xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision 59f3a8acbcdbafeebe816a26d76dfb06e6450f31)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2017-2021 Marvell International Ltd.
3  * Copyright(c) 2017-2021 Semihalf.
4  * All rights reserved.
5  */
6 
7 #include <rte_string_fns.h>
8 #include <ethdev_driver.h>
9 #include <rte_kvargs.h>
10 #include <rte_log.h>
11 #include <rte_malloc.h>
12 #include <rte_bus_vdev.h>
13 
14 #include <fcntl.h>
15 #include <linux/ethtool.h>
16 #include <linux/sockios.h>
17 #include <net/if.h>
18 #include <net/if_arp.h>
19 #include <sys/ioctl.h>
20 #include <sys/socket.h>
21 #include <sys/stat.h>
22 #include <sys/types.h>
23 
24 #include <rte_mvep_common.h>
25 #include "mrvl_ethdev.h"
26 #include "mrvl_qos.h"
27 #include "mrvl_flow.h"
28 #include "mrvl_mtr.h"
29 #include "mrvl_tm.h"
30 
31 /* bitmask with reserved hifs */
32 #define MRVL_MUSDK_HIFS_RESERVED 0x0F
33 /* bitmask with reserved bpools */
34 #define MRVL_MUSDK_BPOOLS_RESERVED 0x07
35 /* bitmask with reserved kernel RSS tables */
36 #define MRVL_MUSDK_RSS_RESERVED 0x0F
37 /* maximum number of available hifs */
38 #define MRVL_MUSDK_HIFS_MAX 9
39 
40 /* prefetch shift */
41 #define MRVL_MUSDK_PREFETCH_SHIFT 2
42 
43 /* TCAM has 25 entries reserved for uc/mc filter entries
44  * + 1 for primary mac address
45  */
46 #define MRVL_MAC_ADDRS_MAX (1 + 25)
47 #define MRVL_MATCH_LEN 16
48 #define MRVL_PKT_EFFEC_OFFS (MRVL_PKT_OFFS + MV_MH_SIZE)
49 /* Maximum allowable packet size */
50 #define MRVL_PKT_SIZE_MAX (10240 - MV_MH_SIZE)
51 
52 #define MRVL_IFACE_NAME_ARG "iface"
53 #define MRVL_CFG_ARG "cfg"
54 
55 #define MRVL_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 (DEV_RX_OFFLOAD_VLAN_FILTER | \
62 			  DEV_RX_OFFLOAD_CHECKSUM)
63 
64 /** Port Tx offloads capabilities */
65 #define MRVL_TX_OFFLOAD_CHECKSUM (DEV_TX_OFFLOAD_IPV4_CKSUM | \
66 				  DEV_TX_OFFLOAD_UDP_CKSUM  | \
67 				  DEV_TX_OFFLOAD_TCP_CKSUM)
68 #define MRVL_TX_OFFLOADS (MRVL_TX_OFFLOAD_CHECKSUM | \
69 			  DEV_TX_OFFLOAD_MULTI_SEGS)
70 
71 #define MRVL_TX_PKT_OFFLOADS (PKT_TX_IP_CKSUM | \
72 			      PKT_TX_TCP_CKSUM | \
73 			      PKT_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 - __builtin_clz(*bitmap);
197 
198 	if (n >= max)
199 		return -1;
200 
201 	*bitmap |= 1 << n;
202 
203 	return n;
204 }
205 
206 static int
207 mrvl_pp2_fixup_init(void)
208 {
209 	struct pp2_bpool_params bpool_params;
210 	char			name[15];
211 	int			err, i;
212 
213 	memset(dummy_pool, 0, sizeof(dummy_pool));
214 	for (i = 0; i < pp2_get_num_inst(); i++) {
215 		dummy_pool_id[i] = mrvl_reserve_bit(&used_bpools[i],
216 					     PP2_BPOOL_NUM_POOLS);
217 		if (dummy_pool_id[i] < 0) {
218 			MRVL_LOG(ERR, "Can't find free pool\n");
219 			return -1;
220 		}
221 
222 		memset(name, 0, sizeof(name));
223 		snprintf(name, sizeof(name), "pool-%d:%d", i, dummy_pool_id[i]);
224 		memset(&bpool_params, 0, sizeof(bpool_params));
225 		bpool_params.match = name;
226 		bpool_params.buff_len = MRVL_PKT_OFFS;
227 		bpool_params.dummy_short_pool = 1;
228 		err = pp2_bpool_init(&bpool_params, &dummy_pool[i]);
229 		if (err != 0 || !dummy_pool[i]) {
230 			MRVL_LOG(ERR, "BPool init failed!\n");
231 			used_bpools[i] &= ~(1 << dummy_pool_id[i]);
232 			return -1;
233 		}
234 	}
235 
236 	return 0;
237 }
238 
239 /**
240  * Initialize packet processor.
241  *
242  * @return
243  *   0 on success, negative error value otherwise.
244  */
245 static int
246 mrvl_init_pp2(void)
247 {
248 	struct pp2_init_params	init_params;
249 	int			err;
250 
251 	memset(&init_params, 0, sizeof(init_params));
252 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
253 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
254 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
255 	if (mrvl_cfg && mrvl_cfg->pp2_cfg.prs_udfs.num_udfs)
256 		memcpy(&init_params.prs_udfs, &mrvl_cfg->pp2_cfg.prs_udfs,
257 		       sizeof(struct pp2_parse_udfs));
258 	err = pp2_init(&init_params);
259 	if (err != 0) {
260 		MRVL_LOG(ERR, "PP2 init failed");
261 		return -1;
262 	}
263 
264 	err = mrvl_pp2_fixup_init();
265 	if (err != 0) {
266 		MRVL_LOG(ERR, "PP2 fixup init failed");
267 		return -1;
268 	}
269 
270 	return 0;
271 }
272 
273 static void
274 mrvl_pp2_fixup_deinit(void)
275 {
276 	int i;
277 
278 	for (i = 0; i < PP2_NUM_PKT_PROC; i++) {
279 		if (!dummy_pool[i])
280 			continue;
281 		pp2_bpool_deinit(dummy_pool[i]);
282 		used_bpools[i] &= ~(1 << dummy_pool_id[i]);
283 	}
284 }
285 
286 /**
287  * Deinitialize packet processor.
288  *
289  * @return
290  *   0 on success, negative error value otherwise.
291  */
292 static void
293 mrvl_deinit_pp2(void)
294 {
295 	mrvl_pp2_fixup_deinit();
296 	pp2_deinit();
297 }
298 
299 static inline void
300 mrvl_fill_shadowq(struct mrvl_shadow_txq *sq, struct rte_mbuf *buf)
301 {
302 	sq->ent[sq->head].buff.cookie = (uint64_t)buf;
303 	sq->ent[sq->head].buff.addr = buf ?
304 		rte_mbuf_data_iova_default(buf) : 0;
305 
306 	sq->ent[sq->head].bpool =
307 		(unlikely(!buf || buf->port >= RTE_MAX_ETHPORTS ||
308 		 buf->refcnt > 1)) ? NULL :
309 		 mrvl_port_to_bpool_lookup[buf->port];
310 
311 	sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
312 	sq->size++;
313 }
314 
315 /**
316  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
317  */
318 static void
319 mrvl_deinit_hifs(void)
320 {
321 	int i;
322 
323 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
324 		if (hifs[i])
325 			pp2_hif_deinit(hifs[i]);
326 	}
327 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
328 	memset(hifs, 0, sizeof(hifs));
329 }
330 
331 static inline void
332 mrvl_fill_desc(struct pp2_ppio_desc *desc, struct rte_mbuf *buf)
333 {
334 	pp2_ppio_outq_desc_reset(desc);
335 	pp2_ppio_outq_desc_set_phys_addr(desc, rte_pktmbuf_iova(buf));
336 	pp2_ppio_outq_desc_set_pkt_offset(desc, 0);
337 	pp2_ppio_outq_desc_set_pkt_len(desc, rte_pktmbuf_data_len(buf));
338 }
339 
340 static inline int
341 mrvl_get_bpool_size(int pp2_id, int pool_id)
342 {
343 	int i;
344 	int size = 0;
345 
346 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
347 		size += mrvl_port_bpool_size[pp2_id][pool_id][i];
348 
349 	return size;
350 }
351 
352 static int
353 mrvl_init_hif(int core_id)
354 {
355 	struct pp2_hif_params params;
356 	char match[MRVL_MATCH_LEN];
357 	int ret;
358 
359 	ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
360 	if (ret < 0) {
361 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
362 		return ret;
363 	}
364 
365 	snprintf(match, sizeof(match), "hif-%d", ret);
366 	memset(&params, 0, sizeof(params));
367 	params.match = match;
368 	params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
369 	ret = pp2_hif_init(&params, &hifs[core_id]);
370 	if (ret) {
371 		MRVL_LOG(ERR, "Failed to initialize hif %d", core_id);
372 		return ret;
373 	}
374 
375 	return 0;
376 }
377 
378 static inline struct pp2_hif*
379 mrvl_get_hif(struct mrvl_priv *priv, int core_id)
380 {
381 	int ret;
382 
383 	if (likely(hifs[core_id] != NULL))
384 		return hifs[core_id];
385 
386 	rte_spinlock_lock(&priv->lock);
387 
388 	ret = mrvl_init_hif(core_id);
389 	if (ret < 0) {
390 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
391 		goto out;
392 	}
393 
394 	if (core_id < mrvl_lcore_first)
395 		mrvl_lcore_first = core_id;
396 
397 	if (core_id > mrvl_lcore_last)
398 		mrvl_lcore_last = core_id;
399 out:
400 	rte_spinlock_unlock(&priv->lock);
401 
402 	return hifs[core_id];
403 }
404 
405 /**
406  * Set tx burst function according to offload flag
407  *
408  * @param dev
409  *   Pointer to Ethernet device structure.
410  */
411 static void
412 mrvl_set_tx_function(struct rte_eth_dev *dev)
413 {
414 	struct mrvl_priv *priv = dev->data->dev_private;
415 
416 	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
417 	if (priv->multiseg) {
418 		RTE_LOG(INFO, PMD, "Using multi-segment tx callback\n");
419 		dev->tx_pkt_burst = mrvl_tx_sg_pkt_burst;
420 	} else {
421 		RTE_LOG(INFO, PMD, "Using single-segment tx callback\n");
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 & 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 & 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 & 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 != ETH_MQ_RX_NONE &&
487 	    dev->data->dev_conf.rxmode.mq_mode != 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.split_hdr_size) {
494 		MRVL_LOG(INFO, "Split headers not supported");
495 		return -EINVAL;
496 	}
497 
498 	if (dev->data->dev_conf.rxmode.mtu > priv->max_mtu) {
499 		MRVL_LOG(ERR, "MTU %u is larger than max_mtu %u\n",
500 			 dev->data->dev_conf.rxmode.mtu,
501 			 priv->max_mtu);
502 		return -EINVAL;
503 	}
504 
505 	if (dev->data->dev_conf.txmode.offloads & DEV_TX_OFFLOAD_MULTI_SEGS)
506 		priv->multiseg = 1;
507 
508 	ret = mrvl_configure_rxqs(priv, dev->data->port_id,
509 				  dev->data->nb_rx_queues);
510 	if (ret < 0)
511 		return ret;
512 
513 	ret = mrvl_configure_txqs(priv, dev->data->port_id,
514 				  dev->data->nb_tx_queues);
515 	if (ret < 0)
516 		return ret;
517 
518 	priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
519 	priv->ppio_params.maintain_stats = 1;
520 	priv->nb_rx_queues = dev->data->nb_rx_queues;
521 
522 	ret = mrvl_tm_init(dev);
523 	if (ret < 0)
524 		return ret;
525 
526 	if (dev->data->nb_rx_queues == 1 &&
527 	    dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
528 		MRVL_LOG(WARNING, "Disabling hash for 1 rx queue");
529 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
530 		priv->configured = 1;
531 		return 0;
532 	}
533 
534 	ret = mrvl_configure_rss(priv,
535 			&dev->data->dev_conf.rx_adv_conf.rss_conf);
536 	if (ret < 0)
537 		return ret;
538 
539 	priv->configured = 1;
540 
541 	return 0;
542 }
543 
544 /**
545  * DPDK callback to change the MTU.
