xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision 3da59f30a23f2e795d2315f3d949e1b3e0ce0c3d)
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 <bus_vdev_driver.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_ARP_LENGTH 28
56 
57 #define MRVL_COOKIE_ADDR_INVALID ~0ULL
58 #define MRVL_COOKIE_HIGH_ADDR_MASK 0xffffff0000000000
59 
60 /** Port Rx offload capabilities */
61 #define MRVL_RX_OFFLOADS (RTE_ETH_RX_OFFLOAD_VLAN_FILTER | \
62 			  RTE_ETH_RX_OFFLOAD_CHECKSUM)
63 
64 /** Port Tx offloads capabilities */
65 #define MRVL_TX_OFFLOAD_CHECKSUM (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM | \
66 				  RTE_ETH_TX_OFFLOAD_UDP_CKSUM  | \
67 				  RTE_ETH_TX_OFFLOAD_TCP_CKSUM)
68 #define MRVL_TX_OFFLOADS (MRVL_TX_OFFLOAD_CHECKSUM | \
69 			  RTE_ETH_TX_OFFLOAD_MULTI_SEGS)
70 
71 #define MRVL_TX_PKT_OFFLOADS (RTE_MBUF_F_TX_IP_CKSUM | \
72 			      RTE_MBUF_F_TX_TCP_CKSUM | \
73 			      RTE_MBUF_F_TX_UDP_CKSUM)
74 
75 static const char * const valid_args[] = {
76 	MRVL_IFACE_NAME_ARG,
77 	MRVL_CFG_ARG,
78 	NULL
79 };
80 
81 static int used_hifs = MRVL_MUSDK_HIFS_RESERVED;
82 static struct pp2_hif *hifs[RTE_MAX_LCORE];
83 static int used_bpools[PP2_NUM_PKT_PROC] = {
84 	[0 ... PP2_NUM_PKT_PROC - 1] = MRVL_MUSDK_BPOOLS_RESERVED
85 };
86 
87 static struct pp2_bpool *mrvl_port_to_bpool_lookup[RTE_MAX_ETHPORTS];
88 static int mrvl_port_bpool_size[PP2_NUM_PKT_PROC][PP2_BPOOL_NUM_POOLS][RTE_MAX_LCORE];
89 static uint64_t cookie_addr_high = MRVL_COOKIE_ADDR_INVALID;
90 static int dummy_pool_id[PP2_NUM_PKT_PROC];
91 struct pp2_bpool *dummy_pool[PP2_NUM_PKT_PROC] = {0};
92 
93 struct mrvl_ifnames {
94 	const char *names[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
95 	int idx;
96 };
97 
98 /*
99  * To use buffer harvesting based on loopback port shadow queue structure
100  * was introduced for buffers information bookkeeping.
101  *
102  * Before sending the packet, related buffer information (pp2_buff_inf) is
103  * stored in shadow queue. After packet is transmitted no longer used
104  * packet buffer is released back to it's original hardware pool,
105  * on condition it originated from interface.
106  * In case it  was generated by application itself i.e: mbuf->port field is
107  * 0xff then its released to software mempool.
108  */
109 struct mrvl_shadow_txq {
110 	int head;           /* write index - used when sending buffers */
111 	int tail;           /* read index - used when releasing buffers */
112 	u16 size;           /* queue occupied size */
113 	u16 num_to_release; /* number of descriptors sent, that can be
114 			     * released
115 			     */
116 	struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
117 };
118 
119 struct mrvl_rxq {
120 	struct mrvl_priv *priv;
121 	struct rte_mempool *mp;
122 	int queue_id;
123 	int port_id;
124 	int cksum_enabled;
125 	uint64_t bytes_recv;
126 	uint64_t drop_mac;
127 };
128 
129 struct mrvl_txq {
130 	struct mrvl_priv *priv;
131 	int queue_id;
132 	int port_id;
133 	uint64_t bytes_sent;
134 	struct mrvl_shadow_txq shadow_txqs[RTE_MAX_LCORE];
135 	int tx_deferred_start;
136 };
137 
138 static int mrvl_lcore_first;
139 static int mrvl_lcore_last;
140 static int mrvl_dev_num;
141 
142 static int mrvl_fill_bpool(struct mrvl_rxq *rxq, int num);
143 static inline void mrvl_free_sent_buffers(struct pp2_ppio *ppio,
144 			struct pp2_hif *hif, unsigned int core_id,
145 			struct mrvl_shadow_txq *sq, int qid, int force);
146 
147 static uint16_t mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
148 				  uint16_t nb_pkts);
149 static uint16_t mrvl_tx_sg_pkt_burst(void *txq,	struct rte_mbuf **tx_pkts,
150 				     uint16_t nb_pkts);
151 static int rte_pmd_mrvl_remove(struct rte_vdev_device *vdev);
152 static void mrvl_deinit_pp2(void);
153 static void mrvl_deinit_hifs(void);
154 
155 static int
156 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
157 		  uint32_t index, uint32_t vmdq __rte_unused);
158 static int
159 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr);
160 static int
161 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on);
162 static int mrvl_promiscuous_enable(struct rte_eth_dev *dev);
163 static int mrvl_allmulticast_enable(struct rte_eth_dev *dev);
164 static int
165 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf);
166 
167 #define MRVL_XSTATS_TBL_ENTRY(name) { \
168 	#name, offsetof(struct pp2_ppio_statistics, name),	\
169 	sizeof(((struct pp2_ppio_statistics *)0)->name)		\
170 }
171 
172 /* Table with xstats data */
173 static struct {
174 	const char *name;
175 	unsigned int offset;
176 	unsigned int size;
177 } mrvl_xstats_tbl[] = {
178 	MRVL_XSTATS_TBL_ENTRY(rx_bytes),
179 	MRVL_XSTATS_TBL_ENTRY(rx_packets),
180 	MRVL_XSTATS_TBL_ENTRY(rx_unicast_packets),
181 	MRVL_XSTATS_TBL_ENTRY(rx_errors),
182 	MRVL_XSTATS_TBL_ENTRY(rx_fullq_dropped),
183 	MRVL_XSTATS_TBL_ENTRY(rx_bm_dropped),
184 	MRVL_XSTATS_TBL_ENTRY(rx_early_dropped),
185 	MRVL_XSTATS_TBL_ENTRY(rx_fifo_dropped),
186 	MRVL_XSTATS_TBL_ENTRY(rx_cls_dropped),
187 	MRVL_XSTATS_TBL_ENTRY(tx_bytes),
188 	MRVL_XSTATS_TBL_ENTRY(tx_packets),
189 	MRVL_XSTATS_TBL_ENTRY(tx_unicast_packets),
190 	MRVL_XSTATS_TBL_ENTRY(tx_errors)
191 };
192 
193 static inline int
194 mrvl_reserve_bit(int *bitmap, int max)
195 {
196 	int n = sizeof(*bitmap) * 8 - rte_clz32(*bitmap);
197 
198 	if (n >= max)
199 		return -1;
200 
201 	*bitmap |= 1 << n;
202 
203 	return n;
204 }
205 
206 static int
207 mrvl_pp2_fixup_init(void)
208 {
209 	struct pp2_bpool_params bpool_params;
210 	char			name[15];
211 	int			err, i;
212 
213 	memset(dummy_pool, 0, sizeof(dummy_pool));
214 	for (i = 0; i < pp2_get_num_inst(); i++) {
215 		dummy_pool_id[i] = mrvl_reserve_bit(&used_bpools[i],
216 					     PP2_BPOOL_NUM_POOLS);
217 		if (dummy_pool_id[i] < 0) {
218 			MRVL_LOG(ERR, "Can't find free pool\n");
219 			return -1;
220 		}
221 
222 		memset(name, 0, sizeof(name));
223 		snprintf(name, sizeof(name), "pool-%d:%d", i, dummy_pool_id[i]);
224 		memset(&bpool_params, 0, sizeof(bpool_params));
225 		bpool_params.match = name;
226 		bpool_params.buff_len = MRVL_PKT_OFFS;
227 		bpool_params.dummy_short_pool = 1;
228 		err = pp2_bpool_init(&bpool_params, &dummy_pool[i]);
229 		if (err != 0 || !dummy_pool[i]) {
230 			MRVL_LOG(ERR, "BPool init failed!\n");
231 			used_bpools[i] &= ~(1 << dummy_pool_id[i]);
232 			return -1;
233 		}
234 	}
235 
236 	return 0;
237 }
238 
239 /**
240  * Initialize packet processor.
241  *
242  * @return
243  *   0 on success, negative error value otherwise.
244  */
245 static int
246 mrvl_init_pp2(void)
247 {
248 	struct pp2_init_params	init_params;
249 	int			err;
250 
251 	memset(&init_params, 0, sizeof(init_params));
252 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
253 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
254 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
255 	if (mrvl_cfg && mrvl_cfg->pp2_cfg.prs_udfs.num_udfs)
256 		memcpy(&init_params.prs_udfs, &mrvl_cfg->pp2_cfg.prs_udfs,
257 		       sizeof(struct pp2_parse_udfs));
258 	err = pp2_init(&init_params);
259 	if (err != 0) {
260 		MRVL_LOG(ERR, "PP2 init failed");
261 		return -1;
262 	}
263 
264 	err = mrvl_pp2_fixup_init();
265 	if (err != 0) {
266 		MRVL_LOG(ERR, "PP2 fixup init failed");
267 		return -1;
268 	}
269 
270 	return 0;
271 }
272 
273 static void
274 mrvl_pp2_fixup_deinit(void)
275 {
276 	int i;
277 
278 	for (i = 0; i < PP2_NUM_PKT_PROC; i++) {
279 		if (!dummy_pool[i])
280 			continue;
281 		pp2_bpool_deinit(dummy_pool[i]);
282 		used_bpools[i] &= ~(1 << dummy_pool_id[i]);
283 	}
284 }
285 
286 /**
287  * Deinitialize packet processor.
288  *
289  * @return
290  *   0 on success, negative error value otherwise.
291  */
292 static void
293 mrvl_deinit_pp2(void)
294 {
295 	mrvl_pp2_fixup_deinit();
296 	pp2_deinit();
297 }
298 
299 static inline void
300 mrvl_fill_shadowq(struct mrvl_shadow_txq *sq, struct rte_mbuf *buf)
301 {
302 	sq->ent[sq->head].buff.cookie = (uint64_t)buf;
303 	sq->ent[sq->head].buff.addr = buf ?
304 		rte_mbuf_data_iova_default(buf) : 0;
305 
306 	sq->ent[sq->head].bpool =
307 		(unlikely(!buf || buf->port >= RTE_MAX_ETHPORTS ||
308 		 buf->refcnt > 1)) ? NULL :
309 		 mrvl_port_to_bpool_lookup[buf->port];
310 
311 	sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
312 	sq->size++;
313 }
314 
315 /**
316  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
317  */
318 static void
319 mrvl_deinit_hifs(void)
320 {
321 	int i;
322 
323 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
324 		if (hifs[i])
325 			pp2_hif_deinit(hifs[i]);
326 	}
327 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
328 	memset(hifs, 0, sizeof(hifs));
329 }
330 
331 static inline void
332 mrvl_fill_desc(struct pp2_ppio_desc *desc, struct rte_mbuf *buf)
333 {
334 	pp2_ppio_outq_desc_reset(desc);
335 	pp2_ppio_outq_desc_set_phys_addr(desc, rte_pktmbuf_iova(buf));
336 	pp2_ppio_outq_desc_set_pkt_offset(desc, 0);
337 	pp2_ppio_outq_desc_set_pkt_len(desc, rte_pktmbuf_data_len(buf));
338 }
339 
340 static inline int
341 mrvl_get_bpool_size(int pp2_id, int pool_id)
342 {
343 	int i;
344 	int size = 0;
345 
346 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
347 		size += mrvl_port_bpool_size[pp2_id][pool_id][i];
348 
349 	return size;
350 }
351 
352 static int
353 mrvl_init_hif(int core_id)
354 {
355 	struct pp2_hif_params params;
356 	char match[MRVL_MATCH_LEN];
357 	int ret;
358 
359 	ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
360 	if (ret < 0) {
361 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
362 		return ret;
363 	}
364 
365 	snprintf(match, sizeof(match), "hif-%d", ret);
366 	memset(&params, 0, sizeof(params));
367 	params.match = match;
368 	params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
369 	ret = pp2_hif_init(&params, &hifs[core_id]);
370 	if (ret) {
371 		MRVL_LOG(ERR, "Failed to initialize hif %d", core_id);
372 		return ret;
373 	}
374 
375 	return 0;
376 }
377 
378 static inline struct pp2_hif*
379 mrvl_get_hif(struct mrvl_priv *priv, int core_id)
380 {
381 	int ret;
382 
383 	if (likely(hifs[core_id] != NULL))
384 		return hifs[core_id];
385 
386 	rte_spinlock_lock(&priv->lock);
387 
388 	ret = mrvl_init_hif(core_id);
389 	if (ret < 0) {
390 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
391 		goto out;
392 	}
393 
394 	if (core_id < mrvl_lcore_first)
395 		mrvl_lcore_first = core_id;
396 
397 	if (core_id > mrvl_lcore_last)
398 		mrvl_lcore_last = core_id;
399 out:
400 	rte_spinlock_unlock(&priv->lock);
401 
402 	return hifs[core_id];
403 }
404 
405 /**
406  * Set tx burst function according to offload flag
407  *
408  * @param dev
409  *   Pointer to Ethernet device structure.
