xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision 41c60f74d85f2e889dfdd155bc4e30cf00913914)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2017-2021 Marvell International Ltd.
3  * Copyright(c) 2017-2021 Semihalf.
4  * All rights reserved.
5  */
6 
7 #include <rte_string_fns.h>
8 #include <ethdev_driver.h>
9 #include <rte_kvargs.h>
10 #include <rte_log.h>
11 #include <rte_malloc.h>
12 #include <rte_bus_vdev.h>
13 
14 #include <fcntl.h>
15 #include <linux/ethtool.h>
16 #include <linux/sockios.h>
17 #include <net/if.h>
18 #include <net/if_arp.h>
19 #include <sys/ioctl.h>
20 #include <sys/socket.h>
21 #include <sys/stat.h>
22 #include <sys/types.h>
23 
24 #include <rte_mvep_common.h>
25 #include "mrvl_ethdev.h"
26 #include "mrvl_qos.h"
27 #include "mrvl_flow.h"
28 #include "mrvl_mtr.h"
29 #include "mrvl_tm.h"
30 
31 /* bitmask with reserved hifs */
32 #define MRVL_MUSDK_HIFS_RESERVED 0x0F
33 /* bitmask with reserved bpools */
34 #define MRVL_MUSDK_BPOOLS_RESERVED 0x07
35 /* bitmask with reserved kernel RSS tables */
36 #define MRVL_MUSDK_RSS_RESERVED 0x0F
37 /* maximum number of available hifs */
38 #define MRVL_MUSDK_HIFS_MAX 9
39 
40 /* prefetch shift */
41 #define MRVL_MUSDK_PREFETCH_SHIFT 2
42 
43 /* TCAM has 25 entries reserved for uc/mc filter entries
44  * + 1 for primary mac address
45  */
46 #define MRVL_MAC_ADDRS_MAX (1 + 25)
47 #define MRVL_MATCH_LEN 16
48 #define MRVL_PKT_EFFEC_OFFS (MRVL_PKT_OFFS + MV_MH_SIZE)
49 /* Maximum allowable packet size */
50 #define MRVL_PKT_SIZE_MAX (10240 - MV_MH_SIZE)
51 
52 #define MRVL_IFACE_NAME_ARG "iface"
53 #define MRVL_CFG_ARG "cfg"
54 
55 #define MRVL_BURST_SIZE 64
56 
57 #define MRVL_ARP_LENGTH 28
58 
59 #define MRVL_COOKIE_ADDR_INVALID ~0ULL
60 #define MRVL_COOKIE_HIGH_ADDR_MASK 0xffffff0000000000
61 
62 /** Port Rx offload capabilities */
63 #define MRVL_RX_OFFLOADS (DEV_RX_OFFLOAD_VLAN_FILTER | \
64 			  DEV_RX_OFFLOAD_JUMBO_FRAME | \
65 			  DEV_RX_OFFLOAD_CHECKSUM)
66 
67 /** Port Tx offloads capabilities */
68 #define MRVL_TX_OFFLOAD_CHECKSUM (DEV_TX_OFFLOAD_IPV4_CKSUM | \
69 				  DEV_TX_OFFLOAD_UDP_CKSUM  | \
70 				  DEV_TX_OFFLOAD_TCP_CKSUM)
71 #define MRVL_TX_OFFLOADS (MRVL_TX_OFFLOAD_CHECKSUM | \
72 			  DEV_TX_OFFLOAD_MULTI_SEGS)
73 
74 #define MRVL_TX_PKT_OFFLOADS (PKT_TX_IP_CKSUM | \
75 			      PKT_TX_TCP_CKSUM | \
76 			      PKT_TX_UDP_CKSUM)
77 
78 static const char * const valid_args[] = {
79 	MRVL_IFACE_NAME_ARG,
80 	MRVL_CFG_ARG,
81 	NULL
82 };
83 
84 static int used_hifs = MRVL_MUSDK_HIFS_RESERVED;
85 static struct pp2_hif *hifs[RTE_MAX_LCORE];
86 static int used_bpools[PP2_NUM_PKT_PROC] = {
87 	[0 ... PP2_NUM_PKT_PROC - 1] = MRVL_MUSDK_BPOOLS_RESERVED
88 };
89 
90 static struct pp2_bpool *mrvl_port_to_bpool_lookup[RTE_MAX_ETHPORTS];
91 static int mrvl_port_bpool_size[PP2_NUM_PKT_PROC][PP2_BPOOL_NUM_POOLS][RTE_MAX_LCORE];
92 static uint64_t cookie_addr_high = MRVL_COOKIE_ADDR_INVALID;
93 static int dummy_pool_id[PP2_NUM_PKT_PROC];
94 struct pp2_bpool *dummy_pool[PP2_NUM_PKT_PROC] = {0};
95 
96 struct mrvl_ifnames {
97 	const char *names[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
98 	int idx;
99 };
100 
101 /*
102  * To use buffer harvesting based on loopback port shadow queue structure
103  * was introduced for buffers information bookkeeping.
104  *
105  * Before sending the packet, related buffer information (pp2_buff_inf) is
106  * stored in shadow queue. After packet is transmitted no longer used
107  * packet buffer is released back to it's original hardware pool,
108  * on condition it originated from interface.
109  * In case it  was generated by application itself i.e: mbuf->port field is
110  * 0xff then its released to software mempool.
111  */
112 struct mrvl_shadow_txq {
113 	int head;           /* write index - used when sending buffers */
114 	int tail;           /* read index - used when releasing buffers */
115 	u16 size;           /* queue occupied size */
116 	u16 num_to_release; /* number of descriptors sent, that can be
117 			     * released
118 			     */
119 	struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
120 };
121 
122 struct mrvl_rxq {
123 	struct mrvl_priv *priv;
124 	struct rte_mempool *mp;
125 	int queue_id;
126 	int port_id;
127 	int cksum_enabled;
128 	uint64_t bytes_recv;
129 	uint64_t drop_mac;
130 };
131 
132 struct mrvl_txq {
133 	struct mrvl_priv *priv;
134 	int queue_id;
135 	int port_id;
136 	uint64_t bytes_sent;
137 	struct mrvl_shadow_txq shadow_txqs[RTE_MAX_LCORE];
138 	int tx_deferred_start;
139 };
140 
141 static int mrvl_lcore_first;
142 static int mrvl_lcore_last;
143 static int mrvl_dev_num;
144 
145 static int mrvl_fill_bpool(struct mrvl_rxq *rxq, int num);
146 static inline void mrvl_free_sent_buffers(struct pp2_ppio *ppio,
147 			struct pp2_hif *hif, unsigned int core_id,
148 			struct mrvl_shadow_txq *sq, int qid, int force);
149 
150 static uint16_t mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
151 				  uint16_t nb_pkts);
152 static uint16_t mrvl_tx_sg_pkt_burst(void *txq,	struct rte_mbuf **tx_pkts,
153 				     uint16_t nb_pkts);
154 static int rte_pmd_mrvl_remove(struct rte_vdev_device *vdev);
155 static void mrvl_deinit_pp2(void);
156 static void mrvl_deinit_hifs(void);
157 
158 static int
159 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
160 		  uint32_t index, uint32_t vmdq __rte_unused);
161 static int
162 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr);
163 static int
164 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on);
165 static int mrvl_promiscuous_enable(struct rte_eth_dev *dev);
166 static int mrvl_allmulticast_enable(struct rte_eth_dev *dev);
167 static int
168 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf);
169 
170 #define MRVL_XSTATS_TBL_ENTRY(name) { \
171 	#name, offsetof(struct pp2_ppio_statistics, name),	\
172 	sizeof(((struct pp2_ppio_statistics *)0)->name)		\
173 }
174 
175 /* Table with xstats data */
176 static struct {
177 	const char *name;
178 	unsigned int offset;
179 	unsigned int size;
180 } mrvl_xstats_tbl[] = {
181 	MRVL_XSTATS_TBL_ENTRY(rx_bytes),
182 	MRVL_XSTATS_TBL_ENTRY(rx_packets),
183 	MRVL_XSTATS_TBL_ENTRY(rx_unicast_packets),
184 	MRVL_XSTATS_TBL_ENTRY(rx_errors),
185 	MRVL_XSTATS_TBL_ENTRY(rx_fullq_dropped),
186 	MRVL_XSTATS_TBL_ENTRY(rx_bm_dropped),
187 	MRVL_XSTATS_TBL_ENTRY(rx_early_dropped),
188 	MRVL_XSTATS_TBL_ENTRY(rx_fifo_dropped),
189 	MRVL_XSTATS_TBL_ENTRY(rx_cls_dropped),
190 	MRVL_XSTATS_TBL_ENTRY(tx_bytes),
191 	MRVL_XSTATS_TBL_ENTRY(tx_packets),
192 	MRVL_XSTATS_TBL_ENTRY(tx_unicast_packets),
193 	MRVL_XSTATS_TBL_ENTRY(tx_errors)
194 };
195 
196 static inline int
197 mrvl_reserve_bit(int *bitmap, int max)
198 {
199 	int n = sizeof(*bitmap) * 8 - __builtin_clz(*bitmap);
200 
201 	if (n >= max)
202 		return -1;
203 
204 	*bitmap |= 1 << n;
205 
206 	return n;
207 }
208 
209 static int
210 mrvl_pp2_fixup_init(void)
211 {
212 	struct pp2_bpool_params bpool_params;
213 	char			name[15];
214 	int			err, i;
215 
216 	memset(dummy_pool, 0, sizeof(dummy_pool));
217 	for (i = 0; i < pp2_get_num_inst(); i++) {
218 		dummy_pool_id[i] = mrvl_reserve_bit(&used_bpools[i],
219 					     PP2_BPOOL_NUM_POOLS);
220 		if (dummy_pool_id[i] < 0) {
221 			MRVL_LOG(ERR, "Can't find free pool\n");
222 			return -1;
223 		}
224 
225 		memset(name, 0, sizeof(name));
226 		snprintf(name, sizeof(name), "pool-%d:%d", i, dummy_pool_id[i]);
227 		memset(&bpool_params, 0, sizeof(bpool_params));
228 		bpool_params.match = name;
229 		bpool_params.buff_len = MRVL_PKT_OFFS;
230 		bpool_params.dummy_short_pool = 1;
231 		err = pp2_bpool_init(&bpool_params, &dummy_pool[i]);
232 		if (err != 0 || !dummy_pool[i]) {
233 			MRVL_LOG(ERR, "BPool init failed!\n");
234 			used_bpools[i] &= ~(1 << dummy_pool_id[i]);
235 			return -1;
236 		}
237 	}
238 
239 	return 0;
240 }
241 
242 /**
243  * Initialize packet processor.
244  *
245  * @return
246  *   0 on success, negative error value otherwise.
247  */
248 static int
249 mrvl_init_pp2(void)
250 {
251 	struct pp2_init_params	init_params;
252 	int			err;
253 
254 	memset(&init_params, 0, sizeof(init_params));
255 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
256 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
257 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
258 	if (mrvl_cfg && mrvl_cfg->pp2_cfg.prs_udfs.num_udfs)
259 		memcpy(&init_params.prs_udfs, &mrvl_cfg->pp2_cfg.prs_udfs,
260 		       sizeof(struct pp2_parse_udfs));
261 	err = pp2_init(&init_params);
262 	if (err != 0) {
263 		MRVL_LOG(ERR, "PP2 init failed");
264 		return -1;
265 	}
266 
267 	err = mrvl_pp2_fixup_init();
268 	if (err != 0) {
269 		MRVL_LOG(ERR, "PP2 fixup init failed");
270 		return -1;
271 	}
272 
273 	return 0;
274 }
275 
276 static void
277 mrvl_pp2_fixup_deinit(void)
278 {
279 	int i;
280 
281 	for (i = 0; i < PP2_NUM_PKT_PROC; i++) {
282 		if (!dummy_pool[i])
283 			continue;
284 		pp2_bpool_deinit(dummy_pool[i]);
285 		used_bpools[i] &= ~(1 << dummy_pool_id[i]);
286 	}
287 }
288 
289 /**
290  * Deinitialize packet processor.
291  *
292  * @return
293  *   0 on success, negative error value otherwise.
294  */
295 static void
296 mrvl_deinit_pp2(void)
297 {
298 	mrvl_pp2_fixup_deinit();
299 	pp2_deinit();
300 }
301 
302 static inline void
303 mrvl_fill_shadowq(struct mrvl_shadow_txq *sq, struct rte_mbuf *buf)
304 {
305 	sq->ent[sq->head].buff.cookie = (uint64_t)buf;
306 	sq->ent[sq->head].buff.addr = buf ?