546  *
547  * Setting the MTU affects hardware MRU (packets larger than the MRU
548  * will be dropped).
549  *
550  * @param dev
551  *   Pointer to Ethernet device structure.
552  * @param mtu
553  *   New MTU.
554  *
555  * @return
556  *   0 on success, negative error value otherwise.
557  */
558 static int
559 mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
560 {
561 	struct mrvl_priv *priv = dev->data->dev_private;
562 	uint16_t mru;
563 	uint16_t mbuf_data_size = 0; /* SW buffer size */
564 	int ret;
565 
566 	mru = MRVL_PP2_MTU_TO_MRU(mtu);
567 	/*
568 	 * min_rx_buf_size is equal to mbuf data size
569 	 * if pmd didn't set it differently
570 	 */
571 	mbuf_data_size = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
572 	/* Prevent PMD from:
573 	 * - setting mru greater than the mbuf size resulting in
574 	 * hw and sw buffer size mismatch
575 	 * - setting mtu that requires the support of scattered packets
576 	 * when this feature has not been enabled/supported so far
577 	 * (TODO check scattered_rx flag here once scattered RX is supported).
578 	 */
579 	if (mru - RTE_ETHER_CRC_LEN + MRVL_PKT_OFFS > mbuf_data_size) {
580 		mru = mbuf_data_size + RTE_ETHER_CRC_LEN - MRVL_PKT_OFFS;
581 		mtu = MRVL_PP2_MRU_TO_MTU(mru);
582 		MRVL_LOG(WARNING, "MTU too big, max MTU possible limitted "
583 			"by current mbuf size: %u. Set MTU to %u, MRU to %u",
584 			mbuf_data_size, mtu, mru);
585 	}
586 
587 	if (mtu < RTE_ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX) {
588 		MRVL_LOG(ERR, "Invalid MTU [%u] or MRU [%u]", mtu, mru);
589 		return -EINVAL;
590 	}
591 
592 	if (!priv->ppio)
593 		return 0;
594 
595 	ret = pp2_ppio_set_mru(priv->ppio, mru);
596 	if (ret) {
597 		MRVL_LOG(ERR, "Failed to change MRU");
598 		return ret;
599 	}
600 
601 	ret = pp2_ppio_set_mtu(priv->ppio, mtu);
602 	if (ret) {
603 		MRVL_LOG(ERR, "Failed to change MTU");
604 		return ret;
605 	}
606 
607 	return 0;
608 }
609 
610 /**
611  * DPDK callback to bring the link up.
612  *
613  * @param dev
614  *   Pointer to Ethernet device structure.
615  *
616  * @return
617  *   0 on success, negative error value otherwise.
618  */
619 static int
620 mrvl_dev_set_link_up(struct rte_eth_dev *dev)
621 {
622 	struct mrvl_priv *priv = dev->data->dev_private;
623 	int ret;
624 
625 	if (!priv->ppio) {
626 		dev->data->dev_link.link_status = ETH_LINK_UP;
627 		return 0;
628 	}
629 
630 	ret = pp2_ppio_enable(priv->ppio);
631 	if (ret)
632 		return ret;
633 
634 	/*
635 	 * mtu/mru can be updated if pp2_ppio_enable() was called at least once
636 	 * as pp2_ppio_enable() changes port->t_mode from default 0 to
637 	 * PP2_TRAFFIC_INGRESS_EGRESS.
638 	 *
639 	 * Set mtu to default DPDK value here.
640 	 */
641 	ret = mrvl_mtu_set(dev, dev->data->mtu);
642 	if (ret) {
643 		pp2_ppio_disable(priv->ppio);
644 		return ret;
645 	}
646 
647 	dev->data->dev_link.link_status = ETH_LINK_UP;
648 	return 0;
649 }
650 
651 /**
652  * DPDK callback to bring the link down.
653  *
654  * @param dev
655  *   Pointer to Ethernet device structure.
656  *
657  * @return
658  *   0 on success, negative error value otherwise.
659  */
660 static int
661 mrvl_dev_set_link_down(struct rte_eth_dev *dev)
662 {
663 	struct mrvl_priv *priv = dev->data->dev_private;
664 	int ret;
665 
666 	if (!priv->ppio) {
667 		dev->data->dev_link.link_status = ETH_LINK_DOWN;
668 		return 0;
669 	}
670 	ret = pp2_ppio_disable(priv->ppio);
671 	if (ret)
672 		return ret;
673 
674 	dev->data->dev_link.link_status = ETH_LINK_DOWN;
675 	return 0;
676 }
677 
678 /**
679  * DPDK callback to start tx queue.
680  *
681  * @param dev
682  *   Pointer to Ethernet device structure.
683  * @param queue_id
684  *   Transmit queue index.
685  *
686  * @return
687  *   0 on success, negative error value otherwise.
688  */
689 static int
690 mrvl_tx_queue_start(struct rte_eth_dev *dev, uint16_t queue_id)
691 {
692 	struct mrvl_priv *priv = dev->data->dev_private;
693 	int ret;
694 
695 	if (!priv)
696 		return -EPERM;
697 
698 	/* passing 1 enables given tx queue */
699 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 1);
700 	if (ret) {
701 		MRVL_LOG(ERR, "Failed to start txq %d", queue_id);
702 		return ret;
703 	}
704 
705 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
706 
707 	return 0;
708 }
709 
710 /**
711  * DPDK callback to stop tx queue.
712  *
713  * @param dev
714  *   Pointer to Ethernet device structure.
715  * @param queue_id
716  *   Transmit queue index.
717  *
718  * @return
719  *   0 on success, negative error value otherwise.
720  */
721 static int
722 mrvl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t queue_id)
723 {
724 	struct mrvl_priv *priv = dev->data->dev_private;
725 	int ret;
726 
727 	if (!priv->ppio)
728 		return -EPERM;
729 
730 	/* passing 0 disables given tx queue */
731 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 0);
732 	if (ret) {
733 		MRVL_LOG(ERR, "Failed to stop txq %d", queue_id);
734 		return ret;
735 	}
736 
737 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
738 
739 	return 0;
740 }
741 
742 /**
743  * Populate VLAN Filter configuration.
744  *
745  * @param dev
746  *   Pointer to Ethernet device structure.
747  * @param on
748  *   Toggle filter.
749  *
750  * @return
751  *   0 on success, negative error value otherwise.
752  */
753 static int mrvl_populate_vlan_table(struct rte_eth_dev *dev, int on)
754 {
755 	uint32_t j;
756 	int ret;
757 	struct rte_vlan_filter_conf *vfc;
758 
759 	vfc = &dev->data->vlan_filter_conf;
760 	for (j = 0; j < RTE_DIM(vfc->ids); j++) {
761 		uint64_t vlan;
762 		uint64_t vbit;
763 		uint64_t ids = vfc->ids[j];
764 
765 		if (ids == 0)
766 			continue;
767 
768 		while (ids) {
769 			vlan = 64 * j;
770 			/* count trailing zeroes */
771 			vbit = ~ids & (ids - 1);
772 			/* clear least significant bit set */
773 			ids ^= (ids ^ (ids - 1)) ^ vbit;
774 			for (; vbit; vlan++)
775 				vbit >>= 1;
776 			ret = mrvl_vlan_filter_set(dev, vlan, on);
777 			if (ret) {
778 				MRVL_LOG(ERR, "Failed to setup VLAN filter\n");
779 				return ret;
780 			}
781 		}
782 	}
783 
784 	return 0;
785 }
786 
787 /**
788  * DPDK callback to start the device.
789  *
790  * @param dev
791  *   Pointer to Ethernet device structure.
792  *
793  * @return
794  *   0 on success, negative errno value on failure.
795  */
796 static int
797 mrvl_dev_start(struct rte_eth_dev *dev)
798 {
799 	struct mrvl_priv *priv = dev->data->dev_private;
800 	char match[MRVL_MATCH_LEN];
801 	int ret = 0, i, def_init_size;
802 	struct rte_ether_addr *mac_addr;
803 
804 	if (priv->ppio)
805 		return mrvl_dev_set_link_up(dev);
806 
807 	snprintf(match, sizeof(match), "ppio-%d:%d",
808 		 priv->pp_id, priv->ppio_id);
809 	priv->ppio_params.match = match;
810 	priv->ppio_params.eth_start_hdr = PP2_PPIO_HDR_ETH;
811 	priv->forward_bad_frames = 0;
812 	priv->fill_bpool_buffs = MRVL_BURST_SIZE;
813 
814 	if (mrvl_cfg) {
815 		priv->ppio_params.eth_start_hdr =
816 			mrvl_cfg->port[dev->data->port_id].eth_start_hdr;
817 		priv->forward_bad_frames =
818 			mrvl_cfg->port[dev->data->port_id].forward_bad_frames;
819 		priv->fill_bpool_buffs =
820 			mrvl_cfg->port[dev->data->port_id].fill_bpool_buffs;
821 	}
822 
823 	/*
824 	 * Calculate the minimum bpool size for refill feature as follows:
825 	 * 2 default burst sizes multiply by number of rx queues.
826 	 * If the bpool size will be below this value, new buffers will
827 	 * be added to the pool.
828 	 */
829 	priv->bpool_min_size = priv->nb_rx_queues * MRVL_BURST_SIZE * 2;
830 
831 	/* In case initial bpool size configured in queues setup is
832 	 * smaller than minimum size add more buffers
833 	 */
834 	def_init_size = priv->bpool_min_size + MRVL_BURST_SIZE * 2;
835 	if (priv->bpool_init_size < def_init_size) {
836 		int buffs_to_add = def_init_size - priv->bpool_init_size;
837 
838 		priv->bpool_init_size += buffs_to_add;
839 		ret = mrvl_fill_bpool(dev->data->rx_queues[0], buffs_to_add);
840 		if (ret)
841 			MRVL_LOG(ERR, "Failed to add buffers to bpool");
842 	}
843 
844 	/*
845 	 * Calculate the maximum bpool size for refill feature as follows:
846 	 * maximum number of descriptors in rx queue multiply by number
847 	 * of rx queues plus minimum bpool size.
848 	 * In case the bpool size will exceed this value, superfluous buffers
849 	 * will be removed
850 	 */
851 	priv->bpool_max_size = (priv->nb_rx_queues * MRVL_PP2_RXD_MAX) +
852 				priv->bpool_min_size;
853 
854 	ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
855 	if (ret) {
856 		MRVL_LOG(ERR, "Failed to init ppio");
857 		return ret;
858 	}
859 
860 	/*
861 	 * In case there are some some stale uc/mc mac addresses flush them
862 	 * here. It cannot be done during mrvl_dev_close() as port information
863 	 * is already gone at that point (due to pp2_ppio_deinit() in
864 	 * mrvl_dev_stop()).