410  */
411 static void
412 mrvl_set_tx_function(struct rte_eth_dev *dev)
413 {
414 	struct mrvl_priv *priv = dev->data->dev_private;
415 
416 	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
417 	if (priv->multiseg) {
418 		MRVL_LOG(INFO, "Using multi-segment tx callback");
419 		dev->tx_pkt_burst = mrvl_tx_sg_pkt_burst;
420 	} else {
421 		MRVL_LOG(INFO, "Using single-segment tx callback");
422 		dev->tx_pkt_burst = mrvl_tx_pkt_burst;
423 	}
424 }
425 
426 /**
427  * Configure rss based on dpdk rss configuration.
428  *
429  * @param priv
430  *   Pointer to private structure.
431  * @param rss_conf
432  *   Pointer to RSS configuration.
433  *
434  * @return
435  *   0 on success, negative error value otherwise.
436  */
437 static int
438 mrvl_configure_rss(struct mrvl_priv *priv, struct rte_eth_rss_conf *rss_conf)
439 {
440 	if (rss_conf->rss_key)
441 		MRVL_LOG(WARNING, "Changing hash key is not supported");
442 
443 	if (rss_conf->rss_hf == 0) {
444 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
445 	} else if (rss_conf->rss_hf & RTE_ETH_RSS_IPV4) {
446 		priv->ppio_params.inqs_params.hash_type =
447 			PP2_PPIO_HASH_T_2_TUPLE;
448 	} else if (rss_conf->rss_hf & RTE_ETH_RSS_NONFRAG_IPV4_TCP) {
449 		priv->ppio_params.inqs_params.hash_type =
450 			PP2_PPIO_HASH_T_5_TUPLE;
451 		priv->rss_hf_tcp = 1;
452 	} else if (rss_conf->rss_hf & RTE_ETH_RSS_NONFRAG_IPV4_UDP) {
453 		priv->ppio_params.inqs_params.hash_type =
454 			PP2_PPIO_HASH_T_5_TUPLE;
455 		priv->rss_hf_tcp = 0;
456 	} else {
457 		return -EINVAL;
458 	}
459 
460 	return 0;
461 }
462 
463 /**
464  * Ethernet device configuration.
465  *
466  * Prepare the driver for a given number of TX and RX queues and
467  * configure RSS.
468  *
469  * @param dev
470  *   Pointer to Ethernet device structure.
471  *
472  * @return
473  *   0 on success, negative error value otherwise.
474  */
475 static int
476 mrvl_dev_configure(struct rte_eth_dev *dev)
477 {
478 	struct mrvl_priv *priv = dev->data->dev_private;
479 	int ret;
480 
481 	if (priv->ppio) {
482 		MRVL_LOG(INFO, "Device reconfiguration is not supported");
483 		return -EINVAL;
484 	}
485 
486 	if (dev->data->dev_conf.rxmode.mq_mode != RTE_ETH_MQ_RX_NONE &&
487 	    dev->data->dev_conf.rxmode.mq_mode != RTE_ETH_MQ_RX_RSS) {
488 		MRVL_LOG(INFO, "Unsupported rx multi queue mode %d",
489 			dev->data->dev_conf.rxmode.mq_mode);
490 		return -EINVAL;
491 	}
492 
493 	if (dev->data->dev_conf.rxmode.mtu > priv->max_mtu) {
494 		MRVL_LOG(ERR, "MTU %u is larger than max_mtu %u\n",
495 			 dev->data->dev_conf.rxmode.mtu,
496 			 priv->max_mtu);
497 		return -EINVAL;
498 	}
499 
500 	if (dev->data->dev_conf.txmode.offloads & RTE_ETH_TX_OFFLOAD_MULTI_SEGS)
501 		priv->multiseg = 1;
502 
503 	ret = mrvl_configure_rxqs(priv, dev->data->port_id,
504 				  dev->data->nb_rx_queues);
505 	if (ret < 0)
506 		return ret;
507 
508 	ret = mrvl_configure_txqs(priv, dev->data->port_id,
509 				  dev->data->nb_tx_queues);
510 	if (ret < 0)
511 		return ret;
512 
513 	priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
514 	priv->ppio_params.maintain_stats = 1;
515 	priv->nb_rx_queues = dev->data->nb_rx_queues;
516 
517 	ret = mrvl_tm_init(dev);
518 	if (ret < 0)
519 		return ret;
520 
521 	if (dev->data->nb_rx_queues == 1 &&
522 	    dev->data->dev_conf.rxmode.mq_mode == RTE_ETH_MQ_RX_RSS) {
523 		MRVL_LOG(WARNING, "Disabling hash for 1 rx queue");
524 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
525 		priv->configured = 1;
526 		return 0;
527 	}
528 
529 	ret = mrvl_configure_rss(priv,
530 			&dev->data->dev_conf.rx_adv_conf.rss_conf);
531 	if (ret < 0)
532 		return ret;
533 
534 	priv->configured = 1;
535 
536 	return 0;
537 }
538 
539 /**
540  * DPDK callback to change the MTU.
541  *
542  * Setting the MTU affects hardware MRU (packets larger than the MRU
543  * will be dropped).
544  *
545  * @param dev
546  *   Pointer to Ethernet device structure.
547  * @param mtu
548  *   New MTU.
549  *
550  * @return
551  *   0 on success, negative error value otherwise.
552  */
553 static int
554 mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
555 {
556 	struct mrvl_priv *priv = dev->data->dev_private;
557 	uint16_t mru;
558 	uint16_t mbuf_data_size = 0; /* SW buffer size */
559 	int ret;
560 
561 	mru = MRVL_PP2_MTU_TO_MRU(mtu);
562 	/*
563 	 * min_rx_buf_size is equal to mbuf data size
564 	 * if pmd didn't set it differently
565 	 */
566 	mbuf_data_size = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
567 	/* Prevent PMD from:
568 	 * - setting mru greater than the mbuf size resulting in
569 	 * hw and sw buffer size mismatch
570 	 * - setting mtu that requires the support of scattered packets
571 	 * when this feature has not been enabled/supported so far
572 	 * (TODO check scattered_rx flag here once scattered RX is supported).
573 	 */
574 	if (mru - RTE_ETHER_CRC_LEN + MRVL_PKT_OFFS > mbuf_data_size) {
575 		mru = mbuf_data_size + RTE_ETHER_CRC_LEN - MRVL_PKT_OFFS;
576 		mtu = MRVL_PP2_MRU_TO_MTU(mru);
577 		MRVL_LOG(WARNING, "MTU too big, max MTU possible limited "
578 			"by current mbuf size: %u. Set MTU to %u, MRU to %u",
579 			mbuf_data_size, mtu, mru);
580 	}
581 
582 	if (mtu < RTE_ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX) {
583 		MRVL_LOG(ERR, "Invalid MTU [%u] or MRU [%u]", mtu, mru);
584 		return -EINVAL;
585 	}
586 
587 	if (!priv->ppio)
588 		return 0;
589 
590 	ret = pp2_ppio_set_mru(priv->ppio, mru);
591 	if (ret) {
592 		MRVL_LOG(ERR, "Failed to change MRU");
593 		return ret;
594 	}
595 
596 	ret = pp2_ppio_set_mtu(priv->ppio, mtu);
597 	if (ret) {
598 		MRVL_LOG(ERR, "Failed to change MTU");
599 		return ret;
600 	}
601 
602 	return 0;
603 }
604 
605 /**
606  * DPDK callback to bring the link up.
607  *
608  * @param dev
609  *   Pointer to Ethernet device structure.
610  *
611  * @return
612  *   0 on success, negative error value otherwise.
613  */
614 static int
615 mrvl_dev_set_link_up(struct rte_eth_dev *dev)
616 {
617 	struct mrvl_priv *priv = dev->data->dev_private;
618 	int ret;
619 
620 	if (!priv->ppio) {
621 		dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
622 		return 0;
623 	}
624 
625 	ret = pp2_ppio_enable(priv->ppio);
626 	if (ret)
627 		return ret;
628 
629 	/*
630 	 * mtu/mru can be updated if pp2_ppio_enable() was called at least once
631 	 * as pp2_ppio_enable() changes port->t_mode from default 0 to
632 	 * PP2_TRAFFIC_INGRESS_EGRESS.
633 	 *
634 	 * Set mtu to default DPDK value here.
635 	 */
636 	ret = mrvl_mtu_set(dev, dev->data->mtu);
637 	if (ret) {
638 		pp2_ppio_disable(priv->ppio);
639 		return ret;
640 	}
641 
642 	dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
643 	return 0;
644 }
645 
646 /**
647  * DPDK callback to bring the link down.
648  *
649  * @param dev
650  *   Pointer to Ethernet device structure.
651  *
652  * @return
653  *   0 on success, negative error value otherwise.
654  */
655 static int
656 mrvl_dev_set_link_down(struct rte_eth_dev *dev)
657 {
658 	struct mrvl_priv *priv = dev->data->dev_private;
659 	int ret;
660 
661 	if (!priv->ppio) {
662 		dev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
663 		return 0;
664 	}
665 	ret = pp2_ppio_disable(priv->ppio);
666 	if (ret)
667 		return ret;
668 
669 	dev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
670 	return 0;
671 }
672 
673 /**
674  * DPDK callback to start tx queue.
675  *
676  * @param dev
677  *   Pointer to Ethernet device structure.
678  * @param queue_id
679  *   Transmit queue index.
680  *
681  * @return
682  *   0 on success, negative error value otherwise.
683  */
684 static int
685 mrvl_tx_queue_start(struct rte_eth_dev *dev, uint16_t queue_id)
686 {
687 	struct mrvl_priv *priv = dev->data->dev_private;
688 	int ret;
689 
690 	if (!priv)
691 		return -EPERM;
692 
693 	/* passing 1 enables given tx queue */
694 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 1);
695 	if (ret) {
696 		MRVL_LOG(ERR, "Failed to start txq %d", queue_id);
697 		return ret;
698 	}
699 
700 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
701 
702 	return 0;
703 }
704 
705 /**
706  * DPDK callback to stop tx queue.
707  *
708  * @param dev
709  *   Pointer to Ethernet device structure.
710  * @param queue_id
711  *   Transmit queue index.
712  *
713  * @return
714  *   0 on success, negative error value otherwise.
715  */
716 static int
717 mrvl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t queue_id)
718 {
719 	struct mrvl_priv *priv = dev->data->dev_private;
720 	int ret;
721 
722 	if (!priv->ppio)
723 		return -EPERM;
724 
725 	/* passing 0 disables given tx queue */
726 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 0);
727 	if (ret) {
728 		MRVL_LOG(ERR, "Failed to stop txq %d", queue_id);
729 		return ret;
730 	}
731 
732 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
733 
734 	return 0;
735 }
736 
737 /**
738  * Populate VLAN Filter configuration.
739  *
740  * @param dev
741  *   Pointer to Ethernet device structure.
742  * @param on
743  *   Toggle filter.
744  *
745  * @return
746  *   0 on success, negative error value otherwise.
747  */
748 static int mrvl_populate_vlan_table(struct rte_eth_dev *dev, int on)
749 {
750 	uint32_t j;
751 	int ret;
752 	struct rte_vlan_filter_conf *vfc;
753 
754 	vfc = &dev->data->vlan_filter_conf;
755 	for (j = 0; j < RTE_DIM(vfc->ids); j++) {
756 		uint64_t vlan;
757 		uint64_t vbit;
758 		uint64_t ids = vfc->ids[j];
759 
760 		if (ids == 0)
761 			continue;
762 
763 		while (ids) {
764 			vlan = 64 * j;
765 			/* count trailing zeroes */
766 			vbit = ~ids & (ids - 1);
767 			/* clear least significant bit set */
768 			ids ^= (ids ^ (ids - 1)) ^ vbit;
769 			for (; vbit; vlan++)
770 				vbit >>= 1;
771 			ret = mrvl_vlan_filter_set(dev, vlan, on);
772 			if (ret) {
773 				MRVL_LOG(ERR, "Failed to setup VLAN filter\n");
774 				return ret;
775 			}
776 		}
777 	}
778 
779 	return 0;
780 }
781 
782 /**
783  * DPDK callback to start the device.
784  *
785  * @param dev
786  *   Pointer to Ethernet device structure.
787  *
788  * @return
789  *   0 on success, negative errno value on failure.
790  */
791 static int
792 mrvl_dev_start(struct rte_eth_dev *dev)
793 {
794 	struct mrvl_priv *priv = dev->data->dev_private;
795 	char match[MRVL_MATCH_LEN];
796 	int ret = 0, i, def_init_size;
797 	struct rte_ether_addr *mac_addr;
798 
799 	if (priv->ppio)
800 		return mrvl_dev_set_link_up(dev);
801 
802 	snprintf(match, sizeof(match), "ppio-%d:%d",
803 		 priv->pp_id, priv->ppio_id);
804 	priv->ppio_params.match = match;
805 	priv->ppio_params.eth_start_hdr = PP2_PPIO_HDR_ETH;
806 	priv->forward_bad_frames = 0;
807 	priv->fill_bpool_buffs = MRVL_BURST_SIZE;
808 
809 	if (mrvl_cfg) {
810 		priv->ppio_params.eth_start_hdr =
811 			mrvl_cfg->port[dev->data->port_id].eth_start_hdr;
812 		priv->forward_bad_frames =
813 			mrvl_cfg->port[dev->data->port_id].forward_bad_frames;
814 		priv->fill_bpool_buffs =
815 			mrvl_cfg->port[dev->data->port_id].fill_bpool_buffs;
816 	}
817 
818 	/*
819 	 * Calculate the minimum bpool size for refill feature as follows:
820 	 * 2 default burst sizes multiply by number of rx queues.