307 		rte_mbuf_data_iova_default(buf) : 0;
308 
309 	sq->ent[sq->head].bpool =
310 		(unlikely(!buf || buf->port >= RTE_MAX_ETHPORTS ||
311 		 buf->refcnt > 1)) ? NULL :
312 		 mrvl_port_to_bpool_lookup[buf->port];
313 
314 	sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
315 	sq->size++;
316 }
317 
318 /**
319  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
320  */
321 static void
322 mrvl_deinit_hifs(void)
323 {
324 	int i;
325 
326 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
327 		if (hifs[i])
328 			pp2_hif_deinit(hifs[i]);
329 	}
330 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
331 	memset(hifs, 0, sizeof(hifs));
332 }
333 
334 static inline void
335 mrvl_fill_desc(struct pp2_ppio_desc *desc, struct rte_mbuf *buf)
336 {
337 	pp2_ppio_outq_desc_reset(desc);
338 	pp2_ppio_outq_desc_set_phys_addr(desc, rte_pktmbuf_iova(buf));
339 	pp2_ppio_outq_desc_set_pkt_offset(desc, 0);
340 	pp2_ppio_outq_desc_set_pkt_len(desc, rte_pktmbuf_data_len(buf));
341 }
342 
343 static inline int
344 mrvl_get_bpool_size(int pp2_id, int pool_id)
345 {
346 	int i;
347 	int size = 0;
348 
349 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
350 		size += mrvl_port_bpool_size[pp2_id][pool_id][i];
351 
352 	return size;
353 }
354 
355 static int
356 mrvl_init_hif(int core_id)
357 {
358 	struct pp2_hif_params params;
359 	char match[MRVL_MATCH_LEN];
360 	int ret;
361 
362 	ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
363 	if (ret < 0) {
364 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
365 		return ret;
366 	}
367 
368 	snprintf(match, sizeof(match), "hif-%d", ret);
369 	memset(&params, 0, sizeof(params));
370 	params.match = match;
371 	params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
372 	ret = pp2_hif_init(&params, &hifs[core_id]);
373 	if (ret) {
374 		MRVL_LOG(ERR, "Failed to initialize hif %d", core_id);
375 		return ret;
376 	}
377 
378 	return 0;
379 }
380 
381 static inline struct pp2_hif*
382 mrvl_get_hif(struct mrvl_priv *priv, int core_id)
383 {
384 	int ret;
385 
386 	if (likely(hifs[core_id] != NULL))
387 		return hifs[core_id];
388 
389 	rte_spinlock_lock(&priv->lock);
390 
391 	ret = mrvl_init_hif(core_id);
392 	if (ret < 0) {
393 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
394 		goto out;
395 	}
396 
397 	if (core_id < mrvl_lcore_first)
398 		mrvl_lcore_first = core_id;
399 
400 	if (core_id > mrvl_lcore_last)
401 		mrvl_lcore_last = core_id;
402 out:
403 	rte_spinlock_unlock(&priv->lock);
404 
405 	return hifs[core_id];
406 }
407 
408 /**
409  * Set tx burst function according to offload flag
410  *
411  * @param dev
412  *   Pointer to Ethernet device structure.
413  */
414 static void
415 mrvl_set_tx_function(struct rte_eth_dev *dev)
416 {
417 	struct mrvl_priv *priv = dev->data->dev_private;
418 
419 	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
420 	if (priv->multiseg) {
421 		RTE_LOG(INFO, PMD, "Using multi-segment tx callback\n");
422 		dev->tx_pkt_burst = mrvl_tx_sg_pkt_burst;
423 	} else {
424 		RTE_LOG(INFO, PMD, "Using single-segment tx callback\n");
425 		dev->tx_pkt_burst = mrvl_tx_pkt_burst;
426 	}
427 }
428 
429 /**
430  * Configure rss based on dpdk rss configuration.
431  *
432  * @param priv
433  *   Pointer to private structure.
434  * @param rss_conf
435  *   Pointer to RSS configuration.
436  *
437  * @return
438  *   0 on success, negative error value otherwise.
439  */
440 static int
441 mrvl_configure_rss(struct mrvl_priv *priv, struct rte_eth_rss_conf *rss_conf)
442 {
443 	if (rss_conf->rss_key)
444 		MRVL_LOG(WARNING, "Changing hash key is not supported");
445 
446 	if (rss_conf->rss_hf == 0) {
447 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
448 	} else if (rss_conf->rss_hf & ETH_RSS_IPV4) {
449 		priv->ppio_params.inqs_params.hash_type =
450 			PP2_PPIO_HASH_T_2_TUPLE;
451 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP) {
452 		priv->ppio_params.inqs_params.hash_type =
453 			PP2_PPIO_HASH_T_5_TUPLE;
454 		priv->rss_hf_tcp = 1;
455 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) {
456 		priv->ppio_params.inqs_params.hash_type =
457 			PP2_PPIO_HASH_T_5_TUPLE;
458 		priv->rss_hf_tcp = 0;
459 	} else {
460 		return -EINVAL;
461 	}
462 
463 	return 0;
464 }
465 
466 /**
467  * Ethernet device configuration.
468  *
469  * Prepare the driver for a given number of TX and RX queues and
470  * configure RSS.
471  *
472  * @param dev
473  *   Pointer to Ethernet device structure.
474  *
475  * @return
476  *   0 on success, negative error value otherwise.
477  */
478 static int
479 mrvl_dev_configure(struct rte_eth_dev *dev)
480 {
481 	struct mrvl_priv *priv = dev->data->dev_private;
482 	int ret;
483 
484 	if (priv->ppio) {
485 		MRVL_LOG(INFO, "Device reconfiguration is not supported");
486 		return -EINVAL;
487 	}
488 
489 	if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_NONE &&
490 	    dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
491 		MRVL_LOG(INFO, "Unsupported rx multi queue mode %d",
492 			dev->data->dev_conf.rxmode.mq_mode);
493 		return -EINVAL;
494 	}
495 
496 	if (dev->data->dev_conf.rxmode.split_hdr_size) {
497 		MRVL_LOG(INFO, "Split headers not supported");
498 		return -EINVAL;
499 	}
500 
501 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) {
502 		dev->data->mtu = dev->data->dev_conf.rxmode.max_rx_pkt_len -
503 				 MRVL_PP2_ETH_HDRS_LEN;
504 		if (dev->data->mtu > priv->max_mtu) {
505 			MRVL_LOG(ERR, "inherit MTU %u from max_rx_pkt_len %u is larger than max_mtu %u\n",
506 				 dev->data->mtu,
507 				 dev->data->dev_conf.rxmode.max_rx_pkt_len,
508 				 priv->max_mtu);
509 			return -EINVAL;
510 		}
511 	}
512 
513 	if (dev->data->dev_conf.txmode.offloads & DEV_TX_OFFLOAD_MULTI_SEGS)
514 		priv->multiseg = 1;
515 
516 	ret = mrvl_configure_rxqs(priv, dev->data->port_id,
517 				  dev->data->nb_rx_queues);
518 	if (ret < 0)
519 		return ret;
520 
521 	ret = mrvl_configure_txqs(priv, dev->data->port_id,
522 				  dev->data->nb_tx_queues);
523 	if (ret < 0)
524 		return ret;
525 
526 	priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
527 	priv->ppio_params.maintain_stats = 1;
528 	priv->nb_rx_queues = dev->data->nb_rx_queues;
529 
530 	ret = mrvl_tm_init(dev);
531 	if (ret < 0)
532 		return ret;
533 
534 	if (dev->data->nb_rx_queues == 1 &&
535 	    dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
536 		MRVL_LOG(WARNING, "Disabling hash for 1 rx queue");
537 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
538 
539 		return 0;
540 	}
541 
542 	return mrvl_configure_rss(priv,
543 			&dev->data->dev_conf.rx_adv_conf.rss_conf);
544 }
545 
546 /**
547  * DPDK callback to change the MTU.
548  *
549  * Setting the MTU affects hardware MRU (packets larger than the MRU
550  * will be dropped).
551  *
552  * @param dev
553  *   Pointer to Ethernet device structure.
554  * @param mtu
555  *   New MTU.
556  *
557  * @return
558  *   0 on success, negative error value otherwise.
559  */
560 static int
561 mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
562 {
563 	struct mrvl_priv *priv = dev->data->dev_private;
564 	uint16_t mru;
565 	uint16_t mbuf_data_size = 0; /* SW buffer size */
566 	int ret;
567 
568 	mru = MRVL_PP2_MTU_TO_MRU(mtu);
569 	/*
570 	 * min_rx_buf_size is equal to mbuf data size
571 	 * if pmd didn't set it differently
572 	 */
573 	mbuf_data_size = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
574 	/* Prevent PMD from:
575 	 * - setting mru greater than the mbuf size resulting in
576 	 * hw and sw buffer size mismatch
577 	 * - setting mtu that requires the support of scattered packets
578 	 * when this feature has not been enabled/supported so far
579 	 * (TODO check scattered_rx flag here once scattered RX is supported).
580 	 */
581 	if (mru - RTE_ETHER_CRC_LEN + MRVL_PKT_OFFS > mbuf_data_size) {
582 		mru = mbuf_data_size + RTE_ETHER_CRC_LEN - MRVL_PKT_OFFS;
583 		mtu = MRVL_PP2_MRU_TO_MTU(mru);
584 		MRVL_LOG(WARNING, "MTU too big, max MTU possible limitted "
585 			"by current mbuf size: %u. Set MTU to %u, MRU to %u",
586 			mbuf_data_size, mtu, mru);
587 	}
588 
589 	if (mtu < RTE_ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX) {
590 		MRVL_LOG(ERR, "Invalid MTU [%u] or MRU [%u]", mtu, mru);
591 		return -EINVAL;
592 	}
593 
594 	dev->data->mtu = mtu;
595 	dev->data->dev_conf.rxmode.max_rx_pkt_len = mru - MV_MH_SIZE;
596 
597 	if (!priv->ppio)
598 		return 0;
599 
600 	ret = pp2_ppio_set_mru(priv->ppio, mru);
601 	if (ret) {
602 		MRVL_LOG(ERR, "Failed to change MRU");
603 		return ret;
604 	}
605 
606 	ret = pp2_ppio_set_mtu(priv->ppio, mtu);
607 	if (ret) {
608 		MRVL_LOG(ERR, "Failed to change MTU");
609 		return ret;
610 	}
611 
612 	return 0;
613 }
614 
615 /**
616  * DPDK callback to bring the link up.
617  *
618  * @param dev
619  *   Pointer to Ethernet device structure.
620  *
621  * @return
622  *   0 on success, negative error value otherwise.
623  */
624 static int
625 mrvl_dev_set_link_up(struct rte_eth_dev *dev)
626 {
627 	struct mrvl_priv *priv = dev->data->dev_private;
628 	int ret;
629 
630 	if (!priv->ppio) {
631 		dev->data->dev_link.link_status = ETH_LINK_UP;
632 		return 0;
633 	}
634 
635 	ret = pp2_ppio_enable(priv->ppio);
636 	if (ret)
637 		return ret;
638 
639 	/*
640 	 * mtu/mru can be updated if pp2_ppio_enable() was called at least once
641 	 * as pp2_ppio_enable() changes port->t_mode from default 0 to
642 	 * PP2_TRAFFIC_INGRESS_EGRESS.
643 	 *
644 	 * Set mtu to default DPDK value here.
645 	 */
646 	ret = mrvl_mtu_set(dev, dev->data->mtu);
647 	if (ret) {
648 		pp2_ppio_disable(priv->ppio);
649 		return ret;
650 	}
651 
652 	dev->data->dev_link.link_status = ETH_LINK_UP;
653 	return 0;
654 }
655 
656 /**
657  * DPDK callback to bring the link down.
658  *
659  * @param dev
660  *   Pointer to Ethernet device structure.
661  *
662  * @return
663  *   0 on success, negative error value otherwise.
664  */
665 static int
666 mrvl_dev_set_link_down(struct rte_eth_dev *dev)
667 {
668 	struct mrvl_priv *priv = dev->data->dev_private;
669 	int ret;
670 
671 	if (!priv->ppio) {
672 		dev->data->dev_link.link_status = ETH_LINK_DOWN;
673 		return 0;
674 	}
675 	ret = pp2_ppio_disable(priv->ppio);
676 	if (ret)
677 		return ret;
678 
679 	dev->data->dev_link.link_status = ETH_LINK_DOWN;
680 	return 0;
681 }
682 
683 /**
684  * DPDK callback to start tx queue.
685  *
686  * @param dev
687  *   Pointer to Ethernet device structure.
688  * @param queue_id
689  *   Transmit queue index.
690  *
691  * @return
692  *   0 on success, negative error value otherwise.
693  */
694 static int
695 mrvl_tx_queue_start(struct rte_eth_dev *dev, uint16_t queue_id)
696 {
697 	struct mrvl_priv *priv = dev->data->dev_private;
698 	int ret;
699 
700 	if (!priv)
701 		return -EPERM;
702 
703 	/* passing 1 enables given tx queue */
704 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 1);
705 	if (ret) {
706 		MRVL_LOG(ERR, "Failed to start txq %d", queue_id);
707 		return ret;
708 	}
709 
710 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
711 
712 	return 0;
713 }
714 
715 /**
716  * DPDK callback to stop tx queue.
717  *
718  * @param dev
719  *   Pointer to Ethernet device structure.
720  * @param queue_id
721  *   Transmit queue index.
722  *
723  * @return
724  *   0 on success, negative error value otherwise.
725  */
726 static int
727 mrvl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t queue_id)
728 {
729 	struct mrvl_priv *priv = dev->data->dev_private;
730 	int ret;
731 
732 	if (!priv->ppio)
733 		return -EPERM;
734 
735 	/* passing 0 disables given tx queue */
736 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 0);
737 	if (ret) {
738 		MRVL_LOG(ERR, "Failed to stop txq %d", queue_id);
739 		return ret;
740 	}
741 
742 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
743 
744 	return 0;
745 }
746 
747 /**
748  * Populate VLAN Filter configuration.
749  *
750  * @param dev
751  *   Pointer to Ethernet device structure.
752  * @param on
753  *   Toggle filter.
754  *
755  * @return
756  *   0 on success, negative error value otherwise.