865 	 */
866 	if (!priv->uc_mc_flushed) {
867 		ret = pp2_ppio_flush_mac_addrs(priv->ppio, 1, 1);
868 		if (ret) {
869 			MRVL_LOG(ERR,
870 				"Failed to flush uc/mc filter list");
871 			goto out;
872 		}
873 		priv->uc_mc_flushed = 1;
874 	}
875 
876 	ret = mrvl_mtu_set(dev, dev->data->mtu);
877 	if (ret)
878 		MRVL_LOG(ERR, "Failed to set MTU to %d", dev->data->mtu);
879 
880 	if (!rte_is_zero_ether_addr(&dev->data->mac_addrs[0]))
881 		mrvl_mac_addr_set(dev, &dev->data->mac_addrs[0]);
882 
883 	for (i = 1; i < MRVL_MAC_ADDRS_MAX; i++) {
884 		mac_addr = &dev->data->mac_addrs[i];
885 
886 		/* skip zero address */
887 		if (rte_is_zero_ether_addr(mac_addr))
888 			continue;
889 
890 		mrvl_mac_addr_add(dev, mac_addr, i, 0);
891 	}
892 
893 	if (dev->data->all_multicast == 1)
894 		mrvl_allmulticast_enable(dev);
895 
896 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_VLAN_FILTER) {
897 		ret = mrvl_populate_vlan_table(dev, 1);
898 		if (ret) {
899 			MRVL_LOG(ERR, "Failed to populate VLAN table");
900 			goto out;
901 		}
902 	}
903 
904 	/* For default QoS config, don't start classifier. */
905 	if (mrvl_cfg  &&
906 	    mrvl_cfg->port[dev->data->port_id].use_qos_global_defaults == 0) {
907 		ret = mrvl_start_qos_mapping(priv);
908 		if (ret) {
909 			MRVL_LOG(ERR, "Failed to setup QoS mapping");
910 			goto out;
911 		}
912 	}
913 
914 	ret = pp2_ppio_set_loopback(priv->ppio, dev->data->dev_conf.lpbk_mode);
915 	if (ret) {
916 		MRVL_LOG(ERR, "Failed to set loopback");
917 		goto out;
918 	}
919 
920 	if (dev->data->promiscuous == 1)
921 		mrvl_promiscuous_enable(dev);
922 
923 	if (priv->flow_ctrl) {
924 		ret = mrvl_flow_ctrl_set(dev, &priv->fc_conf);
925 		if (ret) {
926 			MRVL_LOG(ERR, "Failed to configure flow control");
927 			goto out;
928 		}
929 		priv->flow_ctrl = 0;
930 	}
931 
932 	if (dev->data->dev_link.link_status == ETH_LINK_UP) {
933 		ret = mrvl_dev_set_link_up(dev);
934 		if (ret) {
935 			MRVL_LOG(ERR, "Failed to set link up");
936 			dev->data->dev_link.link_status = ETH_LINK_DOWN;
937 			goto out;
938 		}
939 	}
940 
941 	/* start tx queues */
942 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
943 		struct mrvl_txq *txq = dev->data->tx_queues[i];
944 
945 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
946 
947 		if (!txq->tx_deferred_start)
948 			continue;
949 
950 		/*
951 		 * All txqs are started by default. Stop them
952 		 * so that tx_deferred_start works as expected.
953 		 */
954 		ret = mrvl_tx_queue_stop(dev, i);
955 		if (ret)
956 			goto out;
957 	}
958 
959 	mrvl_flow_init(dev);
960 	mrvl_mtr_init(dev);
961 	mrvl_set_tx_function(dev);
962 
963 	return 0;
964 out:
965 	MRVL_LOG(ERR, "Failed to start device");
966 	pp2_ppio_deinit(priv->ppio);
967 	return ret;
968 }
969 
970 /**
971  * Flush receive queues.
972  *
973  * @param dev
974  *   Pointer to Ethernet device structure.
975  */
976 static void
977 mrvl_flush_rx_queues(struct rte_eth_dev *dev)
978 {
979 	int i;
980 
981 	MRVL_LOG(INFO, "Flushing rx queues");
982 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
983 		int ret, num;
984 
985 		do {
986 			struct mrvl_rxq *q = dev->data->rx_queues[i];
987 			struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];
988 
989 			num = MRVL_PP2_RXD_MAX;
990 			ret = pp2_ppio_recv(q->priv->ppio,
991 					    q->priv->rxq_map[q->queue_id].tc,
992 					    q->priv->rxq_map[q->queue_id].inq,
993 					    descs, (uint16_t *)&num);
994 		} while (ret == 0 && num);
995 	}
996 }
997 
998 /**
999  * Flush transmit shadow queues.
1000  *
1001  * @param dev
1002  *   Pointer to Ethernet device structure.
1003  */
1004 static void
1005 mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
1006 {
1007 	int i, j;
1008 	struct mrvl_txq *txq;
1009 
1010 	MRVL_LOG(INFO, "Flushing tx shadow queues");
1011 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1012 		txq = (struct mrvl_txq *)dev->data->tx_queues[i];
1013 
1014 		for (j = 0; j < RTE_MAX_LCORE; j++) {
1015 			struct mrvl_shadow_txq *sq;
1016 
1017 			if (!hifs[j])
1018 				continue;
1019 
1020 			sq = &txq->shadow_txqs[j];
1021 			mrvl_free_sent_buffers(txq->priv->ppio,
1022 				hifs[j], j, sq, txq->queue_id, 1);
1023 			while (sq->tail != sq->head) {
1024 				uint64_t addr = cookie_addr_high |
1025 					sq->ent[sq->tail].buff.cookie;
1026 				rte_pktmbuf_free(
1027 					(struct rte_mbuf *)addr);
1028 				sq->tail = (sq->tail + 1) &
1029 					    MRVL_PP2_TX_SHADOWQ_MASK;
1030 			}
1031 			memset(sq, 0, sizeof(*sq));
1032 		}
1033 	}
1034 }
1035 
1036 /**
1037  * Flush hardware bpool (buffer-pool).
1038  *
1039  * @param dev
1040  *   Pointer to Ethernet device structure.
1041  */
1042 static void
1043 mrvl_flush_bpool(struct rte_eth_dev *dev)
1044 {
1045 	struct mrvl_priv *priv = dev->data->dev_private;
1046 	struct pp2_hif *hif;
1047 	uint32_t num;
1048 	int ret;
1049 	unsigned int core_id = rte_lcore_id();
1050 
1051 	if (core_id == LCORE_ID_ANY)
1052 		core_id = rte_get_main_lcore();
1053 
1054 	hif = mrvl_get_hif(priv, core_id);
1055 
1056 	ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
1057 	if (ret) {
1058 		MRVL_LOG(ERR, "Failed to get bpool buffers number");
1059 		return;
1060 	}
1061 
1062 	while (num--) {
1063 		struct pp2_buff_inf inf;
1064 		uint64_t addr;
1065 
1066 		ret = pp2_bpool_get_buff(hif, priv->bpool, &inf);
1067 		if (ret)
1068 			break;
1069 
1070 		addr = cookie_addr_high | inf.cookie;
1071 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1072 	}
1073 }
1074 
1075 /**
1076  * DPDK callback to stop the device.
1077  *
1078  * @param dev
1079  *   Pointer to Ethernet device structure.
1080  */
1081 static int
1082 mrvl_dev_stop(struct rte_eth_dev *dev)
1083 {
1084 	return mrvl_dev_set_link_down(dev);
1085 }
1086 
1087 /**
1088  * DPDK callback to close the device.
1089  *
1090  * @param dev
1091  *   Pointer to Ethernet device structure.
1092  */
1093 static int
1094 mrvl_dev_close(struct rte_eth_dev *dev)
1095 {
1096 	struct mrvl_priv *priv = dev->data->dev_private;
1097 	size_t i;
1098 
1099 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1100 		return 0;
1101 
1102 	mrvl_flush_rx_queues(dev);
1103 	mrvl_flush_tx_shadow_queues(dev);
1104 	mrvl_flow_deinit(dev);
1105 	mrvl_mtr_deinit(dev);
1106 
1107 	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
1108 		struct pp2_ppio_tc_params *tc_params =
1109 			&priv->ppio_params.inqs_params.tcs_params[i];
1110 
1111 		if (tc_params->inqs_params) {
1112 			rte_free(tc_params->inqs_params);
1113 			tc_params->inqs_params = NULL;
1114 		}
1115 	}
1116 
1117 	if (priv->cls_tbl) {
1118 		pp2_cls_tbl_deinit(priv->cls_tbl);
1119 		priv->cls_tbl = NULL;
1120 	}
1121 
1122 	if (priv->qos_tbl) {
1123 		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
1124 		priv->qos_tbl = NULL;
1125 	}
1126 
1127 	mrvl_flush_bpool(dev);
1128 	mrvl_tm_deinit(dev);
1129 
1130 	if (priv->ppio) {
1131 		pp2_ppio_deinit(priv->ppio);
1132 		priv->ppio = NULL;
1133 	}
1134 
1135 	/* policer must be released after ppio deinitialization */
1136 	if (priv->default_policer) {
1137 		pp2_cls_plcr_deinit(priv->default_policer);
1138 		priv->default_policer = NULL;
1139 	}
1140 
1141 
1142 	if (priv->bpool) {
1143 		pp2_bpool_deinit(priv->bpool);
1144 		used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
1145 		priv->bpool = NULL;
1146 	}
1147 
1148 	mrvl_dev_num--;
1149 
1150 	if (mrvl_dev_num == 0) {
1151 		MRVL_LOG(INFO, "Perform MUSDK deinit");
1152 		mrvl_deinit_hifs();
1153 		mrvl_deinit_pp2();
1154 		rte_mvep_deinit(MVEP_MOD_T_PP2);
1155 	}
1156 
1157 	return 0;
1158 }
1159 
1160 /**
1161  * DPDK callback to retrieve physical link information.
1162  *
1163  * @param dev
1164  *   Pointer to Ethernet device structure.
1165  * @param wait_to_complete
1166  *   Wait for request completion (ignored).
1167  *
1168  * @return
1169  *   0 on success, negative error value otherwise.
1170  */
1171 static int
1172 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1173 {
1174 	/*
1175 	 * TODO
1176 	 * once MUSDK provides necessary API use it here
1177 	 */
1178 	struct mrvl_priv *priv = dev->data->dev_private;
1179 	struct ethtool_cmd edata;
1180 	struct ifreq req;
1181 	int ret, fd, link_up;
1182 
1183 	if (!priv->ppio)
1184 		return -EPERM;
1185 
1186 	edata.cmd = ETHTOOL_GSET;
1187 
1188 	strcpy(req.ifr_name, dev->data->name);
1189 	req.ifr_data = (void *)&edata;
1190 
1191 	fd = socket(AF_INET, SOCK_DGRAM, 0);
1192 	if (fd == -1)
1193 		return -EFAULT;
1194 
1195 	ret = ioctl(fd, SIOCETHTOOL, &req);
1196 	if (ret == -1) {
1197 		close(fd);
1198 		return -EFAULT;
1199 	}
1200 
1201 	close(fd);
1202 
1203 	switch (ethtool_cmd_speed(&edata)) {
1204 	case SPEED_10:
1205 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
1206 		break;
1207 	case SPEED_100:
1208 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
1209 		break;
1210 	case SPEED_1000:
1211 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
1212 		break;
1213 	case SPEED_2500:
1214 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_2_5G;
1215 		break;
1216 	case SPEED_10000:
1217 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
1218 		break;
1219 	default:
1220 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
1221 	}
1222 
1223 	dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
1224 							 ETH_LINK_HALF_DUPLEX;
1225 	dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
1226 							   ETH_LINK_FIXED;
1227 	pp2_ppio_get_link_state(priv->ppio, &link_up);
1228 	dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
1229 
1230 	return 0;
1231 }
1232 
1233 /**
1234  * DPDK callback to enable promiscuous mode.
1235  *
1236  * @param dev
1237  *   Pointer to Ethernet device structure.