821 	 * If the bpool size will be below this value, new buffers will
822 	 * be added to the pool.
823 	 */
824 	priv->bpool_min_size = priv->nb_rx_queues * MRVL_BURST_SIZE * 2;
825 
826 	/* In case initial bpool size configured in queues setup is
827 	 * smaller than minimum size add more buffers
828 	 */
829 	def_init_size = priv->bpool_min_size + MRVL_BURST_SIZE * 2;
830 	if (priv->bpool_init_size < def_init_size) {
831 		int buffs_to_add = def_init_size - priv->bpool_init_size;
832 
833 		priv->bpool_init_size += buffs_to_add;
834 		ret = mrvl_fill_bpool(dev->data->rx_queues[0], buffs_to_add);
835 		if (ret)
836 			MRVL_LOG(ERR, "Failed to add buffers to bpool");
837 	}
838 
839 	/*
840 	 * Calculate the maximum bpool size for refill feature as follows:
841 	 * maximum number of descriptors in rx queue multiply by number
842 	 * of rx queues plus minimum bpool size.
843 	 * In case the bpool size will exceed this value, superfluous buffers
844 	 * will be removed
845 	 */
846 	priv->bpool_max_size = (priv->nb_rx_queues * MRVL_PP2_RXD_MAX) +
847 				priv->bpool_min_size;
848 
849 	ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
850 	if (ret) {
851 		MRVL_LOG(ERR, "Failed to init ppio");
852 		return ret;
853 	}
854 
855 	/*
856 	 * In case there are some stale uc/mc mac addresses flush them
857 	 * here. It cannot be done during mrvl_dev_close() as port information
858 	 * is already gone at that point (due to pp2_ppio_deinit() in
859 	 * mrvl_dev_stop()).
860 	 */
861 	if (!priv->uc_mc_flushed) {
862 		ret = pp2_ppio_flush_mac_addrs(priv->ppio, 1, 1);
863 		if (ret) {
864 			MRVL_LOG(ERR,
865 				"Failed to flush uc/mc filter list");
866 			goto out;
867 		}
868 		priv->uc_mc_flushed = 1;
869 	}
870 
871 	ret = mrvl_mtu_set(dev, dev->data->mtu);
872 	if (ret)
873 		MRVL_LOG(ERR, "Failed to set MTU to %d", dev->data->mtu);
874 
875 	if (!rte_is_zero_ether_addr(&dev->data->mac_addrs[0]))
876 		mrvl_mac_addr_set(dev, &dev->data->mac_addrs[0]);
877 
878 	for (i = 1; i < MRVL_MAC_ADDRS_MAX; i++) {
879 		mac_addr = &dev->data->mac_addrs[i];
880 
881 		/* skip zero address */
882 		if (rte_is_zero_ether_addr(mac_addr))
883 			continue;
884 
885 		mrvl_mac_addr_add(dev, mac_addr, i, 0);
886 	}
887 
888 	if (dev->data->all_multicast == 1)
889 		mrvl_allmulticast_enable(dev);
890 
891 	if (dev->data->dev_conf.rxmode.offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER) {
892 		ret = mrvl_populate_vlan_table(dev, 1);
893 		if (ret) {
894 			MRVL_LOG(ERR, "Failed to populate VLAN table");
895 			goto out;
896 		}
897 	}
898 
899 	/* For default QoS config, don't start classifier. */
900 	if (mrvl_cfg  &&
901 	    mrvl_cfg->port[dev->data->port_id].use_qos_global_defaults == 0) {
902 		ret = mrvl_start_qos_mapping(priv);
903 		if (ret) {
904 			MRVL_LOG(ERR, "Failed to setup QoS mapping");
905 			goto out;
906 		}
907 	}
908 
909 	ret = pp2_ppio_set_loopback(priv->ppio, dev->data->dev_conf.lpbk_mode);
910 	if (ret) {
911 		MRVL_LOG(ERR, "Failed to set loopback");
912 		goto out;
913 	}
914 
915 	if (dev->data->promiscuous == 1)
916 		mrvl_promiscuous_enable(dev);
917 
918 	if (priv->flow_ctrl) {
919 		ret = mrvl_flow_ctrl_set(dev, &priv->fc_conf);
920 		if (ret) {
921 			MRVL_LOG(ERR, "Failed to configure flow control");
922 			goto out;
923 		}
924 		priv->flow_ctrl = 0;
925 	}
926 
927 	if (dev->data->dev_link.link_status == RTE_ETH_LINK_UP) {
928 		ret = mrvl_dev_set_link_up(dev);
929 		if (ret) {
930 			MRVL_LOG(ERR, "Failed to set link up");
931 			dev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
932 			goto out;
933 		}
934 	}
935 
936 	/* start tx queues */
937 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
938 		struct mrvl_txq *txq = dev->data->tx_queues[i];
939 
940 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
941 
942 		if (!txq->tx_deferred_start)
943 			continue;
944 
945 		/*
946 		 * All txqs are started by default. Stop them
947 		 * so that tx_deferred_start works as expected.
948 		 */
949 		ret = mrvl_tx_queue_stop(dev, i);
950 		if (ret)
951 			goto out;
952 	}
953 
954 	for (i = 0; i < dev->data->nb_rx_queues; i++)
955 		dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
956 
957 	mrvl_flow_init(dev);
958 	mrvl_mtr_init(dev);
959 	mrvl_set_tx_function(dev);
960 
961 	return 0;
962 out:
963 	MRVL_LOG(ERR, "Failed to start device");
964 	pp2_ppio_deinit(priv->ppio);
965 	return ret;
966 }
967 
968 /**
969  * Flush receive queues.
970  *
971  * @param dev
972  *   Pointer to Ethernet device structure.
973  */
974 static void
975 mrvl_flush_rx_queues(struct rte_eth_dev *dev)
976 {
977 	int i;
978 
979 	MRVL_LOG(INFO, "Flushing rx queues");
980 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
981 		int ret, num;
982 
983 		do {
984 			struct mrvl_rxq *q = dev->data->rx_queues[i];
985 			struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];
986 
987 			num = MRVL_PP2_RXD_MAX;
988 			ret = pp2_ppio_recv(q->priv->ppio,
989 					    q->priv->rxq_map[q->queue_id].tc,
990 					    q->priv->rxq_map[q->queue_id].inq,
991 					    descs, (uint16_t *)&num);
992 		} while (ret == 0 && num);
993 	}
994 }
995 
996 /**
997  * Flush transmit shadow queues.
998  *
999  * @param dev
1000  *   Pointer to Ethernet device structure.
1001  */
1002 static void
1003 mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
1004 {
1005 	int i, j;
1006 	struct mrvl_txq *txq;
1007 
1008 	MRVL_LOG(INFO, "Flushing tx shadow queues");
1009 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1010 		txq = (struct mrvl_txq *)dev->data->tx_queues[i];
1011 
1012 		for (j = 0; j < RTE_MAX_LCORE; j++) {
1013 			struct mrvl_shadow_txq *sq;
1014 
1015 			if (!hifs[j])
1016 				continue;
1017 
1018 			sq = &txq->shadow_txqs[j];
1019 			mrvl_free_sent_buffers(txq->priv->ppio,
1020 				hifs[j], j, sq, txq->queue_id, 1);
1021 			while (sq->tail != sq->head) {
1022 				uint64_t addr = cookie_addr_high |
1023 					sq->ent[sq->tail].buff.cookie;
1024 				rte_pktmbuf_free(
1025 					(struct rte_mbuf *)addr);
1026 				sq->tail = (sq->tail + 1) &
1027 					    MRVL_PP2_TX_SHADOWQ_MASK;
1028 			}
1029 			memset(sq, 0, sizeof(*sq));
1030 		}
1031 	}
1032 }
1033 
1034 /**
1035  * Flush hardware bpool (buffer-pool).
1036  *
1037  * @param dev
1038  *   Pointer to Ethernet device structure.
1039  */
1040 static void
1041 mrvl_flush_bpool(struct rte_eth_dev *dev)
1042 {
1043 	struct mrvl_priv *priv = dev->data->dev_private;
1044 	struct pp2_hif *hif;
1045 	uint32_t num;
1046 	int ret;
1047 	unsigned int core_id = rte_lcore_id();
1048 
1049 	if (core_id == LCORE_ID_ANY)
1050 		core_id = rte_get_main_lcore();
1051 
1052 	hif = mrvl_get_hif(priv, core_id);
1053 
1054 	ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
1055 	if (ret) {
1056 		MRVL_LOG(ERR, "Failed to get bpool buffers number");
1057 		return;
1058 	}
1059 
1060 	while (num--) {
1061 		struct pp2_buff_inf inf;
1062 		uint64_t addr;
1063 
1064 		ret = pp2_bpool_get_buff(hif, priv->bpool, &inf);
1065 		if (ret)
1066 			break;
1067 
1068 		addr = cookie_addr_high | inf.cookie;
1069 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1070 	}
1071 }
1072 
1073 /**
1074  * DPDK callback to stop the device.
1075  *
1076  * @param dev
1077  *   Pointer to Ethernet device structure.
1078  */
1079 static int
1080 mrvl_dev_stop(struct rte_eth_dev *dev)
1081 {
1082 	uint16_t i;
1083 
1084 	for (i = 0; i < dev->data->nb_rx_queues; i++)
1085 		dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
1086 	for (i = 0; i < dev->data->nb_tx_queues; i++)
1087 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
1088 
1089 	return mrvl_dev_set_link_down(dev);
1090 }
1091 
1092 /**
1093  * DPDK callback to close the device.
1094  *
1095  * @param dev
1096  *   Pointer to Ethernet device structure.
1097  */
1098 static int
1099 mrvl_dev_close(struct rte_eth_dev *dev)
1100 {
1101 	struct mrvl_priv *priv = dev->data->dev_private;
1102 	size_t i;
1103 
1104 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1105 		return 0;
1106 
1107 	mrvl_flush_rx_queues(dev);
1108 	mrvl_flush_tx_shadow_queues(dev);
1109 	mrvl_flow_deinit(dev);
1110 	mrvl_mtr_deinit(dev);
1111 
1112 	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
1113 		struct pp2_ppio_tc_params *tc_params =
1114 			&priv->ppio_params.inqs_params.tcs_params[i];
1115 
1116 		if (tc_params->inqs_params) {
1117 			rte_free(tc_params->inqs_params);
1118 			tc_params->inqs_params = NULL;
1119 		}
1120 	}
1121 
1122 	if (priv->cls_tbl) {
1123 		pp2_cls_tbl_deinit(priv->cls_tbl);
1124 		priv->cls_tbl = NULL;
1125 	}
1126 
1127 	if (priv->qos_tbl) {
1128 		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
1129 		priv->qos_tbl = NULL;
1130 	}
1131 
1132 	mrvl_flush_bpool(dev);
1133 	mrvl_tm_deinit(dev);
1134 
1135 	if (priv->ppio) {
1136 		pp2_ppio_deinit(priv->ppio);
1137 		priv->ppio = NULL;
1138 	}
1139 
1140 	/* policer must be released after ppio deinitialization */
1141 	if (priv->default_policer) {
1142 		pp2_cls_plcr_deinit(priv->default_policer);
1143 		priv->default_policer = NULL;
1144 	}
1145 
1146 
1147 	if (priv->bpool) {
1148 		pp2_bpool_deinit(priv->bpool);
1149 		used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
1150 		priv->bpool = NULL;
1151 	}
1152 
1153 	mrvl_dev_num--;
1154 
1155 	if (mrvl_dev_num == 0) {
1156 		MRVL_LOG(INFO, "Perform MUSDK deinit");
1157 		mrvl_deinit_hifs();
1158 		mrvl_deinit_pp2();
1159 		rte_mvep_deinit(MVEP_MOD_T_PP2);
1160 	}
1161 
1162 	return 0;
1163 }
1164 
1165 /**
1166  * DPDK callback to retrieve physical link information.
1167  *
1168  * @param dev
1169  *   Pointer to Ethernet device structure.
1170  * @param wait_to_complete
1171  *   Wait for request completion (ignored).
1172  *
1173  * @return
1174  *   0 on success, negative error value otherwise.
