xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision 8df71650e9fdc6346f09b7a57e86cded7b553152)
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");
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!");
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",
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");
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 			      size_t *no_of_elements)
1769 {
1770 	static const uint32_t ptypes[] = {
1771 		RTE_PTYPE_L2_ETHER,
1772 		RTE_PTYPE_L2_ETHER_VLAN,
1773 		RTE_PTYPE_L2_ETHER_QINQ,
1774 		RTE_PTYPE_L3_IPV4,
1775 		RTE_PTYPE_L3_IPV4_EXT,
1776 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1777 		RTE_PTYPE_L3_IPV6,
1778 		RTE_PTYPE_L3_IPV6_EXT,
1779 		RTE_PTYPE_L2_ETHER_ARP,
1780 		RTE_PTYPE_L4_TCP,
1781 		RTE_PTYPE_L4_UDP,
1782 	};
1783 
1784 	*no_of_elements = RTE_DIM(ptypes);
1785 	return ptypes;
1786 }
1787 
1788 /**
1789  * DPDK callback to get information about specific receive queue.
1790  *
1791  * @param dev
1792  *   Pointer to Ethernet device structure.
1793  * @param rx_queue_id
1794  *   Receive queue index.
1795  * @param qinfo
1796  *   Receive queue information structure.
1797  */
1798 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1799 			      struct rte_eth_rxq_info *qinfo)
1800 {
1801 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1802 	struct mrvl_priv *priv = dev->data->dev_private;
1803 	int inq = priv->rxq_map[rx_queue_id].inq;
1804 	int tc = priv->rxq_map[rx_queue_id].tc;
1805 	struct pp2_ppio_tc_params *tc_params =
1806 		&priv->ppio_params.inqs_params.tcs_params[tc];
1807 
1808 	qinfo->mp = q->mp;
1809 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1810 }
1811 
1812 /**
1813  * DPDK callback to get information about specific transmit queue.
1814  *
1815  * @param dev
1816  *   Pointer to Ethernet device structure.
1817  * @param tx_queue_id
1818  *   Transmit queue index.
1819  * @param qinfo
1820  *   Transmit queue information structure.
1821  */
1822 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1823 			      struct rte_eth_txq_info *qinfo)
1824 {
1825 	struct mrvl_priv *priv = dev->data->dev_private;
1826 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1827 
1828 	qinfo->nb_desc =
1829 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1830 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1831 }
1832 
1833 /**
1834  * DPDK callback to Configure a VLAN filter.
1835  *
1836  * @param dev
1837  *   Pointer to Ethernet device structure.
1838  * @param vlan_id
1839  *   VLAN ID to filter.
1840  * @param on
1841  *   Toggle filter.
1842  *
1843  * @return
1844  *   0 on success, negative error value otherwise.
1845  */
1846 static int
1847 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1848 {
1849 	struct mrvl_priv *priv = dev->data->dev_private;
1850 
1851 	if (priv->isolated)
1852 		return -ENOTSUP;
1853 
1854 	if (!priv->ppio)
1855 		return 0;
1856 
1857 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1858 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1859 }
1860 
1861 /**
1862  * DPDK callback to Configure VLAN offload.
1863  *
1864  * @param dev
1865  *   Pointer to Ethernet device structure.
1866  * @param mask
1867  *   VLAN offload mask.
1868  *
1869  * @return
1870  *   0 on success, negative error value otherwise.
1871  */
1872 static int mrvl_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1873 {
1874 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
1875 	int ret;
1876 
1877 	if (mask & RTE_ETH_VLAN_STRIP_MASK) {
1878 		MRVL_LOG(ERR, "VLAN stripping is not supported");
1879 		return -ENOTSUP;
1880 	}
1881 
1882 	if (mask & RTE_ETH_VLAN_FILTER_MASK) {
1883 		if (rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER)
1884 			ret = mrvl_populate_vlan_table(dev, 1);
1885 		else
1886 			ret = mrvl_populate_vlan_table(dev, 0);
1887 
1888 		if (ret)
1889 			return ret;
1890 	}
1891 
1892 	if (mask & RTE_ETH_VLAN_EXTEND_MASK) {
1893 		MRVL_LOG(ERR, "Extend VLAN not supported");
1894 		return -ENOTSUP;
1895 	}
1896 
1897 	return 0;
1898 }
1899 
1900 /**
1901  * Release buffers to hardware bpool (buffer-pool)
1902  *
1903  * @param rxq
1904  *   Receive queue pointer.
1905  * @param num
1906  *   Number of buffers to release to bpool.
1907  *
1908  * @return
1909  *   0 on success, negative error value otherwise.
1910  */
1911 static int
1912 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1913 {
1914 	struct buff_release_entry entries[num];
1915 	struct rte_mbuf *mbufs[num];
1916 	int i, ret;
1917 	unsigned int core_id;
1918 	struct pp2_hif *hif;
1919 	struct pp2_bpool *bpool;
1920 
1921 	core_id = rte_lcore_id();
1922 	if (core_id == LCORE_ID_ANY)
1923 		core_id = rte_get_main_lcore();
1924 
1925 	hif = mrvl_get_hif(rxq->priv, core_id);
1926 	if (!hif)
1927 		return -1;
1928 
1929 	bpool = rxq->priv->bpool;
1930 
1931 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1932 	if (ret)
1933 		return ret;
1934 
1935 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1936 		cookie_addr_high =
1937 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1938 
1939 	for (i = 0; i < num; i++) {
1940 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1941 			!= cookie_addr_high) {
1942 			MRVL_LOG(ERR,
1943 				"mbuf virtual addr high is out of range "
1944 				"0x%x instead of 0x%x",
1945 				(uint32_t)((uint64_t)mbufs[i] >> 32),
1946 				(uint32_t)(cookie_addr_high >> 32));
1947 			goto out;
1948 		}
1949 
1950 		entries[i].buff.addr =
1951 			rte_mbuf_data_iova_default(mbufs[i]);
1952 		entries[i].buff.cookie = (uintptr_t)mbufs[i];
1953 		entries[i].bpool = bpool;
1954 	}
1955 
1956 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1957 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1958 
1959 	if (i != num)
1960 		goto out;
1961 
1962 	return 0;
1963 out:
1964 	for (; i < num; i++)
1965 		rte_pktmbuf_free(mbufs[i]);
1966 
1967 	return -1;
1968 }
1969 
1970 /**
1971  * DPDK callback to configure the receive queue.
1972  *
1973  * @param dev
1974  *   Pointer to Ethernet device structure.
