roc_nix_cq_init(&dev->nix, cq);
311 	if (rc) {
312 		plt_err("Failed to init roc cq for rq=%d, rc=%d", qid, rc);
313 		goto fail;
314 	}
315 
316 	/* Setup ROC RQ */
317 	rq = &dev->rqs[qid];
318 	rq->qid = qid;
319 	rq->aura_handle = mp->pool_id;
320 	rq->flow_tag_width = 32;
321 	rq->sso_ena = false;
322 
323 	/* Calculate first mbuf skip */
324 	first_skip = (sizeof(struct rte_mbuf));
325 	first_skip += RTE_PKTMBUF_HEADROOM;
326 	first_skip += rte_pktmbuf_priv_size(mp);
327 	rq->first_skip = first_skip;
328 	rq->later_skip = sizeof(struct rte_mbuf);
329 	rq->lpb_size = mp->elt_size;
330 
331 	rc = roc_nix_rq_init(&dev->nix, rq, !!eth_dev->data->dev_started);
332 	if (rc) {
333 		plt_err("Failed to init roc rq for rq=%d, rc=%d", qid, rc);
334 		goto cq_fini;
335 	}
336 
337 	/* Allocate and setup fast path rx queue */
338 	rc = -ENOMEM;
339 	rxq_sz = sizeof(struct cnxk_eth_rxq_sp) + fp_rx_q_sz;
340 	rxq_sp = plt_zmalloc(rxq_sz, PLT_CACHE_LINE_SIZE);
341 	if (!rxq_sp) {
342 		plt_err("Failed to alloc rx queue for rq=%d", qid);
343 		goto rq_fini;
344 	}
345 
346 	/* Setup slow path fields */
347 	rxq_sp->dev = dev;
348 	rxq_sp->qid = qid;
349 	rxq_sp->qconf.conf.rx = *rx_conf;
350 	rxq_sp->qconf.nb_desc = nb_desc;
351 	rxq_sp->qconf.mp = mp;
352 
353 	plt_nix_dbg("rq=%d pool=%s nb_desc=%d->%d", qid, mp->name, nb_desc,
354 		    cq->nb_desc);
355 
356 	/* Store start of fast path area */
357 	eth_dev->data->rx_queues[qid] = rxq_sp + 1;
358 	eth_dev->data->rx_queue_state[qid] = RTE_ETH_QUEUE_STATE_STOPPED;
359 
360 	/* Calculating delta and freq mult between PTP HI clock and tsc.
361 	 * These are needed in deriving raw clock value from tsc counter.
362 	 * read_clock eth op returns raw clock value.
363 	 */
364 	if ((dev->rx_offloads & DEV_RX_OFFLOAD_TIMESTAMP) || dev->ptp_en) {
365 		rc = cnxk_nix_tsc_convert(dev);
366 		if (rc) {
367 			plt_err("Failed to calculate delta and freq mult");
368 			goto rq_fini;
369 		}
370 	}
371 
372 	return 0;
373 rq_fini:
374 	rc |= roc_nix_rq_fini(rq);
375 cq_fini:
376 	rc |= roc_nix_cq_fini(cq);
377 fail:
378 	return rc;
379 }
380 
381 static void
382 cnxk_nix_rx_queue_release(void *rxq)
383 {
384 	struct cnxk_eth_rxq_sp *rxq_sp;
385 	struct cnxk_eth_dev *dev;
386 	struct roc_nix_rq *rq;
387 	struct roc_nix_cq *cq;
388 	uint16_t qid;
389 	int rc;
390 
391 	if (!rxq)
392 		return;
393 
394 	rxq_sp = cnxk_eth_rxq_to_sp(rxq);
395 	dev = rxq_sp->dev;
396 	qid = rxq_sp->qid;
397 
398 	plt_nix_dbg("Releasing rxq %u", qid);
399 
400 	/* Cleanup ROC RQ */
401 	rq = &dev->rqs[qid];
402 	rc = roc_nix_rq_fini(rq);
403 	if (rc)
404 		plt_err("Failed to cleanup rq, rc=%d", rc);
405 
406 	/* Cleanup ROC CQ */
407 	cq = &dev->cqs[qid];
408 	rc = roc_nix_cq_fini(cq);
409 	if (rc)
410 		plt_err("Failed to cleanup cq, rc=%d", rc);
411 
412 	/* Finally free fast path area */
413 	plt_free(rxq_sp);
414 }
415 
416 uint32_t
417 cnxk_rss_ethdev_to_nix(struct cnxk_eth_dev *dev, uint64_t ethdev_rss,
418 		       uint8_t rss_level)
419 {
420 	uint32_t flow_key_type[RSS_MAX_LEVELS][6] = {
421 		{FLOW_KEY_TYPE_IPV4, FLOW_KEY_TYPE_IPV6, FLOW_KEY_TYPE_TCP,
422 		 FLOW_KEY_TYPE_UDP, FLOW_KEY_TYPE_SCTP, FLOW_KEY_TYPE_ETH_DMAC},
423 		{FLOW_KEY_TYPE_INNR_IPV4, FLOW_KEY_TYPE_INNR_IPV6,
424 		 FLOW_KEY_TYPE_INNR_TCP, FLOW_KEY_TYPE_INNR_UDP,
425 		 FLOW_KEY_TYPE_INNR_SCTP, FLOW_KEY_TYPE_INNR_ETH_DMAC},
426 		{FLOW_KEY_TYPE_IPV4 | FLOW_KEY_TYPE_INNR_IPV4,
427 		 FLOW_KEY_TYPE_IPV6 | FLOW_KEY_TYPE_INNR_IPV6,
428 		 FLOW_KEY_TYPE_TCP | FLOW_KEY_TYPE_INNR_TCP,
429 		 FLOW_KEY_TYPE_UDP | FLOW_KEY_TYPE_INNR_UDP,
430 		 FLOW_KEY_TYPE_SCTP | FLOW_KEY_TYPE_INNR_SCTP,
431 		 FLOW_KEY_TYPE_ETH_DMAC | FLOW_KEY_TYPE_INNR_ETH_DMAC}
432 	};
433 	uint32_t flowkey_cfg = 0;
434 
435 	dev->ethdev_rss_hf = ethdev_rss;
436 
437 	if (ethdev_rss & ETH_RSS_L2_PAYLOAD &&
438 	    dev->npc.switch_header_type == ROC_PRIV_FLAGS_LEN_90B) {
439 		flowkey_cfg |= FLOW_KEY_TYPE_CH_LEN_90B;
440 	}
441 
442 	if (ethdev_rss & ETH_RSS_C_VLAN)
443 		flowkey_cfg |= FLOW_KEY_TYPE_VLAN;
444 
445 	if (ethdev_rss & ETH_RSS_L3_SRC_ONLY)
446 		flowkey_cfg |= FLOW_KEY_TYPE_L3_SRC;
447 
448 	if (ethdev_rss & ETH_RSS_L3_DST_ONLY)
449 		flowkey_cfg |= FLOW_KEY_TYPE_L3_DST;
450 
451 	if (ethdev_rss & ETH_RSS_L4_SRC_ONLY)
452 		flowkey_cfg |= FLOW_KEY_TYPE_L4_SRC;
453 
454 	if (ethdev_rss & ETH_RSS_L4_DST_ONLY)
455 		flowkey_cfg |= FLOW_KEY_TYPE_L4_DST;
456 
457 	if (ethdev_rss & RSS_IPV4_ENABLE)
458 		flowkey_cfg |= flow_key_type[rss_level][RSS_IPV4_INDEX];