1238  *
1239  * @return
1240  *   0 on success, negative error value otherwise.
1241  */
1242 static int
1243 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
1244 {
1245 	struct mrvl_priv *priv = dev->data->dev_private;
1246 	int ret;
1247 
1248 	if (priv->isolated)
1249 		return -ENOTSUP;
1250 
1251 	if (!priv->ppio)
1252 		return 0;
1253 
1254 	ret = pp2_ppio_set_promisc(priv->ppio, 1);
1255 	if (ret) {
1256 		MRVL_LOG(ERR, "Failed to enable promiscuous mode");
1257 		return -EAGAIN;
1258 	}
1259 
1260 	return 0;
1261 }
1262 
1263 /**
1264  * DPDK callback to enable allmulti mode.
1265  *
1266  * @param dev
1267  *   Pointer to Ethernet device structure.
1268  *
1269  * @return
1270  *   0 on success, negative error value otherwise.
1271  */
1272 static int
1273 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
1274 {
1275 	struct mrvl_priv *priv = dev->data->dev_private;
1276 	int ret;
1277 
1278 	if (priv->isolated)
1279 		return -ENOTSUP;
1280 
1281 	if (!priv->ppio)
1282 		return 0;
1283 
1284 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
1285 	if (ret) {
1286 		MRVL_LOG(ERR, "Failed enable all-multicast mode");
1287 		return -EAGAIN;
1288 	}
1289 
1290 	return 0;
1291 }
1292 
1293 /**
1294  * DPDK callback to disable promiscuous mode.
1295  *
1296  * @param dev
1297  *   Pointer to Ethernet device structure.
1298  *
1299  * @return
1300  *   0 on success, negative error value otherwise.
1301  */
1302 static int
1303 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
1304 {
1305 	struct mrvl_priv *priv = dev->data->dev_private;
1306 	int ret;
1307 
1308 	if (priv->isolated)
1309 		return -ENOTSUP;
1310 
1311 	if (!priv->ppio)
1312 		return 0;
1313 
1314 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
1315 	if (ret) {
1316 		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
1317 		return -EAGAIN;
1318 	}
1319 
1320 	return 0;
1321 }
1322 
1323 /**
1324  * DPDK callback to disable allmulticast mode.
1325  *
1326  * @param dev
1327  *   Pointer to Ethernet device structure.
1328  *
1329  * @return
1330  *   0 on success, negative error value otherwise.
1331  */
1332 static int
1333 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
1334 {
1335 	struct mrvl_priv *priv = dev->data->dev_private;
1336 	int ret;
1337 
1338 	if (priv->isolated)
1339 		return -ENOTSUP;
1340 
1341 	if (!priv->ppio)
1342 		return 0;
1343 
1344 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
1345 	if (ret) {
1346 		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
1347 		return -EAGAIN;
1348 	}
1349 
1350 	return 0;
1351 }
1352 
1353 /**
1354  * DPDK callback to remove a MAC address.
1355  *
1356  * @param dev
1357  *   Pointer to Ethernet device structure.
1358  * @param index
1359  *   MAC address index.
1360  */
1361 static void
1362 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1363 {
1364 	struct mrvl_priv *priv = dev->data->dev_private;
1365 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1366 	int ret;
1367 
1368 	if (priv->isolated)
1369 		return;
1370 
1371 	if (!priv->ppio)
1372 		return;
1373 
1374 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
1375 				       dev->data->mac_addrs[index].addr_bytes);
1376 	if (ret) {
1377 		rte_ether_format_addr(buf, sizeof(buf),
1378 				  &dev->data->mac_addrs[index]);
1379 		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
1380 	}
1381 }
1382 
1383 /**
1384  * DPDK callback to add a MAC address.
1385  *
1386  * @param dev
1387  *   Pointer to Ethernet device structure.
1388  * @param mac_addr
1389  *   MAC address to register.
1390  * @param index
1391  *   MAC address index.
1392  * @param vmdq
1393  *   VMDq pool index to associate address with (unused).
1394  *
1395  * @return
1396  *   0 on success, negative error value otherwise.
1397  */
1398 static int
1399 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1400 		  uint32_t index, uint32_t vmdq __rte_unused)
1401 {
1402 	struct mrvl_priv *priv = dev->data->dev_private;
1403 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1404 	int ret;
1405 
1406 	if (priv->isolated)
1407 		return -ENOTSUP;
1408 
1409 	if (!priv->ppio)
1410 		return 0;
1411 
1412 	if (index == 0)
1413 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1414 		return -1;
1415 
1416 	/*
1417 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1418 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1419 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1420 	 * 21 respectively.
1421 	 *
1422 	 * If more than uc_filter_max uc addresses were added to filter list
1423 	 * then NIC will switch to promiscuous mode automatically.
1424 	 *
1425 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1426 	 * were added to filter list then NIC will switch to all-multicast mode
1427 	 * automatically.
1428 	 */
1429 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1430 	if (ret) {
1431 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1432 		MRVL_LOG(ERR, "Failed to add mac %s", buf);
1433 		return -1;
1434 	}
1435 
1436 	return 0;
1437 }
1438 
1439 /**
1440  * DPDK callback to set the primary MAC address.
1441  *
1442  * @param dev
1443  *   Pointer to Ethernet device structure.
1444  * @param mac_addr
1445  *   MAC address to register.
1446  *
1447  * @return
1448  *   0 on success, negative error value otherwise.
1449  */
1450 static int
1451 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1452 {
1453 	struct mrvl_priv *priv = dev->data->dev_private;
1454 	int ret;
1455 
1456 	if (priv->isolated)
1457 		return -ENOTSUP;
1458 
1459 	if (!priv->ppio)
1460 		return 0;
1461 
1462 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1463 	if (ret) {
1464 		char buf[RTE_ETHER_ADDR_FMT_SIZE];
1465 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1466 		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1467 	}
1468 
1469 	return ret;
1470 }
1471 
1472 /**
1473  * DPDK callback to get device statistics.
1474  *
1475  * @param dev
1476  *   Pointer to Ethernet device structure.
1477  * @param stats
1478  *   Stats structure output buffer.
1479  *
1480  * @return
1481  *   0 on success, negative error value otherwise.
1482  */
1483 static int
1484 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1485 {
1486 	struct mrvl_priv *priv = dev->data->dev_private;
1487 	struct pp2_ppio_statistics ppio_stats;
1488 	uint64_t drop_mac = 0;
1489 	unsigned int i, idx, ret;
1490 
1491 	if (!priv->ppio)
1492 		return -EPERM;
1493 
1494 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1495 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1496 		struct pp2_ppio_inq_statistics rx_stats;
1497 
1498 		if (!rxq)
1499 			continue;
1500 
1501 		idx = rxq->queue_id;
1502 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1503 			MRVL_LOG(ERR,
1504 				"rx queue %d stats out of range (0 - %d)",
1505 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1506 			continue;
1507 		}
1508 
1509 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1510 						  priv->rxq_map[idx].tc,
1511 						  priv->rxq_map[idx].inq,
1512 						  &rx_stats, 0);
1513 		if (unlikely(ret)) {
1514 			MRVL_LOG(ERR,
1515 				"Failed to update rx queue %d stats", idx);
1516 			break;
1517 		}
1518 
1519 		stats->q_ibytes[idx] = rxq->bytes_recv;
1520 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1521 		stats->q_errors[idx] = rx_stats.drop_early +
1522 				       rx_stats.drop_fullq +
1523 				       rx_stats.drop_bm +
1524 				       rxq->drop_mac;
1525 		stats->ibytes += rxq->bytes_recv;
1526 		drop_mac += rxq->drop_mac;
1527 	}
1528 
1529 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1530 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1531 		struct pp2_ppio_outq_statistics tx_stats;
1532 
1533 		if (!txq)
1534 			continue;
1535 
1536 		idx = txq->queue_id;
1537 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1538 			MRVL_LOG(ERR,
1539 				"tx queue %d stats out of range (0 - %d)",
1540 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1541 		}
1542 
1543 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1544 						   &tx_stats, 0);
1545 		if (unlikely(ret)) {
1546 			MRVL_LOG(ERR,
1547 				"Failed to update tx queue %d stats", idx);
1548 			break;
1549 		}
1550 
1551 		stats->q_opackets[idx] = tx_stats.deq_desc;
1552 		stats->q_obytes[idx] = txq->bytes_sent;
1553 		stats->obytes += txq->bytes_sent;
1554 	}
1555 
1556 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1557 	if (unlikely(ret)) {
1558 		MRVL_LOG(ERR, "Failed to update port statistics");
1559 		return ret;
1560 	}
1561 
1562 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1563 	stats->opackets += ppio_stats.tx_packets;
1564 	stats->imissed += ppio_stats.rx_fullq_dropped +
1565 			  ppio_stats.rx_bm_dropped +
1566 			  ppio_stats.rx_early_dropped +
1567 			  ppio_stats.rx_fifo_dropped +
1568 			  ppio_stats.rx_cls_dropped;
1569 	stats->ierrors = drop_mac;
1570 
1571 	return 0;
1572 }
1573 
1574 /**
1575  * DPDK callback to clear device statistics.
1576  *
1577  * @param dev
1578  *   Pointer to Ethernet device structure.
1579  *
1580  * @return
1581  *   0 on success, negative error value otherwise.
1582  */
1583 static int
1584 mrvl_stats_reset(struct rte_eth_dev *dev)
1585 {
1586 	struct mrvl_priv *priv = dev->data->dev_private;
1587 	int i;
1588 
1589 	if (!priv->ppio)
1590 		return 0;
1591 
1592 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1593 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1594 
1595 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1596 					    priv->rxq_map[i].inq, NULL, 1);
1597 		rxq->bytes_recv = 0;
1598 		rxq->drop_mac = 0;
1599 	}
1600 
1601 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1602 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1603 
1604 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1605 		txq->bytes_sent = 0;
1606 	}
1607 
1608 	return pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1609 }
1610 
1611 /**
1612  * DPDK callback to get extended statistics.
1613  *
1614  * @param dev
1615  *   Pointer to Ethernet device structure.
1616  * @param stats
1617  *   Pointer to xstats table.
1618  * @param n
1619  *   Number of entries in xstats table.
1620  * @return
1621  *   Negative value on error, number of read xstats otherwise.
1622  */
1623 static int
1624 mrvl_xstats_get(struct rte_eth_dev *dev,
1625 		struct rte_eth_xstat *stats, unsigned int n)
1626 {
1627 	struct mrvl_priv *priv = dev->data->dev_private;
1628 	struct pp2_ppio_statistics ppio_stats;
1629 	unsigned int i;
1630 
1631 	if (!stats)
1632 		return 0;
1633 
1634 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1635 	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1636 		uint64_t val;
1637 
1638 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1639 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1640 					    mrvl_xstats_tbl[i].offset);
1641 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1642 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1643 					    mrvl_xstats_tbl[i].offset);
1644 		else
1645 			return -EINVAL;
1646 
1647 		stats[i].id = i;
1648 		stats[i].value = val;
1649 	}
1650 
1651 	return n;
1652 }
1653 
1654 /**
1655  * DPDK callback to reset extended statistics.
1656  *
1657  * @param dev
1658  *   Pointer to Ethernet device structure.
1659  *
1660  * @return
1661  *   0 on success, negative error value otherwise.
1662  */
1663 static int
1664 mrvl_xstats_reset(struct rte_eth_dev *dev)
1665 {
1666 	return mrvl_stats_reset(dev);
1667 }
1668 
1669 /**
1670  * DPDK callback to get extended statistics names.
1671  *
1672  * @param dev (unused)
1673  *   Pointer to Ethernet device structure.
1674  * @param xstats_names
1675  *   Pointer to xstats names table.