1975  * @param idx
1976  *   RX queue index.
1977  * @param desc
1978  *   Number of descriptors to configure in queue.
1979  * @param socket
1980  *   NUMA socket on which memory must be allocated.
1981  * @param conf
1982  *   Thresholds parameters.
1983  * @param mp
1984  *   Memory pool for buffer allocations.
1985  *
1986  * @return
1987  *   0 on success, negative error value otherwise.
1988  */
1989 static int
1990 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1991 		    unsigned int socket,
1992 		    const struct rte_eth_rxconf *conf,
1993 		    struct rte_mempool *mp)
1994 {
1995 	struct mrvl_priv *priv = dev->data->dev_private;
1996 	struct mrvl_rxq *rxq;
1997 	uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
1998 	uint32_t max_rx_pktlen = dev->data->mtu + RTE_ETHER_HDR_LEN;
1999 	int ret, tc, inq;
2000 	uint64_t offloads;
2001 
2002 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
2003 
2004 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
2005 		/*
2006 		 * Unknown TC mapping, mapping will not have a correct queue.
2007 		 */
2008 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
2009 			idx, priv->ppio_id);
2010 		return -EFAULT;
2011 	}
2012 
2013 	frame_size = buf_size - RTE_PKTMBUF_HEADROOM - MRVL_PKT_EFFEC_OFFS;
2014 	if (frame_size < max_rx_pktlen) {
2015 		MRVL_LOG(WARNING,
2016 			"Mbuf size must be increased to %u bytes to hold up "
2017 			"to %u bytes of data.",
2018 			max_rx_pktlen + buf_size - frame_size,
2019 			max_rx_pktlen);
2020 		dev->data->mtu = frame_size - RTE_ETHER_HDR_LEN;
2021 		MRVL_LOG(INFO, "Setting MTU to %u", dev->data->mtu);
2022 	}
2023 
2024 	if (dev->data->rx_queues[idx]) {
2025 		rte_free(dev->data->rx_queues[idx]);
2026 		dev->data->rx_queues[idx] = NULL;
2027 	}
2028 
2029 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
2030 	if (!rxq)
2031 		return -ENOMEM;
2032 
2033 	rxq->priv = priv;
2034 	rxq->mp = mp;
2035 	rxq->cksum_enabled = offloads & RTE_ETH_RX_OFFLOAD_IPV4_CKSUM;
2036 	rxq->queue_id = idx;
2037 	rxq->port_id = dev->data->port_id;
2038 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
2039 
2040 	tc = priv->rxq_map[rxq->queue_id].tc,
2041 	inq = priv->rxq_map[rxq->queue_id].inq;
2042 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
2043 		desc;
2044 
2045 	ret = mrvl_fill_bpool(rxq, desc);
2046 	if (ret) {
2047 		rte_free(rxq);
2048 		return ret;
2049 	}
2050 
2051 	priv->bpool_init_size += desc;
2052 
2053 	dev->data->rx_queues[idx] = rxq;
2054 
2055 	return 0;
2056 }
2057 
2058 /**
2059  * DPDK callback to release the receive queue.
2060  *
2061  * @param dev
2062  *   Pointer to Ethernet device structure.
2063  * @param qid
2064  *   Receive queue index.
2065  */
2066 static void
2067 mrvl_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
2068 {
2069 	struct mrvl_rxq *q = dev->data->rx_queues[qid];
2070 	struct pp2_ppio_tc_params *tc_params;
2071 	int i, num, tc, inq;
2072 	struct pp2_hif *hif;
2073 	unsigned int core_id = rte_lcore_id();
2074 
2075 	if (core_id == LCORE_ID_ANY)
2076 		core_id = rte_get_main_lcore();
2077 
2078 	if (!q)
2079 		return;
2080 
2081 	hif = mrvl_get_hif(q->priv, core_id);
2082 
2083 	if (!hif)
2084 		return;
2085 
2086 	tc = q->priv->rxq_map[q->queue_id].tc;
2087 	inq = q->priv->rxq_map[q->queue_id].inq;
2088 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
2089 	num = tc_params->inqs_params[inq].size;
2090 	for (i = 0; i < num; i++) {
2091 		struct pp2_buff_inf inf;
2092 		uint64_t addr;
2093 
2094 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
2095 		addr = cookie_addr_high | inf.cookie;
2096 		rte_pktmbuf_free((struct rte_mbuf *)addr);
2097 	}
2098 
2099 	rte_free(q);
2100 }
2101 
2102 /**
2103  * DPDK callback to configure the transmit queue.
2104  *
2105  * @param dev
2106  *   Pointer to Ethernet device structure.
2107  * @param idx
2108  *   Transmit queue index.
2109  * @param desc
2110  *   Number of descriptors to configure in the queue.
2111  * @param socket
2112  *   NUMA socket on which memory must be allocated.
2113  * @param conf
2114  *   Tx queue configuration parameters.
2115  *
2116  * @return
2117  *   0 on success, negative error value otherwise.
2118  */
2119 static int
2120 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
2121 		    unsigned int socket,
2122 		    const struct rte_eth_txconf *conf)
2123 {
2124 	struct mrvl_priv *priv = dev->data->dev_private;
2125 	struct mrvl_txq *txq;
2126 
2127 	if (dev->data->tx_queues[idx]) {
2128 		rte_free(dev->data->tx_queues[idx]);
2129 		dev->data->tx_queues[idx] = NULL;
2130 	}
2131 
2132 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
2133 	if (!txq)
2134 		return -ENOMEM;
2135 
2136 	txq->priv = priv;
2137 	txq->queue_id = idx;
2138 	txq->port_id = dev->data->port_id;
2139 	txq->tx_deferred_start = conf->tx_deferred_start;
2140 	dev->data->tx_queues[idx] = txq;
2141 
2142 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
2143 
2144 	return 0;
2145 }
2146 
2147 /**
2148  * DPDK callback to release the transmit queue.
2149  *
2150  * @param dev
2151  *   Pointer to Ethernet device structure.
2152  * @param qid
2153  *   Transmit queue index.
2154  */
2155 static void
2156 mrvl_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
2157 {
2158 	struct mrvl_txq *q = dev->data->tx_queues[qid];
2159 
2160 	if (!q)
2161 		return;
2162 
2163 	rte_free(q);
2164 }
2165 
2166 /**
2167  * DPDK callback to get flow control configuration.
2168  *
2169  * @param dev
2170  *  Pointer to Ethernet device structure.