1175  */
1176 static int
1177 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1178 {
1179 	/*
1180 	 * TODO
1181 	 * once MUSDK provides necessary API use it here
1182 	 */
1183 	struct mrvl_priv *priv = dev->data->dev_private;
1184 	struct ethtool_cmd edata;
1185 	struct ifreq req;
1186 	int ret, fd, link_up;
1187 
1188 	if (!priv->ppio)
1189 		return -EPERM;
1190 
1191 	edata.cmd = ETHTOOL_GSET;
1192 
1193 	strcpy(req.ifr_name, dev->data->name);
1194 	req.ifr_data = (void *)&edata;
1195 
1196 	fd = socket(AF_INET, SOCK_DGRAM, 0);
1197 	if (fd == -1)
1198 		return -EFAULT;
1199 
1200 	ret = ioctl(fd, SIOCETHTOOL, &req);
1201 	if (ret == -1) {
1202 		close(fd);
1203 		return -EFAULT;
1204 	}
1205 
1206 	close(fd);
1207 
1208 	switch (ethtool_cmd_speed(&edata)) {
1209 	case SPEED_10:
1210 		dev->data->dev_link.link_speed = RTE_ETH_SPEED_NUM_10M;
1211 		break;
1212 	case SPEED_100:
1213 		dev->data->dev_link.link_speed = RTE_ETH_SPEED_NUM_100M;
1214 		break;
1215 	case SPEED_1000:
1216 		dev->data->dev_link.link_speed = RTE_ETH_SPEED_NUM_1G;
1217 		break;
1218 	case SPEED_2500:
1219 		dev->data->dev_link.link_speed = RTE_ETH_SPEED_NUM_2_5G;
1220 		break;
1221 	case SPEED_10000:
1222 		dev->data->dev_link.link_speed = RTE_ETH_SPEED_NUM_10G;
1223 		break;
1224 	default:
1225 		dev->data->dev_link.link_speed = RTE_ETH_SPEED_NUM_NONE;
1226 	}
1227 
1228 	dev->data->dev_link.link_duplex = edata.duplex ? RTE_ETH_LINK_FULL_DUPLEX :
1229 							 RTE_ETH_LINK_HALF_DUPLEX;
1230 	dev->data->dev_link.link_autoneg = edata.autoneg ? RTE_ETH_LINK_AUTONEG :
1231 							   RTE_ETH_LINK_FIXED;
1232 	pp2_ppio_get_link_state(priv->ppio, &link_up);
1233 	dev->data->dev_link.link_status = link_up ? RTE_ETH_LINK_UP : RTE_ETH_LINK_DOWN;
1234 
1235 	return 0;
1236 }
1237 
1238 /**
1239  * DPDK callback to enable promiscuous mode.
1240  *
1241  * @param dev
1242  *   Pointer to Ethernet device structure.
1243  *
1244  * @return
1245  *   0 on success, negative error value otherwise.
1246  */
1247 static int
1248 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
1249 {
1250 	struct mrvl_priv *priv = dev->data->dev_private;
1251 	int ret;
1252 
1253 	if (priv->isolated)
1254 		return -ENOTSUP;
1255 
1256 	if (!priv->ppio)
1257 		return 0;
1258 
1259 	ret = pp2_ppio_set_promisc(priv->ppio, 1);
1260 	if (ret) {
1261 		MRVL_LOG(ERR, "Failed to enable promiscuous mode");
1262 		return -EAGAIN;
1263 	}
1264 
1265 	return 0;
1266 }
1267 
1268 /**
1269  * DPDK callback to enable allmulti mode.
1270  *
1271  * @param dev
1272  *   Pointer to Ethernet device structure.
1273  *
1274  * @return
1275  *   0 on success, negative error value otherwise.
1276  */
1277 static int
1278 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
1279 {
1280 	struct mrvl_priv *priv = dev->data->dev_private;
1281 	int ret;
1282 
1283 	if (priv->isolated)
1284 		return -ENOTSUP;
1285 
1286 	if (!priv->ppio)
1287 		return 0;
1288 
1289 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
1290 	if (ret) {
1291 		MRVL_LOG(ERR, "Failed enable all-multicast mode");
1292 		return -EAGAIN;
1293 	}
1294 
1295 	return 0;
1296 }
1297 
1298 /**
1299  * DPDK callback to disable promiscuous mode.
1300  *
1301  * @param dev
1302  *   Pointer to Ethernet device structure.
1303  *
1304  * @return
1305  *   0 on success, negative error value otherwise.
1306  */
1307 static int
1308 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
1309 {
1310 	struct mrvl_priv *priv = dev->data->dev_private;
1311 	int ret;
1312 
1313 	if (priv->isolated)
1314 		return -ENOTSUP;
1315 
1316 	if (!priv->ppio)
1317 		return 0;
1318 
1319 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
1320 	if (ret) {
1321 		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
1322 		return -EAGAIN;
1323 	}
1324 
1325 	return 0;
1326 }
1327 
1328 /**
1329  * DPDK callback to disable allmulticast mode.
1330  *
1331  * @param dev
1332  *   Pointer to Ethernet device structure.
1333  *
1334  * @return
1335  *   0 on success, negative error value otherwise.
1336  */
1337 static int
1338 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
1339 {
1340 	struct mrvl_priv *priv = dev->data->dev_private;
1341 	int ret;
1342 
1343 	if (priv->isolated)
1344 		return -ENOTSUP;
1345 
1346 	if (!priv->ppio)
1347 		return 0;
1348 
1349 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
1350 	if (ret) {
1351 		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
1352 		return -EAGAIN;
1353 	}
1354 
1355 	return 0;
1356 }
1357 
1358 /**
1359  * DPDK callback to remove a MAC address.
1360  *
1361  * @param dev
1362  *   Pointer to Ethernet device structure.
1363  * @param index
1364  *   MAC address index.
1365  */
1366 static void
1367 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1368 {
1369 	struct mrvl_priv *priv = dev->data->dev_private;
1370 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1371 	int ret;
1372 
1373 	if (priv->isolated)
1374 		return;
1375 
1376 	if (!priv->ppio)
1377 		return;
1378 
1379 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
1380 				       dev->data->mac_addrs[index].addr_bytes);
1381 	if (ret) {
1382 		rte_ether_format_addr(buf, sizeof(buf),
1383 				  &dev->data->mac_addrs[index]);
1384 		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
1385 	}
1386 }
1387 
1388 /**
1389  * DPDK callback to add a MAC address.
1390  *
1391  * @param dev
1392  *   Pointer to Ethernet device structure.
1393  * @param mac_addr
1394  *   MAC address to register.
1395  * @param index
1396  *   MAC address index.
1397  * @param vmdq
1398  *   VMDq pool index to associate address with (unused).
1399  *
1400  * @return
1401  *   0 on success, negative error value otherwise.
1402  */
1403 static int
1404 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1405 		  uint32_t index, uint32_t vmdq __rte_unused)
1406 {
1407 	struct mrvl_priv *priv = dev->data->dev_private;
1408 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1409 	int ret;
1410 
1411 	if (priv->isolated)
1412 		return -ENOTSUP;
1413 
1414 	if (!priv->ppio)
1415 		return 0;
1416 
1417 	if (index == 0)
1418 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1419 		return -1;
1420 
1421 	/*
1422 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1423 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1424 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1425 	 * 21 respectively.
1426 	 *
1427 	 * If more than uc_filter_max uc addresses were added to filter list
1428 	 * then NIC will switch to promiscuous mode automatically.
1429 	 *
1430 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1431 	 * were added to filter list then NIC will switch to all-multicast mode
1432 	 * automatically.
1433 	 */
1434 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1435 	if (ret) {
1436 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1437 		MRVL_LOG(ERR, "Failed to add mac %s", buf);
1438 		return -1;
1439 	}
1440 
1441 	return 0;
1442 }
1443 
1444 /**
1445  * DPDK callback to set the primary MAC address.
1446  *
1447  * @param dev
1448  *   Pointer to Ethernet device structure.
1449  * @param mac_addr
1450  *   MAC address to register.
1451  *
1452  * @return
1453  *   0 on success, negative error value otherwise.
1454  */
1455 static int
1456 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1457 {
1458 	struct mrvl_priv *priv = dev->data->dev_private;
1459 	int ret;
1460 
1461 	if (priv->isolated)
1462 		return -ENOTSUP;
1463 
1464 	if (!priv->ppio)
1465 		return 0;
1466 
1467 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1468 	if (ret) {
1469 		char buf[RTE_ETHER_ADDR_FMT_SIZE];
1470 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1471 		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1472 	}
1473 
1474 	return ret;
1475 }
1476 
1477 /**
1478  * DPDK callback to get device statistics.
1479  *
1480  * @param dev
1481  *   Pointer to Ethernet device structure.
1482  * @param stats
1483  *   Stats structure output buffer.
1484  *
1485  * @return
1486  *   0 on success, negative error value otherwise.
1487  */
1488 static int
1489 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1490 {
1491 	struct mrvl_priv *priv = dev->data->dev_private;
1492 	struct pp2_ppio_statistics ppio_stats;
1493 	uint64_t drop_mac = 0;
1494 	unsigned int i, idx, ret;
1495 
1496 	if (!priv->ppio)
1497 		return -EPERM;
1498 
1499 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1500 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1501 		struct pp2_ppio_inq_statistics rx_stats;
1502 
1503 		if (!rxq)
1504 			continue;
1505 
1506 		idx = rxq->queue_id;
1507 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1508 			MRVL_LOG(ERR,
1509 				"rx queue %d stats out of range (0 - %d)",
1510 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1511 			continue;
1512 		}
1513 
1514 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1515 						  priv->rxq_map[idx].tc,
1516 						  priv->rxq_map[idx].inq,
1517 						  &rx_stats, 0);
1518 		if (unlikely(ret)) {
1519 			MRVL_LOG(ERR,
1520 				"Failed to update rx queue %d stats", idx);
1521 			break;
1522 		}
1523 
1524 		stats->q_ibytes[idx] = rxq->bytes_recv;
1525 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1526 		stats->q_errors[idx] = rx_stats.drop_early +
1527 				       rx_stats.drop_fullq +
1528 				       rx_stats.drop_bm +
1529 				       rxq->drop_mac;
1530 		stats->ibytes += rxq->bytes_recv;
1531 		drop_mac += rxq->drop_mac;
1532 	}
1533 
1534 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1535 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1536 		struct pp2_ppio_outq_statistics tx_stats;
1537 
1538 		if (!txq)
1539 			continue;
1540 
1541 		idx = txq->queue_id;
1542 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1543 			MRVL_LOG(ERR,
1544 				"tx queue %d stats out of range (0 - %d)",
1545 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1546 		}
1547 
1548 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1549 						   &tx_stats, 0);
1550 		if (unlikely(ret)) {
1551 			MRVL_LOG(ERR,
1552 				"Failed to update tx queue %d stats", idx);
1553 			break;
1554 		}
1555 
1556 		stats->q_opackets[idx] = tx_stats.deq_desc;
1557 		stats->q_obytes[idx] = txq->bytes_sent;
1558 		stats->obytes += txq->bytes_sent;
1559 	}
1560 
1561 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1562 	if (unlikely(ret)) {
1563 		MRVL_LOG(ERR, "Failed to update port statistics");
1564 		return ret;
1565 	}
1566 
1567 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1568 	stats->opackets += ppio_stats.tx_packets;
1569 	stats->imissed += ppio_stats.rx_fullq_dropped +
1570 			  ppio_stats.rx_bm_dropped +
1571 			  ppio_stats.rx_early_dropped +
1572 			  ppio_stats.rx_fifo_dropped +
1573 			  ppio_stats.rx_cls_dropped;
1574 	stats->ierrors = drop_mac;
1575 
1576 	return 0;
1577 }
1578 
1579 /**
1580  * DPDK callback to clear device statistics.
1581  *
1582  * @param dev
1583  *   Pointer to Ethernet device structure.
1584  *
1585  * @return
1586  *   0 on success, negative error value otherwise.
1587  */
1588 static int
1589 mrvl_stats_reset(struct rte_eth_dev *dev)
1590 {
1591 	struct mrvl_priv *priv = dev->data->dev_private;
1592 	int i;
1593 
1594 	if (!priv->ppio)
1595 		return 0;
1596 
1597 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1598 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1599 
1600 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1601 					    priv->rxq_map[i].inq, NULL, 1);
1602 		rxq->bytes_recv = 0;
1603 		rxq->drop_mac = 0;
1604 	}
1605 
1606 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1607 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1608 
1609 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1610 		txq->bytes_sent = 0;
1611 	}
1612 
1613 	return pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1614 }
1615 
1616 /**
1617  * DPDK callback to get extended statistics.
1618  *
1619  * @param dev
1620  *   Pointer to Ethernet device structure.
1621  * @param stats
1622  *   Pointer to xstats table.
1623  * @param n
1624  *   Number of entries in xstats table.
1625  * @return
1626  *   Negative value on error, number of read xstats otherwise.
1627  */
1628 static int
1629 mrvl_xstats_get(struct rte_eth_dev *dev,
1630 		struct rte_eth_xstat *stats, unsigned int n)
1631 {
1632 	struct mrvl_priv *priv = dev->data->dev_private;
1633 	struct pp2_ppio_statistics ppio_stats;
1634 	unsigned int i, count;
1635 
1636 	count = RTE_DIM(mrvl_xstats_tbl);
1637 	if (n < count)
1638 		return count;
1639 
1640 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1641 	for (i = 0; i < count; i++) {
1642 		uint64_t val;
1643 
1644 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1645 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1646 					    mrvl_xstats_tbl[i].offset);
1647 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1648 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1649 					    mrvl_xstats_tbl[i].offset);
1650 		else
1651 			return -EINVAL;
1652 
1653 		stats[i].id = i;
1654 		stats[i].value = val;
1655 	}
1656 
1657 	return count;
1658 }
1659 
1660 /**
1661  * DPDK callback to reset extended statistics.
1662  *
1663  * @param dev
1664  *   Pointer to Ethernet device structure.
1665  *
1666  * @return
1667  *   0 on success, negative error value otherwise.
1668  */
1669 static int
1670 mrvl_xstats_reset(struct rte_eth_dev *dev)
1671 {
1672 	return mrvl_stats_reset(dev);
1673 }
1674 
1675 /**
1676  * DPDK callback to get extended statistics names.