1676  * @param size
1677  *   Size of the xstats names table.
1678  * @return
1679  *   Number of read names.
1680  */
1681 static int
1682 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1683 		      struct rte_eth_xstat_name *xstats_names,
1684 		      unsigned int size)
1685 {
1686 	unsigned int i;
1687 
1688 	if (!xstats_names)
1689 		return RTE_DIM(mrvl_xstats_tbl);
1690 
1691 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1692 		strlcpy(xstats_names[i].name, mrvl_xstats_tbl[i].name,
1693 			RTE_ETH_XSTATS_NAME_SIZE);
1694 
1695 	return size;
1696 }
1697 
1698 /**
1699  * DPDK callback to get information about the device.
1700  *
1701  * @param dev
1702  *   Pointer to Ethernet device structure (unused).
1703  * @param info
1704  *   Info structure output buffer.
1705  */
1706 static int
1707 mrvl_dev_infos_get(struct rte_eth_dev *dev,
1708 		   struct rte_eth_dev_info *info)
1709 {
1710 	struct mrvl_priv *priv = dev->data->dev_private;
1711 
1712 	info->speed_capa = ETH_LINK_SPEED_10M |
1713 			   ETH_LINK_SPEED_100M |
1714 			   ETH_LINK_SPEED_1G |
1715 			   ETH_LINK_SPEED_2_5G |
1716 			   ETH_LINK_SPEED_10G;
1717 
1718 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1719 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1720 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1721 
1722 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1723 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1724 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1725 
1726 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1727 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1728 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1729 
1730 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1731 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1732 
1733 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1734 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1735 
1736 	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1737 				       ETH_RSS_NONFRAG_IPV4_TCP |
1738 				       ETH_RSS_NONFRAG_IPV4_UDP;
1739 
1740 	/* By default packets are dropped if no descriptors are available */
1741 	info->default_rxconf.rx_drop_en = 1;
1742 
1743 	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1744 	info->max_mtu = priv->max_mtu;
1745 
1746 	return 0;
1747 }
1748 
1749 /**
1750  * Return supported packet types.
1751  *
1752  * @param dev
1753  *   Pointer to Ethernet device structure (unused).
1754  *
1755  * @return
1756  *   Const pointer to the table with supported packet types.
1757  */
1758 static const uint32_t *
1759 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1760 {
1761 	static const uint32_t ptypes[] = {
1762 		RTE_PTYPE_L2_ETHER,
1763 		RTE_PTYPE_L2_ETHER_VLAN,
1764 		RTE_PTYPE_L2_ETHER_QINQ,
1765 		RTE_PTYPE_L3_IPV4,
1766 		RTE_PTYPE_L3_IPV4_EXT,
1767 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1768 		RTE_PTYPE_L3_IPV6,
1769 		RTE_PTYPE_L3_IPV6_EXT,
1770 		RTE_PTYPE_L2_ETHER_ARP,
1771 		RTE_PTYPE_L4_TCP,
1772 		RTE_PTYPE_L4_UDP
1773 	};
1774 
1775 	return ptypes;
1776 }
1777 
1778 /**
1779  * DPDK callback to get information about specific receive queue.
1780  *
1781  * @param dev
1782  *   Pointer to Ethernet device structure.
1783  * @param rx_queue_id
1784  *   Receive queue index.
1785  * @param qinfo
1786  *   Receive queue information structure.
1787  */
1788 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1789 			      struct rte_eth_rxq_info *qinfo)
1790 {
1791 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1792 	struct mrvl_priv *priv = dev->data->dev_private;
1793 	int inq = priv->rxq_map[rx_queue_id].inq;
1794 	int tc = priv->rxq_map[rx_queue_id].tc;
1795 	struct pp2_ppio_tc_params *tc_params =
1796 		&priv->ppio_params.inqs_params.tcs_params[tc];
1797 
1798 	qinfo->mp = q->mp;
1799 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1800 }
1801 
1802 /**
1803  * DPDK callback to get information about specific transmit queue.
1804  *
1805  * @param dev
1806  *   Pointer to Ethernet device structure.
1807  * @param tx_queue_id
1808  *   Transmit queue index.
1809  * @param qinfo
1810  *   Transmit queue information structure.
1811  */
1812 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1813 			      struct rte_eth_txq_info *qinfo)
1814 {
1815 	struct mrvl_priv *priv = dev->data->dev_private;
1816 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1817 
1818 	qinfo->nb_desc =
1819 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1820 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1821 }
1822 
1823 /**
1824  * DPDK callback to Configure a VLAN filter.
1825  *
1826  * @param dev
1827  *   Pointer to Ethernet device structure.
1828  * @param vlan_id
1829  *   VLAN ID to filter.
1830  * @param on
1831  *   Toggle filter.
1832  *
1833  * @return
1834  *   0 on success, negative error value otherwise.
1835  */
1836 static int
1837 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1838 {
1839 	struct mrvl_priv *priv = dev->data->dev_private;
1840 
1841 	if (priv->isolated)
1842 		return -ENOTSUP;
1843 
1844 	if (!priv->ppio)
1845 		return 0;
1846 
1847 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1848 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1849 }
1850 
1851 /**
1852  * DPDK callback to Configure VLAN offload.
1853  *
1854  * @param dev
1855  *   Pointer to Ethernet device structure.
1856  * @param mask
1857  *   VLAN offload mask.
1858  *
1859  * @return
1860  *   0 on success, negative error value otherwise.
1861  */
1862 static int mrvl_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1863 {
1864 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
1865 	int ret;
1866 
1867 	if (mask & ETH_VLAN_STRIP_MASK) {
1868 		MRVL_LOG(ERR, "VLAN stripping is not supported\n");
1869 		return -ENOTSUP;
1870 	}
1871 
1872 	if (mask & ETH_VLAN_FILTER_MASK) {
1873 		if (rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1874 			ret = mrvl_populate_vlan_table(dev, 1);
1875 		else
1876 			ret = mrvl_populate_vlan_table(dev, 0);
1877 
1878 		if (ret)
1879 			return ret;
1880 	}
1881 
1882 	if (mask & ETH_VLAN_EXTEND_MASK) {
1883 		MRVL_LOG(ERR, "Extend VLAN not supported\n");
1884 		return -ENOTSUP;
1885 	}
1886 
1887 	return 0;
1888 }
1889 
1890 /**
1891  * Release buffers to hardware bpool (buffer-pool)
1892  *
1893  * @param rxq
1894  *   Receive queue pointer.
1895  * @param num
1896  *   Number of buffers to release to bpool.
1897  *
1898  * @return
1899  *   0 on success, negative error value otherwise.
1900  */
1901 static int
1902 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1903 {
1904 	struct buff_release_entry entries[num];
1905 	struct rte_mbuf *mbufs[num];
1906 	int i, ret;
1907 	unsigned int core_id;
1908 	struct pp2_hif *hif;
1909 	struct pp2_bpool *bpool;
1910 
1911 	core_id = rte_lcore_id();
1912 	if (core_id == LCORE_ID_ANY)
1913 		core_id = rte_get_main_lcore();
1914 
1915 	hif = mrvl_get_hif(rxq->priv, core_id);
1916 	if (!hif)
1917 		return -1;
1918 
1919 	bpool = rxq->priv->bpool;
1920 
1921 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1922 	if (ret)
1923 		return ret;
1924 
1925 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1926 		cookie_addr_high =
1927 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1928 
1929 	for (i = 0; i < num; i++) {
1930 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1931 			!= cookie_addr_high) {
1932 			MRVL_LOG(ERR,
1933 				"mbuf virtual addr high is out of range "
1934 				"0x%x instead of 0x%x\n",
1935 				(uint32_t)((uint64_t)mbufs[i] >> 32),
1936 				(uint32_t)(cookie_addr_high >> 32));
1937 			goto out;
1938 		}
1939 
1940 		entries[i].buff.addr =
1941 			rte_mbuf_data_iova_default(mbufs[i]);
1942 		entries[i].buff.cookie = (uintptr_t)mbufs[i];
1943 		entries[i].bpool = bpool;
1944 	}
1945 
1946 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1947 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1948 
1949 	if (i != num)
1950 		goto out;
1951 
1952 	return 0;
1953 out:
1954 	for (; i < num; i++)
1955 		rte_pktmbuf_free(mbufs[i]);
1956 
1957 	return -1;
1958 }
1959 
1960 /**
1961  * DPDK callback to configure the receive queue.
1962  *
1963  * @param dev
1964  *   Pointer to Ethernet device structure.
1965  * @param idx
1966  *   RX queue index.
1967  * @param desc
1968  *   Number of descriptors to configure in queue.
1969  * @param socket
1970  *   NUMA socket on which memory must be allocated.
1971  * @param conf
1972  *   Thresholds parameters.
1973  * @param mp
1974  *   Memory pool for buffer allocations.
1975  *
1976  * @return
1977  *   0 on success, negative error value otherwise.
1978  */
1979 static int
1980 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1981 		    unsigned int socket,
1982 		    const struct rte_eth_rxconf *conf,
1983 		    struct rte_mempool *mp)
1984 {
1985 	struct mrvl_priv *priv = dev->data->dev_private;
1986 	struct mrvl_rxq *rxq;
1987 	uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
1988 	uint32_t max_rx_pktlen = dev->data->mtu + RTE_ETHER_HDR_LEN;
1989 	int ret, tc, inq;
1990 	uint64_t offloads;
1991 
1992 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
1993 
1994 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1995 		/*
1996 		 * Unknown TC mapping, mapping will not have a correct queue.
1997 		 */
1998 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
1999 			idx, priv->ppio_id);
2000 		return -EFAULT;
2001 	}
2002 
2003 	frame_size = buf_size - RTE_PKTMBUF_HEADROOM - MRVL_PKT_EFFEC_OFFS;
2004 	if (frame_size < max_rx_pktlen) {
2005 		MRVL_LOG(WARNING,
2006 			"Mbuf size must be increased to %u bytes to hold up "
2007 			"to %u bytes of data.",
2008 			max_rx_pktlen + buf_size - frame_size,
2009 			max_rx_pktlen);
2010 		dev->data->mtu = frame_size - RTE_ETHER_HDR_LEN;
2011 		MRVL_LOG(INFO, "Setting MTU to %u", dev->data->mtu);
2012 	}
2013 
2014 	if (dev->data->rx_queues[idx]) {
2015 		rte_free(dev->data->rx_queues[idx]);
2016 		dev->data->rx_queues[idx] = NULL;
2017 	}
2018 
2019 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
2020 	if (!rxq)
2021 		return -ENOMEM;
2022 
2023 	rxq->priv = priv;
2024 	rxq->mp = mp;
2025 	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
2026 	rxq->queue_id = idx;
2027 	rxq->port_id = dev->data->port_id;
2028 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
2029 
2030 	tc = priv->rxq_map[rxq->queue_id].tc,
2031 	inq = priv->rxq_map[rxq->queue_id].inq;
2032 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
2033 		desc;
2034 
2035 	ret = mrvl_fill_bpool(rxq, desc);
2036 	if (ret) {
2037 		rte_free(rxq);
2038 		return ret;
2039 	}
2040 
2041 	priv->bpool_init_size += desc;
2042 
2043 	dev->data->rx_queues[idx] = rxq;
2044 
2045 	return 0;
2046 }
2047 
2048 /**
2049  * DPDK callback to release the receive queue.
2050  *
2051  * @param dev
2052  *   Pointer to Ethernet device structure.
2053  * @param qid
2054  *   Receive queue index.