2171  * @param fc_conf
2172  *  Pointer to the flow control configuration.
2173  *
2174  * @return
2175  *  0 on success, negative error value otherwise.
2176  */
2177 static int
2178 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2179 {
2180 	struct mrvl_priv *priv = dev->data->dev_private;
2181 	int ret, en;
2182 
2183 	if (!priv->ppio) {
2184 		memcpy(fc_conf, &priv->fc_conf, sizeof(struct rte_eth_fc_conf));
2185 		return 0;
2186 	}
2187 
2188 	fc_conf->autoneg = 1;
2189 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
2190 	if (ret) {
2191 		MRVL_LOG(ERR, "Failed to read rx pause state");
2192 		return ret;
2193 	}
2194 
2195 	fc_conf->mode = en ? RTE_ETH_FC_RX_PAUSE : RTE_ETH_FC_NONE;
2196 
2197 	ret = pp2_ppio_get_tx_pause(priv->ppio, &en);
2198 	if (ret) {
2199 		MRVL_LOG(ERR, "Failed to read tx pause state");
2200 		return ret;
2201 	}
2202 
2203 	if (en) {
2204 		if (fc_conf->mode == RTE_ETH_FC_NONE)
2205 			fc_conf->mode = RTE_ETH_FC_TX_PAUSE;
2206 		else
2207 			fc_conf->mode = RTE_ETH_FC_FULL;
2208 	}
2209 
2210 	return 0;
2211 }
2212 
2213 /**
2214  * DPDK callback to set flow control configuration.
2215  *
2216  * @param dev
2217  *  Pointer to Ethernet device structure.
2218  * @param fc_conf
2219  *  Pointer to the flow control configuration.
2220  *
2221  * @return
2222  *  0 on success, negative error value otherwise.
2223  */
2224 static int
2225 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2226 {
2227 	struct mrvl_priv *priv = dev->data->dev_private;
2228 	struct pp2_ppio_tx_pause_params mrvl_pause_params;
2229 	int ret;
2230 	int rx_en, tx_en;
2231 
2232 	if (fc_conf->high_water ||
2233 	    fc_conf->low_water ||
2234 	    fc_conf->pause_time ||
2235 	    fc_conf->mac_ctrl_frame_fwd) {
2236 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
2237 
2238 		return -EINVAL;
2239 	}
2240 
2241 	if (fc_conf->autoneg == 0) {
2242 		MRVL_LOG(ERR, "Flowctrl Autoneg disable is not supported");
2243 		return -EINVAL;
2244 	}
2245 
2246 	if (!priv->ppio) {
2247 		memcpy(&priv->fc_conf, fc_conf, sizeof(struct rte_eth_fc_conf));
2248 		priv->flow_ctrl = 1;
2249 		return 0;
2250 	}
2251 
2252 	switch (fc_conf->mode) {
2253 	case RTE_ETH_FC_FULL:
2254 		rx_en = 1;
2255 		tx_en = 1;
2256 		break;
2257 	case RTE_ETH_FC_TX_PAUSE:
2258 		rx_en = 0;
2259 		tx_en = 1;
2260 		break;
2261 	case RTE_ETH_FC_RX_PAUSE:
2262 		rx_en = 1;
2263 		tx_en = 0;
2264 		break;
2265 	case RTE_ETH_FC_NONE:
2266 		rx_en = 0;
2267 		tx_en = 0;
2268 		break;
2269 	default:
2270 		MRVL_LOG(ERR, "Incorrect Flow control flag (%d)",
2271 			 fc_conf->mode);
2272 		return -EINVAL;
2273 	}
2274 
2275 	/* Set RX flow control */
2276 	ret = pp2_ppio_set_rx_pause(priv->ppio, rx_en);
2277 	if (ret) {
2278 		MRVL_LOG(ERR, "Failed to change RX flowctrl");
2279 		return ret;
2280 	}
2281 
2282 	/* Set TX flow control */
2283 	mrvl_pause_params.en = tx_en;
2284 	/* all inqs participate in xon/xoff decision */
2285 	mrvl_pause_params.use_tc_pause_inqs = 0;
2286 	ret = pp2_ppio_set_tx_pause(priv->ppio, &mrvl_pause_params);
2287 	if (ret) {
2288 		MRVL_LOG(ERR, "Failed to change TX flowctrl");
2289 		return ret;
2290 	}
2291 
2292 	return 0;
2293 }
2294 
2295 /**
2296  * Update RSS hash configuration
2297  *
2298  * @param dev
2299  *   Pointer to Ethernet device structure.
2300  * @param rss_conf
2301  *   Pointer to RSS configuration.
2302  *
2303  * @return
2304  *   0 on success, negative error value otherwise.
2305  */
2306 static int
2307 mrvl_rss_hash_update(struct rte_eth_dev *dev,
2308 		     struct rte_eth_rss_conf *rss_conf)
2309 {
2310 	struct mrvl_priv *priv = dev->data->dev_private;
2311 
2312 	if (priv->isolated)
2313 		return -ENOTSUP;
2314 
2315 	return mrvl_configure_rss(priv, rss_conf);
2316 }
2317 
2318 /**
2319  * DPDK callback to get RSS hash configuration.
2320  *
2321  * @param dev
2322  *   Pointer to Ethernet device structure.
2323  * @rss_conf
2324  *   Pointer to RSS configuration.
2325  *
2326  * @return
2327  *   Always 0.
2328  */
2329 static int
2330 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
2331 		       struct rte_eth_rss_conf *rss_conf)
2332 {
2333 	struct mrvl_priv *priv = dev->data->dev_private;
2334 	enum pp2_ppio_hash_type hash_type =
2335 		priv->ppio_params.inqs_params.hash_type;
2336 
2337 	rss_conf->rss_key = NULL;
2338 
2339 	if (hash_type == PP2_PPIO_HASH_T_NONE)
2340 		rss_conf->rss_hf = 0;
2341 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
2342 		rss_conf->rss_hf = RTE_ETH_RSS_IPV4;
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_TCP;
2345 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
2346 		rss_conf->rss_hf = RTE_ETH_RSS_NONFRAG_IPV4_UDP;
2347 
2348 	return 0;
2349 }
2350 
2351 /**
2352  * DPDK callback to get rte_flow callbacks.
2353  *
2354  * @param dev
2355  *   Pointer to the device structure.
2356  * @param ops
2357  *   Pointer to pass the flow ops.
2358  *
2359  * @return
2360  *   0 on success, negative error value otherwise.