1677  *
1678  * @param dev (unused)
1679  *   Pointer to Ethernet device structure.
1680  * @param xstats_names
1681  *   Pointer to xstats names table.
1682  * @param size
1683  *   Size of the xstats names table.
1684  * @return
1685  *   Number of read names.
1686  */
1687 static int
1688 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1689 		      struct rte_eth_xstat_name *xstats_names,
1690 		      unsigned int size)
1691 {
1692 	unsigned int i;
1693 
1694 	if (!xstats_names)
1695 		return RTE_DIM(mrvl_xstats_tbl);
1696 
1697 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1698 		strlcpy(xstats_names[i].name, mrvl_xstats_tbl[i].name,
1699 			RTE_ETH_XSTATS_NAME_SIZE);
1700 
1701 	return size;
1702 }
1703 
1704 /**
1705  * DPDK callback to get information about the device.
1706  *
1707  * @param dev
1708  *   Pointer to Ethernet device structure (unused).
1709  * @param info
1710  *   Info structure output buffer.
1711  */
1712 static int
1713 mrvl_dev_infos_get(struct rte_eth_dev *dev,
1714 		   struct rte_eth_dev_info *info)
1715 {
1716 	struct mrvl_priv *priv = dev->data->dev_private;
1717 
1718 	info->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
1719 
1720 	info->speed_capa = RTE_ETH_LINK_SPEED_10M |
1721 			   RTE_ETH_LINK_SPEED_100M |
1722 			   RTE_ETH_LINK_SPEED_1G |
1723 			   RTE_ETH_LINK_SPEED_2_5G |
1724 			   RTE_ETH_LINK_SPEED_10G;
1725 
1726 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1727 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1728 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1729 
1730 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1731 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1732 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1733 
1734 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1735 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1736 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1737 
1738 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1739 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1740 
1741 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1742 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1743 
1744 	info->flow_type_rss_offloads = RTE_ETH_RSS_IPV4 |
1745 				       RTE_ETH_RSS_NONFRAG_IPV4_TCP |
1746 				       RTE_ETH_RSS_NONFRAG_IPV4_UDP;
1747 
1748 	/* By default packets are dropped if no descriptors are available */
1749 	info->default_rxconf.rx_drop_en = 1;
1750 
1751 	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1752 	info->max_mtu = priv->max_mtu;
1753 
1754 	return 0;
1755 }
1756 
1757 /**
1758  * Return supported packet types.
1759  *
1760  * @param dev
1761  *   Pointer to Ethernet device structure (unused).
1762  *
1763  * @return
1764  *   Const pointer to the table with supported packet types.
1765  */
1766 static const uint32_t *
1767 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1768 {
1769 	static const uint32_t ptypes[] = {
1770 		RTE_PTYPE_L2_ETHER,
1771 		RTE_PTYPE_L2_ETHER_VLAN,
1772 		RTE_PTYPE_L2_ETHER_QINQ,
1773 		RTE_PTYPE_L3_IPV4,
1774 		RTE_PTYPE_L3_IPV4_EXT,
1775 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1776 		RTE_PTYPE_L3_IPV6,
1777 		RTE_PTYPE_L3_IPV6_EXT,
1778 		RTE_PTYPE_L2_ETHER_ARP,
1779 		RTE_PTYPE_L4_TCP,
1780 		RTE_PTYPE_L4_UDP
1781 	};
1782 
1783 	return ptypes;
1784 }
1785 
1786 /**
1787  * DPDK callback to get information about specific receive queue.
1788  *
1789  * @param dev
1790  *   Pointer to Ethernet device structure.
1791  * @param rx_queue_id
1792  *   Receive queue index.
1793  * @param qinfo
1794  *   Receive queue information structure.
1795  */
1796 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1797 			      struct rte_eth_rxq_info *qinfo)
1798 {
1799 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1800 	struct mrvl_priv *priv = dev->data->dev_private;
1801 	int inq = priv->rxq_map[rx_queue_id].inq;
1802 	int tc = priv->rxq_map[rx_queue_id].tc;
1803 	struct pp2_ppio_tc_params *tc_params =
1804 		&priv->ppio_params.inqs_params.tcs_params[tc];
1805 
1806 	qinfo->mp = q->mp;
1807 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1808 }
1809 
1810 /**
1811  * DPDK callback to get information about specific transmit queue.
1812  *
1813  * @param dev
1814  *   Pointer to Ethernet device structure.
1815  * @param tx_queue_id
1816  *   Transmit queue index.
1817  * @param qinfo
1818  *   Transmit queue information structure.
1819  */
1820 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1821 			      struct rte_eth_txq_info *qinfo)
1822 {
1823 	struct mrvl_priv *priv = dev->data->dev_private;
1824 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1825 
1826 	qinfo->nb_desc =
1827 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1828 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1829 }
1830 
1831 /**
1832  * DPDK callback to Configure a VLAN filter.
1833  *
1834  * @param dev
1835  *   Pointer to Ethernet device structure.
1836  * @param vlan_id
1837  *   VLAN ID to filter.
1838  * @param on
1839  *   Toggle filter.
1840  *
1841  * @return
1842  *   0 on success, negative error value otherwise.
1843  */
1844 static int
1845 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1846 {
1847 	struct mrvl_priv *priv = dev->data->dev_private;
1848 
1849 	if (priv->isolated)
1850 		return -ENOTSUP;
1851 
1852 	if (!priv->ppio)
1853 		return 0;
1854 
1855 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1856 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1857 }
1858 
1859 /**
1860  * DPDK callback to Configure VLAN offload.
1861  *
1862  * @param dev
1863  *   Pointer to Ethernet device structure.
1864  * @param mask
1865  *   VLAN offload mask.
1866  *
1867  * @return
1868  *   0 on success, negative error value otherwise.
1869  */
1870 static int mrvl_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1871 {
1872 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
1873 	int ret;
1874 
1875 	if (mask & RTE_ETH_VLAN_STRIP_MASK) {
1876 		MRVL_LOG(ERR, "VLAN stripping is not supported\n");
1877 		return -ENOTSUP;
1878 	}
1879 
1880 	if (mask & RTE_ETH_VLAN_FILTER_MASK) {
1881 		if (rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER)
1882 			ret = mrvl_populate_vlan_table(dev, 1);
1883 		else
1884 			ret = mrvl_populate_vlan_table(dev, 0);
1885 
1886 		if (ret)
1887 			return ret;
1888 	}
1889 
1890 	if (mask & RTE_ETH_VLAN_EXTEND_MASK) {
1891 		MRVL_LOG(ERR, "Extend VLAN not supported\n");
1892 		return -ENOTSUP;
1893 	}
1894 
1895 	return 0;
1896 }
1897 
1898 /**
1899  * Release buffers to hardware bpool (buffer-pool)
1900  *
1901  * @param rxq
1902  *   Receive queue pointer.
1903  * @param num
1904  *   Number of buffers to release to bpool.
1905  *
1906  * @return
1907  *   0 on success, negative error value otherwise.
1908  */
1909 static int
1910 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1911 {
1912 	struct buff_release_entry entries[num];
1913 	struct rte_mbuf *mbufs[num];
1914 	int i, ret;
1915 	unsigned int core_id;
1916 	struct pp2_hif *hif;
1917 	struct pp2_bpool *bpool;
1918 
1919 	core_id = rte_lcore_id();
1920 	if (core_id == LCORE_ID_ANY)
1921 		core_id = rte_get_main_lcore();
1922 
1923 	hif = mrvl_get_hif(rxq->priv, core_id);
1924 	if (!hif)
1925 		return -1;
1926 
1927 	bpool = rxq->priv->bpool;
1928 
1929 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1930 	if (ret)
1931 		return ret;
1932 
1933 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1934 		cookie_addr_high =
1935 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1936 
1937 	for (i = 0; i < num; i++) {
1938 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1939 			!= cookie_addr_high) {
1940 			MRVL_LOG(ERR,
1941 				"mbuf virtual addr high is out of range "
1942 				"0x%x instead of 0x%x\n",
1943 				(uint32_t)((uint64_t)mbufs[i] >> 32),
1944 				(uint32_t)(cookie_addr_high >> 32));
1945 			goto out;
1946 		}
1947 
1948 		entries[i].buff.addr =
1949 			rte_mbuf_data_iova_default(mbufs[i]);
1950 		entries[i].buff.cookie = (uintptr_t)mbufs[i];
1951 		entries[i].bpool = bpool;
1952 	}
1953 
1954 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1955 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1956 
1957 	if (i != num)
1958 		goto out;
1959 
1960 	return 0;
1961 out:
1962 	for (; i < num; i++)
1963 		rte_pktmbuf_free(mbufs[i]);
1964 
1965 	return -1;
1966 }
1967 
1968 /**
1969  * DPDK callback to configure the receive queue.
1970  *
1971  * @param dev
1972  *   Pointer to Ethernet device structure.
1973  * @param idx
1974  *   RX queue index.
1975  * @param desc
1976  *   Number of descriptors to configure in queue.
1977  * @param socket
1978  *   NUMA socket on which memory must be allocated.
1979  * @param conf
1980  *   Thresholds parameters.
1981  * @param mp
1982  *   Memory pool for buffer allocations.
1983  *
1984  * @return
1985  *   0 on success, negative error value otherwise.
1986  */
1987 static int
1988 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1989 		    unsigned int socket,
1990 		    const struct rte_eth_rxconf *conf,
1991 		    struct rte_mempool *mp)
1992 {
1993 	struct mrvl_priv *priv = dev->data->dev_private;
1994 	struct mrvl_rxq *rxq;
1995 	uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
1996 	uint32_t max_rx_pktlen = dev->data->mtu + RTE_ETHER_HDR_LEN;
1997 	int ret, tc, inq;
1998 	uint64_t offloads;
1999 
2000 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
2001 
2002 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
2003 		/*
2004 		 * Unknown TC mapping, mapping will not have a correct queue.
2005 		 */
2006 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
2007 			idx, priv->ppio_id);
2008 		return -EFAULT;
2009 	}
2010 
2011 	frame_size = buf_size - RTE_PKTMBUF_HEADROOM - MRVL_PKT_EFFEC_OFFS;
2012 	if (frame_size < max_rx_pktlen) {
2013 		MRVL_LOG(WARNING,
2014 			"Mbuf size must be increased to %u bytes to hold up "
2015 			"to %u bytes of data.",
2016 			max_rx_pktlen + buf_size - frame_size,
2017 			max_rx_pktlen);
2018 		dev->data->mtu = frame_size - RTE_ETHER_HDR_LEN;
2019 		MRVL_LOG(INFO, "Setting MTU to %u", dev->data->mtu);
2020 	}
2021 
2022 	if (dev->data->rx_queues[idx]) {
2023 		rte_free(dev->data->rx_queues[idx]);
2024 		dev->data->rx_queues[idx] = NULL;
2025 	}
2026 
2027 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
2028 	if (!rxq)
2029 		return -ENOMEM;
2030 
2031 	rxq->priv = priv;
2032 	rxq->mp = mp;
2033 	rxq->cksum_enabled = offloads & RTE_ETH_RX_OFFLOAD_IPV4_CKSUM;
2034 	rxq->queue_id = idx;
2035 	rxq->port_id = dev->data->port_id;
2036 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
2037 
2038 	tc = priv->rxq_map[rxq->queue_id].tc,
2039 	inq = priv->rxq_map[rxq->queue_id].inq;
2040 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
2041 		desc;
2042 
2043 	ret = mrvl_fill_bpool(rxq, desc);
2044 	if (ret) {
2045 		rte_free(rxq);
2046 		return ret;
2047 	}
2048 
2049 	priv->bpool_init_size += desc;
2050 
2051 	dev->data->rx_queues[idx] = rxq;
2052 
2053 	return 0;
2054 }
2055 
2056 /**
2057  * DPDK callback to release the receive queue.
2058  *
2059  * @param dev
2060  *   Pointer to Ethernet device structure.
2061  * @param qid
2062  *   Receive queue index.
2063  */
2064 static void
2065 mrvl_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
2066 {
2067 	struct mrvl_rxq *q = dev->data->rx_queues[qid];
2068 	struct pp2_ppio_tc_params *tc_params;
2069 	int i, num, tc, inq;
2070 	struct pp2_hif *hif;
2071 	unsigned int core_id = rte_lcore_id();
2072 
2073 	if (core_id == LCORE_ID_ANY)
2074 		core_id = rte_get_main_lcore();
2075 
2076 	if (!q)
2077 		return;
2078 
2079 	hif = mrvl_get_hif(q->priv, core_id);
2080 
2081 	if (!hif)
2082 		return;
2083 
2084 	tc = q->priv->rxq_map[q->queue_id].tc;
2085 	inq = q->priv->rxq_map[q->queue_id].inq;
2086 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
2087 	num = tc_params->inqs_params[inq].size;
2088 	for (i = 0; i < num; i++) {
2089 		struct pp2_buff_inf inf;
2090 		uint64_t addr;
2091 
2092 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
2093 		addr = cookie_addr_high | inf.cookie;
2094 		rte_pktmbuf_free((struct rte_mbuf *)addr);
2095 	}
2096 
2097 	rte_free(q);
2098 }
2099 
2100 /**
2101  * DPDK callback to configure the transmit queue.
2102  *
2103  * @param dev
2104  *   Pointer to Ethernet device structure.
2105  * @param idx
2106  *   Transmit queue index.