2055  */
2056 static void
2057 mrvl_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
2058 {
2059 	struct mrvl_rxq *q = dev->data->rx_queues[qid];
2060 	struct pp2_ppio_tc_params *tc_params;
2061 	int i, num, tc, inq;
2062 	struct pp2_hif *hif;
2063 	unsigned int core_id = rte_lcore_id();
2064 
2065 	if (core_id == LCORE_ID_ANY)
2066 		core_id = rte_get_main_lcore();
2067 
2068 	if (!q)
2069 		return;
2070 
2071 	hif = mrvl_get_hif(q->priv, core_id);
2072 
2073 	if (!hif)
2074 		return;
2075 
2076 	tc = q->priv->rxq_map[q->queue_id].tc;
2077 	inq = q->priv->rxq_map[q->queue_id].inq;
2078 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
2079 	num = tc_params->inqs_params[inq].size;
2080 	for (i = 0; i < num; i++) {
2081 		struct pp2_buff_inf inf;
2082 		uint64_t addr;
2083 
2084 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
2085 		addr = cookie_addr_high | inf.cookie;
2086 		rte_pktmbuf_free((struct rte_mbuf *)addr);
2087 	}
2088 
2089 	rte_free(q);
2090 }
2091 
2092 /**
2093  * DPDK callback to configure the transmit queue.
2094  *
2095  * @param dev
2096  *   Pointer to Ethernet device structure.
2097  * @param idx
2098  *   Transmit queue index.
2099  * @param desc
2100  *   Number of descriptors to configure in the queue.
2101  * @param socket
2102  *   NUMA socket on which memory must be allocated.
2103  * @param conf
2104  *   Tx queue configuration parameters.
2105  *
2106  * @return
2107  *   0 on success, negative error value otherwise.
2108  */
2109 static int
2110 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
2111 		    unsigned int socket,
2112 		    const struct rte_eth_txconf *conf)
2113 {
2114 	struct mrvl_priv *priv = dev->data->dev_private;
2115 	struct mrvl_txq *txq;
2116 
2117 	if (dev->data->tx_queues[idx]) {
2118 		rte_free(dev->data->tx_queues[idx]);
2119 		dev->data->tx_queues[idx] = NULL;
2120 	}
2121 
2122 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
2123 	if (!txq)
2124 		return -ENOMEM;
2125 
2126 	txq->priv = priv;
2127 	txq->queue_id = idx;
2128 	txq->port_id = dev->data->port_id;
2129 	txq->tx_deferred_start = conf->tx_deferred_start;
2130 	dev->data->tx_queues[idx] = txq;
2131 
2132 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
2133 
2134 	return 0;
2135 }
2136 
2137 /**
2138  * DPDK callback to release the transmit queue.
2139  *
2140  * @param dev
2141  *   Pointer to Ethernet device structure.
2142  * @param qid
2143  *   Transmit queue index.
2144  */
2145 static void
2146 mrvl_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
2147 {
2148 	struct mrvl_txq *q = dev->data->tx_queues[qid];
2149 
2150 	if (!q)
2151 		return;
2152 
2153 	rte_free(q);
2154 }
2155 
2156 /**
2157  * DPDK callback to get flow control configuration.
2158  *
2159  * @param dev
2160  *  Pointer to Ethernet device structure.
2161  * @param fc_conf
2162  *  Pointer to the flow control configuration.
2163  *
2164  * @return
2165  *  0 on success, negative error value otherwise.
2166  */
2167 static int
2168 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2169 {
2170 	struct mrvl_priv *priv = dev->data->dev_private;
2171 	int ret, en;
2172 
2173 	if (!priv->ppio) {
2174 		memcpy(fc_conf, &priv->fc_conf, sizeof(struct rte_eth_fc_conf));
2175 		return 0;
2176 	}
2177 
2178 	fc_conf->autoneg = 1;
2179 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
2180 	if (ret) {
2181 		MRVL_LOG(ERR, "Failed to read rx pause state");
2182 		return ret;
2183 	}
2184 
2185 	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
2186 
2187 	ret = pp2_ppio_get_tx_pause(priv->ppio, &en);
2188 	if (ret) {
2189 		MRVL_LOG(ERR, "Failed to read tx pause state");
2190 		return ret;
2191 	}
2192 
2193 	if (en) {
2194 		if (fc_conf->mode == RTE_FC_NONE)
2195 			fc_conf->mode = RTE_FC_TX_PAUSE;
2196 		else
2197 			fc_conf->mode = RTE_FC_FULL;
2198 	}
2199 
2200 	return 0;
2201 }
2202 
2203 /**
2204  * DPDK callback to set flow control configuration.
2205  *
2206  * @param dev
2207  *  Pointer to Ethernet device structure.
2208  * @param fc_conf
2209  *  Pointer to the flow control configuration.
2210  *
2211  * @return
2212  *  0 on success, negative error value otherwise.
2213  */
2214 static int
2215 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2216 {
2217 	struct mrvl_priv *priv = dev->data->dev_private;
2218 	struct pp2_ppio_tx_pause_params mrvl_pause_params;
2219 	int ret;
2220 	int rx_en, tx_en;
2221 
2222 	if (fc_conf->high_water ||
2223 	    fc_conf->low_water ||
2224 	    fc_conf->pause_time ||
2225 	    fc_conf->mac_ctrl_frame_fwd) {
2226 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
2227 
2228 		return -EINVAL;
2229 	}
2230 
2231 	if (fc_conf->autoneg == 0) {
2232 		MRVL_LOG(ERR, "Flowctrl Autoneg disable is not supported");
2233 		return -EINVAL;
2234 	}
2235 
2236 	if (!priv->ppio) {
2237 		memcpy(&priv->fc_conf, fc_conf, sizeof(struct rte_eth_fc_conf));
2238 		priv->flow_ctrl = 1;
2239 		return 0;
2240 	}
2241 
2242 	switch (fc_conf->mode) {
2243 	case RTE_FC_FULL:
2244 		rx_en = 1;
2245 		tx_en = 1;
2246 		break;
2247 	case RTE_FC_TX_PAUSE:
2248 		rx_en = 0;
2249 		tx_en = 1;
2250 		break;
2251 	case RTE_FC_RX_PAUSE:
2252 		rx_en = 1;
2253 		tx_en = 0;
2254 		break;
2255 	case RTE_FC_NONE:
2256 		rx_en = 0;
2257 		tx_en = 0;
2258 		break;
2259 	default:
2260 		MRVL_LOG(ERR, "Incorrect Flow control flag (%d)",
2261 			 fc_conf->mode);
2262 		return -EINVAL;
2263 	}
2264 
2265 	/* Set RX flow control */
2266 	ret = pp2_ppio_set_rx_pause(priv->ppio, rx_en);
2267 	if (ret) {
2268 		MRVL_LOG(ERR, "Failed to change RX flowctrl");
2269 		return ret;
2270 	}
2271 
2272 	/* Set TX flow control */
2273 	mrvl_pause_params.en = tx_en;
2274 	/* all inqs participate in xon/xoff decision */
2275 	mrvl_pause_params.use_tc_pause_inqs = 0;
2276 	ret = pp2_ppio_set_tx_pause(priv->ppio, &mrvl_pause_params);
2277 	if (ret) {
2278 		MRVL_LOG(ERR, "Failed to change TX flowctrl");
2279 		return ret;
2280 	}
2281 
2282 	return 0;
2283 }
2284 
2285 /**
2286  * Update RSS hash configuration
2287  *
2288  * @param dev
2289  *   Pointer to Ethernet device structure.
2290  * @param rss_conf
2291  *   Pointer to RSS configuration.
2292  *
2293  * @return
2294  *   0 on success, negative error value otherwise.
2295  */
2296 static int
2297 mrvl_rss_hash_update(struct rte_eth_dev *dev,
2298 		     struct rte_eth_rss_conf *rss_conf)
2299 {
2300 	struct mrvl_priv *priv = dev->data->dev_private;
2301 
2302 	if (priv->isolated)
2303 		return -ENOTSUP;
2304 
2305 	return mrvl_configure_rss(priv, rss_conf);
2306 }
2307 
2308 /**
2309  * DPDK callback to get RSS hash configuration.
2310  *
2311  * @param dev
2312  *   Pointer to Ethernet device structure.
2313  * @rss_conf
2314  *   Pointer to RSS configuration.
2315  *
2316  * @return
2317  *   Always 0.
2318  */
2319 static int
2320 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
2321 		       struct rte_eth_rss_conf *rss_conf)
2322 {
2323 	struct mrvl_priv *priv = dev->data->dev_private;
2324 	enum pp2_ppio_hash_type hash_type =
2325 		priv->ppio_params.inqs_params.hash_type;
2326 
2327 	rss_conf->rss_key = NULL;
2328 
2329 	if (hash_type == PP2_PPIO_HASH_T_NONE)
2330 		rss_conf->rss_hf = 0;
2331 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
2332 		rss_conf->rss_hf = ETH_RSS_IPV4;
2333 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
2334 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
2335 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
2336 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
2337 
2338 	return 0;
2339 }
2340 
2341 /**
2342  * DPDK callback to get rte_flow callbacks.
2343  *
2344  * @param dev
2345  *   Pointer to the device structure.
2346  * @param ops
2347  *   Pointer to pass the flow ops.
2348  *
2349  * @return
2350  *   0 on success, negative error value otherwise.
2351  */
2352 static int
2353 mrvl_eth_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
2354 		      const struct rte_flow_ops **ops)
2355 {
2356 	*ops = &mrvl_flow_ops;
2357 	return 0;
2358 }
2359 
2360 /**
2361  * DPDK callback to get rte_mtr callbacks.
2362  *
2363  * @param dev
2364  *   Pointer to the device structure.
2365  * @param ops
2366  *   Pointer to pass the mtr ops.
2367  *
2368  * @return
2369  *   Always 0.
2370  */
2371 static int
2372 mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2373 {
2374 	*(const void **)ops = &mrvl_mtr_ops;
2375 
2376 	return 0;
2377 }
2378 
2379 /**
2380  * DPDK callback to get rte_tm callbacks.
2381  *
2382  * @param dev
2383  *   Pointer to the device structure.
2384  * @param ops
2385  *   Pointer to pass the tm ops.
2386  *
2387  * @return
2388  *   Always 0.
2389  */
2390 static int
2391 mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2392 {
2393 	*(const void **)ops = &mrvl_tm_ops;
2394 
2395 	return 0;
2396 }
2397 
2398 static const struct eth_dev_ops mrvl_ops = {
2399 	.dev_configure = mrvl_dev_configure,
2400 	.dev_start = mrvl_dev_start,
2401 	.dev_stop = mrvl_dev_stop,
2402 	.dev_set_link_up = mrvl_dev_set_link_up,
2403 	.dev_set_link_down = mrvl_dev_set_link_down,
2404 	.dev_close = mrvl_dev_close,
2405 	.link_update = mrvl_link_update,
2406 	.promiscuous_enable = mrvl_promiscuous_enable,
2407 	.allmulticast_enable = mrvl_allmulticast_enable,
2408 	.promiscuous_disable = mrvl_promiscuous_disable,
2409 	.allmulticast_disable = mrvl_allmulticast_disable,
2410 	.mac_addr_remove = mrvl_mac_addr_remove,
2411 	.mac_addr_add = mrvl_mac_addr_add,
2412 	.mac_addr_set = mrvl_mac_addr_set,
2413 	.mtu_set = mrvl_mtu_set,
2414 	.stats_get = mrvl_stats_get,
2415 	.stats_reset = mrvl_stats_reset,
2416 	.xstats_get = mrvl_xstats_get,
2417 	.xstats_reset = mrvl_xstats_reset,
2418 	.xstats_get_names = mrvl_xstats_get_names,
2419 	.dev_infos_get = mrvl_dev_infos_get,
2420 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
2421 	.rxq_info_get = mrvl_rxq_info_get,
2422 	.txq_info_get = mrvl_txq_info_get,
2423 	.vlan_filter_set = mrvl_vlan_filter_set,
2424 	.vlan_offload_set = mrvl_vlan_offload_set,
2425 	.tx_queue_start = mrvl_tx_queue_start,
2426 	.tx_queue_stop = mrvl_tx_queue_stop,
2427 	.rx_queue_setup = mrvl_rx_queue_setup,
2428 	.rx_queue_release = mrvl_rx_queue_release,
2429 	.tx_queue_setup = mrvl_tx_queue_setup,
2430 	.tx_queue_release = mrvl_tx_queue_release,
2431 	.flow_ctrl_get = mrvl_flow_ctrl_get,
2432 	.flow_ctrl_set = mrvl_flow_ctrl_set,
2433 	.rss_hash_update = mrvl_rss_hash_update,
2434 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
2435 	.flow_ops_get = mrvl_eth_flow_ops_get,
2436 	.mtr_ops_get = mrvl_mtr_ops_get,
2437 	.tm_ops_get = mrvl_tm_ops_get,
2438 };
2439 
2440 /**
2441  * Return packet type information and l3/l4 offsets.