2361  */
2362 static int
2363 mrvl_eth_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
2364 		      const struct rte_flow_ops **ops)
2365 {
2366 	*ops = &mrvl_flow_ops;
2367 	return 0;
2368 }
2369 
2370 /**
2371  * DPDK callback to get rte_mtr callbacks.
2372  *
2373  * @param dev
2374  *   Pointer to the device structure.
2375  * @param ops
2376  *   Pointer to pass the mtr ops.
2377  *
2378  * @return
2379  *   Always 0.
2380  */
2381 static int
2382 mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2383 {
2384 	*(const void **)ops = &mrvl_mtr_ops;
2385 
2386 	return 0;
2387 }
2388 
2389 /**
2390  * DPDK callback to get rte_tm callbacks.
2391  *
2392  * @param dev
2393  *   Pointer to the device structure.
2394  * @param ops
2395  *   Pointer to pass the tm ops.
2396  *
2397  * @return
2398  *   Always 0.
2399  */
2400 static int
2401 mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2402 {
2403 	*(const void **)ops = &mrvl_tm_ops;
2404 
2405 	return 0;
2406 }
2407 
2408 static const struct eth_dev_ops mrvl_ops = {
2409 	.dev_configure = mrvl_dev_configure,
2410 	.dev_start = mrvl_dev_start,
2411 	.dev_stop = mrvl_dev_stop,
2412 	.dev_set_link_up = mrvl_dev_set_link_up,
2413 	.dev_set_link_down = mrvl_dev_set_link_down,
2414 	.dev_close = mrvl_dev_close,
2415 	.link_update = mrvl_link_update,
2416 	.promiscuous_enable = mrvl_promiscuous_enable,
2417 	.allmulticast_enable = mrvl_allmulticast_enable,
2418 	.promiscuous_disable = mrvl_promiscuous_disable,
2419 	.allmulticast_disable = mrvl_allmulticast_disable,
2420 	.mac_addr_remove = mrvl_mac_addr_remove,
2421 	.mac_addr_add = mrvl_mac_addr_add,
2422 	.mac_addr_set = mrvl_mac_addr_set,
2423 	.mtu_set = mrvl_mtu_set,
2424 	.stats_get = mrvl_stats_get,
2425 	.stats_reset = mrvl_stats_reset,
2426 	.xstats_get = mrvl_xstats_get,
2427 	.xstats_reset = mrvl_xstats_reset,
2428 	.xstats_get_names = mrvl_xstats_get_names,
2429 	.dev_infos_get = mrvl_dev_infos_get,
2430 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
2431 	.rxq_info_get = mrvl_rxq_info_get,
2432 	.txq_info_get = mrvl_txq_info_get,
2433 	.vlan_filter_set = mrvl_vlan_filter_set,
2434 	.vlan_offload_set = mrvl_vlan_offload_set,
2435 	.tx_queue_start = mrvl_tx_queue_start,
2436 	.tx_queue_stop = mrvl_tx_queue_stop,
2437 	.rx_queue_setup = mrvl_rx_queue_setup,
2438 	.rx_queue_release = mrvl_rx_queue_release,
2439 	.tx_queue_setup = mrvl_tx_queue_setup,
2440 	.tx_queue_release = mrvl_tx_queue_release,
2441 	.flow_ctrl_get = mrvl_flow_ctrl_get,
2442 	.flow_ctrl_set = mrvl_flow_ctrl_set,
2443 	.rss_hash_update = mrvl_rss_hash_update,
2444 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
2445 	.flow_ops_get = mrvl_eth_flow_ops_get,
2446 	.mtr_ops_get = mrvl_mtr_ops_get,
2447 	.tm_ops_get = mrvl_tm_ops_get,
2448 };
2449 
2450 /**
2451  * Return packet type information and l3/l4 offsets.
2452  *
2453  * @param desc
2454  *   Pointer to the received packet descriptor.
2455  * @param l3_offset
2456  *   l3 packet offset.
2457  * @param l4_offset
2458  *   l4 packet offset.
2459  *
2460  * @return
2461  *   Packet type information.
2462  */
2463 static inline uint64_t
2464 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
2465 				    uint8_t *l3_offset, uint8_t *l4_offset)
2466 {
2467 	enum pp2_inq_l3_type l3_type;
2468 	enum pp2_inq_l4_type l4_type;
2469 	enum pp2_inq_vlan_tag vlan_tag;
2470 	uint64_t packet_type;
2471 
2472 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
2473 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
2474 	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
2475 
2476 	packet_type = RTE_PTYPE_L2_ETHER;
2477 
2478 	switch (vlan_tag) {
2479 	case PP2_INQ_VLAN_TAG_SINGLE:
2480 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
2481 		break;
2482 	case PP2_INQ_VLAN_TAG_DOUBLE:
2483 	case PP2_INQ_VLAN_TAG_TRIPLE:
2484 		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
2485 		break;
2486 	default:
2487 		break;
2488 	}
2489 
2490 	switch (l3_type) {
2491 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
2492 		packet_type |= RTE_PTYPE_L3_IPV4;
2493 		break;
2494 	case PP2_INQ_L3_TYPE_IPV4_OK:
2495 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
2496 		break;
2497 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
2498 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
2499 		break;
2500 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
2501 		packet_type |= RTE_PTYPE_L3_IPV6;
2502 		break;
2503 	case PP2_INQ_L3_TYPE_IPV6_EXT:
2504 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
2505 		break;
2506 	case PP2_INQ_L3_TYPE_ARP:
2507 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
2508 		/*
2509 		 * In case of ARP l4_offset is set to wrong value.
2510 		 * Set it to proper one so that later on mbuf->l3_len can be
2511 		 * calculated subtracting l4_offset and l3_offset.
2512 		 */
2513 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
2514 		break;
2515 	default:
2516 		break;
2517 	}
2518 
2519 	switch (l4_type) {
2520 	case PP2_INQ_L4_TYPE_TCP:
2521 		packet_type |= RTE_PTYPE_L4_TCP;
2522 		break;
2523 	case PP2_INQ_L4_TYPE_UDP:
2524 		packet_type |= RTE_PTYPE_L4_UDP;
2525 		break;
2526 	default:
2527 		break;
2528 	}
2529 
2530 	return packet_type;
2531 }
2532 
2533 /**
2534  * Get offload information from the received packet descriptor.
2535  *
2536  * @param desc
2537  *   Pointer to the received packet descriptor.
2538  *
2539  * @return
2540  *   Mbuf offload flags.