459 
460 	if (ethdev_rss & RSS_IPV6_ENABLE)
461 		flowkey_cfg |= flow_key_type[rss_level][RSS_IPV6_INDEX];
462 
463 	if (ethdev_rss & ETH_RSS_TCP)
464 		flowkey_cfg |= flow_key_type[rss_level][RSS_TCP_INDEX];
465 
466 	if (ethdev_rss & ETH_RSS_UDP)
467 		flowkey_cfg |= flow_key_type[rss_level][RSS_UDP_INDEX];
468 
469 	if (ethdev_rss & ETH_RSS_SCTP)
470 		flowkey_cfg |= flow_key_type[rss_level][RSS_SCTP_INDEX];
471 
472 	if (ethdev_rss & ETH_RSS_L2_PAYLOAD)
473 		flowkey_cfg |= flow_key_type[rss_level][RSS_DMAC_INDEX];
474 
475 	if (ethdev_rss & RSS_IPV6_EX_ENABLE)
476 		flowkey_cfg |= FLOW_KEY_TYPE_IPV6_EXT;
477 
478 	if (ethdev_rss & ETH_RSS_PORT)
479 		flowkey_cfg |= FLOW_KEY_TYPE_PORT;
480 
481 	if (ethdev_rss & ETH_RSS_NVGRE)
482 		flowkey_cfg |= FLOW_KEY_TYPE_NVGRE;
483 
484 	if (ethdev_rss & ETH_RSS_VXLAN)
485 		flowkey_cfg |= FLOW_KEY_TYPE_VXLAN;
486 
487 	if (ethdev_rss & ETH_RSS_GENEVE)
488 		flowkey_cfg |= FLOW_KEY_TYPE_GENEVE;
489 
490 	if (ethdev_rss & ETH_RSS_GTPU)
491 		flowkey_cfg |= FLOW_KEY_TYPE_GTPU;
492 
493 	return flowkey_cfg;
494 }
495 
496 static void
497 nix_free_queue_mem(struct cnxk_eth_dev *dev)
498 {
499 	plt_free(dev->rqs);
500 	plt_free(dev->cqs);
501 	plt_free(dev->sqs);
502 	dev->rqs = NULL;
503 	dev->cqs = NULL;
504 	dev->sqs = NULL;
505 }
506 
507 static int
508 nix_rss_default_setup(struct cnxk_eth_dev *dev)
509 {
510 	struct rte_eth_dev *eth_dev = dev->eth_dev;
511 	uint8_t rss_hash_level;
512 	uint32_t flowkey_cfg;
513 	uint64_t rss_hf;
514 
515 	rss_hf = eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
516 	rss_hash_level = ETH_RSS_LEVEL(rss_hf);
517 	if (rss_hash_level)
518 		rss_hash_level -= 1;
519 
520 	flowkey_cfg = cnxk_rss_ethdev_to_nix(dev, rss_hf, rss_hash_level);
521 	return roc_nix_rss_default_setup(&dev->nix, flowkey_cfg);
522 }
523 
524 static int
525 nix_store_queue_cfg_and_then_release(struct rte_eth_dev *eth_dev)
526 {
527 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
528 	const struct eth_dev_ops *dev_ops = eth_dev->dev_ops;
529 	struct cnxk_eth_qconf *tx_qconf = NULL;
530 	struct cnxk_eth_qconf *rx_qconf = NULL;
531 	struct cnxk_eth_rxq_sp *rxq_sp;
532 	struct cnxk_eth_txq_sp *txq_sp;
533 	int i, nb_rxq, nb_txq;
534 	void **txq, **rxq;
535 
536 	nb_rxq = RTE_MIN(dev->nb_rxq, eth_dev->data->nb_rx_queues);
537 	nb_txq = RTE_MIN(dev->nb_txq, eth_dev->data->nb_tx_queues);
538 
539 	tx_qconf = malloc(nb_txq * sizeof(*tx_qconf));
540 	if (tx_qconf == NULL) {
541 		plt_err("Failed to allocate memory for tx_qconf");
542 		goto fail;
543 	}
544 
545 	rx_qconf = malloc(nb_rxq * sizeof(*rx_qconf));
546 	if (rx_qconf == NULL) {
547 		plt_err("Failed to allocate memory for rx_qconf");
548 		goto fail;
549 	}
550 
551 	txq = eth_dev->data->tx_queues;
552 	for (i = 0; i < nb_txq; i++) {
553 		if (txq[i] == NULL) {
554 			tx_qconf[i].valid = false;
555 			plt_info("txq[%d] is already released", i);
556 			continue;
557 		}
558 		txq_sp = cnxk_eth_txq_to_sp(txq[i]);
559 		memcpy(&tx_qconf[i], &txq_sp->qconf, sizeof(*tx_qconf));
560 		tx_qconf[i].valid = true;
561 		dev_ops->tx_queue_release(txq[i]);
562 		eth_dev->data->tx_queues[i] = NULL;
563 	}
564 
565 	rxq = eth_dev->data->rx_queues;
566 	for (i = 0; i < nb_rxq; i++) {
567 		if (rxq[i] == NULL) {
568 			rx_qconf[i].valid = false;
569 			plt_info("rxq[%d] is already released", i);
570 			continue;
571 		}
572 		rxq_sp = cnxk_eth_rxq_to_sp(rxq[i]);
573 		memcpy(&rx_qconf[i], &rxq_sp->qconf, sizeof(*rx_qconf));
574 		rx_qconf[i].valid = true;
575 		dev_ops->rx_queue_release(rxq[i]);
576 		eth_dev->data->rx_queues[i] = NULL;
577 	}
578 
579 	dev->tx_qconf = tx_qconf;
580 	dev->rx_qconf = rx_qconf;
581 	return 0;
582 
583 fail:
584 	free(tx_qconf);
585 	free(rx_qconf);
586 	return -ENOMEM;
587 }
588 
589 static int
590 nix_restore_queue_cfg(struct rte_eth_dev *eth_dev)
591 {
592 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
593 	const struct eth_dev_ops *dev_ops = eth_dev->dev_ops;
594 	struct cnxk_eth_qconf *tx_qconf = dev->tx_qconf;
595 	struct cnxk_eth_qconf *rx_qconf = dev->rx_qconf;
596 	int rc, i, nb_rxq, nb_txq;
597 	void **txq, **rxq;
598 
599 	nb_rxq = RTE_MIN(dev->nb_rxq, eth_dev->data->nb_rx_queues);
600 	nb_txq = RTE_MIN(dev->nb_txq, eth_dev->data->nb_tx_queues);
601 
602 	rc = -ENOMEM;
603 	/* Setup tx & rx queues with previous configuration so
604 	 * that the queues can be functional in cases like ports
605 	 * are started without re configuring queues.