757  */
758 static int mrvl_populate_vlan_table(struct rte_eth_dev *dev, int on)
759 {
760 	uint32_t j;
761 	int ret;
762 	struct rte_vlan_filter_conf *vfc;
763 
764 	vfc = &dev->data->vlan_filter_conf;
765 	for (j = 0; j < RTE_DIM(vfc->ids); j++) {
766 		uint64_t vlan;
767 		uint64_t vbit;
768 		uint64_t ids = vfc->ids[j];
769 
770 		if (ids == 0)
771 			continue;
772 
773 		while (ids) {
774 			vlan = 64 * j;
775 			/* count trailing zeroes */
776 			vbit = ~ids & (ids - 1);
777 			/* clear least significant bit set */
778 			ids ^= (ids ^ (ids - 1)) ^ vbit;
779 			for (; vbit; vlan++)
780 				vbit >>= 1;
781 			ret = mrvl_vlan_filter_set(dev, vlan, on);
782 			if (ret) {
783 				MRVL_LOG(ERR, "Failed to setup VLAN filter\n");
784 				return ret;
785 			}
786 		}
787 	}
788 
789 	return 0;
790 }
791 
792 /**
793  * DPDK callback to start the device.
794  *
795  * @param dev
796  *   Pointer to Ethernet device structure.
797  *
798  * @return
799  *   0 on success, negative errno value on failure.
800  */
801 static int
802 mrvl_dev_start(struct rte_eth_dev *dev)
803 {
804 	struct mrvl_priv *priv = dev->data->dev_private;
805 	char match[MRVL_MATCH_LEN];
806 	int ret = 0, i, def_init_size;
807 	struct rte_ether_addr *mac_addr;
808 
809 	if (priv->ppio)
810 		return mrvl_dev_set_link_up(dev);
811 
812 	snprintf(match, sizeof(match), "ppio-%d:%d",
813 		 priv->pp_id, priv->ppio_id);
814 	priv->ppio_params.match = match;
815 	priv->ppio_params.eth_start_hdr = PP2_PPIO_HDR_ETH;
816 	priv->forward_bad_frames = 0;
817 
818 	if (mrvl_cfg) {
819 		priv->ppio_params.eth_start_hdr =
820 			mrvl_cfg->port[dev->data->port_id].eth_start_hdr;
821 		priv->forward_bad_frames =
822 			mrvl_cfg->port[dev->data->port_id].forward_bad_frames;
823 	}
824 
825 	/*
826 	 * Calculate the minimum bpool size for refill feature as follows:
827 	 * 2 default burst sizes multiply by number of rx queues.
828 	 * If the bpool size will be below this value, new buffers will
829 	 * be added to the pool.
830 	 */
831 	priv->bpool_min_size = priv->nb_rx_queues * MRVL_BURST_SIZE * 2;
832 
833 	/* In case initial bpool size configured in queues setup is
834 	 * smaller than minimum size add more buffers
835 	 */
836 	def_init_size = priv->bpool_min_size + MRVL_BURST_SIZE * 2;
837 	if (priv->bpool_init_size < def_init_size) {
838 		int buffs_to_add = def_init_size - priv->bpool_init_size;
839 
840 		priv->bpool_init_size += buffs_to_add;
841 		ret = mrvl_fill_bpool(dev->data->rx_queues[0], buffs_to_add);
842 		if (ret)
843 			MRVL_LOG(ERR, "Failed to add buffers to bpool");
844 	}
845 
846 	/*
847 	 * Calculate the maximum bpool size for refill feature as follows:
848 	 * maximum number of descriptors in rx queue multiply by number
849 	 * of rx queues plus minimum bpool size.
850 	 * In case the bpool size will exceed this value, superfluous buffers
851 	 * will be removed
852 	 */
853 	priv->bpool_max_size = (priv->nb_rx_queues * MRVL_PP2_RXD_MAX) +
854 				priv->bpool_min_size;
855 
856 	ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
857 	if (ret) {
858 		MRVL_LOG(ERR, "Failed to init ppio");
859 		return ret;
860 	}
861 
862 	/*
863 	 * In case there are some some stale uc/mc mac addresses flush them
864 	 * here. It cannot be done during mrvl_dev_close() as port information
865 	 * is already gone at that point (due to pp2_ppio_deinit() in
866 	 * mrvl_dev_stop()).
867 	 */
868 	if (!priv->uc_mc_flushed) {
869 		ret = pp2_ppio_flush_mac_addrs(priv->ppio, 1, 1);
870 		if (ret) {
871 			MRVL_LOG(ERR,
872 				"Failed to flush uc/mc filter list");
873 			goto out;
874 		}
875 		priv->uc_mc_flushed = 1;
876 	}
877 
878 	ret = mrvl_mtu_set(dev, dev->data->mtu);
879 	if (ret)
880 		MRVL_LOG(ERR, "Failed to set MTU to %d", dev->data->mtu);
881 
882 	if (!rte_is_zero_ether_addr(&dev->data->mac_addrs[0]))
883 		mrvl_mac_addr_set(dev, &dev->data->mac_addrs[0]);
884 
885 	for (i = 1; i < MRVL_MAC_ADDRS_MAX; i++) {
886 		mac_addr = &dev->data->mac_addrs[i];
887 
888 		/* skip zero address */
889 		if (rte_is_zero_ether_addr(mac_addr))
890 			continue;
891 
892 		mrvl_mac_addr_add(dev, mac_addr, i, 0);
893 	}
894 
895 	if (dev->data->all_multicast == 1)
896 		mrvl_allmulticast_enable(dev);
897 
898 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_VLAN_FILTER) {
899 		ret = mrvl_populate_vlan_table(dev, 1);
900 		if (ret) {
901 			MRVL_LOG(ERR, "Failed to populate VLAN table");
902 			goto out;
903 		}
904 	}
905 
906 	/* For default QoS config, don't start classifier. */
907 	if (mrvl_cfg  &&
908 	    mrvl_cfg->port[dev->data->port_id].use_qos_global_defaults == 0) {
909 		ret = mrvl_start_qos_mapping(priv);
910 		if (ret) {
911 			MRVL_LOG(ERR, "Failed to setup QoS mapping");
912 			goto out;
913 		}
914 	}
915 
916 	ret = pp2_ppio_set_loopback(priv->ppio, dev->data->dev_conf.lpbk_mode);
917 	if (ret) {
918 		MRVL_LOG(ERR, "Failed to set loopback");
919 		goto out;
920 	}
921 
922 	if (dev->data->promiscuous == 1)
923 		mrvl_promiscuous_enable(dev);
924 
925 	if (priv->flow_ctrl) {
926 		ret = mrvl_flow_ctrl_set(dev, &priv->fc_conf);
927 		if (ret) {
928 			MRVL_LOG(ERR, "Failed to configure flow control");
929 			goto out;
930 		}
931 		priv->flow_ctrl = 0;
932 	}
933 
934 	if (dev->data->dev_link.link_status == ETH_LINK_UP) {
935 		ret = mrvl_dev_set_link_up(dev);
936 		if (ret) {
937 			MRVL_LOG(ERR, "Failed to set link up");
938 			dev->data->dev_link.link_status = ETH_LINK_DOWN;
939 			goto out;
940 		}
941 	}
942 
943 	/* start tx queues */
944 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
945 		struct mrvl_txq *txq = dev->data->tx_queues[i];
946 
947 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
948 
949 		if (!txq->tx_deferred_start)
950 			continue;
951 
952 		/*
953 		 * All txqs are started by default. Stop them
954 		 * so that tx_deferred_start works as expected.
955 		 */
956 		ret = mrvl_tx_queue_stop(dev, i);
957 		if (ret)
958 			goto out;
959 	}
960 
961 	mrvl_flow_init(dev);
962 	mrvl_mtr_init(dev);
963 	mrvl_set_tx_function(dev);
964 
965 	return 0;
966 out:
967 	MRVL_LOG(ERR, "Failed to start device");
968 	pp2_ppio_deinit(priv->ppio);
969 	return ret;
970 }
971 
972 /**
973  * Flush receive queues.
974  *
975  * @param dev
976  *   Pointer to Ethernet device structure.
977  */
978 static void
979 mrvl_flush_rx_queues(struct rte_eth_dev *dev)
980 {
981 	int i;
982 
983 	MRVL_LOG(INFO, "Flushing rx queues");
984 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
985 		int ret, num;
986 
987 		do {
988 			struct mrvl_rxq *q = dev->data->rx_queues[i];
989 			struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];
990 
991 			num = MRVL_PP2_RXD_MAX;
992 			ret = pp2_ppio_recv(q->priv->ppio,
993 					    q->priv->rxq_map[q->queue_id].tc,
994 					    q->priv->rxq_map[q->queue_id].inq,
995 					    descs, (uint16_t *)&num);
996 		} while (ret == 0 && num);
997 	}
998 }
999 
1000 /**
1001  * Flush transmit shadow queues.
1002  *
1003  * @param dev
1004  *   Pointer to Ethernet device structure.
1005  */
1006 static void
1007 mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
1008 {
1009 	int i, j;
1010 	struct mrvl_txq *txq;
1011 
1012 	MRVL_LOG(INFO, "Flushing tx shadow queues");
1013 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1014 		txq = (struct mrvl_txq *)dev->data->tx_queues[i];
1015 
1016 		for (j = 0; j < RTE_MAX_LCORE; j++) {
1017 			struct mrvl_shadow_txq *sq;
1018 
1019 			if (!hifs[j])
1020 				continue;
1021 
1022 			sq = &txq->shadow_txqs[j];
1023 			mrvl_free_sent_buffers(txq->priv->ppio,
1024 				hifs[j], j, sq, txq->queue_id, 1);
1025 			while (sq->tail != sq->head) {
1026 				uint64_t addr = cookie_addr_high |
1027 					sq->ent[sq->tail].buff.cookie;
1028 				rte_pktmbuf_free(
1029 					(struct rte_mbuf *)addr);
1030 				sq->tail = (sq->tail + 1) &
1031 					    MRVL_PP2_TX_SHADOWQ_MASK;
1032 			}
1033 			memset(sq, 0, sizeof(*sq));
1034 		}
1035 	}
1036 }
1037 
1038 /**
1039  * Flush hardware bpool (buffer-pool).
1040  *
1041  * @param dev
1042  *   Pointer to Ethernet device structure.
1043  */
1044 static void
1045 mrvl_flush_bpool(struct rte_eth_dev *dev)
1046 {
1047 	struct mrvl_priv *priv = dev->data->dev_private;
1048 	struct pp2_hif *hif;
1049 	uint32_t num;
1050 	int ret;
1051 	unsigned int core_id = rte_lcore_id();
1052 
1053 	if (core_id == LCORE_ID_ANY)
1054 		core_id = rte_get_main_lcore();
1055 
1056 	hif = mrvl_get_hif(priv, core_id);
1057 
1058 	ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
1059 	if (ret) {
1060 		MRVL_LOG(ERR, "Failed to get bpool buffers number");
1061 		return;
1062 	}
1063 
1064 	while (num--) {
1065 		struct pp2_buff_inf inf;
1066 		uint64_t addr;
1067 
1068 		ret = pp2_bpool_get_buff(hif, priv->bpool, &inf);
1069 		if (ret)
1070 			break;
1071 
1072 		addr = cookie_addr_high | inf.cookie;
1073 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1074 	}
1075 }
1076 
1077 /**
1078  * DPDK callback to stop the device.
1079  *
1080  * @param dev
1081  *   Pointer to Ethernet device structure.
1082  */
1083 static int
1084 mrvl_dev_stop(struct rte_eth_dev *dev)
1085 {
1086 	return mrvl_dev_set_link_down(dev);
1087 }
1088 
1089 /**
1090  * DPDK callback to close the device.
1091  *
1092  * @param dev
1093  *   Pointer to Ethernet device structure.
1094  */
1095 static int
1096 mrvl_dev_close(struct rte_eth_dev *dev)
1097 {
1098 	struct mrvl_priv *priv = dev->data->dev_private;
1099 	size_t i;
1100 
1101 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1102 		return 0;
1103 
1104 	mrvl_flush_rx_queues(dev);
1105 	mrvl_flush_tx_shadow_queues(dev);
1106 	mrvl_flow_deinit(dev);
1107 	mrvl_mtr_deinit(dev);
1108 
1109 	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
1110 		struct pp2_ppio_tc_params *tc_params =
1111 			&priv->ppio_params.inqs_params.tcs_params[i];
1112 
1113 		if (tc_params->inqs_params) {
1114 			rte_free(tc_params->inqs_params);
1115 			tc_params->inqs_params = NULL;
1116 		}
1117 	}
1118 
1119 	if (priv->cls_tbl) {
1120 		pp2_cls_tbl_deinit(priv->cls_tbl);
1121 		priv->cls_tbl = NULL;
1122 	}
1123 
1124 	if (priv->qos_tbl) {
1125 		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
1126 		priv->qos_tbl = NULL;
1127 	}
1128 
1129 	mrvl_flush_bpool(dev);
1130 	mrvl_tm_deinit(dev);
1131 
1132 	if (priv->ppio) {
1133 		pp2_ppio_deinit(priv->ppio);
1134 		priv->ppio = NULL;
1135 	}
1136 
1137 	/* policer must be released after ppio deinitialization */
1138 	if (priv->default_policer) {
1139 		pp2_cls_plcr_deinit(priv->default_policer);
1140 		priv->default_policer = NULL;
1141 	}
1142 
1143 
1144 	if (priv->bpool) {
1145 		pp2_bpool_deinit(priv->bpool);
1146 		used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
1147 		priv->bpool = NULL;
1148 	}
1149 
1150 	mrvl_dev_num--;
1151 
1152 	if (mrvl_dev_num == 0) {
1153 		MRVL_LOG(INFO, "Perform MUSDK deinit");
1154 		mrvl_deinit_hifs();
1155 		mrvl_deinit_pp2();
1156 		rte_mvep_deinit(MVEP_MOD_T_PP2);
1157 	}
1158 
1159 	return 0;
1160 }
1161 
1162 /**
1163  * DPDK callback to retrieve physical link information.