2442  *
2443  * @param desc
2444  *   Pointer to the received packet descriptor.
2445  * @param l3_offset
2446  *   l3 packet offset.
2447  * @param l4_offset
2448  *   l4 packet offset.
2449  *
2450  * @return
2451  *   Packet type information.
2452  */
2453 static inline uint64_t
2454 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
2455 				    uint8_t *l3_offset, uint8_t *l4_offset)
2456 {
2457 	enum pp2_inq_l3_type l3_type;
2458 	enum pp2_inq_l4_type l4_type;
2459 	enum pp2_inq_vlan_tag vlan_tag;
2460 	uint64_t packet_type;
2461 
2462 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
2463 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
2464 	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
2465 
2466 	packet_type = RTE_PTYPE_L2_ETHER;
2467 
2468 	switch (vlan_tag) {
2469 	case PP2_INQ_VLAN_TAG_SINGLE:
2470 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
2471 		break;
2472 	case PP2_INQ_VLAN_TAG_DOUBLE:
2473 	case PP2_INQ_VLAN_TAG_TRIPLE:
2474 		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
2475 		break;
2476 	default:
2477 		break;
2478 	}
2479 
2480 	switch (l3_type) {
2481 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
2482 		packet_type |= RTE_PTYPE_L3_IPV4;
2483 		break;
2484 	case PP2_INQ_L3_TYPE_IPV4_OK:
2485 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
2486 		break;
2487 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
2488 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
2489 		break;
2490 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
2491 		packet_type |= RTE_PTYPE_L3_IPV6;
2492 		break;
2493 	case PP2_INQ_L3_TYPE_IPV6_EXT:
2494 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
2495 		break;
2496 	case PP2_INQ_L3_TYPE_ARP:
2497 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
2498 		/*
2499 		 * In case of ARP l4_offset is set to wrong value.
2500 		 * Set it to proper one so that later on mbuf->l3_len can be
2501 		 * calculated subtracting l4_offset and l3_offset.
2502 		 */
2503 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
2504 		break;
2505 	default:
2506 		break;
2507 	}
2508 
2509 	switch (l4_type) {
2510 	case PP2_INQ_L4_TYPE_TCP:
2511 		packet_type |= RTE_PTYPE_L4_TCP;
2512 		break;
2513 	case PP2_INQ_L4_TYPE_UDP:
2514 		packet_type |= RTE_PTYPE_L4_UDP;
2515 		break;
2516 	default:
2517 		break;
2518 	}
2519 
2520 	return packet_type;
2521 }
2522 
2523 /**
2524  * Get offload information from the received packet descriptor.
2525  *
2526  * @param desc
2527  *   Pointer to the received packet descriptor.
2528  *
2529  * @return
2530  *   Mbuf offload flags.
2531  */
2532 static inline uint64_t
2533 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc, uint64_t packet_type)
2534 {
2535 	uint64_t flags = 0;
2536 	enum pp2_inq_desc_status status;
2537 
2538 	if (RTE_ETH_IS_IPV4_HDR(packet_type)) {
2539 		status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2540 		if (unlikely(status != PP2_DESC_ERR_OK))
2541 			flags |= PKT_RX_IP_CKSUM_BAD;
2542 		else
2543 			flags |= PKT_RX_IP_CKSUM_GOOD;
2544 	}
2545 
2546 	if (((packet_type & RTE_PTYPE_L4_UDP) == RTE_PTYPE_L4_UDP) ||
2547 	    ((packet_type & RTE_PTYPE_L4_TCP) == RTE_PTYPE_L4_TCP)) {
2548 		status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2549 		if (unlikely(status != PP2_DESC_ERR_OK))
2550 			flags |= PKT_RX_L4_CKSUM_BAD;
2551 		else
2552 			flags |= PKT_RX_L4_CKSUM_GOOD;
2553 	}
2554 
2555 	return flags;
2556 }
2557 
2558 /**
2559  * DPDK callback for receive.
2560  *
2561  * @param rxq
2562  *   Generic pointer to the receive queue.
2563  * @param rx_pkts
2564  *   Array to store received packets.
2565  * @param nb_pkts
2566  *   Maximum number of packets in array.
2567  *
2568  * @return
2569  *   Number of packets successfully received.
2570  */
2571 static uint16_t
2572 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2573 {
2574 	struct mrvl_rxq *q = rxq;
2575 	struct pp2_ppio_desc descs[nb_pkts];
2576 	struct pp2_bpool *bpool;
2577 	int i, ret, rx_done = 0;
2578 	int num;
2579 	struct pp2_hif *hif;
2580 	unsigned int core_id = rte_lcore_id();
2581 
2582 	hif = mrvl_get_hif(q->priv, core_id);
2583 
2584 	if (unlikely(!q->priv->ppio || !hif))
2585 		return 0;
2586 
2587 	bpool = q->priv->bpool;
2588 
2589 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2590 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2591 	if (unlikely(ret < 0))
2592 		return 0;
2593 
2594 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2595 
2596 	for (i = 0; i < nb_pkts; i++) {
2597 		struct rte_mbuf *mbuf;
2598 		uint8_t l3_offset, l4_offset;
2599 		enum pp2_inq_desc_status status;
2600 		uint64_t addr;
2601 
2602 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2603 			struct pp2_ppio_desc *pref_desc;
2604 			u64 pref_addr;
2605 
2606 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2607 			pref_addr = cookie_addr_high |
2608 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2609 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2610 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2611 		}
2612 
2613 		addr = cookie_addr_high |
2614 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2615 		mbuf = (struct rte_mbuf *)addr;
2616 		rte_pktmbuf_reset(mbuf);
2617 
2618 		/* drop packet in case of mac, overrun or resource error */
2619 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2620 		if ((unlikely(status != PP2_DESC_ERR_OK)) &&
2621 			!(q->priv->forward_bad_frames)) {
2622 			struct pp2_buff_inf binf = {
2623 				.addr = rte_mbuf_data_iova_default(mbuf),
2624 				.cookie = (uint64_t)mbuf,
2625 			};
2626 
2627 			pp2_bpool_put_buff(hif, bpool, &binf);
2628 			mrvl_port_bpool_size
2629 				[bpool->pp2_id][bpool->id][core_id]++;
2630 			q->drop_mac++;
2631 			continue;
2632 		}
2633 
2634 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2635 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2636 		mbuf->data_len = mbuf->pkt_len;
2637 		mbuf->port = q->port_id;
2638 		mbuf->packet_type =
2639 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2640 							    &l3_offset,
2641 							    &l4_offset);
2642 		mbuf->l2_len = l3_offset;
2643 		mbuf->l3_len = l4_offset - l3_offset;
2644 
2645 		if (likely(q->cksum_enabled))
2646 			mbuf->ol_flags =
2647 				mrvl_desc_to_ol_flags(&descs[i],
2648 						      mbuf->packet_type);
2649 
2650 		rx_pkts[rx_done++] = mbuf;
2651 		q->bytes_recv += mbuf->pkt_len;
2652 	}
2653 
2654 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2655 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2656 
2657 		if (unlikely(num <= q->priv->bpool_min_size ||
2658 			     (!rx_done && num < q->priv->bpool_init_size))) {
2659 			mrvl_fill_bpool(q, q->priv->fill_bpool_buffs);
2660 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2661 			int i;
2662 			int pkt_to_remove = num - q->priv->bpool_init_size;
2663 			struct rte_mbuf *mbuf;
2664 			struct pp2_buff_inf buff;
2665 
2666 			for (i = 0; i < pkt_to_remove; i++) {
2667 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2668 				if (ret)
2669 					break;
2670 				mbuf = (struct rte_mbuf *)
2671 					(cookie_addr_high | buff.cookie);
2672 				rte_pktmbuf_free(mbuf);
2673 			}
2674 			mrvl_port_bpool_size
2675 				[bpool->pp2_id][bpool->id][core_id] -= i;
2676 		}
2677 		rte_spinlock_unlock(&q->priv->lock);
2678 	}
2679 
2680 	return rx_done;
2681 }
2682 
2683 /**
2684  * Prepare offload information.
2685  *
2686  * @param ol_flags
2687  *   Offload flags.
2688  * @param l3_type
2689  *   Pointer to the pp2_ouq_l3_type structure.
2690  * @param l4_type
2691  *   Pointer to the pp2_outq_l4_type structure.
2692  * @param gen_l3_cksum
2693  *   Will be set to 1 in case l3 checksum is computed.
2694  * @param l4_cksum
2695  *   Will be set to 1 in case l4 checksum is computed.
2696  */
2697 static inline void
2698 mrvl_prepare_proto_info(uint64_t ol_flags,
2699 			enum pp2_outq_l3_type *l3_type,
2700 			enum pp2_outq_l4_type *l4_type,
2701 			int *gen_l3_cksum,
2702 			int *gen_l4_cksum)
2703 {
2704 	/*
2705 	 * Based on ol_flags prepare information
2706 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2707 	 * for offloading.
2708 	 * in most of the checksum cases ipv4 must be set, so this is the
2709 	 * default value
2710 	 */
2711 	*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2712 	*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2713 
2714 	if (ol_flags & PKT_TX_IPV6) {
2715 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2716 		/* no checksum for ipv6 header */
2717 		*gen_l3_cksum = 0;
2718 	}
2719 
2720 	if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM) {
2721 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2722 		*gen_l4_cksum = 1;
2723 	} else if ((ol_flags & PKT_TX_L4_MASK) ==  PKT_TX_UDP_CKSUM) {
2724 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2725 		*gen_l4_cksum = 1;
2726 	} else {
2727 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2728 		/* no checksum for other type */
2729 		*gen_l4_cksum = 0;
2730 	}
2731 }
2732 
2733 /**
2734  * Release already sent buffers to bpool (buffer-pool).
2735  *
2736  * @param ppio
2737  *   Pointer to the port structure.
2738  * @param hif
2739  *   Pointer to the MUSDK hardware interface.
2740  * @param sq
2741  *   Pointer to the shadow queue.
2742  * @param qid
2743  *   Queue id number.
2744  * @param force
2745  *   Force releasing packets.