606 	 *
607 	 * Usual re config sequence is like below:
608 	 * port_configure() {
609 	 *      if(reconfigure) {
610 	 *              queue_release()
611 	 *              queue_setup()
612 	 *      }
613 	 *      queue_configure() {
614 	 *              queue_release()
615 	 *              queue_setup()
616 	 *      }
617 	 * }
618 	 * port_start()
619 	 *
620 	 * In some application's control path, queue_configure() would
621 	 * NOT be invoked for TXQs/RXQs in port_configure().
622 	 * In such cases, queues can be functional after start as the
623 	 * queues are already setup in port_configure().
624 	 */
625 	for (i = 0; i < nb_txq; i++) {
626 		if (!tx_qconf[i].valid)
627 			continue;
628 		rc = dev_ops->tx_queue_setup(eth_dev, i, tx_qconf[i].nb_desc, 0,
629 					     &tx_qconf[i].conf.tx);
630 		if (rc) {
631 			plt_err("Failed to setup tx queue rc=%d", rc);
632 			txq = eth_dev->data->tx_queues;
633 			for (i -= 1; i >= 0; i--)
634 				dev_ops->tx_queue_release(txq[i]);
635 			goto fail;
636 		}
637 	}
638 
639 	free(tx_qconf);
640 	tx_qconf = NULL;
641 
642 	for (i = 0; i < nb_rxq; i++) {
643 		if (!rx_qconf[i].valid)
644 			continue;
645 		rc = dev_ops->rx_queue_setup(eth_dev, i, rx_qconf[i].nb_desc, 0,
646 					     &rx_qconf[i].conf.rx,
647 					     rx_qconf[i].mp);
648 		if (rc) {
649 			plt_err("Failed to setup rx queue rc=%d", rc);
650 			rxq = eth_dev->data->rx_queues;
651 			for (i -= 1; i >= 0; i--)
652 				dev_ops->rx_queue_release(rxq[i]);
653 			goto tx_queue_release;
654 		}
655 	}
656 
657 	free(rx_qconf);
658 	rx_qconf = NULL;
659 
660 	return 0;
661 
662 tx_queue_release:
663 	txq = eth_dev->data->tx_queues;
664 	for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
665 		dev_ops->tx_queue_release(txq[i]);
666 fail:
667 	if (tx_qconf)
668 		free(tx_qconf);
669 	if (rx_qconf)
670 		free(rx_qconf);
671 
672 	return rc;
673 }
674 
675 static uint16_t
676 nix_eth_nop_burst(void *queue, struct rte_mbuf **mbufs, uint16_t pkts)
677 {
678 	RTE_SET_USED(queue);
679 	RTE_SET_USED(mbufs);
680 	RTE_SET_USED(pkts);
681 
682 	return 0;
683 }
684 
685 static void
686 nix_set_nop_rxtx_function(struct rte_eth_dev *eth_dev)
687 {
688 	/* These dummy functions are required for supporting
689 	 * some applications which reconfigure queues without
690 	 * stopping tx burst and rx burst threads(eg kni app)
691 	 * When the queues context is saved, txq/rxqs are released
692 	 * which caused app crash since rx/tx burst is still
693 	 * on different lcores
694 	 */
695 	eth_dev->tx_pkt_burst = nix_eth_nop_burst;
696 	eth_dev->rx_pkt_burst = nix_eth_nop_burst;
697 	rte_mb();
698 }
699 
700 static int
701 nix_lso_tun_fmt_update(struct cnxk_eth_dev *dev)
702 {
703 	uint8_t udp_tun[ROC_NIX_LSO_TUN_MAX];
704 	uint8_t tun[ROC_NIX_LSO_TUN_MAX];
705 	struct roc_nix *nix = &dev->nix;
706 	int rc;
707 
708 	rc = roc_nix_lso_fmt_get(nix, udp_tun, tun);
709 	if (rc)
710 		return rc;
711 
712 	dev->lso_tun_fmt = ((uint64_t)tun[ROC_NIX_LSO_TUN_V4V4] |
713 			    (uint64_t)tun[ROC_NIX_LSO_TUN_V4V6] << 8 |
714 			    (uint64_t)tun[ROC_NIX_LSO_TUN_V6V4] << 16 |
715 			    (uint64_t)tun[ROC_NIX_LSO_TUN_V6V6] << 24);
716 
717 	dev->lso_tun_fmt |= ((uint64_t)udp_tun[ROC_NIX_LSO_TUN_V4V4] << 32 |
718 			     (uint64_t)udp_tun[ROC_NIX_LSO_TUN_V4V6] << 40 |
719 			     (uint64_t)udp_tun[ROC_NIX_LSO_TUN_V6V4] << 48 |
720 			     (uint64_t)udp_tun[ROC_NIX_LSO_TUN_V6V6] << 56);
721 	return 0;
722 }
723 
724 static int
725 nix_lso_fmt_setup(struct cnxk_eth_dev *dev)
726 {
727 	struct roc_nix *nix = &dev->nix;
728 	int rc;
729 
730 	/* Nothing much to do if offload is not enabled */
731 	if (!(dev->tx_offloads &
732 	      (DEV_TX_OFFLOAD_TCP_TSO | DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
733 	       DEV_TX_OFFLOAD_GENEVE_TNL_TSO | DEV_TX_OFFLOAD_GRE_TNL_TSO)))
734 		return 0;
735 
736 	/* Setup LSO formats in AF. Its a no-op if other ethdev has
737 	 * already set it up
738 	 */
739 	rc = roc_nix_lso_fmt_setup(nix);
740 	if (rc)
741 		return rc;
742 
743 	return nix_lso_tun_fmt_update(dev);
744 }
745 
746 int
747 cnxk_nix_configure(struct rte_eth_dev *eth_dev)
748 {
749 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
750 	struct rte_eth_dev_data *data = eth_dev->data;
751 	struct rte_eth_conf *conf = &data->dev_conf;
752 	struct rte_eth_rxmode *rxmode = &conf->rxmode;
753 	struct rte_eth_txmode *txmode = &conf->txmode;
754 	char ea_fmt[RTE_ETHER_ADDR_FMT_SIZE];
755 	struct roc_nix_fc_cfg fc_cfg = {0};
756 	struct roc_nix *nix = &dev->nix;
757 	struct rte_ether_addr *ea;
758 	uint8_t nb_rxq, nb_txq;
759 	uint64_t rx_cfg;
760 	void *qs;
761 	int rc;
762 
763 	rc = -EINVAL;
764 
765 	/* Sanity checks */
766 	if (rte_eal_has_hugepages() == 0) {
767 		plt_err("Huge page is not configured");
768 		goto fail_configure;