2107  * @param desc
2108  *   Number of descriptors to configure in the queue.
2109  * @param socket
2110  *   NUMA socket on which memory must be allocated.
2111  * @param conf
2112  *   Tx queue configuration parameters.
2113  *
2114  * @return
2115  *   0 on success, negative error value otherwise.
2116  */
2117 static int
2118 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
2119 		    unsigned int socket,
2120 		    const struct rte_eth_txconf *conf)
2121 {
2122 	struct mrvl_priv *priv = dev->data->dev_private;
2123 	struct mrvl_txq *txq;
2124 
2125 	if (dev->data->tx_queues[idx]) {
2126 		rte_free(dev->data->tx_queues[idx]);
2127 		dev->data->tx_queues[idx] = NULL;
2128 	}
2129 
2130 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
2131 	if (!txq)
2132 		return -ENOMEM;
2133 
2134 	txq->priv = priv;
2135 	txq->queue_id = idx;
2136 	txq->port_id = dev->data->port_id;
2137 	txq->tx_deferred_start = conf->tx_deferred_start;
2138 	dev->data->tx_queues[idx] = txq;
2139 
2140 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
2141 
2142 	return 0;
2143 }
2144 
2145 /**
2146  * DPDK callback to release the transmit queue.
2147  *
2148  * @param dev
2149  *   Pointer to Ethernet device structure.
2150  * @param qid
2151  *   Transmit queue index.
2152  */
2153 static void
2154 mrvl_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
2155 {
2156 	struct mrvl_txq *q = dev->data->tx_queues[qid];
2157 
2158 	if (!q)
2159 		return;
2160 
2161 	rte_free(q);
2162 }
2163 
2164 /**
2165  * DPDK callback to get flow control configuration.
2166  *
2167  * @param dev
2168  *  Pointer to Ethernet device structure.
2169  * @param fc_conf
2170  *  Pointer to the flow control configuration.
2171  *
2172  * @return
2173  *  0 on success, negative error value otherwise.
2174  */
2175 static int
2176 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2177 {
2178 	struct mrvl_priv *priv = dev->data->dev_private;
2179 	int ret, en;
2180 
2181 	if (!priv->ppio) {
2182 		memcpy(fc_conf, &priv->fc_conf, sizeof(struct rte_eth_fc_conf));
2183 		return 0;
2184 	}
2185 
2186 	fc_conf->autoneg = 1;
2187 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
2188 	if (ret) {
2189 		MRVL_LOG(ERR, "Failed to read rx pause state");
2190 		return ret;
2191 	}
2192 
2193 	fc_conf->mode = en ? RTE_ETH_FC_RX_PAUSE : RTE_ETH_FC_NONE;
2194 
2195 	ret = pp2_ppio_get_tx_pause(priv->ppio, &en);
2196 	if (ret) {
2197 		MRVL_LOG(ERR, "Failed to read tx pause state");
2198 		return ret;
2199 	}
2200 
2201 	if (en) {
2202 		if (fc_conf->mode == RTE_ETH_FC_NONE)
2203 			fc_conf->mode = RTE_ETH_FC_TX_PAUSE;
2204 		else
2205 			fc_conf->mode = RTE_ETH_FC_FULL;
2206 	}
2207 
2208 	return 0;
2209 }
2210 
2211 /**
2212  * DPDK callback to set flow control configuration.
2213  *
2214  * @param dev
2215  *  Pointer to Ethernet device structure.
2216  * @param fc_conf
2217  *  Pointer to the flow control configuration.
2218  *
2219  * @return
2220  *  0 on success, negative error value otherwise.
2221  */
2222 static int
2223 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2224 {
2225 	struct mrvl_priv *priv = dev->data->dev_private;
2226 	struct pp2_ppio_tx_pause_params mrvl_pause_params;
2227 	int ret;
2228 	int rx_en, tx_en;
2229 
2230 	if (fc_conf->high_water ||
2231 	    fc_conf->low_water ||
2232 	    fc_conf->pause_time ||
2233 	    fc_conf->mac_ctrl_frame_fwd) {
2234 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
2235 
2236 		return -EINVAL;
2237 	}
2238 
2239 	if (fc_conf->autoneg == 0) {
2240 		MRVL_LOG(ERR, "Flowctrl Autoneg disable is not supported");
2241 		return -EINVAL;
2242 	}
2243 
2244 	if (!priv->ppio) {
2245 		memcpy(&priv->fc_conf, fc_conf, sizeof(struct rte_eth_fc_conf));
2246 		priv->flow_ctrl = 1;
2247 		return 0;
2248 	}
2249 
2250 	switch (fc_conf->mode) {
2251 	case RTE_ETH_FC_FULL:
2252 		rx_en = 1;
2253 		tx_en = 1;
2254 		break;
2255 	case RTE_ETH_FC_TX_PAUSE:
2256 		rx_en = 0;
2257 		tx_en = 1;
2258 		break;
2259 	case RTE_ETH_FC_RX_PAUSE:
2260 		rx_en = 1;
2261 		tx_en = 0;
2262 		break;
2263 	case RTE_ETH_FC_NONE:
2264 		rx_en = 0;
2265 		tx_en = 0;
2266 		break;
2267 	default:
2268 		MRVL_LOG(ERR, "Incorrect Flow control flag (%d)",
2269 			 fc_conf->mode);
2270 		return -EINVAL;
2271 	}
2272 
2273 	/* Set RX flow control */
2274 	ret = pp2_ppio_set_rx_pause(priv->ppio, rx_en);
2275 	if (ret) {
2276 		MRVL_LOG(ERR, "Failed to change RX flowctrl");
2277 		return ret;
2278 	}
2279 
2280 	/* Set TX flow control */
2281 	mrvl_pause_params.en = tx_en;
2282 	/* all inqs participate in xon/xoff decision */
2283 	mrvl_pause_params.use_tc_pause_inqs = 0;
2284 	ret = pp2_ppio_set_tx_pause(priv->ppio, &mrvl_pause_params);
2285 	if (ret) {
2286 		MRVL_LOG(ERR, "Failed to change TX flowctrl");
2287 		return ret;
2288 	}
2289 
2290 	return 0;
2291 }
2292 
2293 /**
2294  * Update RSS hash configuration
2295  *
2296  * @param dev
2297  *   Pointer to Ethernet device structure.
2298  * @param rss_conf
2299  *   Pointer to RSS configuration.
2300  *
2301  * @return
2302  *   0 on success, negative error value otherwise.
2303  */
2304 static int
2305 mrvl_rss_hash_update(struct rte_eth_dev *dev,
2306 		     struct rte_eth_rss_conf *rss_conf)
2307 {
2308 	struct mrvl_priv *priv = dev->data->dev_private;
2309 
2310 	if (priv->isolated)
2311 		return -ENOTSUP;
2312 
2313 	return mrvl_configure_rss(priv, rss_conf);
2314 }
2315 
2316 /**
2317  * DPDK callback to get RSS hash configuration.
2318  *
2319  * @param dev
2320  *   Pointer to Ethernet device structure.
2321  * @rss_conf
2322  *   Pointer to RSS configuration.
2323  *
2324  * @return
2325  *   Always 0.
2326  */
2327 static int
2328 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
2329 		       struct rte_eth_rss_conf *rss_conf)
2330 {
2331 	struct mrvl_priv *priv = dev->data->dev_private;
2332 	enum pp2_ppio_hash_type hash_type =
2333 		priv->ppio_params.inqs_params.hash_type;
2334 
2335 	rss_conf->rss_key = NULL;
2336 
2337 	if (hash_type == PP2_PPIO_HASH_T_NONE)
2338 		rss_conf->rss_hf = 0;
2339 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
2340 		rss_conf->rss_hf = RTE_ETH_RSS_IPV4;
2341 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
2342 		rss_conf->rss_hf = RTE_ETH_RSS_NONFRAG_IPV4_TCP;
2343 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
2344 		rss_conf->rss_hf = RTE_ETH_RSS_NONFRAG_IPV4_UDP;
2345 
2346 	return 0;
2347 }
2348 
2349 /**
2350  * DPDK callback to get rte_flow callbacks.
2351  *
2352  * @param dev
2353  *   Pointer to the device structure.
2354  * @param ops
2355  *   Pointer to pass the flow ops.
2356  *
2357  * @return
2358  *   0 on success, negative error value otherwise.
2359  */
2360 static int
2361 mrvl_eth_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
2362 		      const struct rte_flow_ops **ops)
2363 {
2364 	*ops = &mrvl_flow_ops;
2365 	return 0;
2366 }
2367 
2368 /**
2369  * DPDK callback to get rte_mtr callbacks.
2370  *
2371  * @param dev
2372  *   Pointer to the device structure.
2373  * @param ops
2374  *   Pointer to pass the mtr ops.
2375  *
2376  * @return
2377  *   Always 0.
2378  */
2379 static int
2380 mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2381 {
2382 	*(const void **)ops = &mrvl_mtr_ops;
2383 
2384 	return 0;
2385 }
2386 
2387 /**
2388  * DPDK callback to get rte_tm callbacks.
2389  *
2390  * @param dev
2391  *   Pointer to the device structure.
2392  * @param ops
2393  *   Pointer to pass the tm ops.
2394  *
2395  * @return
2396  *   Always 0.
2397  */
2398 static int
2399 mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2400 {
2401 	*(const void **)ops = &mrvl_tm_ops;
2402 
2403 	return 0;
2404 }
2405 
2406 static const struct eth_dev_ops mrvl_ops = {
2407 	.dev_configure = mrvl_dev_configure,
2408 	.dev_start = mrvl_dev_start,
2409 	.dev_stop = mrvl_dev_stop,
2410 	.dev_set_link_up = mrvl_dev_set_link_up,
2411 	.dev_set_link_down = mrvl_dev_set_link_down,
2412 	.dev_close = mrvl_dev_close,
2413 	.link_update = mrvl_link_update,
2414 	.promiscuous_enable = mrvl_promiscuous_enable,
2415 	.allmulticast_enable = mrvl_allmulticast_enable,
2416 	.promiscuous_disable = mrvl_promiscuous_disable,
2417 	.allmulticast_disable = mrvl_allmulticast_disable,
2418 	.mac_addr_remove = mrvl_mac_addr_remove,
2419 	.mac_addr_add = mrvl_mac_addr_add,
2420 	.mac_addr_set = mrvl_mac_addr_set,
2421 	.mtu_set = mrvl_mtu_set,
2422 	.stats_get = mrvl_stats_get,
2423 	.stats_reset = mrvl_stats_reset,
2424 	.xstats_get = mrvl_xstats_get,
2425 	.xstats_reset = mrvl_xstats_reset,
2426 	.xstats_get_names = mrvl_xstats_get_names,
2427 	.dev_infos_get = mrvl_dev_infos_get,
2428 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
2429 	.rxq_info_get = mrvl_rxq_info_get,
2430 	.txq_info_get = mrvl_txq_info_get,
2431 	.vlan_filter_set = mrvl_vlan_filter_set,
2432 	.vlan_offload_set = mrvl_vlan_offload_set,
2433 	.tx_queue_start = mrvl_tx_queue_start,
2434 	.tx_queue_stop = mrvl_tx_queue_stop,
2435 	.rx_queue_setup = mrvl_rx_queue_setup,
2436 	.rx_queue_release = mrvl_rx_queue_release,
2437 	.tx_queue_setup = mrvl_tx_queue_setup,
2438 	.tx_queue_release = mrvl_tx_queue_release,
2439 	.flow_ctrl_get = mrvl_flow_ctrl_get,
2440 	.flow_ctrl_set = mrvl_flow_ctrl_set,
2441 	.rss_hash_update = mrvl_rss_hash_update,
2442 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
2443 	.flow_ops_get = mrvl_eth_flow_ops_get,
2444 	.mtr_ops_get = mrvl_mtr_ops_get,
2445 	.tm_ops_get = mrvl_tm_ops_get,
2446 };
2447 
2448 /**
2449  * Return packet type information and l3/l4 offsets.
2450  *
2451  * @param desc
2452  *   Pointer to the received packet descriptor.
2453  * @param l3_offset
2454  *   l3 packet offset.
2455  * @param l4_offset
2456  *   l4 packet offset.
2457  *
2458  * @return
2459  *   Packet type information.
2460  */
2461 static inline uint64_t
2462 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
2463 				    uint8_t *l3_offset, uint8_t *l4_offset)
2464 {
2465 	enum pp2_inq_l3_type l3_type;
2466 	enum pp2_inq_l4_type l4_type;
2467 	enum pp2_inq_vlan_tag vlan_tag;
2468 	uint64_t packet_type;
2469 
2470 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
2471 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
2472 	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
2473 
2474 	packet_type = RTE_PTYPE_L2_ETHER;
2475 
2476 	switch (vlan_tag) {
2477 	case PP2_INQ_VLAN_TAG_SINGLE:
2478 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
2479 		break;
2480 	case PP2_INQ_VLAN_TAG_DOUBLE:
2481 	case PP2_INQ_VLAN_TAG_TRIPLE:
2482 		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
2483 		break;
2484 	default:
2485 		break;
2486 	}
2487 
2488 	switch (l3_type) {
2489 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
2490 		packet_type |= RTE_PTYPE_L3_IPV4;
2491 		break;
2492 	case PP2_INQ_L3_TYPE_IPV4_OK:
2493 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
2494 		break;
2495 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
2496 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
2497 		break;
2498 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
2499 		packet_type |= RTE_PTYPE_L3_IPV6;
2500 		break;
2501 	case PP2_INQ_L3_TYPE_IPV6_EXT:
2502 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
2503 		break;
2504 	case PP2_INQ_L3_TYPE_ARP:
2505 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
2506 		/*
2507 		 * In case of ARP l4_offset is set to wrong value.