2541  */
2542 static inline uint64_t
2543 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc, uint64_t packet_type)
2544 {
2545 	uint64_t flags = 0;
2546 	enum pp2_inq_desc_status status;
2547 
2548 	if (RTE_ETH_IS_IPV4_HDR(packet_type)) {
2549 		status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2550 		if (unlikely(status != PP2_DESC_ERR_OK))
2551 			flags |= RTE_MBUF_F_RX_IP_CKSUM_BAD;
2552 		else
2553 			flags |= RTE_MBUF_F_RX_IP_CKSUM_GOOD;
2554 	}
2555 
2556 	if (((packet_type & RTE_PTYPE_L4_UDP) == RTE_PTYPE_L4_UDP) ||
2557 	    ((packet_type & RTE_PTYPE_L4_TCP) == RTE_PTYPE_L4_TCP)) {
2558 		status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2559 		if (unlikely(status != PP2_DESC_ERR_OK))
2560 			flags |= RTE_MBUF_F_RX_L4_CKSUM_BAD;
2561 		else
2562 			flags |= RTE_MBUF_F_RX_L4_CKSUM_GOOD;
2563 	}
2564 
2565 	return flags;
2566 }
2567 
2568 /**
2569  * DPDK callback for receive.
2570  *
2571  * @param rxq
2572  *   Generic pointer to the receive queue.
2573  * @param rx_pkts
2574  *   Array to store received packets.
2575  * @param nb_pkts
2576  *   Maximum number of packets in array.
2577  *
2578  * @return
2579  *   Number of packets successfully received.
2580  */
2581 static uint16_t
2582 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2583 {
2584 	struct mrvl_rxq *q = rxq;
2585 	struct pp2_ppio_desc descs[nb_pkts];
2586 	struct pp2_bpool *bpool;
2587 	int i, ret, rx_done = 0;
2588 	int num;
2589 	struct pp2_hif *hif;
2590 	unsigned int core_id = rte_lcore_id();
2591 
2592 	hif = mrvl_get_hif(q->priv, core_id);
2593 
2594 	if (unlikely(!q->priv->ppio || !hif))
2595 		return 0;
2596 
2597 	bpool = q->priv->bpool;
2598 
2599 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2600 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2601 	if (unlikely(ret < 0))
2602 		return 0;
2603 
2604 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2605 
2606 	for (i = 0; i < nb_pkts; i++) {
2607 		struct rte_mbuf *mbuf;
2608 		uint8_t l3_offset, l4_offset;
2609 		enum pp2_inq_desc_status status;
2610 		uint64_t addr;
2611 
2612 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2613 			struct pp2_ppio_desc *pref_desc;
2614 			u64 pref_addr;
2615 
2616 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2617 			pref_addr = cookie_addr_high |
2618 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2619 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2620 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2621 		}
2622 
2623 		addr = cookie_addr_high |
2624 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2625 		mbuf = (struct rte_mbuf *)addr;
2626 		rte_pktmbuf_reset(mbuf);
2627 
2628 		/* drop packet in case of mac, overrun or resource error */
2629 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2630 		if ((unlikely(status != PP2_DESC_ERR_OK)) &&
2631 			!(q->priv->forward_bad_frames)) {
2632 			struct pp2_buff_inf binf = {
2633 				.addr = rte_mbuf_data_iova_default(mbuf),
2634 				.cookie = (uint64_t)mbuf,
2635 			};
2636 
2637 			pp2_bpool_put_buff(hif, bpool, &binf);
2638 			mrvl_port_bpool_size
2639 				[bpool->pp2_id][bpool->id][core_id]++;
2640 			q->drop_mac++;
2641 			continue;
2642 		}
2643 
2644 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2645 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2646 		mbuf->data_len = mbuf->pkt_len;
2647 		mbuf->port = q->port_id;
2648 		mbuf->packet_type =
2649 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2650 							    &l3_offset,
2651 							    &l4_offset);
2652 		mbuf->l2_len = l3_offset;
2653 		mbuf->l3_len = l4_offset - l3_offset;
2654 
2655 		if (likely(q->cksum_enabled))
2656 			mbuf->ol_flags =
2657 				mrvl_desc_to_ol_flags(&descs[i],
2658 						      mbuf->packet_type);
2659 
2660 		rx_pkts[rx_done++] = mbuf;
2661 		q->bytes_recv += mbuf->pkt_len;
2662 	}
2663 
2664 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2665 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2666 
2667 		if (unlikely(num <= q->priv->bpool_min_size ||
2668 			     (!rx_done && num < q->priv->bpool_init_size))) {
2669 			mrvl_fill_bpool(q, q->priv->fill_bpool_buffs);
2670 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2671 			int i;
2672 			int pkt_to_remove = num - q->priv->bpool_init_size;
2673 			struct rte_mbuf *mbuf;
2674 			struct pp2_buff_inf buff;
2675 
2676 			for (i = 0; i < pkt_to_remove; i++) {
2677 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2678 				if (ret)
2679 					break;
2680 				mbuf = (struct rte_mbuf *)
2681 					(cookie_addr_high | buff.cookie);
2682 				rte_pktmbuf_free(mbuf);
2683 			}
2684 			mrvl_port_bpool_size
2685 				[bpool->pp2_id][bpool->id][core_id] -= i;
2686 		}
2687 		rte_spinlock_unlock(&q->priv->lock);
2688 	}
2689 
2690 	return rx_done;
2691 }
2692 
2693 /**
2694  * Prepare offload information.
2695  *
2696  * @param ol_flags
2697  *   Offload flags.
2698  * @param l3_type
2699  *   Pointer to the pp2_ouq_l3_type structure.
2700  * @param l4_type
2701  *   Pointer to the pp2_outq_l4_type structure.
2702  * @param gen_l3_cksum
2703  *   Will be set to 1 in case l3 checksum is computed.
2704  * @param l4_cksum
2705  *   Will be set to 1 in case l4 checksum is computed.
2706  */
2707 static inline void
2708 mrvl_prepare_proto_info(uint64_t ol_flags,
2709 			enum pp2_outq_l3_type *l3_type,
2710 			enum pp2_outq_l4_type *l4_type,
2711 			int *gen_l3_cksum,
2712 			int *gen_l4_cksum)
2713 {
2714 	/*
2715 	 * Based on ol_flags prepare information
2716 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2717 	 * for offloading.