769 	}
770 
771 	if (conf->dcb_capability_en == 1) {
772 		plt_err("dcb enable is not supported");
773 		goto fail_configure;
774 	}
775 
776 	if (conf->fdir_conf.mode != RTE_FDIR_MODE_NONE) {
777 		plt_err("Flow director is not supported");
778 		goto fail_configure;
779 	}
780 
781 	if (rxmode->mq_mode != ETH_MQ_RX_NONE &&
782 	    rxmode->mq_mode != ETH_MQ_RX_RSS) {
783 		plt_err("Unsupported mq rx mode %d", rxmode->mq_mode);
784 		goto fail_configure;
785 	}
786 
787 	if (txmode->mq_mode != ETH_MQ_TX_NONE) {
788 		plt_err("Unsupported mq tx mode %d", txmode->mq_mode);
789 		goto fail_configure;
790 	}
791 
792 	/* Free the resources allocated from the previous configure */
793 	if (dev->configured == 1) {
794 		/* Unregister queue irq's */
795 		roc_nix_unregister_queue_irqs(nix);
796 
797 		/* Unregister CQ irqs if present */
798 		if (eth_dev->data->dev_conf.intr_conf.rxq)
799 			roc_nix_unregister_cq_irqs(nix);
800 
801 		/* Set no-op functions */
802 		nix_set_nop_rxtx_function(eth_dev);
803 		/* Store queue config for later */
804 		rc = nix_store_queue_cfg_and_then_release(eth_dev);
805 		if (rc)
806 			goto fail_configure;
807 		roc_nix_tm_fini(nix);
808 		roc_nix_lf_free(nix);
809 	}
810 
811 	dev->rx_offloads = rxmode->offloads;
812 	dev->tx_offloads = txmode->offloads;
813 
814 	/* Prepare rx cfg */
815 	rx_cfg = ROC_NIX_LF_RX_CFG_DIS_APAD;
816 	if (dev->rx_offloads &
817 	    (DEV_RX_OFFLOAD_TCP_CKSUM | DEV_RX_OFFLOAD_UDP_CKSUM)) {
818 		rx_cfg |= ROC_NIX_LF_RX_CFG_CSUM_OL4;
819 		rx_cfg |= ROC_NIX_LF_RX_CFG_CSUM_IL4;
820 	}
821 	rx_cfg |= (ROC_NIX_LF_RX_CFG_DROP_RE | ROC_NIX_LF_RX_CFG_L2_LEN_ERR |
822 		   ROC_NIX_LF_RX_CFG_LEN_IL4 | ROC_NIX_LF_RX_CFG_LEN_IL3 |
823 		   ROC_NIX_LF_RX_CFG_LEN_OL4 | ROC_NIX_LF_RX_CFG_LEN_OL3);
824 
825 	nb_rxq = RTE_MAX(data->nb_rx_queues, 1);
826 	nb_txq = RTE_MAX(data->nb_tx_queues, 1);
827 
828 	/* Alloc a nix lf */
829 	rc = roc_nix_lf_alloc(nix, nb_rxq, nb_txq, rx_cfg);
830 	if (rc) {
831 		plt_err("Failed to init nix_lf rc=%d", rc);
832 		goto fail_configure;
833 	}
834 
835 	dev->npc.channel = roc_nix_get_base_chan(nix);
836 
837 	nb_rxq = data->nb_rx_queues;
838 	nb_txq = data->nb_tx_queues;
839 	rc = -ENOMEM;
840 	if (nb_rxq) {
841 		/* Allocate memory for roc rq's and cq's */
842 		qs = plt_zmalloc(sizeof(struct roc_nix_rq) * nb_rxq, 0);
843 		if (!qs) {
844 			plt_err("Failed to alloc rqs");
845 			goto free_nix_lf;
846 		}
847 		dev->rqs = qs;
848 
849 		qs = plt_zmalloc(sizeof(struct roc_nix_cq) * nb_rxq, 0);
850 		if (!qs) {
851 			plt_err("Failed to alloc cqs");
852 			goto free_nix_lf;
853 		}
854 		dev->cqs = qs;
855 	}
856 
857 	if (nb_txq) {
858 		/* Allocate memory for roc sq's */
859 		qs = plt_zmalloc(sizeof(struct roc_nix_sq) * nb_txq, 0);
860 		if (!qs) {
861 			plt_err("Failed to alloc sqs");
862 			goto free_nix_lf;
863 		}
864 		dev->sqs = qs;
865 	}
866 
867 	/* Re-enable NIX LF error interrupts */
868 	roc_nix_err_intr_ena_dis(nix, true);
869 	roc_nix_ras_intr_ena_dis(nix, true);
870 
871 	if (nix->rx_ptp_ena &&
872 	    dev->npc.switch_header_type == ROC_PRIV_FLAGS_HIGIG) {
873 		plt_err("Both PTP and switch header enabled");
874 		goto free_nix_lf;
875 	}
876 
877 	rc = roc_nix_switch_hdr_set(nix, dev->npc.switch_header_type);
878 	if (rc) {
879 		plt_err("Failed to enable switch type nix_lf rc=%d", rc);
880 		goto free_nix_lf;
881 	}
882 
883 	/* Setup LSO if needed */
884 	rc = nix_lso_fmt_setup(dev);
885 	if (rc) {
886 		plt_err("Failed to setup nix lso format fields, rc=%d", rc);
887 		goto free_nix_lf;
888 	}
889 
890 	/* Configure RSS */
891 	rc = nix_rss_default_setup(dev);
892 	if (rc) {
893 		plt_err("Failed to configure rss rc=%d", rc);
894 		goto free_nix_lf;
895 	}
896 
897 	/* Init the default TM scheduler hierarchy */
898 	rc = roc_nix_tm_init(nix);
899 	if (rc) {
900 		plt_err("Failed to init traffic manager, rc=%d", rc);
901 		goto free_nix_lf;
902 	}
903 
904 	rc = roc_nix_tm_hierarchy_enable(nix, ROC_NIX_TM_DEFAULT, false);
905 	if (rc) {
906 		plt_err("Failed to enable default tm hierarchy, rc=%d", rc);
907 		goto tm_fini;
908 	}
909 
910 	/* Register queue IRQs */
911 	rc = roc_nix_register_queue_irqs(nix);
912 	if (rc) {
913 		plt_err("Failed to register queue interrupts rc=%d", rc);
914 		goto tm_fini;
915 	}
916 
917 	/* Register cq IRQs */
918 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
919 		if (eth_dev->data->nb_rx_queues > dev->nix.cints) {
920 			plt_err("Rx interrupt cannot be enabled, rxq > %d",
921 				dev->nix.cints);
922 			goto q_irq_fini;
923 		}
924 		/* Rx interrupt feature cannot work with vector mode because,
925 		 * vector mode does not process packets unless min 4 pkts are
926 		 * received, while cq interrupts are generated even for 1 pkt
927 		 * in the CQ.