1164  *
1165  * @param dev
1166  *   Pointer to Ethernet device structure.
1167  * @param wait_to_complete
1168  *   Wait for request completion (ignored).
1169  *
1170  * @return
1171  *   0 on success, negative error value otherwise.
1172  */
1173 static int
1174 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1175 {
1176 	/*
1177 	 * TODO
1178 	 * once MUSDK provides necessary API use it here
1179 	 */
1180 	struct mrvl_priv *priv = dev->data->dev_private;
1181 	struct ethtool_cmd edata;
1182 	struct ifreq req;
1183 	int ret, fd, link_up;
1184 
1185 	if (!priv->ppio)
1186 		return -EPERM;
1187 
1188 	edata.cmd = ETHTOOL_GSET;
1189 
1190 	strcpy(req.ifr_name, dev->data->name);
1191 	req.ifr_data = (void *)&edata;
1192 
1193 	fd = socket(AF_INET, SOCK_DGRAM, 0);
1194 	if (fd == -1)
1195 		return -EFAULT;
1196 
1197 	ret = ioctl(fd, SIOCETHTOOL, &req);
1198 	if (ret == -1) {
1199 		close(fd);
1200 		return -EFAULT;
1201 	}
1202 
1203 	close(fd);
1204 
1205 	switch (ethtool_cmd_speed(&edata)) {
1206 	case SPEED_10:
1207 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
1208 		break;
1209 	case SPEED_100:
1210 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
1211 		break;
1212 	case SPEED_1000:
1213 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
1214 		break;
1215 	case SPEED_2500:
1216 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_2_5G;
1217 		break;
1218 	case SPEED_10000:
1219 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
1220 		break;
1221 	default:
1222 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
1223 	}
1224 
1225 	dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
1226 							 ETH_LINK_HALF_DUPLEX;
1227 	dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
1228 							   ETH_LINK_FIXED;
1229 	pp2_ppio_get_link_state(priv->ppio, &link_up);
1230 	dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
1231 
1232 	return 0;
1233 }
1234 
1235 /**
1236  * DPDK callback to enable promiscuous mode.
1237  *
1238  * @param dev
1239  *   Pointer to Ethernet device structure.
1240  *
1241  * @return
1242  *   0 on success, negative error value otherwise.
1243  */
1244 static int
1245 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
1246 {
1247 	struct mrvl_priv *priv = dev->data->dev_private;
1248 	int ret;
1249 
1250 	if (priv->isolated)
1251 		return -ENOTSUP;
1252 
1253 	if (!priv->ppio)
1254 		return 0;
1255 
1256 	ret = pp2_ppio_set_promisc(priv->ppio, 1);
1257 	if (ret) {
1258 		MRVL_LOG(ERR, "Failed to enable promiscuous mode");
1259 		return -EAGAIN;
1260 	}
1261 
1262 	return 0;
1263 }
1264 
1265 /**
1266  * DPDK callback to enable allmulti mode.
1267  *
1268  * @param dev
1269  *   Pointer to Ethernet device structure.
1270  *
1271  * @return
1272  *   0 on success, negative error value otherwise.
1273  */
1274 static int
1275 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
1276 {
1277 	struct mrvl_priv *priv = dev->data->dev_private;
1278 	int ret;
1279 
1280 	if (priv->isolated)
1281 		return -ENOTSUP;
1282 
1283 	if (!priv->ppio)
1284 		return 0;
1285 
1286 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
1287 	if (ret) {
1288 		MRVL_LOG(ERR, "Failed enable all-multicast mode");
1289 		return -EAGAIN;
1290 	}
1291 
1292 	return 0;
1293 }
1294 
1295 /**
1296  * DPDK callback to disable promiscuous mode.
1297  *
1298  * @param dev
1299  *   Pointer to Ethernet device structure.
1300  *
1301  * @return
1302  *   0 on success, negative error value otherwise.
1303  */
1304 static int
1305 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
1306 {
1307 	struct mrvl_priv *priv = dev->data->dev_private;
1308 	int ret;
1309 
1310 	if (priv->isolated)
1311 		return -ENOTSUP;
1312 
1313 	if (!priv->ppio)
1314 		return 0;
1315 
1316 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
1317 	if (ret) {
1318 		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
1319 		return -EAGAIN;
1320 	}
1321 
1322 	return 0;
1323 }
1324 
1325 /**
1326  * DPDK callback to disable allmulticast mode.
1327  *
1328  * @param dev
1329  *   Pointer to Ethernet device structure.
1330  *
1331  * @return
1332  *   0 on success, negative error value otherwise.
1333  */
1334 static int
1335 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
1336 {
1337 	struct mrvl_priv *priv = dev->data->dev_private;
1338 	int ret;
1339 
1340 	if (priv->isolated)
1341 		return -ENOTSUP;
1342 
1343 	if (!priv->ppio)
1344 		return 0;
1345 
1346 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
1347 	if (ret) {
1348 		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
1349 		return -EAGAIN;
1350 	}
1351 
1352 	return 0;
1353 }
1354 
1355 /**
1356  * DPDK callback to remove a MAC address.
1357  *
1358  * @param dev
1359  *   Pointer to Ethernet device structure.
1360  * @param index
1361  *   MAC address index.
1362  */
1363 static void
1364 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1365 {
1366 	struct mrvl_priv *priv = dev->data->dev_private;
1367 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1368 	int ret;
1369 
1370 	if (priv->isolated)
1371 		return;
1372 
1373 	if (!priv->ppio)
1374 		return;
1375 
1376 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
1377 				       dev->data->mac_addrs[index].addr_bytes);
1378 	if (ret) {
1379 		rte_ether_format_addr(buf, sizeof(buf),
1380 				  &dev->data->mac_addrs[index]);
1381 		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
1382 	}
1383 }
1384 
1385 /**
1386  * DPDK callback to add a MAC address.
1387  *
1388  * @param dev
1389  *   Pointer to Ethernet device structure.
1390  * @param mac_addr
1391  *   MAC address to register.
1392  * @param index
1393  *   MAC address index.
1394  * @param vmdq
1395  *   VMDq pool index to associate address with (unused).
1396  *
1397  * @return
1398  *   0 on success, negative error value otherwise.
1399  */
1400 static int
1401 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1402 		  uint32_t index, uint32_t vmdq __rte_unused)
1403 {
1404 	struct mrvl_priv *priv = dev->data->dev_private;
1405 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1406 	int ret;
1407 
1408 	if (priv->isolated)
1409 		return -ENOTSUP;
1410 
1411 	if (!priv->ppio)
1412 		return 0;
1413 
1414 	if (index == 0)
1415 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1416 		return -1;
1417 
1418 	/*
1419 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1420 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1421 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1422 	 * 21 respectively.
1423 	 *
1424 	 * If more than uc_filter_max uc addresses were added to filter list
1425 	 * then NIC will switch to promiscuous mode automatically.
1426 	 *
1427 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1428 	 * were added to filter list then NIC will switch to all-multicast mode
1429 	 * automatically.
1430 	 */
1431 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1432 	if (ret) {
1433 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1434 		MRVL_LOG(ERR, "Failed to add mac %s", buf);
1435 		return -1;
1436 	}
1437 
1438 	return 0;
1439 }
1440 
1441 /**
1442  * DPDK callback to set the primary MAC address.
1443  *
1444  * @param dev
1445  *   Pointer to Ethernet device structure.
1446  * @param mac_addr
1447  *   MAC address to register.
1448  *
1449  * @return
1450  *   0 on success, negative error value otherwise.
1451  */
1452 static int
1453 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1454 {
1455 	struct mrvl_priv *priv = dev->data->dev_private;
1456 	int ret;
1457 
1458 	if (priv->isolated)
1459 		return -ENOTSUP;
1460 
1461 	if (!priv->ppio)
1462 		return 0;
1463 
1464 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1465 	if (ret) {
1466 		char buf[RTE_ETHER_ADDR_FMT_SIZE];
1467 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1468 		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1469 	}
1470 
1471 	return ret;
1472 }
1473 
1474 /**
1475  * DPDK callback to get device statistics.
1476  *
1477  * @param dev
1478  *   Pointer to Ethernet device structure.
1479  * @param stats
1480  *   Stats structure output buffer.
1481  *
1482  * @return
1483  *   0 on success, negative error value otherwise.
1484  */
1485 static int
1486 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1487 {
1488 	struct mrvl_priv *priv = dev->data->dev_private;
1489 	struct pp2_ppio_statistics ppio_stats;
1490 	uint64_t drop_mac = 0;
1491 	unsigned int i, idx, ret;
1492 
1493 	if (!priv->ppio)
1494 		return -EPERM;
1495 
1496 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1497 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1498 		struct pp2_ppio_inq_statistics rx_stats;
1499 
1500 		if (!rxq)
1501 			continue;
1502 
1503 		idx = rxq->queue_id;
1504 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1505 			MRVL_LOG(ERR,
1506 				"rx queue %d stats out of range (0 - %d)",
1507 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1508 			continue;
1509 		}
1510 
1511 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1512 						  priv->rxq_map[idx].tc,
1513 						  priv->rxq_map[idx].inq,
1514 						  &rx_stats, 0);
1515 		if (unlikely(ret)) {
1516 			MRVL_LOG(ERR,
1517 				"Failed to update rx queue %d stats", idx);
1518 			break;
1519 		}
1520 
1521 		stats->q_ibytes[idx] = rxq->bytes_recv;
1522 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1523 		stats->q_errors[idx] = rx_stats.drop_early +
1524 				       rx_stats.drop_fullq +
1525 				       rx_stats.drop_bm +
1526 				       rxq->drop_mac;
1527 		stats->ibytes += rxq->bytes_recv;
1528 		drop_mac += rxq->drop_mac;
1529 	}
1530 
1531 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1532 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1533 		struct pp2_ppio_outq_statistics tx_stats;
1534 
1535 		if (!txq)
1536 			continue;
1537 
1538 		idx = txq->queue_id;
1539 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1540 			MRVL_LOG(ERR,
1541 				"tx queue %d stats out of range (0 - %d)",
1542 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1543 		}
1544 
1545 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1546 						   &tx_stats, 0);
1547 		if (unlikely(ret)) {
1548 			MRVL_LOG(ERR,
1549 				"Failed to update tx queue %d stats", idx);
1550 			break;
1551 		}
1552 
1553 		stats->q_opackets[idx] = tx_stats.deq_desc;
1554 		stats->q_obytes[idx] = txq->bytes_sent;
1555 		stats->obytes += txq->bytes_sent;
1556 	}
1557 
1558 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1559 	if (unlikely(ret)) {
1560 		MRVL_LOG(ERR, "Failed to update port statistics");
1561 		return ret;
1562 	}
1563 
1564 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1565 	stats->opackets += ppio_stats.tx_packets;
1566 	stats->imissed += ppio_stats.rx_fullq_dropped +
1567 			  ppio_stats.rx_bm_dropped +
1568 			  ppio_stats.rx_early_dropped +
1569 			  ppio_stats.rx_fifo_dropped +
1570 			  ppio_stats.rx_cls_dropped;
1571 	stats->ierrors = drop_mac;
1572 
1573 	return 0;
1574 }
1575 
1576 /**
1577  * DPDK callback to clear device statistics.
1578  *
1579  * @param dev
1580  *   Pointer to Ethernet device structure.
1581  *
1582  * @return
1583  *   0 on success, negative error value otherwise.
1584  */
1585 static int
1586 mrvl_stats_reset(struct rte_eth_dev *dev)
1587 {
1588 	struct mrvl_priv *priv = dev->data->dev_private;
1589 	int i;
1590 
1591 	if (!priv->ppio)
1592 		return 0;
1593 
1594 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1595 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1596 
1597 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1598 					    priv->rxq_map[i].inq, NULL, 1);
1599 		rxq->bytes_recv = 0;
1600 		rxq->drop_mac = 0;
1601 	}
1602 
1603 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1604 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1605 
1606 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1607 		txq->bytes_sent = 0;
1608 	}
1609 
1610 	return pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1611 }
1612 
1613 /**
1614  * DPDK callback to get extended statistics.
1615  *
1616  * @param dev
1617  *   Pointer to Ethernet device structure.
1618  * @param stats
1619  *   Pointer to xstats table.
1620  * @param n
1621  *   Number of entries in xstats table.
1622  * @return
1623  *   Negative value on error, number of read xstats otherwise.
1624  */
1625 static int
1626 mrvl_xstats_get(struct rte_eth_dev *dev,
1627 		struct rte_eth_xstat *stats, unsigned int n)
1628 {
1629 	struct mrvl_priv *priv = dev->data->dev_private;
1630 	struct pp2_ppio_statistics ppio_stats;
1631 	unsigned int i;
1632 
1633 	if (!stats)
1634 		return 0;
1635 
1636 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1637 	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1638 		uint64_t val;
1639 
1640 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1641 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1642 					    mrvl_xstats_tbl[i].offset);
1643 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1644 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1645 					    mrvl_xstats_tbl[i].offset);
1646 		else
1647 			return -EINVAL;
1648 
1649 		stats[i].id = i;
1650 		stats[i].value = val;
1651 	}
1652 
1653 	return n;
1654 }
1655 
1656 /**
1657  * DPDK callback to reset extended statistics.