2508 		 * Set it to proper one so that later on mbuf->l3_len can be
2509 		 * calculated subtracting l4_offset and l3_offset.
2510 		 */
2511 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
2512 		break;
2513 	default:
2514 		break;
2515 	}
2516 
2517 	switch (l4_type) {
2518 	case PP2_INQ_L4_TYPE_TCP:
2519 		packet_type |= RTE_PTYPE_L4_TCP;
2520 		break;
2521 	case PP2_INQ_L4_TYPE_UDP:
2522 		packet_type |= RTE_PTYPE_L4_UDP;
2523 		break;
2524 	default:
2525 		break;
2526 	}
2527 
2528 	return packet_type;
2529 }
2530 
2531 /**
2532  * Get offload information from the received packet descriptor.
2533  *
2534  * @param desc
2535  *   Pointer to the received packet descriptor.
2536  *
2537  * @return
2538  *   Mbuf offload flags.
2539  */
2540 static inline uint64_t
2541 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc, uint64_t packet_type)
2542 {
2543 	uint64_t flags = 0;
2544 	enum pp2_inq_desc_status status;
2545 
2546 	if (RTE_ETH_IS_IPV4_HDR(packet_type)) {
2547 		status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2548 		if (unlikely(status != PP2_DESC_ERR_OK))
2549 			flags |= RTE_MBUF_F_RX_IP_CKSUM_BAD;
2550 		else
2551 			flags |= RTE_MBUF_F_RX_IP_CKSUM_GOOD;
2552 	}
2553 
2554 	if (((packet_type & RTE_PTYPE_L4_UDP) == RTE_PTYPE_L4_UDP) ||
2555 	    ((packet_type & RTE_PTYPE_L4_TCP) == RTE_PTYPE_L4_TCP)) {
2556 		status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2557 		if (unlikely(status != PP2_DESC_ERR_OK))
2558 			flags |= RTE_MBUF_F_RX_L4_CKSUM_BAD;
2559 		else
2560 			flags |= RTE_MBUF_F_RX_L4_CKSUM_GOOD;
2561 	}
2562 
2563 	return flags;
2564 }
2565 
2566 /**
2567  * DPDK callback for receive.
2568  *
2569  * @param rxq
2570  *   Generic pointer to the receive queue.
2571  * @param rx_pkts
2572  *   Array to store received packets.
2573  * @param nb_pkts
2574  *   Maximum number of packets in array.
2575  *
2576  * @return
2577  *   Number of packets successfully received.
2578  */
2579 static uint16_t
2580 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2581 {
2582 	struct mrvl_rxq *q = rxq;
2583 	struct pp2_ppio_desc descs[nb_pkts];
2584 	struct pp2_bpool *bpool;
2585 	int i, ret, rx_done = 0;
2586 	int num;
2587 	struct pp2_hif *hif;
2588 	unsigned int core_id = rte_lcore_id();
2589 
2590 	hif = mrvl_get_hif(q->priv, core_id);
2591 
2592 	if (unlikely(!q->priv->ppio || !hif))
2593 		return 0;
2594 
2595 	bpool = q->priv->bpool;
2596 
2597 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2598 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2599 	if (unlikely(ret < 0))
2600 		return 0;
2601 
2602 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2603 
2604 	for (i = 0; i < nb_pkts; i++) {
2605 		struct rte_mbuf *mbuf;
2606 		uint8_t l3_offset, l4_offset;
2607 		enum pp2_inq_desc_status status;
2608 		uint64_t addr;
2609 
2610 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2611 			struct pp2_ppio_desc *pref_desc;
2612 			u64 pref_addr;
2613 
2614 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2615 			pref_addr = cookie_addr_high |
2616 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2617 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2618 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2619 		}
2620 
2621 		addr = cookie_addr_high |
2622 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2623 		mbuf = (struct rte_mbuf *)addr;
2624 		rte_pktmbuf_reset(mbuf);
2625 
2626 		/* drop packet in case of mac, overrun or resource error */
2627 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2628 		if ((unlikely(status != PP2_DESC_ERR_OK)) &&
2629 			!(q->priv->forward_bad_frames)) {
2630 			struct pp2_buff_inf binf = {
2631 				.addr = rte_mbuf_data_iova_default(mbuf),
2632 				.cookie = (uint64_t)mbuf,
2633 			};
2634 
2635 			pp2_bpool_put_buff(hif, bpool, &binf);
2636 			mrvl_port_bpool_size
2637 				[bpool->pp2_id][bpool->id][core_id]++;
2638 			q->drop_mac++;
2639 			continue;
2640 		}
2641 
2642 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2643 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2644 		mbuf->data_len = mbuf->pkt_len;
2645 		mbuf->port = q->port_id;
2646 		mbuf->packet_type =
2647 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2648 							    &l3_offset,
2649 							    &l4_offset);
2650 		mbuf->l2_len = l3_offset;
2651 		mbuf->l3_len = l4_offset - l3_offset;
2652 
2653 		if (likely(q->cksum_enabled))
2654 			mbuf->ol_flags =
2655 				mrvl_desc_to_ol_flags(&descs[i],
2656 						      mbuf->packet_type);
2657 
2658 		rx_pkts[rx_done++] = mbuf;
2659 		q->bytes_recv += mbuf->pkt_len;
2660 	}
2661 
2662 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2663 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2664 
2665 		if (unlikely(num <= q->priv->bpool_min_size ||
2666 			     (!rx_done && num < q->priv->bpool_init_size))) {
2667 			mrvl_fill_bpool(q, q->priv->fill_bpool_buffs);
2668 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2669 			int i;
2670 			int pkt_to_remove = num - q->priv->bpool_init_size;
2671 			struct rte_mbuf *mbuf;
2672 			struct pp2_buff_inf buff;
2673 
2674 			for (i = 0; i < pkt_to_remove; i++) {
2675 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2676 				if (ret)
2677 					break;
2678 				mbuf = (struct rte_mbuf *)
2679 					(cookie_addr_high | buff.cookie);
2680 				rte_pktmbuf_free(mbuf);
2681 			}
2682 			mrvl_port_bpool_size
2683 				[bpool->pp2_id][bpool->id][core_id] -= i;
2684 		}
2685 		rte_spinlock_unlock(&q->priv->lock);
2686 	}
2687 
2688 	return rx_done;
2689 }
2690 
2691 /**
2692  * Prepare offload information.
2693  *
2694  * @param ol_flags
2695  *   Offload flags.
2696  * @param l3_type
2697  *   Pointer to the pp2_ouq_l3_type structure.
2698  * @param l4_type
2699  *   Pointer to the pp2_outq_l4_type structure.
2700  * @param gen_l3_cksum
2701  *   Will be set to 1 in case l3 checksum is computed.
2702  * @param l4_cksum
2703  *   Will be set to 1 in case l4 checksum is computed.
2704  */
2705 static inline void
2706 mrvl_prepare_proto_info(uint64_t ol_flags,
2707 			enum pp2_outq_l3_type *l3_type,
2708 			enum pp2_outq_l4_type *l4_type,
2709 			int *gen_l3_cksum,
2710 			int *gen_l4_cksum)
2711 {
2712 	/*
2713 	 * Based on ol_flags prepare information
2714 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2715 	 * for offloading.
2716 	 * in most of the checksum cases ipv4 must be set, so this is the
2717 	 * default value
2718 	 */
2719 	*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2720 	*gen_l3_cksum = ol_flags & RTE_MBUF_F_TX_IP_CKSUM ? 1 : 0;
2721 
2722 	if (ol_flags & RTE_MBUF_F_TX_IPV6) {
2723 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2724 		/* no checksum for ipv6 header */
2725 		*gen_l3_cksum = 0;
2726 	}
2727 
2728 	if ((ol_flags & RTE_MBUF_F_TX_L4_MASK) == RTE_MBUF_F_TX_TCP_CKSUM) {
2729 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2730 		*gen_l4_cksum = 1;
2731 	} else if ((ol_flags & RTE_MBUF_F_TX_L4_MASK) ==  RTE_MBUF_F_TX_UDP_CKSUM) {
2732 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2733 		*gen_l4_cksum = 1;
2734 	} else {
2735 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2736 		/* no checksum for other type */
2737 		*gen_l4_cksum = 0;
2738 	}
2739 }
2740 
2741 /**
2742  * Release already sent buffers to bpool (buffer-pool).
2743  *
2744  * @param ppio
2745  *   Pointer to the port structure.
2746  * @param hif
2747  *   Pointer to the MUSDK hardware interface.
2748  * @param sq
2749  *   Pointer to the shadow queue.
2750  * @param qid
2751  *   Queue id number.
2752  * @param force
2753  *   Force releasing packets.
2754  */
2755 static inline void
2756 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2757 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2758 		       int qid, int force)
2759 {
2760 	struct buff_release_entry *entry;
2761 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2762 	int i;
2763 
2764 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2765 
2766 	sq->num_to_release += nb_done;
2767 
2768 	if (likely(!force &&
2769 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2770 		return;
2771 
2772 	nb_done = sq->num_to_release;
2773 	sq->num_to_release = 0;
2774 
2775 	for (i = 0; i < nb_done; i++) {
2776 		entry = &sq->ent[sq->tail + num];
2777 		if (unlikely(!entry->buff.addr)) {
2778 			MRVL_LOG(ERR,
2779 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2780 				sq->tail, (u64)entry->buff.cookie,
2781 				(u64)entry->buff.addr);
2782 			skip_bufs = 1;
2783 			goto skip;
2784 		}
2785 
2786 		if (unlikely(!entry->bpool)) {
2787 			struct rte_mbuf *mbuf;
2788 
2789 			mbuf = (struct rte_mbuf *)entry->buff.cookie;
2790 			rte_pktmbuf_free(mbuf);
2791 			skip_bufs = 1;
2792 			goto skip;
2793 		}
2794 
2795 		mrvl_port_bpool_size
2796 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2797 		num++;
2798 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2799 			goto skip;
2800 		continue;
2801 skip:
2802 		if (likely(num))
2803 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2804 		num += skip_bufs;
2805 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2806 		sq->size -= num;
2807 		num = 0;
2808 		skip_bufs = 0;
2809 	}
2810 
2811 	if (likely(num)) {
2812 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2813 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2814 		sq->size -= num;
2815 	}
2816 }
2817 
2818 /**
2819  * DPDK callback for transmit.
2820  *
2821  * @param txq
2822  *   Generic pointer transmit queue.
2823  * @param tx_pkts
2824  *   Packets to transmit.
2825  * @param nb_pkts
2826  *   Number of packets in array.
2827  *
2828  * @return
2829  *   Number of packets successfully transmitted.
2830  */
2831 static uint16_t
2832 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2833 {
2834 	struct mrvl_txq *q = txq;
2835 	struct mrvl_shadow_txq *sq;
2836 	struct pp2_hif *hif;
2837 	struct pp2_ppio_desc descs[nb_pkts];
2838 	unsigned int core_id = rte_lcore_id();
2839 	int i, bytes_sent = 0;
2840 	uint16_t num, sq_free_size;
2841 	uint64_t addr;
2842 
2843 	hif = mrvl_get_hif(q->priv, core_id);
2844 	sq = &q->shadow_txqs[core_id];
2845 
2846 	if (unlikely(!q->priv->ppio || !hif))
2847 		return 0;
2848 
2849 	if (sq->size)
2850 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2851 				       sq, q->queue_id, 0);
2852 
2853 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2854 	if (unlikely(nb_pkts > sq_free_size))
2855 		nb_pkts = sq_free_size;
2856 
2857 	for (i = 0; i < nb_pkts; i++) {
2858 		struct rte_mbuf *mbuf = tx_pkts[i];
2859 		int gen_l3_cksum, gen_l4_cksum;
2860 		enum pp2_outq_l3_type l3_type;
2861 		enum pp2_outq_l4_type l4_type;
2862 
2863 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2864 			struct rte_mbuf *pref_pkt_hdr;
2865 
2866 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2867 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2868 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2869 		}
2870 
2871 		mrvl_fill_shadowq(sq, mbuf);
2872 		mrvl_fill_desc(&descs[i], mbuf);
2873 
2874 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2875 		/*
2876 		 * in case unsupported ol_flags were passed
2877 		 * do not update descriptor offload information
2878 		 */
2879 		if (!(mbuf->ol_flags & MRVL_TX_PKT_OFFLOADS))
2880 			continue;
2881 		mrvl_prepare_proto_info(mbuf->ol_flags, &l3_type, &l4_type,
2882 					&gen_l3_cksum, &gen_l4_cksum);
2883 
2884 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2885 						  mbuf->l2_len,
2886 						  mbuf->l2_len + mbuf->l3_len,
2887 						  gen_l3_cksum, gen_l4_cksum);
2888 	}
2889 
2890 	num = nb_pkts;
2891 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2892 	/* number of packets that were not sent */
2893 	if (unlikely(num > nb_pkts)) {
2894 		for (i = nb_pkts; i < num; i++) {
2895 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2896 				MRVL_PP2_TX_SHADOWQ_MASK;
2897 			addr = sq->ent[sq->head].buff.cookie;
2898 			bytes_sent -=
2899 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2900 		}
2901 		sq->size -= num - nb_pkts;
2902 	}
2903 
2904 	q->bytes_sent += bytes_sent;
2905 
2906 	return nb_pkts;
2907 }
2908 
2909 /** DPDK callback for S/G transmit.