928 		 */
929 		dev->scalar_ena = true;
930 
931 		rc = roc_nix_register_cq_irqs(nix);
932 		if (rc) {
933 			plt_err("Failed to register CQ interrupts rc=%d", rc);
934 			goto q_irq_fini;
935 		}
936 	}
937 
938 	/* Configure loop back mode */
939 	rc = roc_nix_mac_loopback_enable(nix,
940 					 eth_dev->data->dev_conf.lpbk_mode);
941 	if (rc) {
942 		plt_err("Failed to configure cgx loop back mode rc=%d", rc);
943 		goto cq_fini;
944 	}
945 
946 	/* Init flow control configuration */
947 	fc_cfg.cq_cfg_valid = false;
948 	fc_cfg.rxchan_cfg.enable = true;
949 	rc = roc_nix_fc_config_set(nix, &fc_cfg);
950 	if (rc) {
951 		plt_err("Failed to initialize flow control rc=%d", rc);
952 		goto cq_fini;
953 	}
954 
955 	/* Update flow control configuration to PMD */
956 	rc = nix_init_flow_ctrl_config(eth_dev);
957 	if (rc) {
958 		plt_err("Failed to initialize flow control rc=%d", rc);
959 		goto cq_fini;
960 	}
961 	/*
962 	 * Restore queue config when reconfigure followed by
963 	 * reconfigure and no queue configure invoked from application case.
964 	 */
965 	if (dev->configured == 1) {
966 		rc = nix_restore_queue_cfg(eth_dev);
967 		if (rc)
968 			goto cq_fini;
969 	}
970 
971 	/* Update the mac address */
972 	ea = eth_dev->data->mac_addrs;
973 	memcpy(ea, dev->mac_addr, RTE_ETHER_ADDR_LEN);
974 	if (rte_is_zero_ether_addr(ea))
975 		rte_eth_random_addr((uint8_t *)ea);
976 
977 	rte_ether_format_addr(ea_fmt, RTE_ETHER_ADDR_FMT_SIZE, ea);
978 
979 	plt_nix_dbg("Configured port%d mac=%s nb_rxq=%d nb_txq=%d"
980 		    " rx_offloads=0x%" PRIx64 " tx_offloads=0x%" PRIx64 "",
981 		    eth_dev->data->port_id, ea_fmt, nb_rxq, nb_txq,
982 		    dev->rx_offloads, dev->tx_offloads);
983 
984 	/* All good */
985 	dev->configured = 1;
986 	dev->nb_rxq = data->nb_rx_queues;
987 	dev->nb_txq = data->nb_tx_queues;
988 	return 0;
989 
990 cq_fini:
991 	roc_nix_unregister_cq_irqs(nix);
992 q_irq_fini:
993 	roc_nix_unregister_queue_irqs(nix);
994 tm_fini:
995 	roc_nix_tm_fini(nix);
996 free_nix_lf:
997 	nix_free_queue_mem(dev);
998 	rc |= roc_nix_lf_free(nix);
999 fail_configure:
1000 	dev->configured = 0;
1001 	return rc;
1002 }
1003 
1004 int
1005 cnxk_nix_tx_queue_start(struct rte_eth_dev *eth_dev, uint16_t qid)
1006 {
1007 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
1008 	struct rte_eth_dev_data *data = eth_dev->data;
1009 	struct roc_nix_sq *sq = &dev->sqs[qid];
1010 	int rc = -EINVAL;
1011 
1012 	if (data->tx_queue_state[qid] == RTE_ETH_QUEUE_STATE_STARTED)
1013 		return 0;
1014 
1015 	rc = roc_nix_tm_sq_aura_fc(sq, true);
1016 	if (rc) {
1017 		plt_err("Failed to enable sq aura fc, txq=%u, rc=%d", qid, rc);
1018 		goto done;
1019 	}
1020 
1021 	data->tx_queue_state[qid] = RTE_ETH_QUEUE_STATE_STARTED;
1022 done:
1023 	return rc;
1024 }
1025 
1026 int
1027 cnxk_nix_tx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t qid)
1028 {
1029 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
1030 	struct rte_eth_dev_data *data = eth_dev->data;
1031 	struct roc_nix_sq *sq = &dev->sqs[qid];
1032 	int rc;
1033 
1034 	if (data->tx_queue_state[qid] == RTE_ETH_QUEUE_STATE_STOPPED)
1035 		return 0;
1036 
1037 	rc = roc_nix_tm_sq_aura_fc(sq, false);
1038 	if (rc) {
1039 		plt_err("Failed to disable sqb aura fc, txq=%u, rc=%d", qid,
1040 			rc);
1041 		goto done;
1042 	}
1043 
1044 	data->tx_queue_state[qid] = RTE_ETH_QUEUE_STATE_STOPPED;
1045 done:
1046 	return rc;
1047 }
1048 
1049 static int
1050 cnxk_nix_rx_queue_start(struct rte_eth_dev *eth_dev, uint16_t qid)
1051 {
1052 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
1053 	struct rte_eth_dev_data *data = eth_dev->data;
1054 	struct roc_nix_rq *rq = &dev->rqs[qid];
1055 	int rc;
1056 
1057 	if (data->rx_queue_state[qid] == RTE_ETH_QUEUE_STATE_STARTED)
1058 		return 0;
1059 
1060 	rc = roc_nix_rq_ena_dis(rq, true);
1061 	if (rc) {
1062 		plt_err("Failed to enable rxq=%u, rc=%d", qid, rc);
1063 		goto done;
1064 	}
1065 
1066 	data->rx_queue_state[qid] = RTE_ETH_QUEUE_STATE_STARTED;
1067 done:
1068 	return rc;
1069 }
1070 
1071 static int
1072 cnxk_nix_rx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t qid)
1073 {
1074 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
1075 	struct rte_eth_dev_data *data = eth_dev->data;
1076 	struct roc_nix_rq *rq = &dev->rqs[qid];
1077 	int rc;
1078 
1079 	if (data->rx_queue_state[qid] == RTE_ETH_QUEUE_STATE_STOPPED)
1080 		return 0;
1081 
1082 	rc = roc_nix_rq_ena_dis(rq, false);
1083 	if (rc) {
1084 		plt_err("Failed to disable rxq=%u, rc=%d", qid, rc);
1085 		goto done;
1086 	}
1087 
1088 	data->rx_queue_state[qid] = RTE_ETH_QUEUE_STATE_STOPPED;
1089 done:
1090 	return rc;
1091 }
1092 
1093 static int
1094 cnxk_nix_dev_stop(struct rte_eth_dev *eth_dev)
1095 {