1658  *
1659  * @param dev
1660  *   Pointer to Ethernet device structure.
1661  *
1662  * @return
1663  *   0 on success, negative error value otherwise.
1664  */
1665 static int
1666 mrvl_xstats_reset(struct rte_eth_dev *dev)
1667 {
1668 	return mrvl_stats_reset(dev);
1669 }
1670 
1671 /**
1672  * DPDK callback to get extended statistics names.
1673  *
1674  * @param dev (unused)
1675  *   Pointer to Ethernet device structure.
1676  * @param xstats_names
1677  *   Pointer to xstats names table.
1678  * @param size
1679  *   Size of the xstats names table.
1680  * @return
1681  *   Number of read names.
1682  */
1683 static int
1684 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1685 		      struct rte_eth_xstat_name *xstats_names,
1686 		      unsigned int size)
1687 {
1688 	unsigned int i;
1689 
1690 	if (!xstats_names)
1691 		return RTE_DIM(mrvl_xstats_tbl);
1692 
1693 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1694 		strlcpy(xstats_names[i].name, mrvl_xstats_tbl[i].name,
1695 			RTE_ETH_XSTATS_NAME_SIZE);
1696 
1697 	return size;
1698 }
1699 
1700 /**
1701  * DPDK callback to get information about the device.
1702  *
1703  * @param dev
1704  *   Pointer to Ethernet device structure (unused).
1705  * @param info
1706  *   Info structure output buffer.
1707  */
1708 static int
1709 mrvl_dev_infos_get(struct rte_eth_dev *dev,
1710 		   struct rte_eth_dev_info *info)
1711 {
1712 	struct mrvl_priv *priv = dev->data->dev_private;
1713 
1714 	info->speed_capa = ETH_LINK_SPEED_10M |
1715 			   ETH_LINK_SPEED_100M |
1716 			   ETH_LINK_SPEED_1G |
1717 			   ETH_LINK_SPEED_2_5G |
1718 			   ETH_LINK_SPEED_10G;
1719 
1720 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1721 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1722 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1723 
1724 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1725 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1726 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1727 
1728 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1729 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1730 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1731 
1732 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1733 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1734 
1735 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1736 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1737 
1738 	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1739 				       ETH_RSS_NONFRAG_IPV4_TCP |
1740 				       ETH_RSS_NONFRAG_IPV4_UDP;
1741 
1742 	/* By default packets are dropped if no descriptors are available */
1743 	info->default_rxconf.rx_drop_en = 1;
1744 
1745 	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1746 	info->max_mtu = priv->max_mtu;
1747 
1748 	return 0;
1749 }
1750 
1751 /**
1752  * Return supported packet types.
1753  *
1754  * @param dev
1755  *   Pointer to Ethernet device structure (unused).
1756  *
1757  * @return
1758  *   Const pointer to the table with supported packet types.
1759  */
1760 static const uint32_t *
1761 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1762 {
1763 	static const uint32_t ptypes[] = {
1764 		RTE_PTYPE_L2_ETHER,
1765 		RTE_PTYPE_L2_ETHER_VLAN,
1766 		RTE_PTYPE_L2_ETHER_QINQ,
1767 		RTE_PTYPE_L3_IPV4,
1768 		RTE_PTYPE_L3_IPV4_EXT,
1769 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1770 		RTE_PTYPE_L3_IPV6,
1771 		RTE_PTYPE_L3_IPV6_EXT,
1772 		RTE_PTYPE_L2_ETHER_ARP,
1773 		RTE_PTYPE_L4_TCP,
1774 		RTE_PTYPE_L4_UDP
1775 	};
1776 
1777 	return ptypes;
1778 }
1779 
1780 /**
1781  * DPDK callback to get information about specific receive queue.
1782  *
1783  * @param dev
1784  *   Pointer to Ethernet device structure.
1785  * @param rx_queue_id
1786  *   Receive queue index.
1787  * @param qinfo
1788  *   Receive queue information structure.
1789  */
1790 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1791 			      struct rte_eth_rxq_info *qinfo)
1792 {
1793 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1794 	struct mrvl_priv *priv = dev->data->dev_private;
1795 	int inq = priv->rxq_map[rx_queue_id].inq;
1796 	int tc = priv->rxq_map[rx_queue_id].tc;
1797 	struct pp2_ppio_tc_params *tc_params =
1798 		&priv->ppio_params.inqs_params.tcs_params[tc];
1799 
1800 	qinfo->mp = q->mp;
1801 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1802 }
1803 
1804 /**
1805  * DPDK callback to get information about specific transmit queue.
1806  *
1807  * @param dev
1808  *   Pointer to Ethernet device structure.
1809  * @param tx_queue_id
1810  *   Transmit queue index.
1811  * @param qinfo
1812  *   Transmit queue information structure.
1813  */
1814 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1815 			      struct rte_eth_txq_info *qinfo)
1816 {
1817 	struct mrvl_priv *priv = dev->data->dev_private;
1818 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1819 
1820 	qinfo->nb_desc =
1821 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1822 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1823 }
1824 
1825 /**
1826  * DPDK callback to Configure a VLAN filter.
1827  *
1828  * @param dev
1829  *   Pointer to Ethernet device structure.
1830  * @param vlan_id
1831  *   VLAN ID to filter.
1832  * @param on
1833  *   Toggle filter.
1834  *
1835  * @return
1836  *   0 on success, negative error value otherwise.
1837  */
1838 static int
1839 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1840 {
1841 	struct mrvl_priv *priv = dev->data->dev_private;
1842 
1843 	if (priv->isolated)
1844 		return -ENOTSUP;
1845 
1846 	if (!priv->ppio)
1847 		return 0;
1848 
1849 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1850 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1851 }
1852 
1853 /**
1854  * DPDK callback to Configure VLAN offload.
1855  *
1856  * @param dev
1857  *   Pointer to Ethernet device structure.
1858  * @param mask
1859  *   VLAN offload mask.
1860  *
1861  * @return
1862  *   0 on success, negative error value otherwise.
1863  */
1864 static int mrvl_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1865 {
1866 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
1867 	int ret;
1868 
1869 	if (mask & ETH_VLAN_STRIP_MASK)
1870 		MRVL_LOG(ERR, "VLAN stripping is not supported\n");
1871 
1872 	if (mask & ETH_VLAN_FILTER_MASK) {
1873 		if (rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1874 			ret = mrvl_populate_vlan_table(dev, 1);
1875 		else
1876 			ret = mrvl_populate_vlan_table(dev, 0);
1877 
1878 		if (ret)
1879 			return ret;
1880 	}
1881 
1882 	if (mask & ETH_VLAN_EXTEND_MASK)
1883 		MRVL_LOG(ERR, "Extend VLAN not supported\n");
1884 
1885 	return 0;
1886 }
1887 
1888 /**
1889  * Release buffers to hardware bpool (buffer-pool)
1890  *
1891  * @param rxq
1892  *   Receive queue pointer.
1893  * @param num
1894  *   Number of buffers to release to bpool.
1895  *
1896  * @return
1897  *   0 on success, negative error value otherwise.
1898  */
1899 static int
1900 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1901 {
1902 	struct buff_release_entry entries[num];
1903 	struct rte_mbuf *mbufs[num];
1904 	int i, ret;
1905 	unsigned int core_id;
1906 	struct pp2_hif *hif;
1907 	struct pp2_bpool *bpool;
1908 
1909 	core_id = rte_lcore_id();
1910 	if (core_id == LCORE_ID_ANY)
1911 		core_id = rte_get_main_lcore();
1912 
1913 	hif = mrvl_get_hif(rxq->priv, core_id);
1914 	if (!hif)
1915 		return -1;
1916 
1917 	bpool = rxq->priv->bpool;
1918 
1919 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1920 	if (ret)
1921 		return ret;
1922 
1923 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1924 		cookie_addr_high =
1925 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1926 
1927 	for (i = 0; i < num; i++) {
1928 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1929 			!= cookie_addr_high) {
1930 			MRVL_LOG(ERR,
1931 				"mbuf virtual addr high is out of range "
1932 				"0x%x instead of 0x%x\n",
1933 				(uint32_t)((uint64_t)mbufs[i] >> 32),
1934 				(uint32_t)(cookie_addr_high >> 32));
1935 			goto out;
1936 		}
1937 
1938 		entries[i].buff.addr =
1939 			rte_mbuf_data_iova_default(mbufs[i]);
1940 		entries[i].buff.cookie = (uintptr_t)mbufs[i];
1941 		entries[i].bpool = bpool;
1942 	}
1943 
1944 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1945 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1946 
1947 	if (i != num)
1948 		goto out;
1949 
1950 	return 0;
1951 out:
1952 	for (; i < num; i++)
1953 		rte_pktmbuf_free(mbufs[i]);
1954 
1955 	return -1;
1956 }
1957 
1958 /**
1959  * DPDK callback to configure the receive queue.
1960  *
1961  * @param dev
1962  *   Pointer to Ethernet device structure.
1963  * @param idx
1964  *   RX queue index.
1965  * @param desc
1966  *   Number of descriptors to configure in queue.
1967  * @param socket
1968  *   NUMA socket on which memory must be allocated.
1969  * @param conf
1970  *   Thresholds parameters.
1971  * @param mp
1972  *   Memory pool for buffer allocations.
1973  *
1974  * @return
1975  *   0 on success, negative error value otherwise.
1976  */
1977 static int
1978 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1979 		    unsigned int socket,
1980 		    const struct rte_eth_rxconf *conf,
1981 		    struct rte_mempool *mp)
1982 {
1983 	struct mrvl_priv *priv = dev->data->dev_private;
1984 	struct mrvl_rxq *rxq;
1985 	uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
1986 	uint32_t max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1987 	int ret, tc, inq;
1988 	uint64_t offloads;
1989 
1990 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
1991 
1992 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1993 		/*
1994 		 * Unknown TC mapping, mapping will not have a correct queue.
1995 		 */
1996 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
1997 			idx, priv->ppio_id);
1998 		return -EFAULT;
1999 	}
2000 
2001 	frame_size = buf_size - RTE_PKTMBUF_HEADROOM -
2002 		     MRVL_PKT_EFFEC_OFFS + RTE_ETHER_CRC_LEN;
2003 	if (frame_size < max_rx_pkt_len) {
2004 		MRVL_LOG(WARNING,
2005 			"Mbuf size must be increased to %u bytes to hold up "
2006 			"to %u bytes of data.",
2007 			buf_size + max_rx_pkt_len - frame_size,
2008 			max_rx_pkt_len);
2009 		dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
2010 		MRVL_LOG(INFO, "Setting max rx pkt len to %u",
2011 			dev->data->dev_conf.rxmode.max_rx_pkt_len);
2012 	}
2013 
2014 	if (dev->data->rx_queues[idx]) {
2015 		rte_free(dev->data->rx_queues[idx]);
2016 		dev->data->rx_queues[idx] = NULL;
2017 	}
2018 
2019 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
2020 	if (!rxq)
2021 		return -ENOMEM;
2022 
2023 	rxq->priv = priv;
2024 	rxq->mp = mp;
2025 	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
2026 	rxq->queue_id = idx;
2027 	rxq->port_id = dev->data->port_id;
2028 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
2029 
2030 	tc = priv->rxq_map[rxq->queue_id].tc,
2031 	inq = priv->rxq_map[rxq->queue_id].inq;
2032 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
2033 		desc;
2034 
2035 	ret = mrvl_fill_bpool(rxq, desc);
2036 	if (ret) {
2037 		rte_free(rxq);
2038 		return ret;
2039 	}
2040 
2041 	priv->bpool_init_size += desc;
2042 
2043 	dev->data->rx_queues[idx] = rxq;
2044 
2045 	return 0;
2046 }
2047 
2048 /**
2049  * DPDK callback to release the receive queue.
2050  *
2051  * @param rxq
2052  *   Generic receive queue pointer.
2053  */
2054 static void
2055 mrvl_rx_queue_release(void *rxq)
2056 {
2057 	struct mrvl_rxq *q = rxq;
2058 	struct pp2_ppio_tc_params *tc_params;
2059 	int i, num, tc, inq;
2060 	struct pp2_hif *hif;
2061 	unsigned int core_id = rte_lcore_id();
2062 
2063 	if (core_id == LCORE_ID_ANY)
2064 		core_id = rte_get_main_lcore();
2065 
2066 	if (!q)
2067 		return;
2068 
2069 	hif = mrvl_get_hif(q->priv, core_id);
2070 
2071 	if (!hif)
2072 		return;
2073 
2074 	tc = q->priv->rxq_map[q->queue_id].tc;
2075 	inq = q->priv->rxq_map[q->queue_id].inq;
2076 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
2077 	num = tc_params->inqs_params[inq].size;
2078 	for (i = 0; i < num; i++) {
2079 		struct pp2_buff_inf inf;
2080 		uint64_t addr;
2081 
2082 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
2083 		addr = cookie_addr_high | inf.cookie;
2084 		rte_pktmbuf_free((struct rte_mbuf *)addr);
2085 	}
2086 
2087 	rte_free(q);
2088 }
2089 
2090 /**
2091  * DPDK callback to configure the transmit queue.