2718 	 * in most of the checksum cases ipv4 must be set, so this is the
2719 	 * default value
2720 	 */
2721 	*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2722 	*gen_l3_cksum = ol_flags & RTE_MBUF_F_TX_IP_CKSUM ? 1 : 0;
2723 
2724 	if (ol_flags & RTE_MBUF_F_TX_IPV6) {
2725 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2726 		/* no checksum for ipv6 header */
2727 		*gen_l3_cksum = 0;
2728 	}
2729 
2730 	if ((ol_flags & RTE_MBUF_F_TX_L4_MASK) == RTE_MBUF_F_TX_TCP_CKSUM) {
2731 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2732 		*gen_l4_cksum = 1;
2733 	} else if ((ol_flags & RTE_MBUF_F_TX_L4_MASK) ==  RTE_MBUF_F_TX_UDP_CKSUM) {
2734 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2735 		*gen_l4_cksum = 1;
2736 	} else {
2737 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2738 		/* no checksum for other type */
2739 		*gen_l4_cksum = 0;
2740 	}
2741 }
2742 
2743 /**
2744  * Release already sent buffers to bpool (buffer-pool).
2745  *
2746  * @param ppio
2747  *   Pointer to the port structure.
2748  * @param hif
2749  *   Pointer to the MUSDK hardware interface.
2750  * @param sq
2751  *   Pointer to the shadow queue.
2752  * @param qid
2753  *   Queue id number.
2754  * @param force
2755  *   Force releasing packets.
2756  */
2757 static inline void
2758 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2759 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2760 		       int qid, int force)
2761 {
2762 	struct buff_release_entry *entry;
2763 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2764 	int i;
2765 
2766 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2767 
2768 	sq->num_to_release += nb_done;
2769 
2770 	if (likely(!force &&
2771 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2772 		return;
2773 
2774 	nb_done = sq->num_to_release;
2775 	sq->num_to_release = 0;
2776 
2777 	for (i = 0; i < nb_done; i++) {
2778 		entry = &sq->ent[sq->tail + num];
2779 		if (unlikely(!entry->buff.addr)) {
2780 			MRVL_LOG(ERR,
2781 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2782 				sq->tail, (u64)entry->buff.cookie,
2783 				(u64)entry->buff.addr);
2784 			skip_bufs = 1;
2785 			goto skip;
2786 		}
2787 
2788 		if (unlikely(!entry->bpool)) {
2789 			struct rte_mbuf *mbuf;
2790 
2791 			mbuf = (struct rte_mbuf *)entry->buff.cookie;
2792 			rte_pktmbuf_free(mbuf);
2793 			skip_bufs = 1;
2794 			goto skip;
2795 		}
2796 
2797 		mrvl_port_bpool_size
2798 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2799 		num++;
2800 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2801 			goto skip;
2802 		continue;
2803 skip:
2804 		if (likely(num))
2805 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2806 		num += skip_bufs;
2807 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2808 		sq->size -= num;
2809 		num = 0;
2810 		skip_bufs = 0;
2811 	}
2812 
2813 	if (likely(num)) {
2814 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2815 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2816 		sq->size -= num;
2817 	}
2818 }
2819 
2820 /**
2821  * DPDK callback for transmit.
2822  *
2823  * @param txq
2824  *   Generic pointer transmit queue.
2825  * @param tx_pkts
2826  *   Packets to transmit.
2827  * @param nb_pkts
2828  *   Number of packets in array.
2829  *
2830  * @return
2831  *   Number of packets successfully transmitted.
2832  */
2833 static uint16_t
2834 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2835 {
2836 	struct mrvl_txq *q = txq;
2837 	struct mrvl_shadow_txq *sq;
2838 	struct pp2_hif *hif;
2839 	struct pp2_ppio_desc descs[nb_pkts];
2840 	unsigned int core_id = rte_lcore_id();
2841 	int i, bytes_sent = 0;
2842 	uint16_t num, sq_free_size;
2843 	uint64_t addr;
2844 
2845 	hif = mrvl_get_hif(q->priv, core_id);
2846 	sq = &q->shadow_txqs[core_id];
2847 
2848 	if (unlikely(!q->priv->ppio || !hif))
2849 		return 0;
2850 
2851 	if (sq->size)
2852 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2853 				       sq, q->queue_id, 0);
2854 
2855 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2856 	if (unlikely(nb_pkts > sq_free_size))
2857 		nb_pkts = sq_free_size;
2858 
2859 	for (i = 0; i < nb_pkts; i++) {
2860 		struct rte_mbuf *mbuf = tx_pkts[i];
2861 		int gen_l3_cksum, gen_l4_cksum;
2862 		enum pp2_outq_l3_type l3_type;
2863 		enum pp2_outq_l4_type l4_type;
2864 
2865 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2866 			struct rte_mbuf *pref_pkt_hdr;
2867 
2868 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2869 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2870 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2871 		}
2872 
2873 		mrvl_fill_shadowq(sq, mbuf);
2874 		mrvl_fill_desc(&descs[i], mbuf);
2875 
2876 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2877 		/*
2878 		 * in case unsupported ol_flags were passed
2879 		 * do not update descriptor offload information
2880 		 */
2881 		if (!(mbuf->ol_flags & MRVL_TX_PKT_OFFLOADS))
2882 			continue;
2883 		mrvl_prepare_proto_info(mbuf->ol_flags, &l3_type, &l4_type,
2884 					&gen_l3_cksum, &gen_l4_cksum);
2885 
2886 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2887 						  mbuf->l2_len,
2888 						  mbuf->l2_len + mbuf->l3_len,
2889 						  gen_l3_cksum, gen_l4_cksum);
2890 	}
2891 
2892 	num = nb_pkts;
2893 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2894 	/* number of packets that were not sent */
2895 	if (unlikely(num > nb_pkts)) {
2896 		for (i = nb_pkts; i < num; i++) {
2897 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2898 				MRVL_PP2_TX_SHADOWQ_MASK;
2899 			addr = sq->ent[sq->head].buff.cookie;
2900 			bytes_sent -=
2901 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2902 		}
2903 		sq->size -= num - nb_pkts;
2904 	}
2905 
2906 	q->bytes_sent += bytes_sent;
2907 
2908 	return nb_pkts;
2909 }
2910 
2911 /** DPDK callback for S/G transmit.
2912  *
2913  * @param txq
2914  *   Generic pointer transmit queue.
2915  * @param tx_pkts
2916  *   Packets to transmit.
2917  * @param nb_pkts
2918  *   Number of packets in array.
2919  *
2920  * @return
2921  *   Number of packets successfully transmitted.