2910  *
2911  * @param txq
2912  *   Generic pointer transmit queue.
2913  * @param tx_pkts
2914  *   Packets to transmit.
2915  * @param nb_pkts
2916  *   Number of packets in array.
2917  *
2918  * @return
2919  *   Number of packets successfully transmitted.
2920  */
2921 static uint16_t
2922 mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
2923 		     uint16_t nb_pkts)
2924 {
2925 	struct mrvl_txq *q = txq;
2926 	struct mrvl_shadow_txq *sq;
2927 	struct pp2_hif *hif;
2928 	struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
2929 	struct pp2_ppio_sg_pkts pkts;
2930 	uint8_t frags[nb_pkts];
2931 	unsigned int core_id = rte_lcore_id();
2932 	int i, j, bytes_sent = 0;
2933 	int tail, tail_first;
2934 	uint16_t num, sq_free_size;
2935 	uint16_t nb_segs, total_descs = 0;
2936 	uint64_t addr;
2937 
2938 	hif = mrvl_get_hif(q->priv, core_id);
2939 	sq = &q->shadow_txqs[core_id];
2940 	pkts.frags = frags;
2941 	pkts.num = 0;
2942 
2943 	if (unlikely(!q->priv->ppio || !hif))
2944 		return 0;
2945 
2946 	if (sq->size)
2947 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2948 				       sq, q->queue_id, 0);
2949 
2950 	/* Save shadow queue free size */
2951 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2952 
2953 	tail = 0;
2954 	for (i = 0; i < nb_pkts; i++) {
2955 		struct rte_mbuf *mbuf = tx_pkts[i];
2956 		struct rte_mbuf *seg = NULL;
2957 		int gen_l3_cksum, gen_l4_cksum;
2958 		enum pp2_outq_l3_type l3_type;
2959 		enum pp2_outq_l4_type l4_type;
2960 
2961 		nb_segs = mbuf->nb_segs;
2962 		tail_first = tail;
2963 		total_descs += nb_segs;
2964 
2965 		/*
2966 		 * Check if total_descs does not exceed
2967 		 * shadow queue free size
2968 		 */
2969 		if (unlikely(total_descs > sq_free_size)) {
2970 			total_descs -= nb_segs;
2971 			break;
2972 		}
2973 
2974 		/* Check if nb_segs does not exceed the max nb of desc per
2975 		 * fragmented packet
2976 		 */
2977 		if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
2978 			total_descs -= nb_segs;
2979 			MRVL_LOG(ERR, "Too many segments. Packet won't be sent.");
2980 			break;
2981 		}
2982 
2983 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2984 			struct rte_mbuf *pref_pkt_hdr;
2985 
2986 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2987 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2988 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2989 		}
2990 
2991 		pkts.frags[pkts.num] = nb_segs;
2992 		pkts.num++;
2993 
2994 		seg = mbuf;
2995 		for (j = 0; j < nb_segs - 1; j++) {
2996 			/* For the subsequent segments, set shadow queue
2997 			 * buffer to NULL
2998 			 */
2999 			mrvl_fill_shadowq(sq, NULL);
3000 			mrvl_fill_desc(&descs[tail], seg);
3001 
3002 			tail++;
3003 			seg = seg->next;
3004 		}
3005 		/* Put first mbuf info in last shadow queue entry */
3006 		mrvl_fill_shadowq(sq, mbuf);
3007 		/* Update descriptor with last segment */
3008 		mrvl_fill_desc(&descs[tail++], seg);
3009 
3010 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
3011 		/* In case unsupported ol_flags were passed
3012 		 * do not update descriptor offload information
3013 		 */
3014 		if (!(mbuf->ol_flags & MRVL_TX_PKT_OFFLOADS))
3015 			continue;
3016 		mrvl_prepare_proto_info(mbuf->ol_flags, &l3_type, &l4_type,
3017 					&gen_l3_cksum, &gen_l4_cksum);
3018 
3019 		pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
3020 						  l4_type, mbuf->l2_len,
3021 						  mbuf->l2_len + mbuf->l3_len,
3022 						  gen_l3_cksum, gen_l4_cksum);
3023 	}
3024 
3025 	num = total_descs;
3026 	pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
3027 			 &total_descs, &pkts);
3028 	/* number of packets that were not sent */
3029 	if (unlikely(num > total_descs)) {
3030 		for (i = total_descs; i < num; i++) {
3031 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
3032 				MRVL_PP2_TX_SHADOWQ_MASK;
3033 
3034 			addr = sq->ent[sq->head].buff.cookie;
3035 			if (addr)
3036 				bytes_sent -=
3037 					rte_pktmbuf_pkt_len((struct rte_mbuf *)
3038 						(cookie_addr_high | addr));
3039 		}
3040 		sq->size -= num - total_descs;
3041 		nb_pkts = pkts.num;
3042 	}
3043 
3044 	q->bytes_sent += bytes_sent;
3045 
3046 	return nb_pkts;
3047 }
3048 
3049 /**
3050  * Create private device structure.
3051  *
3052  * @param dev_name
3053  *   Pointer to the port name passed in the initialization parameters.
3054  *
3055  * @return
3056  *   Pointer to the newly allocated private device structure.
3057  */
3058 static struct mrvl_priv *
3059 mrvl_priv_create(const char *dev_name)
3060 {
3061 	struct pp2_bpool_params bpool_params;
3062 	char match[MRVL_MATCH_LEN];
3063 	struct mrvl_priv *priv;
3064 	uint16_t max_frame_size;
3065 	int ret, bpool_bit;
3066 
3067 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
3068 	if (!priv)
3069 		return NULL;
3070 
3071 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
3072 				       &priv->pp_id, &priv->ppio_id);
3073 	if (ret)
3074 		goto out_free_priv;
3075 
3076 	ret = pp2_ppio_get_l4_cksum_max_frame_size(priv->pp_id, priv->ppio_id,
3077 						   &max_frame_size);
3078 	if (ret)
3079 		goto out_free_priv;
3080 
3081 	priv->max_mtu = max_frame_size + RTE_ETHER_CRC_LEN -
3082 		MRVL_PP2_ETH_HDRS_LEN;
3083 
3084 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
3085 				     PP2_BPOOL_NUM_POOLS);
3086 	if (bpool_bit < 0)
3087 		goto out_free_priv;
3088 	priv->bpool_bit = bpool_bit;
3089 
3090 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
3091 		 priv->bpool_bit);
3092 	memset(&bpool_params, 0, sizeof(bpool_params));
3093 	bpool_params.match = match;
3094 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
3095 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
3096 	if (ret)
3097 		goto out_clear_bpool_bit;
3098 
3099 	priv->ppio_params.type = PP2_PPIO_T_NIC;
3100 	rte_spinlock_init(&priv->lock);
3101 
3102 	return priv;
3103 out_clear_bpool_bit:
3104 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
3105 out_free_priv:
3106 	rte_free(priv);
3107 	return NULL;
3108 }
3109 
3110 /**
3111  * Create device representing Ethernet port.
3112  *
3113  * @param name
3114  *   Pointer to the port's name.
3115  *
3116  * @return
3117  *   0 on success, negative error value otherwise.
3118  */
3119 static int
3120 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
3121 {
3122 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
3123 	struct rte_eth_dev *eth_dev;
3124 	struct mrvl_priv *priv;
3125 	struct ifreq req;
3126 
3127 	eth_dev = rte_eth_dev_allocate(name);
3128 	if (!eth_dev)
3129 		return -ENOMEM;
3130 
3131 	priv = mrvl_priv_create(name);
3132 	if (!priv) {
3133 		ret = -ENOMEM;
3134 		goto out_free;
3135 	}
3136 	eth_dev->data->dev_private = priv;
3137 
3138 	eth_dev->data->mac_addrs =
3139 		rte_zmalloc("mac_addrs",
3140 			    RTE_ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
3141 	if (!eth_dev->data->mac_addrs) {
3142 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
3143 		ret = -ENOMEM;
3144 		goto out_free;
3145 	}
3146 
3147 	memset(&req, 0, sizeof(req));
3148 	strcpy(req.ifr_name, name);
3149 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
3150 	if (ret)
3151 		goto out_free;
3152 
3153 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
3154 	       req.ifr_addr.sa_data, RTE_ETHER_ADDR_LEN);
3155 
3156 	eth_dev->device = &vdev->device;
3157 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
3158 	mrvl_set_tx_function(eth_dev);
3159 	eth_dev->dev_ops = &mrvl_ops;
3160 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
3161 
3162 	eth_dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
3163 
3164 	rte_eth_dev_probing_finish(eth_dev);
3165 	return 0;
3166 out_free:
3167 	rte_eth_dev_release_port(eth_dev);
3168 
3169 	return ret;
3170 }
3171 
3172 /**
3173  * Callback used by rte_kvargs_process() during argument parsing.
3174  *
3175  * @param key
3176  *   Pointer to the parsed key (unused).
3177  * @param value
3178  *   Pointer to the parsed value.
3179  * @param extra_args
3180  *   Pointer to the extra arguments which contains address of the
3181  *   table of pointers to parsed interface names.
3182  *
3183  * @return
3184  *   Always 0.
3185  */
3186 static int
3187 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
3188 		 void *extra_args)
3189 {
3190 	struct mrvl_ifnames *ifnames = extra_args;
3191 
3192 	ifnames->names[ifnames->idx++] = value;
3193 
3194 	return 0;
3195 }
3196 
3197 /**
3198  * DPDK callback to register the virtual device.
3199  *
3200  * @param vdev
3201  *   Pointer to the virtual device.
3202  *
3203  * @return
3204  *   0 on success, negative error value otherwise.
3205  */
3206 static int
3207 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
3208 {
3209 	struct rte_kvargs *kvlist;
3210 	struct mrvl_ifnames ifnames;
3211 	int ret = -EINVAL;
3212 	uint32_t i, ifnum, cfgnum;
3213 	const char *params;
3214 
3215 	params = rte_vdev_device_args(vdev);
3216 	if (!params)
3217 		return -EINVAL;
3218 
3219 	kvlist = rte_kvargs_parse(params, valid_args);
3220 	if (!kvlist)
3221 		return -EINVAL;
3222 
3223 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
3224 	if (ifnum > RTE_DIM(ifnames.names))
3225 		goto out_free_kvlist;
3226 
3227 	ifnames.idx = 0;
3228 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
3229 			   mrvl_get_ifnames, &ifnames);
3230 
3231 
3232 	/*
3233 	 * The below system initialization should be done only once,
3234 	 * on the first provided configuration file
3235 	 */
3236 	if (!mrvl_cfg) {
3237 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
3238 		MRVL_LOG(INFO, "Parsing config file!");
3239 		if (cfgnum > 1) {
3240 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
3241 			goto out_free_kvlist;
3242 		} else if (cfgnum == 1) {
3243 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
3244 					   mrvl_get_cfg, &mrvl_cfg);
3245 		}
3246 	}
3247 
3248 	if (mrvl_dev_num)
3249 		goto init_devices;
3250 
3251 	MRVL_LOG(INFO, "Perform MUSDK initializations");
3252 
3253 	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
3254 	if (ret)
3255 		goto out_free_kvlist;
3256 
3257 	ret = mrvl_init_pp2();
3258 	if (ret) {
3259 		MRVL_LOG(ERR, "Failed to init PP!");
3260 		rte_mvep_deinit(MVEP_MOD_T_PP2);
3261 		goto out_free_kvlist;
3262 	}
3263 
3264 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
3265 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
3266 
3267 	mrvl_lcore_first = RTE_MAX_LCORE;
3268 	mrvl_lcore_last = 0;
3269 
3270 init_devices:
3271 	for (i = 0; i < ifnum; i++) {
3272 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
3273 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
3274 		if (ret)
3275 			goto out_cleanup;
3276 		mrvl_dev_num++;
3277 	}
3278 
3279 	rte_kvargs_free(kvlist);
3280 
3281 	return 0;
3282 out_cleanup:
3283 	rte_pmd_mrvl_remove(vdev);
3284 
3285 out_free_kvlist:
3286 	rte_kvargs_free(kvlist);
3287 
3288 	return ret;
3289 }
3290 
3291 /**
3292  * DPDK callback to remove virtual device.
3293  *
3294  * @param vdev
3295  *   Pointer to the removed virtual device.
3296  *
3297  * @return
3298  *   0 on success, negative error value otherwise.
3299  */
3300 static int
3301 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
3302 {
3303 	uint16_t port_id;
3304 	int ret = 0;
3305 
3306 	RTE_ETH_FOREACH_DEV(port_id) {
3307 		if (rte_eth_devices[port_id].device != &vdev->device)
3308 			continue;
3309 		ret |= rte_eth_dev_close(port_id);
3310 	}
3311 
3312 	return ret == 0 ? 0 : -EIO;
3313 }
3314 
3315 static struct rte_vdev_driver pmd_mrvl_drv = {
3316 	.probe = rte_pmd_mrvl_probe,
3317 	.remove = rte_pmd_mrvl_remove,
3318 };
3319 
3320 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
3321 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
3322 RTE_LOG_REGISTER_DEFAULT(mrvl_logtype, NOTICE);
3323