2092  *
2093  * @param dev
2094  *   Pointer to Ethernet device structure.
2095  * @param idx
2096  *   Transmit queue index.
2097  * @param desc
2098  *   Number of descriptors to configure in the queue.
2099  * @param socket
2100  *   NUMA socket on which memory must be allocated.
2101  * @param conf
2102  *   Tx queue configuration parameters.
2103  *
2104  * @return
2105  *   0 on success, negative error value otherwise.
2106  */
2107 static int
2108 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
2109 		    unsigned int socket,
2110 		    const struct rte_eth_txconf *conf)
2111 {
2112 	struct mrvl_priv *priv = dev->data->dev_private;
2113 	struct mrvl_txq *txq;
2114 
2115 	if (dev->data->tx_queues[idx]) {
2116 		rte_free(dev->data->tx_queues[idx]);
2117 		dev->data->tx_queues[idx] = NULL;
2118 	}
2119 
2120 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
2121 	if (!txq)
2122 		return -ENOMEM;
2123 
2124 	txq->priv = priv;
2125 	txq->queue_id = idx;
2126 	txq->port_id = dev->data->port_id;
2127 	txq->tx_deferred_start = conf->tx_deferred_start;
2128 	dev->data->tx_queues[idx] = txq;
2129 
2130 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
2131 
2132 	return 0;
2133 }
2134 
2135 /**
2136  * DPDK callback to release the transmit queue.
2137  *
2138  * @param txq
2139  *   Generic transmit queue pointer.
2140  */
2141 static void
2142 mrvl_tx_queue_release(void *txq)
2143 {
2144 	struct mrvl_txq *q = txq;
2145 
2146 	if (!q)
2147 		return;
2148 
2149 	rte_free(q);
2150 }
2151 
2152 /**
2153  * DPDK callback to get flow control configuration.
2154  *
2155  * @param dev
2156  *  Pointer to Ethernet device structure.
2157  * @param fc_conf
2158  *  Pointer to the flow control configuration.
2159  *
2160  * @return
2161  *  0 on success, negative error value otherwise.
2162  */
2163 static int
2164 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2165 {
2166 	struct mrvl_priv *priv = dev->data->dev_private;
2167 	int ret, en;
2168 
2169 	if (!priv->ppio) {
2170 		memcpy(fc_conf, &priv->fc_conf, sizeof(struct rte_eth_fc_conf));
2171 		return 0;
2172 	}
2173 
2174 	fc_conf->autoneg = 1;
2175 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
2176 	if (ret) {
2177 		MRVL_LOG(ERR, "Failed to read rx pause state");
2178 		return ret;
2179 	}
2180 
2181 	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
2182 
2183 	ret = pp2_ppio_get_tx_pause(priv->ppio, &en);
2184 	if (ret) {
2185 		MRVL_LOG(ERR, "Failed to read tx pause state");
2186 		return ret;
2187 	}
2188 
2189 	if (en) {
2190 		if (fc_conf->mode == RTE_FC_NONE)
2191 			fc_conf->mode = RTE_FC_TX_PAUSE;
2192 		else
2193 			fc_conf->mode = RTE_FC_FULL;
2194 	}
2195 
2196 	return 0;
2197 }
2198 
2199 /**
2200  * DPDK callback to set flow control configuration.
2201  *
2202  * @param dev
2203  *  Pointer to Ethernet device structure.
2204  * @param fc_conf
2205  *  Pointer to the flow control configuration.
2206  *
2207  * @return
2208  *  0 on success, negative error value otherwise.
2209  */
2210 static int
2211 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2212 {
2213 	struct mrvl_priv *priv = dev->data->dev_private;
2214 	struct pp2_ppio_tx_pause_params mrvl_pause_params;
2215 	int ret;
2216 	int rx_en, tx_en;
2217 
2218 	if (fc_conf->high_water ||
2219 	    fc_conf->low_water ||
2220 	    fc_conf->pause_time ||
2221 	    fc_conf->mac_ctrl_frame_fwd) {
2222 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
2223 
2224 		return -EINVAL;
2225 	}
2226 
2227 	if (fc_conf->autoneg == 0) {
2228 		MRVL_LOG(ERR, "Flowctrl Autoneg disable is not supported");
2229 		return -EINVAL;
2230 	}
2231 
2232 	if (!priv->ppio) {
2233 		memcpy(&priv->fc_conf, fc_conf, sizeof(struct rte_eth_fc_conf));
2234 		priv->flow_ctrl = 1;
2235 		return 0;
2236 	}
2237 
2238 	switch (fc_conf->mode) {
2239 	case RTE_FC_FULL:
2240 		rx_en = 1;
2241 		tx_en = 1;
2242 		break;
2243 	case RTE_FC_TX_PAUSE:
2244 		rx_en = 0;
2245 		tx_en = 1;
2246 		break;
2247 	case RTE_FC_RX_PAUSE:
2248 		rx_en = 1;
2249 		tx_en = 0;
2250 		break;
2251 	case RTE_FC_NONE:
2252 		rx_en = 0;
2253 		tx_en = 0;
2254 		break;
2255 	default:
2256 		MRVL_LOG(ERR, "Incorrect Flow control flag (%d)",
2257 			 fc_conf->mode);
2258 		return -EINVAL;
2259 	}
2260 
2261 	/* Set RX flow control */
2262 	ret = pp2_ppio_set_rx_pause(priv->ppio, rx_en);
2263 	if (ret) {
2264 		MRVL_LOG(ERR, "Failed to change RX flowctrl");
2265 		return ret;
2266 	}
2267 
2268 	/* Set TX flow control */
2269 	mrvl_pause_params.en = tx_en;
2270 	/* all inqs participate in xon/xoff decision */
2271 	mrvl_pause_params.use_tc_pause_inqs = 0;
2272 	ret = pp2_ppio_set_tx_pause(priv->ppio, &mrvl_pause_params);
2273 	if (ret) {
2274 		MRVL_LOG(ERR, "Failed to change TX flowctrl");
2275 		return ret;
2276 	}
2277 
2278 	return 0;
2279 }
2280 
2281 /**
2282  * Update RSS hash configuration
2283  *
2284  * @param dev
2285  *   Pointer to Ethernet device structure.
2286  * @param rss_conf
2287  *   Pointer to RSS configuration.
2288  *
2289  * @return
2290  *   0 on success, negative error value otherwise.
2291  */
2292 static int
2293 mrvl_rss_hash_update(struct rte_eth_dev *dev,
2294 		     struct rte_eth_rss_conf *rss_conf)
2295 {
2296 	struct mrvl_priv *priv = dev->data->dev_private;
2297 
2298 	if (priv->isolated)
2299 		return -ENOTSUP;
2300 
2301 	return mrvl_configure_rss(priv, rss_conf);
2302 }
2303 
2304 /**
2305  * DPDK callback to get RSS hash configuration.
2306  *
2307  * @param dev
2308  *   Pointer to Ethernet device structure.
2309  * @rss_conf
2310  *   Pointer to RSS configuration.
2311  *
2312  * @return
2313  *   Always 0.
2314  */
2315 static int
2316 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
2317 		       struct rte_eth_rss_conf *rss_conf)
2318 {
2319 	struct mrvl_priv *priv = dev->data->dev_private;
2320 	enum pp2_ppio_hash_type hash_type =
2321 		priv->ppio_params.inqs_params.hash_type;
2322 
2323 	rss_conf->rss_key = NULL;
2324 
2325 	if (hash_type == PP2_PPIO_HASH_T_NONE)
2326 		rss_conf->rss_hf = 0;
2327 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
2328 		rss_conf->rss_hf = ETH_RSS_IPV4;
2329 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
2330 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
2331 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
2332 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
2333 
2334 	return 0;
2335 }
2336 
2337 /**
2338  * DPDK callback to get rte_flow callbacks.
2339  *
2340  * @param dev
2341  *   Pointer to the device structure.
2342  * @param filer_type
2343  *   Flow filter type.
2344  * @param filter_op
2345  *   Flow filter operation.
2346  * @param arg
2347  *   Pointer to pass the flow ops.
2348  *
2349  * @return
2350  *   0 on success, negative error value otherwise.
2351  */
2352 static int
2353 mrvl_eth_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
2354 		     enum rte_filter_type filter_type,
2355 		     enum rte_filter_op filter_op, void *arg)
2356 {
2357 	switch (filter_type) {
2358 	case RTE_ETH_FILTER_GENERIC:
2359 		if (filter_op != RTE_ETH_FILTER_GET)
2360 			return -EINVAL;
2361 		*(const void **)arg = &mrvl_flow_ops;
2362 		return 0;
2363 	default:
2364 		MRVL_LOG(WARNING, "Filter type (%d) not supported",
2365 				filter_type);
2366 		return -EINVAL;
2367 	}
2368 }
2369 
2370 /**
2371  * DPDK callback to get rte_mtr callbacks.
2372  *
2373  * @param dev
2374  *   Pointer to the device structure.
2375  * @param ops
2376  *   Pointer to pass the mtr ops.
2377  *
2378  * @return
2379  *   Always 0.
2380  */
2381 static int
2382 mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2383 {
2384 	*(const void **)ops = &mrvl_mtr_ops;
2385 
2386 	return 0;
2387 }
2388 
2389 /**
2390  * DPDK callback to get rte_tm callbacks.
2391  *
2392  * @param dev
2393  *   Pointer to the device structure.
2394  * @param ops
2395  *   Pointer to pass the tm ops.
2396  *
2397  * @return
2398  *   Always 0.
2399  */
2400 static int
2401 mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2402 {
2403 	*(const void **)ops = &mrvl_tm_ops;
2404 
2405 	return 0;
2406 }
2407 
2408 static const struct eth_dev_ops mrvl_ops = {
2409 	.dev_configure = mrvl_dev_configure,
2410 	.dev_start = mrvl_dev_start,
2411 	.dev_stop = mrvl_dev_stop,
2412 	.dev_set_link_up = mrvl_dev_set_link_up,
2413 	.dev_set_link_down = mrvl_dev_set_link_down,
2414 	.dev_close = mrvl_dev_close,
2415 	.link_update = mrvl_link_update,
2416 	.promiscuous_enable = mrvl_promiscuous_enable,
2417 	.allmulticast_enable = mrvl_allmulticast_enable,
2418 	.promiscuous_disable = mrvl_promiscuous_disable,
2419 	.allmulticast_disable = mrvl_allmulticast_disable,
2420 	.mac_addr_remove = mrvl_mac_addr_remove,
2421 	.mac_addr_add = mrvl_mac_addr_add,
2422 	.mac_addr_set = mrvl_mac_addr_set,
2423 	.mtu_set = mrvl_mtu_set,
2424 	.stats_get = mrvl_stats_get,
2425 	.stats_reset = mrvl_stats_reset,
2426 	.xstats_get = mrvl_xstats_get,
2427 	.xstats_reset = mrvl_xstats_reset,
2428 	.xstats_get_names = mrvl_xstats_get_names,
2429 	.dev_infos_get = mrvl_dev_infos_get,
2430 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
2431 	.rxq_info_get = mrvl_rxq_info_get,
2432 	.txq_info_get = mrvl_txq_info_get,
2433 	.vlan_filter_set = mrvl_vlan_filter_set,
2434 	.vlan_offload_set = mrvl_vlan_offload_set,
2435 	.tx_queue_start = mrvl_tx_queue_start,
2436 	.tx_queue_stop = mrvl_tx_queue_stop,
2437 	.rx_queue_setup = mrvl_rx_queue_setup,
2438 	.rx_queue_release = mrvl_rx_queue_release,
2439 	.tx_queue_setup = mrvl_tx_queue_setup,
2440 	.tx_queue_release = mrvl_tx_queue_release,
2441 	.flow_ctrl_get = mrvl_flow_ctrl_get,
2442 	.flow_ctrl_set = mrvl_flow_ctrl_set,
2443 	.rss_hash_update = mrvl_rss_hash_update,
2444 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
2445 	.filter_ctrl = mrvl_eth_filter_ctrl,
2446 	.mtr_ops_get = mrvl_mtr_ops_get,
2447 	.tm_ops_get = mrvl_tm_ops_get,
2448 };
2449 
2450 /**
2451  * Return packet type information and l3/l4 offsets.
2452  *
2453  * @param desc
2454  *   Pointer to the received packet descriptor.
2455  * @param l3_offset
2456  *   l3 packet offset.
2457  * @param l4_offset
2458  *   l4 packet offset.
2459  *
2460  * @return
2461  *   Packet type information.
2462  */
2463 static inline uint64_t
2464 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
2465 				    uint8_t *l3_offset, uint8_t *l4_offset)
2466 {
2467 	enum pp2_inq_l3_type l3_type;
2468 	enum pp2_inq_l4_type l4_type;
2469 	enum pp2_inq_vlan_tag vlan_tag;
2470 	uint64_t packet_type;
2471 
2472 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
2473 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
2474 	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
2475 
2476 	packet_type = RTE_PTYPE_L2_ETHER;
2477 
2478 	switch (vlan_tag) {
2479 	case PP2_INQ_VLAN_TAG_SINGLE:
2480 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
2481 		break;
2482 	case PP2_INQ_VLAN_TAG_DOUBLE:
2483 	case PP2_INQ_VLAN_TAG_TRIPLE:
2484 		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
2485 		break;
2486 	default:
2487 		break;
2488 	}
2489 
2490 	switch (l3_type) {
2491 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
2492 		packet_type |= RTE_PTYPE_L3_IPV4;
2493 		break;
2494 	case PP2_INQ_L3_TYPE_IPV4_OK:
2495 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
2496 		break;
2497 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
2498 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
2499 		break;
2500 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
2501 		packet_type |= RTE_PTYPE_L3_IPV6;
2502 		break;
2503 	case PP2_INQ_L3_TYPE_IPV6_EXT:
2504 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
2505 		break;
2506 	case PP2_INQ_L3_TYPE_ARP:
2507 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
2508 		/*
2509 		 * In case of ARP l4_offset is set to wrong value.