2922  */
2923 static uint16_t
2924 mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
2925 		     uint16_t nb_pkts)
2926 {
2927 	struct mrvl_txq *q = txq;
2928 	struct mrvl_shadow_txq *sq;
2929 	struct pp2_hif *hif;
2930 	struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
2931 	struct pp2_ppio_sg_pkts pkts;
2932 	uint8_t frags[nb_pkts];
2933 	unsigned int core_id = rte_lcore_id();
2934 	int i, j, bytes_sent = 0;
2935 	int tail, tail_first;
2936 	uint16_t num, sq_free_size;
2937 	uint16_t nb_segs, total_descs = 0;
2938 	uint64_t addr;
2939 
2940 	hif = mrvl_get_hif(q->priv, core_id);
2941 	sq = &q->shadow_txqs[core_id];
2942 	pkts.frags = frags;
2943 	pkts.num = 0;
2944 
2945 	if (unlikely(!q->priv->ppio || !hif))
2946 		return 0;
2947 
2948 	if (sq->size)
2949 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2950 				       sq, q->queue_id, 0);
2951 
2952 	/* Save shadow queue free size */
2953 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2954 
2955 	tail = 0;
2956 	for (i = 0; i < nb_pkts; i++) {
2957 		struct rte_mbuf *mbuf = tx_pkts[i];
2958 		struct rte_mbuf *seg = NULL;
2959 		int gen_l3_cksum, gen_l4_cksum;
2960 		enum pp2_outq_l3_type l3_type;
2961 		enum pp2_outq_l4_type l4_type;
2962 
2963 		nb_segs = mbuf->nb_segs;
2964 		tail_first = tail;
2965 		total_descs += nb_segs;
2966 
2967 		/*
2968 		 * Check if total_descs does not exceed
2969 		 * shadow queue free size
2970 		 */
2971 		if (unlikely(total_descs > sq_free_size)) {
2972 			total_descs -= nb_segs;
2973 			break;
2974 		}
2975 
2976 		/* Check if nb_segs does not exceed the max nb of desc per
2977 		 * fragmented packet
2978 		 */
2979 		if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
2980 			total_descs -= nb_segs;
2981 			MRVL_LOG(ERR, "Too many segments. Packet won't be sent.");
2982 			break;
2983 		}
2984 
2985 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2986 			struct rte_mbuf *pref_pkt_hdr;
2987 
2988 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2989 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2990 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2991 		}
2992 
2993 		pkts.frags[pkts.num] = nb_segs;
2994 		pkts.num++;
2995 
2996 		seg = mbuf;
2997 		for (j = 0; j < nb_segs - 1; j++) {
2998 			/* For the subsequent segments, set shadow queue
2999 			 * buffer to NULL
3000 			 */
3001 			mrvl_fill_shadowq(sq, NULL);
3002 			mrvl_fill_desc(&descs[tail], seg);
3003 
3004 			tail++;
3005 			seg = seg->next;
3006 		}
3007 		/* Put first mbuf info in last shadow queue entry */
3008 		mrvl_fill_shadowq(sq, mbuf);
3009 		/* Update descriptor with last segment */
3010 		mrvl_fill_desc(&descs[tail++], seg);
3011 
3012 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
3013 		/* In case unsupported ol_flags were passed
3014 		 * do not update descriptor offload information
3015 		 */
3016 		if (!(mbuf->ol_flags & MRVL_TX_PKT_OFFLOADS))
3017 			continue;
3018 		mrvl_prepare_proto_info(mbuf->ol_flags, &l3_type, &l4_type,
3019 					&gen_l3_cksum, &gen_l4_cksum);
3020 
3021 		pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
3022 						  l4_type, mbuf->l2_len,
3023 						  mbuf->l2_len + mbuf->l3_len,
3024 						  gen_l3_cksum, gen_l4_cksum);
3025 	}
3026 
3027 	num = total_descs;
3028 	pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
3029 			 &total_descs, &pkts);
3030 	/* number of packets that were not sent */
3031 	if (unlikely(num > total_descs)) {
3032 		for (i = total_descs; i < num; i++) {
3033 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
3034 				MRVL_PP2_TX_SHADOWQ_MASK;
3035 
3036 			addr = sq->ent[sq->head].buff.cookie;
3037 			if (addr)
3038 				bytes_sent -=
3039 					rte_pktmbuf_pkt_len((struct rte_mbuf *)
3040 						(cookie_addr_high | addr));
3041 		}
3042 		sq->size -= num - total_descs;
3043 		nb_pkts = pkts.num;
3044 	}
3045 
3046 	q->bytes_sent += bytes_sent;
3047 
3048 	return nb_pkts;
3049 }
3050 
3051 /**
3052  * Create private device structure.
3053  *
3054  * @param dev_name
3055  *   Pointer to the port name passed in the initialization parameters.
3056  *
3057  * @return
3058  *   Pointer to the newly allocated private device structure.
3059  */
3060 static struct mrvl_priv *
3061 mrvl_priv_create(const char *dev_name)
3062 {
3063 	struct pp2_bpool_params bpool_params;
3064 	char match[MRVL_MATCH_LEN];
3065 	struct mrvl_priv *priv;
3066 	uint16_t max_frame_size;
3067 	int ret, bpool_bit;
3068 
3069 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
3070 	if (!priv)
3071 		return NULL;
3072 
3073 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
3074 				       &priv->pp_id, &priv->ppio_id);
3075 	if (ret)
3076 		goto out_free_priv;
3077 
3078 	ret = pp2_ppio_get_l4_cksum_max_frame_size(priv->pp_id, priv->ppio_id,
3079 						   &max_frame_size);
3080 	if (ret)
3081 		goto out_free_priv;
3082 
3083 	priv->max_mtu = max_frame_size + RTE_ETHER_CRC_LEN -
3084 		MRVL_PP2_ETH_HDRS_LEN;
3085 
3086 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
3087 				     PP2_BPOOL_NUM_POOLS);
3088 	if (bpool_bit < 0)
3089 		goto out_free_priv;
3090 	priv->bpool_bit = bpool_bit;
3091 
3092 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
3093 		 priv->bpool_bit);
3094 	memset(&bpool_params, 0, sizeof(bpool_params));
3095 	bpool_params.match = match;
3096 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
3097 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
3098 	if (ret)
3099 		goto out_clear_bpool_bit;
3100 
3101 	priv->ppio_params.type = PP2_PPIO_T_NIC;
3102 	rte_spinlock_init(&priv->lock);
3103 
3104 	return priv;
3105 out_clear_bpool_bit:
3106 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
3107 out_free_priv:
3108 	rte_free(priv);
3109 	return NULL;
3110 }
3111 
3112 /**
3113  * Create device representing Ethernet port.