1096 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
1097 	const struct eth_dev_ops *dev_ops = eth_dev->dev_ops;
1098 	struct rte_mbuf *rx_pkts[32];
1099 	struct rte_eth_link link;
1100 	int count, i, j, rc;
1101 	void *rxq;
1102 
1103 	/* Disable switch hdr pkind */
1104 	roc_nix_switch_hdr_set(&dev->nix, 0);
1105 
1106 	/* Stop link change events */
1107 	if (!roc_nix_is_vf_or_sdp(&dev->nix))
1108 		roc_nix_mac_link_event_start_stop(&dev->nix, false);
1109 
1110 	/* Disable Rx via NPC */
1111 	roc_nix_npc_rx_ena_dis(&dev->nix, false);
1112 
1113 	/* Stop rx queues and free up pkts pending */
1114 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
1115 		rc = dev_ops->rx_queue_stop(eth_dev, i);
1116 		if (rc)
1117 			continue;
1118 
1119 		rxq = eth_dev->data->rx_queues[i];
1120 		count = dev->rx_pkt_burst_no_offload(rxq, rx_pkts, 32);
1121 		while (count) {
1122 			for (j = 0; j < count; j++)
1123 				rte_pktmbuf_free(rx_pkts[j]);
1124 			count = dev->rx_pkt_burst_no_offload(rxq, rx_pkts, 32);
1125 		}
1126 	}
1127 
1128 	/* Stop tx queues  */
1129 	for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
1130 		dev_ops->tx_queue_stop(eth_dev, i);
1131 
1132 	/* Bring down link status internally */
1133 	memset(&link, 0, sizeof(link));
1134 	rte_eth_linkstatus_set(eth_dev, &link);
1135 
1136 	return 0;
1137 }
1138 
1139 int
1140 cnxk_nix_dev_start(struct rte_eth_dev *eth_dev)
1141 {
1142 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
1143 	int rc, i;
1144 
1145 	if (eth_dev->data->nb_rx_queues != 0 && !dev->ptp_en) {
1146 		rc = nix_recalc_mtu(eth_dev);
1147 		if (rc)
1148 			return rc;
1149 	}
1150 
1151 	/* Start rx queues */
1152 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
1153 		rc = cnxk_nix_rx_queue_start(eth_dev, i);
1154 		if (rc)
1155 			return rc;
1156 	}
1157 
1158 	/* Start tx queues  */
1159 	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
1160 		rc = cnxk_nix_tx_queue_start(eth_dev, i);
1161 		if (rc)
1162 			return rc;
1163 	}
1164 
1165 	/* Update Flow control configuration */
1166 	rc = nix_update_flow_ctrl_config(eth_dev);
1167 	if (rc) {
1168 		plt_err("Failed to enable flow control. error code(%d)", rc);
1169 		return rc;
1170 	}
1171 
1172 	/* Enable Rx in NPC */
1173 	rc = roc_nix_npc_rx_ena_dis(&dev->nix, true);
1174 	if (rc) {
1175 		plt_err("Failed to enable NPC rx %d", rc);
1176 		return rc;
1177 	}
1178 
1179 	cnxk_nix_toggle_flag_link_cfg(dev, true);
1180 
1181 	/* Start link change events */
1182 	if (!roc_nix_is_vf_or_sdp(&dev->nix)) {
1183 		rc = roc_nix_mac_link_event_start_stop(&dev->nix, true);
1184 		if (rc) {
1185 			plt_err("Failed to start cgx link event %d", rc);
1186 			goto rx_disable;
1187 		}
1188 	}
1189 
1190 	/* Enable PTP if it is requested by the user or already
1191 	 * enabled on PF owning this VF
1192 	 */
1193 	memset(&dev->tstamp, 0, sizeof(struct cnxk_timesync_info));
1194 	if ((dev->rx_offloads & DEV_RX_OFFLOAD_TIMESTAMP) || dev->ptp_en)
1195 		cnxk_eth_dev_ops.timesync_enable(eth_dev);
1196 	else
1197 		cnxk_eth_dev_ops.timesync_disable(eth_dev);
1198 
1199 	if (dev->rx_offloads & DEV_RX_OFFLOAD_TIMESTAMP) {
1200 		rc = rte_mbuf_dyn_rx_timestamp_register
1201 			(&dev->tstamp.tstamp_dynfield_offset,
1202 			 &dev->tstamp.rx_tstamp_dynflag);
1203 		if (rc != 0) {
1204 			plt_err("Failed to register Rx timestamp field/flag");
1205 			goto rx_disable;
1206 		}
1207 	}
1208 
1209 	cnxk_nix_toggle_flag_link_cfg(dev, false);
1210 
1211 	return 0;
1212 
1213 rx_disable:
1214 	roc_nix_npc_rx_ena_dis(&dev->nix, false);
1215 	cnxk_nix_toggle_flag_link_cfg(dev, false);
1216 	return rc;
1217 }
1218 
1219 static int cnxk_nix_dev_reset(struct rte_eth_dev *eth_dev);
1220 static int cnxk_nix_dev_close(struct rte_eth_dev *eth_dev);
1221 
1222 /* CNXK platform independent eth dev ops */
1223 struct eth_dev_ops cnxk_eth_dev_ops = {
1224 	.mtu_set = cnxk_nix_mtu_set,
1225 	.mac_addr_add = cnxk_nix_mac_addr_add,
1226 	.mac_addr_remove = cnxk_nix_mac_addr_del,
1227 	.mac_addr_set = cnxk_nix_mac_addr_set,
1228 	.dev_infos_get = cnxk_nix_info_get,
1229 	.link_update = cnxk_nix_link_update,
1230 	.tx_queue_release = cnxk_nix_tx_queue_release,
1231 	.rx_queue_release = cnxk_nix_rx_queue_release,
1232 	.dev_stop = cnxk_nix_dev_stop,
1233 	.dev_close = cnxk_nix_dev_close,
1234 	.dev_reset = cnxk_nix_dev_reset,
1235 	.tx_queue_start = cnxk_nix_tx_queue_start,
1236 	.rx_queue_start = cnxk_nix_rx_queue_start,
1237 	.rx_queue_stop = cnxk_nix_rx_queue_stop,
1238 	.dev_supported_ptypes_get = cnxk_nix_supported_ptypes_get,
1239 	.promiscuous_enable = cnxk_nix_promisc_enable,
1240 	.promiscuous_disable = cnxk_nix_promisc_disable,
1241 	.allmulticast_enable = cnxk_nix_allmulticast_enable,
1242 	.allmulticast_disable = cnxk_nix_allmulticast_disable,