2510 		 * Set it to proper one so that later on mbuf->l3_len can be
2511 		 * calculated subtracting l4_offset and l3_offset.
2512 		 */
2513 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
2514 		break;
2515 	default:
2516 		break;
2517 	}
2518 
2519 	switch (l4_type) {
2520 	case PP2_INQ_L4_TYPE_TCP:
2521 		packet_type |= RTE_PTYPE_L4_TCP;
2522 		break;
2523 	case PP2_INQ_L4_TYPE_UDP:
2524 		packet_type |= RTE_PTYPE_L4_UDP;
2525 		break;
2526 	default:
2527 		break;
2528 	}
2529 
2530 	return packet_type;
2531 }
2532 
2533 /**
2534  * Get offload information from the received packet descriptor.
2535  *
2536  * @param desc
2537  *   Pointer to the received packet descriptor.
2538  *
2539  * @return
2540  *   Mbuf offload flags.
2541  */
2542 static inline uint64_t
2543 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc, uint64_t packet_type)
2544 {
2545 	uint64_t flags = 0;
2546 	enum pp2_inq_desc_status status;
2547 
2548 	if (RTE_ETH_IS_IPV4_HDR(packet_type)) {
2549 		status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2550 		if (unlikely(status != PP2_DESC_ERR_OK))
2551 			flags |= PKT_RX_IP_CKSUM_BAD;
2552 		else
2553 			flags |= PKT_RX_IP_CKSUM_GOOD;
2554 	}
2555 
2556 	if (((packet_type & RTE_PTYPE_L4_UDP) == RTE_PTYPE_L4_UDP) ||
2557 	    ((packet_type & RTE_PTYPE_L4_TCP) == RTE_PTYPE_L4_TCP)) {
2558 		status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2559 		if (unlikely(status != PP2_DESC_ERR_OK))
2560 			flags |= PKT_RX_L4_CKSUM_BAD;
2561 		else
2562 			flags |= PKT_RX_L4_CKSUM_GOOD;
2563 	}
2564 
2565 	return flags;
2566 }
2567 
2568 /**
2569  * DPDK callback for receive.
2570  *
2571  * @param rxq
2572  *   Generic pointer to the receive queue.
2573  * @param rx_pkts
2574  *   Array to store received packets.
2575  * @param nb_pkts
2576  *   Maximum number of packets in array.
2577  *
2578  * @return
2579  *   Number of packets successfully received.
2580  */
2581 static uint16_t
2582 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2583 {
2584 	struct mrvl_rxq *q = rxq;
2585 	struct pp2_ppio_desc descs[nb_pkts];
2586 	struct pp2_bpool *bpool;
2587 	int i, ret, rx_done = 0;
2588 	int num;
2589 	struct pp2_hif *hif;
2590 	unsigned int core_id = rte_lcore_id();
2591 
2592 	hif = mrvl_get_hif(q->priv, core_id);
2593 
2594 	if (unlikely(!q->priv->ppio || !hif))
2595 		return 0;
2596 
2597 	bpool = q->priv->bpool;
2598 
2599 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2600 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2601 	if (unlikely(ret < 0))
2602 		return 0;
2603 
2604 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2605 
2606 	for (i = 0; i < nb_pkts; i++) {
2607 		struct rte_mbuf *mbuf;
2608 		uint8_t l3_offset, l4_offset;
2609 		enum pp2_inq_desc_status status;
2610 		uint64_t addr;
2611 
2612 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2613 			struct pp2_ppio_desc *pref_desc;
2614 			u64 pref_addr;
2615 
2616 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2617 			pref_addr = cookie_addr_high |
2618 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2619 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2620 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2621 		}
2622 
2623 		addr = cookie_addr_high |
2624 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2625 		mbuf = (struct rte_mbuf *)addr;
2626 		rte_pktmbuf_reset(mbuf);
2627 
2628 		/* drop packet in case of mac, overrun or resource error */
2629 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2630 		if ((unlikely(status != PP2_DESC_ERR_OK)) &&
2631 			!(q->priv->forward_bad_frames)) {
2632 			struct pp2_buff_inf binf = {
2633 				.addr = rte_mbuf_data_iova_default(mbuf),
2634 				.cookie = (uint64_t)mbuf,
2635 			};
2636 
2637 			pp2_bpool_put_buff(hif, bpool, &binf);
2638 			mrvl_port_bpool_size
2639 				[bpool->pp2_id][bpool->id][core_id]++;
2640 			q->drop_mac++;
2641 			continue;
2642 		}
2643 
2644 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2645 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2646 		mbuf->data_len = mbuf->pkt_len;
2647 		mbuf->port = q->port_id;
2648 		mbuf->packet_type =
2649 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2650 							    &l3_offset,
2651 							    &l4_offset);
2652 		mbuf->l2_len = l3_offset;
2653 		mbuf->l3_len = l4_offset - l3_offset;
2654 
2655 		if (likely(q->cksum_enabled))
2656 			mbuf->ol_flags =
2657 				mrvl_desc_to_ol_flags(&descs[i],
2658 						      mbuf->packet_type);
2659 
2660 		rx_pkts[rx_done++] = mbuf;
2661 		q->bytes_recv += mbuf->pkt_len;
2662 	}
2663 
2664 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2665 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2666 
2667 		if (unlikely(num <= q->priv->bpool_min_size ||
2668 			     (!rx_done && num < q->priv->bpool_init_size))) {
2669 			mrvl_fill_bpool(q, MRVL_BURST_SIZE);
2670 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2671 			int i;
2672 			int pkt_to_remove = num - q->priv->bpool_init_size;
2673 			struct rte_mbuf *mbuf;
2674 			struct pp2_buff_inf buff;
2675 
2676 			for (i = 0; i < pkt_to_remove; i++) {
2677 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2678 				if (ret)
2679 					break;
2680 				mbuf = (struct rte_mbuf *)
2681 					(cookie_addr_high | buff.cookie);
2682 				rte_pktmbuf_free(mbuf);
2683 			}
2684 			mrvl_port_bpool_size
2685 				[bpool->pp2_id][bpool->id][core_id] -= i;
2686 		}
2687 		rte_spinlock_unlock(&q->priv->lock);
2688 	}
2689 
2690 	return rx_done;
2691 }
2692 
2693 /**
2694  * Prepare offload information.
2695  *
2696  * @param ol_flags
2697  *   Offload flags.
2698  * @param l3_type
2699  *   Pointer to the pp2_ouq_l3_type structure.
2700  * @param l4_type
2701  *   Pointer to the pp2_outq_l4_type structure.
2702  * @param gen_l3_cksum
2703  *   Will be set to 1 in case l3 checksum is computed.
2704  * @param l4_cksum
2705  *   Will be set to 1 in case l4 checksum is computed.
2706  */
2707 static inline void
2708 mrvl_prepare_proto_info(uint64_t ol_flags,
2709 			enum pp2_outq_l3_type *l3_type,
2710 			enum pp2_outq_l4_type *l4_type,
2711 			int *gen_l3_cksum,
2712 			int *gen_l4_cksum)
2713 {
2714 	/*
2715 	 * Based on ol_flags prepare information
2716 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2717 	 * for offloading.
2718 	 * in most of the checksum cases ipv4 must be set, so this is the
2719 	 * default value
2720 	 */
2721 	*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2722 	*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2723 
2724 	if (ol_flags & PKT_TX_IPV6) {
2725 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2726 		/* no checksum for ipv6 header */
2727 		*gen_l3_cksum = 0;
2728 	}
2729 
2730 	if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM) {
2731 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2732 		*gen_l4_cksum = 1;
2733 	} else if ((ol_flags & PKT_TX_L4_MASK) ==  PKT_TX_UDP_CKSUM) {
2734 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2735 		*gen_l4_cksum = 1;
2736 	} else {
2737 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2738 		/* no checksum for other type */
2739 		*gen_l4_cksum = 0;
2740 	}
2741 }
2742 
2743 /**
2744  * Release already sent buffers to bpool (buffer-pool).
2745  *
2746  * @param ppio
2747  *   Pointer to the port structure.
2748  * @param hif
2749  *   Pointer to the MUSDK hardware interface.
2750  * @param sq
2751  *   Pointer to the shadow queue.
2752  * @param qid
2753  *   Queue id number.
2754  * @param force
2755  *   Force releasing packets.
2756  */
2757 static inline void
2758 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2759 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2760 		       int qid, int force)
2761 {
2762 	struct buff_release_entry *entry;
2763 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2764 	int i;
2765 
2766 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2767 
2768 	sq->num_to_release += nb_done;
2769 
2770 	if (likely(!force &&
2771 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2772 		return;
2773 
2774 	nb_done = sq->num_to_release;
2775 	sq->num_to_release = 0;
2776 
2777 	for (i = 0; i < nb_done; i++) {
2778 		entry = &sq->ent[sq->tail + num];
2779 		if (unlikely(!entry->buff.addr)) {
2780 			MRVL_LOG(ERR,
2781 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2782 				sq->tail, (u64)entry->buff.cookie,
2783 				(u64)entry->buff.addr);
2784 			skip_bufs = 1;
2785 			goto skip;
2786 		}
2787 
2788 		if (unlikely(!entry->bpool)) {
2789 			struct rte_mbuf *mbuf;
2790 
2791 			mbuf = (struct rte_mbuf *)entry->buff.cookie;
2792 			rte_pktmbuf_free(mbuf);
2793 			skip_bufs = 1;
2794 			goto skip;
2795 		}
2796 
2797 		mrvl_port_bpool_size
2798 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2799 		num++;
2800 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2801 			goto skip;
2802 		continue;
2803 skip:
2804 		if (likely(num))
2805 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2806 		num += skip_bufs;
2807 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2808 		sq->size -= num;
2809 		num = 0;
2810 		skip_bufs = 0;
2811 	}
2812 
2813 	if (likely(num)) {
2814 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2815 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2816 		sq->size -= num;
2817 	}
2818 }
2819 
2820 /**
2821  * DPDK callback for transmit.
2822  *
2823  * @param txq
2824  *   Generic pointer transmit queue.
2825  * @param tx_pkts
2826  *   Packets to transmit.
2827  * @param nb_pkts
2828  *   Number of packets in array.
2829  *
2830  * @return
2831  *   Number of packets successfully transmitted.
2832  */
2833 static uint16_t
2834 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2835 {
2836 	struct mrvl_txq *q = txq;
2837 	struct mrvl_shadow_txq *sq;
2838 	struct pp2_hif *hif;
2839 	struct pp2_ppio_desc descs[nb_pkts];
2840 	unsigned int core_id = rte_lcore_id();
2841 	int i, bytes_sent = 0;
2842 	uint16_t num, sq_free_size;
2843 	uint64_t addr;
2844 
2845 	hif = mrvl_get_hif(q->priv, core_id);
2846 	sq = &q->shadow_txqs[core_id];
2847 
2848 	if (unlikely(!q->priv->ppio || !hif))
2849 		return 0;
2850 
2851 	if (sq->size)
2852 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2853 				       sq, q->queue_id, 0);
2854 
2855 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2856 	if (unlikely(nb_pkts > sq_free_size))
2857 		nb_pkts = sq_free_size;
2858 
2859 	for (i = 0; i < nb_pkts; i++) {
2860 		struct rte_mbuf *mbuf = tx_pkts[i];
2861 		int gen_l3_cksum, gen_l4_cksum;
2862 		enum pp2_outq_l3_type l3_type;
2863 		enum pp2_outq_l4_type l4_type;
2864 
2865 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2866 			struct rte_mbuf *pref_pkt_hdr;
2867 
2868 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2869 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2870 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2871 		}
2872 
2873 		mrvl_fill_shadowq(sq, mbuf);
2874 		mrvl_fill_desc(&descs[i], mbuf);
2875 
2876 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2877 		/*
2878 		 * in case unsupported ol_flags were passed
2879 		 * do not update descriptor offload information
2880 		 */
2881 		if (!(mbuf->ol_flags & MRVL_TX_PKT_OFFLOADS))
2882 			continue;
2883 		mrvl_prepare_proto_info(mbuf->ol_flags, &l3_type, &l4_type,
2884 					&gen_l3_cksum, &gen_l4_cksum);
2885 
2886 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2887 						  mbuf->l2_len,
2888 						  mbuf->l2_len + mbuf->l3_len,
2889 						  gen_l3_cksum, gen_l4_cksum);
2890 	}
2891 
2892 	num = nb_pkts;
2893 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2894 	/* number of packets that were not sent */
2895 	if (unlikely(num > nb_pkts)) {
2896 		for (i = nb_pkts; i < num; i++) {
2897 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2898 				MRVL_PP2_TX_SHADOWQ_MASK;
2899 			addr = sq->ent[sq->head].buff.cookie;
2900 			bytes_sent -=
2901 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2902 		}
2903 		sq->size -= num - nb_pkts;
2904 	}
2905 
2906 	q->bytes_sent += bytes_sent;
2907 
2908 	return nb_pkts;
2909 }
2910 
2911 /** DPDK callback for S/G transmit.