3114  *
3115  * @param name
3116  *   Pointer to the port's name.
3117  *
3118  * @return
3119  *   0 on success, negative error value otherwise.
3120  */
3121 static int
3122 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
3123 {
3124 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
3125 	struct rte_eth_dev *eth_dev;
3126 	struct mrvl_priv *priv;
3127 	struct ifreq req;
3128 
3129 	eth_dev = rte_eth_dev_allocate(name);
3130 	if (!eth_dev)
3131 		return -ENOMEM;
3132 
3133 	priv = mrvl_priv_create(name);
3134 	if (!priv) {
3135 		ret = -ENOMEM;
3136 		goto out_free;
3137 	}
3138 	eth_dev->data->dev_private = priv;
3139 
3140 	eth_dev->data->mac_addrs =
3141 		rte_zmalloc("mac_addrs",
3142 			    RTE_ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
3143 	if (!eth_dev->data->mac_addrs) {
3144 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
3145 		ret = -ENOMEM;
3146 		goto out_free;
3147 	}
3148 
3149 	memset(&req, 0, sizeof(req));
3150 	strcpy(req.ifr_name, name);
3151 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
3152 	if (ret)
3153 		goto out_free;
3154 
3155 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
3156 	       req.ifr_addr.sa_data, RTE_ETHER_ADDR_LEN);
3157 
3158 	eth_dev->device = &vdev->device;
3159 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
3160 	mrvl_set_tx_function(eth_dev);
3161 	eth_dev->dev_ops = &mrvl_ops;
3162 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
3163 
3164 	eth_dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
3165 
3166 	rte_eth_dev_probing_finish(eth_dev);
3167 	return 0;
3168 out_free:
3169 	rte_eth_dev_release_port(eth_dev);
3170 
3171 	return ret;
3172 }
3173 
3174 /**
3175  * Callback used by rte_kvargs_process() during argument parsing.
3176  *
3177  * @param key
3178  *   Pointer to the parsed key (unused).
3179  * @param value
3180  *   Pointer to the parsed value.
3181  * @param extra_args
3182  *   Pointer to the extra arguments which contains address of the
3183  *   table of pointers to parsed interface names.
3184  *
3185  * @return
3186  *   Always 0.
3187  */
3188 static int
3189 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
3190 		 void *extra_args)
3191 {
3192 	struct mrvl_ifnames *ifnames = extra_args;
3193 
3194 	ifnames->names[ifnames->idx++] = value;
3195 
3196 	return 0;
3197 }
3198 
3199 /**
3200  * DPDK callback to register the virtual device.
3201  *
3202  * @param vdev
3203  *   Pointer to the virtual device.
3204  *
3205  * @return
3206  *   0 on success, negative error value otherwise.
3207  */
3208 static int
3209 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
3210 {
3211 	struct rte_kvargs *kvlist;
3212 	struct mrvl_ifnames ifnames;
3213 	int ret = -EINVAL;
3214 	uint32_t i, ifnum, cfgnum;
3215 	const char *params;
3216 
3217 	params = rte_vdev_device_args(vdev);
3218 	if (!params)
3219 		return -EINVAL;
3220 
3221 	kvlist = rte_kvargs_parse(params, valid_args);
3222 	if (!kvlist)
3223 		return -EINVAL;
3224 
3225 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
3226 	if (ifnum > RTE_DIM(ifnames.names))
3227 		goto out_free_kvlist;
3228 
3229 	ifnames.idx = 0;
3230 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
3231 			   mrvl_get_ifnames, &ifnames);
3232 
3233 
3234 	/*
3235 	 * The below system initialization should be done only once,
3236 	 * on the first provided configuration file
3237 	 */
3238 	if (!mrvl_cfg) {
3239 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
3240 		MRVL_LOG(INFO, "Parsing config file!");
3241 		if (cfgnum > 1) {
3242 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
3243 			goto out_free_kvlist;
3244 		} else if (cfgnum == 1) {
3245 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
3246 					   mrvl_get_cfg, &mrvl_cfg);
3247 		}
3248 	}
3249 
3250 	if (mrvl_dev_num)
3251 		goto init_devices;
3252 
3253 	MRVL_LOG(INFO, "Perform MUSDK initializations");
3254 
3255 	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
3256 	if (ret)
3257 		goto out_free_kvlist;
3258 
3259 	ret = mrvl_init_pp2();
3260 	if (ret) {
3261 		MRVL_LOG(ERR, "Failed to init PP!");
3262 		rte_mvep_deinit(MVEP_MOD_T_PP2);
3263 		goto out_free_kvlist;
3264 	}
3265 
3266 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
3267 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
3268 
3269 	mrvl_lcore_first = RTE_MAX_LCORE;
3270 	mrvl_lcore_last = 0;
3271 
3272 init_devices:
3273 	for (i = 0; i < ifnum; i++) {
3274 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
3275 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
3276 		if (ret)
3277 			goto out_cleanup;
3278 		mrvl_dev_num++;
3279 	}
3280 
3281 	rte_kvargs_free(kvlist);
3282 
3283 	return 0;
3284 out_cleanup:
3285 	rte_pmd_mrvl_remove(vdev);
3286 
3287 out_free_kvlist:
3288 	rte_kvargs_free(kvlist);
3289 
3290 	return ret;
3291 }
3292 
3293 /**
3294  * DPDK callback to remove virtual device.
3295  *
3296  * @param vdev
3297  *   Pointer to the removed virtual device.
3298  *
3299  * @return
3300  *   0 on success, negative error value otherwise.
3301  */
3302 static int
3303 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
3304 {
3305 	uint16_t port_id;
3306 	int ret = 0;
3307 
3308 	RTE_ETH_FOREACH_DEV(port_id) {
3309 		if (rte_eth_devices[port_id].device != &vdev->device)
3310 			continue;
3311 		ret |= rte_eth_dev_close(port_id);
3312 	}
3313 
3314 	return ret == 0 ? 0 : -EIO;
3315 }
3316 
3317 static struct rte_vdev_driver pmd_mrvl_drv = {
3318 	.probe = rte_pmd_mrvl_probe,
3319 	.remove = rte_pmd_mrvl_remove,
3320 };
3321 
3322 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
3323 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
3324 RTE_LOG_REGISTER_DEFAULT(mrvl_logtype, NOTICE);
3325