1243 	.rx_burst_mode_get = cnxk_nix_rx_burst_mode_get,
1244 	.tx_burst_mode_get = cnxk_nix_tx_burst_mode_get,
1245 	.flow_ctrl_get = cnxk_nix_flow_ctrl_get,
1246 	.flow_ctrl_set = cnxk_nix_flow_ctrl_set,
1247 	.dev_set_link_up = cnxk_nix_set_link_up,
1248 	.dev_set_link_down = cnxk_nix_set_link_down,
1249 	.get_module_info = cnxk_nix_get_module_info,
1250 	.get_module_eeprom = cnxk_nix_get_module_eeprom,
1251 	.rx_queue_intr_enable = cnxk_nix_rx_queue_intr_enable,
1252 	.rx_queue_intr_disable = cnxk_nix_rx_queue_intr_disable,
1253 	.pool_ops_supported = cnxk_nix_pool_ops_supported,
1254 	.queue_stats_mapping_set = cnxk_nix_queue_stats_mapping,
1255 	.stats_get = cnxk_nix_stats_get,
1256 	.stats_reset = cnxk_nix_stats_reset,
1257 	.xstats_get = cnxk_nix_xstats_get,
1258 	.xstats_get_names = cnxk_nix_xstats_get_names,
1259 	.xstats_reset = cnxk_nix_xstats_reset,
1260 	.xstats_get_by_id = cnxk_nix_xstats_get_by_id,
1261 	.xstats_get_names_by_id = cnxk_nix_xstats_get_names_by_id,
1262 	.fw_version_get = cnxk_nix_fw_version_get,
1263 	.rxq_info_get = cnxk_nix_rxq_info_get,
1264 	.txq_info_get = cnxk_nix_txq_info_get,
1265 	.tx_done_cleanup = cnxk_nix_tx_done_cleanup,
1266 	.flow_ops_get = cnxk_nix_flow_ops_get,
1267 	.get_reg = cnxk_nix_dev_get_reg,
1268 	.timesync_read_rx_timestamp = cnxk_nix_timesync_read_rx_timestamp,
1269 	.timesync_read_tx_timestamp = cnxk_nix_timesync_read_tx_timestamp,
1270 	.timesync_read_time = cnxk_nix_timesync_read_time,
1271 	.timesync_write_time = cnxk_nix_timesync_write_time,
1272 	.timesync_adjust_time = cnxk_nix_timesync_adjust_time,
1273 	.read_clock = cnxk_nix_read_clock,
1274 	.reta_update = cnxk_nix_reta_update,
1275 	.reta_query = cnxk_nix_reta_query,
1276 	.rss_hash_update = cnxk_nix_rss_hash_update,
1277 	.rss_hash_conf_get = cnxk_nix_rss_hash_conf_get,
1278 	.set_mc_addr_list = cnxk_nix_mc_addr_list_configure,
1279 };
1280 
1281 static int
1282 cnxk_eth_dev_init(struct rte_eth_dev *eth_dev)
1283 {
1284 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
1285 	struct roc_nix *nix = &dev->nix;
1286 	struct rte_pci_device *pci_dev;
1287 	int rc, max_entries;
1288 
1289 	eth_dev->dev_ops = &cnxk_eth_dev_ops;
1290 
1291 	/* For secondary processes, the primary has done all the work */
1292 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1293 		return 0;
1294 
1295 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1296 	rte_eth_copy_pci_info(eth_dev, pci_dev);
1297 
1298 	/* Parse devargs string */
1299 	rc = cnxk_ethdev_parse_devargs(eth_dev->device->devargs, dev);
1300 	if (rc) {
1301 		plt_err("Failed to parse devargs rc=%d", rc);
1302 		goto error;
1303 	}
1304 
1305 	/* Initialize base roc nix */
1306 	nix->pci_dev = pci_dev;
1307 	nix->hw_vlan_ins = true;
1308 	rc = roc_nix_dev_init(nix);
1309 	if (rc) {
1310 		plt_err("Failed to initialize roc nix rc=%d", rc);
1311 		goto error;
1312 	}
1313 
1314 	/* Register up msg callbacks */
1315 	roc_nix_mac_link_cb_register(nix, cnxk_eth_dev_link_status_cb);
1316 
1317 	/* Register up msg callbacks */
1318 	roc_nix_mac_link_info_get_cb_register(nix,
1319 					      cnxk_eth_dev_link_status_get_cb);
1320 
1321 	dev->eth_dev = eth_dev;
1322 	dev->configured = 0;
1323 	dev->ptype_disable = 0;
1324 
1325 	/* For vfs, returned max_entries will be 0. but to keep default mac
1326 	 * address, one entry must be allocated. so setting up to 1.
1327 	 */
1328 	if (roc_nix_is_vf_or_sdp(nix))
1329 		max_entries = 1;
1330 	else
1331 		max_entries = roc_nix_mac_max_entries_get(nix);
1332 
1333 	if (max_entries <= 0) {
1334 		plt_err("Failed to get max entries for mac addr");
1335 		rc = -ENOTSUP;
1336 		goto dev_fini;
1337 	}
1338 
1339 	eth_dev->data->mac_addrs =
1340 		rte_zmalloc("mac_addr", max_entries * RTE_ETHER_ADDR_LEN, 0);
1341 	if (eth_dev->data->mac_addrs == NULL) {
1342 		plt_err("Failed to allocate memory for mac addr");
1343 		rc = -ENOMEM;
1344 		goto dev_fini;
1345 	}
1346 
1347 	dev->max_mac_entries = max_entries;
1348 	dev->dmac_filter_count = 1;
1349 
1350 	/* Get mac address */
1351 	rc = roc_nix_npc_mac_addr_get(nix, dev->mac_addr);
1352 	if (rc) {
1353 		plt_err("Failed to get mac addr, rc=%d", rc);
1354 		goto free_mac_addrs;
1355 	}
1356 
1357 	/* Update the mac address */
1358 	memcpy(eth_dev->data->mac_addrs, dev->mac_addr, RTE_ETHER_ADDR_LEN);
1359 
1360 	if (!roc_nix_is_vf_or_sdp(nix)) {
1361 		/* Sync same MAC address to CGX/RPM table */
1362 		rc = roc_nix_mac_addr_set(nix, dev->mac_addr);
1363 		if (rc) {
1364 			plt_err("Failed to set mac addr, rc=%d", rc);
1365 			goto free_mac_addrs;
1366 		}
1367 	}
1368 
1369 	/* Union of all capabilities supported by CNXK.
1370 	 * Platform specific capabilities will be
1371 	 * updated later.