2912  *
2913  * @param txq
2914  *   Generic pointer transmit queue.
2915  * @param tx_pkts
2916  *   Packets to transmit.
2917  * @param nb_pkts
2918  *   Number of packets in array.
2919  *
2920  * @return
2921  *   Number of packets successfully transmitted.
2922  */
2923 static uint16_t
2924 mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
2925 		     uint16_t nb_pkts)
2926 {
2927 	struct mrvl_txq *q = txq;
2928 	struct mrvl_shadow_txq *sq;
2929 	struct pp2_hif *hif;
2930 	struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
2931 	struct pp2_ppio_sg_pkts pkts;
2932 	uint8_t frags[nb_pkts];
2933 	unsigned int core_id = rte_lcore_id();
2934 	int i, j, bytes_sent = 0;
2935 	int tail, tail_first;
2936 	uint16_t num, sq_free_size;
2937 	uint16_t nb_segs, total_descs = 0;
2938 	uint64_t addr;
2939 
2940 	hif = mrvl_get_hif(q->priv, core_id);
2941 	sq = &q->shadow_txqs[core_id];
2942 	pkts.frags = frags;
2943 	pkts.num = 0;
2944 
2945 	if (unlikely(!q->priv->ppio || !hif))
2946 		return 0;
2947 
2948 	if (sq->size)
2949 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2950 				       sq, q->queue_id, 0);
2951 
2952 	/* Save shadow queue free size */
2953 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2954 
2955 	tail = 0;
2956 	for (i = 0; i < nb_pkts; i++) {
2957 		struct rte_mbuf *mbuf = tx_pkts[i];
2958 		struct rte_mbuf *seg = NULL;
2959 		int gen_l3_cksum, gen_l4_cksum;
2960 		enum pp2_outq_l3_type l3_type;
2961 		enum pp2_outq_l4_type l4_type;
2962 
2963 		nb_segs = mbuf->nb_segs;
2964 		tail_first = tail;
2965 		total_descs += nb_segs;
2966 
2967 		/*
2968 		 * Check if total_descs does not exceed
2969 		 * shadow queue free size
2970 		 */
2971 		if (unlikely(total_descs > sq_free_size)) {
2972 			total_descs -= nb_segs;
2973 			break;
2974 		}
2975 
2976 		/* Check if nb_segs does not exceed the max nb of desc per
2977 		 * fragmented packet
2978 		 */
2979 		if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
2980 			total_descs -= nb_segs;
2981 			RTE_LOG(ERR, PMD,
2982 				"Too many segments. Packet won't be sent.\n");
2983 			break;
2984 		}
2985 
2986 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2987 			struct rte_mbuf *pref_pkt_hdr;
2988 
2989 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2990 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2991 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2992 		}
2993 
2994 		pkts.frags[pkts.num] = nb_segs;
2995 		pkts.num++;
2996 
2997 		seg = mbuf;
2998 		for (j = 0; j < nb_segs - 1; j++) {
2999 			/* For the subsequent segments, set shadow queue
3000 			 * buffer to NULL
3001 			 */
3002 			mrvl_fill_shadowq(sq, NULL);
3003 			mrvl_fill_desc(&descs[tail], seg);
3004 
3005 			tail++;
3006 			seg = seg->next;
3007 		}
3008 		/* Put first mbuf info in last shadow queue entry */
3009 		mrvl_fill_shadowq(sq, mbuf);
3010 		/* Update descriptor with last segment */
3011 		mrvl_fill_desc(&descs[tail++], seg);
3012 
3013 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
3014 		/* In case unsupported ol_flags were passed
3015 		 * do not update descriptor offload information
3016 		 */
3017 		if (!(mbuf->ol_flags & MRVL_TX_PKT_OFFLOADS))
3018 			continue;
3019 		mrvl_prepare_proto_info(mbuf->ol_flags, &l3_type, &l4_type,
3020 					&gen_l3_cksum, &gen_l4_cksum);
3021 
3022 		pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
3023 						  l4_type, mbuf->l2_len,
3024 						  mbuf->l2_len + mbuf->l3_len,
3025 						  gen_l3_cksum, gen_l4_cksum);
3026 	}
3027 
3028 	num = total_descs;
3029 	pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
3030 			 &total_descs, &pkts);
3031 	/* number of packets that were not sent */
3032 	if (unlikely(num > total_descs)) {
3033 		for (i = total_descs; i < num; i++) {
3034 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
3035 				MRVL_PP2_TX_SHADOWQ_MASK;
3036 
3037 			addr = sq->ent[sq->head].buff.cookie;
3038 			if (addr)
3039 				bytes_sent -=
3040 					rte_pktmbuf_pkt_len((struct rte_mbuf *)
3041 						(cookie_addr_high | addr));
3042 		}
3043 		sq->size -= num - total_descs;
3044 		nb_pkts = pkts.num;
3045 	}
3046 
3047 	q->bytes_sent += bytes_sent;
3048 
3049 	return nb_pkts;
3050 }
3051 
3052 /**
3053  * Create private device structure.
3054  *
3055  * @param dev_name
3056  *   Pointer to the port name passed in the initialization parameters.
3057  *
3058  * @return
3059  *   Pointer to the newly allocated private device structure.
3060  */
3061 static struct mrvl_priv *
3062 mrvl_priv_create(const char *dev_name)
3063 {
3064 	struct pp2_bpool_params bpool_params;
3065 	char match[MRVL_MATCH_LEN];
3066 	struct mrvl_priv *priv;
3067 	uint16_t max_frame_size;
3068 	int ret, bpool_bit;
3069 
3070 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
3071 	if (!priv)
3072 		return NULL;
3073 
3074 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
3075 				       &priv->pp_id, &priv->ppio_id);
3076 	if (ret)
3077 		goto out_free_priv;
3078 
3079 	ret = pp2_ppio_get_l4_cksum_max_frame_size(priv->pp_id, priv->ppio_id,
3080 						   &max_frame_size);
3081 	if (ret)
3082 		goto out_free_priv;
3083 
3084 	priv->max_mtu = max_frame_size + RTE_ETHER_CRC_LEN -
3085 		MRVL_PP2_ETH_HDRS_LEN;
3086 
3087 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
3088 				     PP2_BPOOL_NUM_POOLS);
3089 	if (bpool_bit < 0)
3090 		goto out_free_priv;
3091 	priv->bpool_bit = bpool_bit;
3092 
3093 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
3094 		 priv->bpool_bit);
3095 	memset(&bpool_params, 0, sizeof(bpool_params));
3096 	bpool_params.match = match;
3097 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
3098 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
3099 	if (ret)
3100 		goto out_clear_bpool_bit;
3101 
3102 	priv->ppio_params.type = PP2_PPIO_T_NIC;
3103 	rte_spinlock_init(&priv->lock);
3104 
3105 	return priv;
3106 out_clear_bpool_bit:
3107 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
3108 out_free_priv:
3109 	rte_free(priv);
3110 	return NULL;
3111 }
3112 
3113 /**
3114  * Create device representing Ethernet port.
3115  *
3116  * @param name
3117  *   Pointer to the port's name.
3118  *
3119  * @return
3120  *   0 on success, negative error value otherwise.
3121  */
3122 static int
3123 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
3124 {
3125 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
3126 	struct rte_eth_dev *eth_dev;
3127 	struct mrvl_priv *priv;
3128 	struct ifreq req;
3129 
3130 	eth_dev = rte_eth_dev_allocate(name);
3131 	if (!eth_dev)
3132 		return -ENOMEM;
3133 
3134 	priv = mrvl_priv_create(name);
3135 	if (!priv) {
3136 		ret = -ENOMEM;
3137 		goto out_free;
3138 	}
3139 	eth_dev->data->dev_private = priv;
3140 
3141 	eth_dev->data->mac_addrs =
3142 		rte_zmalloc("mac_addrs",
3143 			    RTE_ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
3144 	if (!eth_dev->data->mac_addrs) {
3145 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
3146 		ret = -ENOMEM;
3147 		goto out_free;
3148 	}
3149 
3150 	memset(&req, 0, sizeof(req));
3151 	strcpy(req.ifr_name, name);
3152 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
3153 	if (ret)
3154 		goto out_free;
3155 
3156 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
3157 	       req.ifr_addr.sa_data, RTE_ETHER_ADDR_LEN);
3158 
3159 	eth_dev->device = &vdev->device;
3160 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
3161 	mrvl_set_tx_function(eth_dev);
3162 	eth_dev->dev_ops = &mrvl_ops;
3163 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
3164 
3165 	eth_dev->data->dev_link.link_status = ETH_LINK_UP;
3166 
3167 	rte_eth_dev_probing_finish(eth_dev);
3168 	return 0;
3169 out_free:
3170 	rte_eth_dev_release_port(eth_dev);
3171 
3172 	return ret;
3173 }
3174 
3175 /**
3176  * Callback used by rte_kvargs_process() during argument parsing.
3177  *
3178  * @param key
3179  *   Pointer to the parsed key (unused).
3180  * @param value
3181  *   Pointer to the parsed value.
3182  * @param extra_args
3183  *   Pointer to the extra arguments which contains address of the
3184  *   table of pointers to parsed interface names.
3185  *
3186  * @return
3187  *   Always 0.
3188  */
3189 static int
3190 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
3191 		 void *extra_args)
3192 {
3193 	struct mrvl_ifnames *ifnames = extra_args;
3194 
3195 	ifnames->names[ifnames->idx++] = value;
3196 
3197 	return 0;
3198 }
3199 
3200 /**
3201  * DPDK callback to register the virtual device.
3202  *
3203  * @param vdev
3204  *   Pointer to the virtual device.
3205  *
3206  * @return
3207  *   0 on success, negative error value otherwise.
3208  */
3209 static int
3210 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
3211 {
3212 	struct rte_kvargs *kvlist;
3213 	struct mrvl_ifnames ifnames;
3214 	int ret = -EINVAL;
3215 	uint32_t i, ifnum, cfgnum;
3216 	const char *params;
3217 
3218 	params = rte_vdev_device_args(vdev);
3219 	if (!params)
3220 		return -EINVAL;
3221 
3222 	kvlist = rte_kvargs_parse(params, valid_args);
3223 	if (!kvlist)
3224 		return -EINVAL;
3225 
3226 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
3227 	if (ifnum > RTE_DIM(ifnames.names))
3228 		goto out_free_kvlist;
3229 
3230 	ifnames.idx = 0;
3231 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
3232 			   mrvl_get_ifnames, &ifnames);
3233 
3234 
3235 	/*
3236 	 * The below system initialization should be done only once,
3237 	 * on the first provided configuration file
3238 	 */
3239 	if (!mrvl_cfg) {
3240 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
3241 		MRVL_LOG(INFO, "Parsing config file!");
3242 		if (cfgnum > 1) {
3243 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
3244 			goto out_free_kvlist;
3245 		} else if (cfgnum == 1) {
3246 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
3247 					   mrvl_get_cfg, &mrvl_cfg);
3248 		}
3249 	}
3250 
3251 	if (mrvl_dev_num)
3252 		goto init_devices;
3253 
3254 	MRVL_LOG(INFO, "Perform MUSDK initializations");
3255 
3256 	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
3257 	if (ret)
3258 		goto out_free_kvlist;
3259 
3260 	ret = mrvl_init_pp2();
3261 	if (ret) {
3262 		MRVL_LOG(ERR, "Failed to init PP!");
3263 		rte_mvep_deinit(MVEP_MOD_T_PP2);
3264 		goto out_free_kvlist;
3265 	}
3266 
3267 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
3268 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
3269 
3270 	mrvl_lcore_first = RTE_MAX_LCORE;
3271 	mrvl_lcore_last = 0;
3272 
3273 init_devices:
3274 	for (i = 0; i < ifnum; i++) {
3275 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
3276 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
3277 		if (ret)
3278 			goto out_cleanup;
3279 		mrvl_dev_num++;
3280 	}
3281 
3282 	rte_kvargs_free(kvlist);
3283 
3284 	return 0;
3285 out_cleanup:
3286 	rte_pmd_mrvl_remove(vdev);
3287 
3288 out_free_kvlist:
3289 	rte_kvargs_free(kvlist);
3290 
3291 	return ret;
3292 }
3293 
3294 /**
3295  * DPDK callback to remove virtual device.
3296  *
3297  * @param vdev
3298  *   Pointer to the removed virtual device.
3299  *
3300  * @return
3301  *   0 on success, negative error value otherwise.
3302  */
3303 static int
3304 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
3305 {
3306 	uint16_t port_id;
3307 	int ret = 0;
3308 
3309 	RTE_ETH_FOREACH_DEV(port_id) {
3310 		if (rte_eth_devices[port_id].device != &vdev->device)
3311 			continue;
3312 		ret |= rte_eth_dev_close(port_id);
3313 	}
3314 
3315 	return ret == 0 ? 0 : -EIO;
3316 }
3317 
3318 static struct rte_vdev_driver pmd_mrvl_drv = {
3319 	.probe = rte_pmd_mrvl_probe,
3320 	.remove = rte_pmd_mrvl_remove,
3321 };
3322 
3323 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
3324 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
3325 RTE_LOG_REGISTER(mrvl_logtype, pmd.net.mvpp2, NOTICE);
3326