1372 	 */
1373 	dev->rx_offload_capa = nix_get_rx_offload_capa(dev);
1374 	dev->tx_offload_capa = nix_get_tx_offload_capa(dev);
1375 	dev->speed_capa = nix_get_speed_capa(dev);
1376 
1377 	/* Initialize roc npc */
1378 	dev->npc.roc_nix = nix;
1379 	rc = roc_npc_init(&dev->npc);
1380 	if (rc)
1381 		goto free_mac_addrs;
1382 
1383 	plt_nix_dbg("Port=%d pf=%d vf=%d ver=%s hwcap=0x%" PRIx64
1384 		    " rxoffload_capa=0x%" PRIx64 " txoffload_capa=0x%" PRIx64,
1385 		    eth_dev->data->port_id, roc_nix_get_pf(nix),
1386 		    roc_nix_get_vf(nix), CNXK_ETH_DEV_PMD_VERSION, dev->hwcap,
1387 		    dev->rx_offload_capa, dev->tx_offload_capa);
1388 	return 0;
1389 
1390 free_mac_addrs:
1391 	rte_free(eth_dev->data->mac_addrs);
1392 dev_fini:
1393 	roc_nix_dev_fini(nix);
1394 error:
1395 	plt_err("Failed to init nix eth_dev rc=%d", rc);
1396 	return rc;
1397 }
1398 
1399 static int
1400 cnxk_eth_dev_uninit(struct rte_eth_dev *eth_dev, bool reset)
1401 {
1402 	struct cnxk_eth_dev *dev = cnxk_eth_pmd_priv(eth_dev);
1403 	const struct eth_dev_ops *dev_ops = eth_dev->dev_ops;
1404 	struct roc_nix *nix = &dev->nix;
1405 	int rc, i;
1406 
1407 	/* Nothing to be done for secondary processes */
1408 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1409 		return 0;
1410 
1411 	/* Clear the flag since we are closing down */
1412 	dev->configured = 0;
1413 
1414 	roc_nix_npc_rx_ena_dis(nix, false);
1415 
1416 	/* Disable and free rte_flow entries */
1417 	roc_npc_fini(&dev->npc);
1418 
1419 	/* Disable link status events */
1420 	roc_nix_mac_link_event_start_stop(nix, false);
1421 
1422 	/* Unregister the link update op, this is required to stop VFs from
1423 	 * receiving link status updates on exit path.
1424 	 */
1425 	roc_nix_mac_link_cb_unregister(nix);
1426 
1427 	/* Free up SQs */
1428 	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
1429 		dev_ops->tx_queue_release(eth_dev->data->tx_queues[i]);
1430 		eth_dev->data->tx_queues[i] = NULL;
1431 	}
1432 	eth_dev->data->nb_tx_queues = 0;
1433 
1434 	/* Free up RQ's and CQ's */
1435 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
1436 		dev_ops->rx_queue_release(eth_dev->data->rx_queues[i]);
1437 		eth_dev->data->rx_queues[i] = NULL;
1438 	}
1439 	eth_dev->data->nb_rx_queues = 0;
1440 
1441 	/* Free tm resources */
1442 	roc_nix_tm_fini(nix);
1443 
1444 	/* Unregister queue irqs */
1445 	roc_nix_unregister_queue_irqs(nix);
1446 
1447 	/* Unregister cq irqs */
1448 	if (eth_dev->data->dev_conf.intr_conf.rxq)
1449 		roc_nix_unregister_cq_irqs(nix);
1450 
1451 	/* Free ROC RQ's, SQ's and CQ's memory */
1452 	nix_free_queue_mem(dev);
1453 
1454 	/* Free nix lf resources */
1455 	rc = roc_nix_lf_free(nix);
1456 	if (rc)
1457 		plt_err("Failed to free nix lf, rc=%d", rc);
1458 
1459 	rte_free(eth_dev->data->mac_addrs);
1460 	eth_dev->data->mac_addrs = NULL;
1461 
1462 	rc = roc_nix_dev_fini(nix);
1463 	/* Can be freed later by PMD if NPA LF is in use */
1464 	if (rc == -EAGAIN) {
1465 		if (!reset)
1466 			eth_dev->data->dev_private = NULL;
1467 		return 0;
1468 	} else if (rc) {
1469 		plt_err("Failed in nix dev fini, rc=%d", rc);
1470 	}
1471 
1472 	return rc;
1473 }
1474 
1475 static int
1476 cnxk_nix_dev_close(struct rte_eth_dev *eth_dev)
1477 {
1478 	cnxk_eth_dev_uninit(eth_dev, false);
1479 	return 0;
1480 }
1481 
1482 static int
1483 cnxk_nix_dev_reset(struct rte_eth_dev *eth_dev)
1484 {
1485 	int rc;
1486 
1487 	rc = cnxk_eth_dev_uninit(eth_dev, true);
1488 	if (rc)
1489 		return rc;
1490 
1491 	return cnxk_eth_dev_init(eth_dev);
1492 }
1493 
1494 int
1495 cnxk_nix_remove(struct rte_pci_device *pci_dev)
1496 {
1497 	struct rte_eth_dev *eth_dev;
1498 	struct roc_nix *nix;
1499 	int rc = -EINVAL;
1500 
1501 	eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
1502 	if (eth_dev) {
1503 		/* Cleanup eth dev */
1504 		rc = cnxk_eth_dev_uninit(eth_dev, false);
1505 		if (rc)
1506 			return rc;
1507 
1508 		rte_eth_dev_release_port(eth_dev);
1509 	}
1510 
1511 	/* Nothing to be done for secondary processes */
1512 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1513 		return 0;
1514 
1515 	/* Check if this device is hosting common resource */
1516 	nix = roc_idev_npa_nix_get();
1517 	if (nix->pci_dev != pci_dev)
1518 		return 0;
1519 
1520 	/* Try nix fini now */
1521 	rc = roc_nix_dev_fini(nix);
1522 	if (rc == -EAGAIN) {
1523 		plt_info("%s: common resource in use by other devices",
1524 			 pci_dev->name);
1525 		goto exit;
1526 	} else if (rc) {
1527 		plt_err("Failed in nix dev fini, rc=%d", rc);
1528 		goto exit;
1529 	}
1530 
1531 	/* Free device pointer as rte_ethdev does not have it anymore */
1532 	rte_free(nix);
1533 exit:
1534 	return rc;
1535 }
1536 
1537 int
1538 cnxk_nix_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
1539 {
1540 	int rc;
1541 
1542 	RTE_SET_USED(pci_drv);
1543 
1544 	rc = rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct cnxk_eth_dev),
1545 					   cnxk_eth_dev_init);
1546 
1547 	/* On error on secondary, recheck if port exists in primary or
1548 	 * in mid of detach state.
1549 	 */
1550 	if (rte_eal_process_type() != RTE_PROC_PRIMARY && rc)
1551 		if (!rte_eth_dev_allocated(pci_dev->device.name))
1552 			return 0;
1553 	return rc;
1554 }
1555