2746  */
2747 static inline void
2748 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2749 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2750 		       int qid, int force)
2751 {
2752 	struct buff_release_entry *entry;
2753 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2754 	int i;
2755 
2756 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2757 
2758 	sq->num_to_release += nb_done;
2759 
2760 	if (likely(!force &&
2761 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2762 		return;
2763 
2764 	nb_done = sq->num_to_release;
2765 	sq->num_to_release = 0;
2766 
2767 	for (i = 0; i < nb_done; i++) {
2768 		entry = &sq->ent[sq->tail + num];
2769 		if (unlikely(!entry->buff.addr)) {
2770 			MRVL_LOG(ERR,
2771 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2772 				sq->tail, (u64)entry->buff.cookie,
2773 				(u64)entry->buff.addr);
2774 			skip_bufs = 1;
2775 			goto skip;
2776 		}
2777 
2778 		if (unlikely(!entry->bpool)) {
2779 			struct rte_mbuf *mbuf;
2780 
2781 			mbuf = (struct rte_mbuf *)entry->buff.cookie;
2782 			rte_pktmbuf_free(mbuf);
2783 			skip_bufs = 1;
2784 			goto skip;
2785 		}
2786 
2787 		mrvl_port_bpool_size
2788 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2789 		num++;
2790 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2791 			goto skip;
2792 		continue;
2793 skip:
2794 		if (likely(num))
2795 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2796 		num += skip_bufs;
2797 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2798 		sq->size -= num;
2799 		num = 0;
2800 		skip_bufs = 0;
2801 	}
2802 
2803 	if (likely(num)) {
2804 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2805 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2806 		sq->size -= num;
2807 	}
2808 }
2809 
2810 /**
2811  * DPDK callback for transmit.
2812  *
2813  * @param txq
2814  *   Generic pointer transmit queue.
2815  * @param tx_pkts
2816  *   Packets to transmit.
2817  * @param nb_pkts
2818  *   Number of packets in array.
2819  *
2820  * @return
2821  *   Number of packets successfully transmitted.
2822  */
2823 static uint16_t
2824 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2825 {
2826 	struct mrvl_txq *q = txq;
2827 	struct mrvl_shadow_txq *sq;
2828 	struct pp2_hif *hif;
2829 	struct pp2_ppio_desc descs[nb_pkts];
2830 	unsigned int core_id = rte_lcore_id();
2831 	int i, bytes_sent = 0;
2832 	uint16_t num, sq_free_size;
2833 	uint64_t addr;
2834 
2835 	hif = mrvl_get_hif(q->priv, core_id);
2836 	sq = &q->shadow_txqs[core_id];
2837 
2838 	if (unlikely(!q->priv->ppio || !hif))
2839 		return 0;
2840 
2841 	if (sq->size)
2842 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2843 				       sq, q->queue_id, 0);
2844 
2845 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2846 	if (unlikely(nb_pkts > sq_free_size))
2847 		nb_pkts = sq_free_size;
2848 
2849 	for (i = 0; i < nb_pkts; i++) {
2850 		struct rte_mbuf *mbuf = tx_pkts[i];
2851 		int gen_l3_cksum, gen_l4_cksum;
2852 		enum pp2_outq_l3_type l3_type;
2853 		enum pp2_outq_l4_type l4_type;
2854 
2855 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2856 			struct rte_mbuf *pref_pkt_hdr;
2857 
2858 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2859 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2860 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2861 		}
2862 
2863 		mrvl_fill_shadowq(sq, mbuf);
2864 		mrvl_fill_desc(&descs[i], mbuf);
2865 
2866 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2867 		/*
2868 		 * in case unsupported ol_flags were passed
2869 		 * do not update descriptor offload information
2870 		 */
2871 		if (!(mbuf->ol_flags & MRVL_TX_PKT_OFFLOADS))
2872 			continue;
2873 		mrvl_prepare_proto_info(mbuf->ol_flags, &l3_type, &l4_type,
2874 					&gen_l3_cksum, &gen_l4_cksum);
2875 
2876 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2877 						  mbuf->l2_len,
2878 						  mbuf->l2_len + mbuf->l3_len,
2879 						  gen_l3_cksum, gen_l4_cksum);
2880 	}
2881 
2882 	num = nb_pkts;
2883 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2884 	/* number of packets that were not sent */
2885 	if (unlikely(num > nb_pkts)) {
2886 		for (i = nb_pkts; i < num; i++) {
2887 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2888 				MRVL_PP2_TX_SHADOWQ_MASK;
2889 			addr = sq->ent[sq->head].buff.cookie;
2890 			bytes_sent -=
2891 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2892 		}
2893 		sq->size -= num - nb_pkts;
2894 	}
2895 
2896 	q->bytes_sent += bytes_sent;
2897 
2898 	return nb_pkts;
2899 }
2900 
2901 /** DPDK callback for S/G transmit.
2902  *
2903  * @param txq
2904  *   Generic pointer transmit queue.
2905  * @param tx_pkts
2906  *   Packets to transmit.
2907  * @param nb_pkts
2908  *   Number of packets in array.
2909  *
2910  * @return
2911  *   Number of packets successfully transmitted.
2912  */
2913 static uint16_t
2914 mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
2915 		     uint16_t nb_pkts)
2916 {
2917 	struct mrvl_txq *q = txq;
2918 	struct mrvl_shadow_txq *sq;
2919 	struct pp2_hif *hif;
2920 	struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
2921 	struct pp2_ppio_sg_pkts pkts;
2922 	uint8_t frags[nb_pkts];
2923 	unsigned int core_id = rte_lcore_id();
2924 	int i, j, bytes_sent = 0;
2925 	int tail, tail_first;
2926 	uint16_t num, sq_free_size;
2927 	uint16_t nb_segs, total_descs = 0;
2928 	uint64_t addr;
2929 
2930 	hif = mrvl_get_hif(q->priv, core_id);
2931 	sq = &q->shadow_txqs[core_id];
2932 	pkts.frags = frags;
2933 	pkts.num = 0;
2934 
2935 	if (unlikely(!q->priv->ppio || !hif))
2936 		return 0;
2937 
2938 	if (sq->size)
2939 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2940 				       sq, q->queue_id, 0);
2941 
2942 	/* Save shadow queue free size */
2943 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2944 
2945 	tail = 0;
2946 	for (i = 0; i < nb_pkts; i++) {
2947 		struct rte_mbuf *mbuf = tx_pkts[i];
2948 		struct rte_mbuf *seg = NULL;
2949 		int gen_l3_cksum, gen_l4_cksum;
2950 		enum pp2_outq_l3_type l3_type;
2951 		enum pp2_outq_l4_type l4_type;
2952 
2953 		nb_segs = mbuf->nb_segs;
2954 		tail_first = tail;
2955 		total_descs += nb_segs;
2956 
2957 		/*
2958 		 * Check if total_descs does not exceed
2959 		 * shadow queue free size
2960 		 */
2961 		if (unlikely(total_descs > sq_free_size)) {
2962 			total_descs -= nb_segs;
2963 			break;
2964 		}
2965 
2966 		/* Check if nb_segs does not exceed the max nb of desc per
2967 		 * fragmented packet
2968 		 */
2969 		if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
2970 			total_descs -= nb_segs;
2971 			RTE_LOG(ERR, PMD,
2972 				"Too many segments. Packet won't be sent.\n");
2973 			break;
2974 		}
2975 
2976 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2977 			struct rte_mbuf *pref_pkt_hdr;
2978 
2979 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2980 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2981 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2982 		}
2983 
2984 		pkts.frags[pkts.num] = nb_segs;
2985 		pkts.num++;
2986 
2987 		seg = mbuf;
2988 		for (j = 0; j < nb_segs - 1; j++) {
2989 			/* For the subsequent segments, set shadow queue
2990 			 * buffer to NULL
2991 			 */
2992 			mrvl_fill_shadowq(sq, NULL);
2993 			mrvl_fill_desc(&descs[tail], seg);
2994 
2995 			tail++;
2996 			seg = seg->next;
2997 		}
2998 		/* Put first mbuf info in last shadow queue entry */
2999 		mrvl_fill_shadowq(sq, mbuf);
3000 		/* Update descriptor with last segment */
3001 		mrvl_fill_desc(&descs[tail++], seg);
3002 
3003 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
3004 		/* In case unsupported ol_flags were passed
3005 		 * do not update descriptor offload information
3006 		 */
3007 		if (!(mbuf->ol_flags & MRVL_TX_PKT_OFFLOADS))
3008 			continue;
3009 		mrvl_prepare_proto_info(mbuf->ol_flags, &l3_type, &l4_type,
3010 					&gen_l3_cksum, &gen_l4_cksum);
3011 
3012 		pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
3013 						  l4_type, mbuf->l2_len,
3014 						  mbuf->l2_len + mbuf->l3_len,
3015 						  gen_l3_cksum, gen_l4_cksum);
3016 	}
3017 
3018 	num = total_descs;
3019 	pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
3020 			 &total_descs, &pkts);
3021 	/* number of packets that were not sent */
3022 	if (unlikely(num > total_descs)) {
3023 		for (i = total_descs; i < num; i++) {
3024 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
3025 				MRVL_PP2_TX_SHADOWQ_MASK;
3026 
3027 			addr = sq->ent[sq->head].buff.cookie;
3028 			if (addr)
3029 				bytes_sent -=
3030 					rte_pktmbuf_pkt_len((struct rte_mbuf *)
3031 						(cookie_addr_high | addr));
3032 		}
3033 		sq->size -= num - total_descs;
3034 		nb_pkts = pkts.num;
3035 	}
3036 
3037 	q->bytes_sent += bytes_sent;
3038 
3039 	return nb_pkts;
3040 }
3041 
3042 /**
3043  * Create private device structure.
3044  *
3045  * @param dev_name
3046  *   Pointer to the port name passed in the initialization parameters.
3047  *
3048  * @return
3049  *   Pointer to the newly allocated private device structure.
3050  */
3051 static struct mrvl_priv *
3052 mrvl_priv_create(const char *dev_name)
3053 {
3054 	struct pp2_bpool_params bpool_params;
3055 	char match[MRVL_MATCH_LEN];
3056 	struct mrvl_priv *priv;
3057 	uint16_t max_frame_size;
3058 	int ret, bpool_bit;
3059 
3060 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
3061 	if (!priv)
3062 		return NULL;
3063 
3064 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
3065 				       &priv->pp_id, &priv->ppio_id);
3066 	if (ret)
3067 		goto out_free_priv;
3068 
3069 	ret = pp2_ppio_get_l4_cksum_max_frame_size(priv->pp_id, priv->ppio_id,
3070 						   &max_frame_size);
3071 	if (ret)
3072 		goto out_free_priv;
3073 
3074 	priv->max_mtu = max_frame_size + RTE_ETHER_CRC_LEN -
3075 		MRVL_PP2_ETH_HDRS_LEN;
3076 
3077 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
3078 				     PP2_BPOOL_NUM_POOLS);
3079 	if (bpool_bit < 0)
3080 		goto out_free_priv;
3081 	priv->bpool_bit = bpool_bit;
3082 
3083 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
3084 		 priv->bpool_bit);
3085 	memset(&bpool_params, 0, sizeof(bpool_params));
3086 	bpool_params.match = match;
3087 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
3088 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
3089 	if (ret)
3090 		goto out_clear_bpool_bit;
3091 
3092 	priv->ppio_params.type = PP2_PPIO_T_NIC;
3093 	rte_spinlock_init(&priv->lock);
3094 
3095 	return priv;
3096 out_clear_bpool_bit:
3097 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
3098 out_free_priv:
3099 	rte_free(priv);
3100 	return NULL;
3101 }
3102 
3103 /**
3104  * Create device representing Ethernet port.
3105  *
3106  * @param name
3107  *   Pointer to the port's name.
3108  *
3109  * @return
3110  *   0 on success, negative error value otherwise.
3111  */
3112 static int
3113 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
3114 {
3115 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
3116 	struct rte_eth_dev *eth_dev;
3117 	struct mrvl_priv *priv;
3118 	struct ifreq req;
3119 
3120 	eth_dev = rte_eth_dev_allocate(name);
3121 	if (!eth_dev)
3122 		return -ENOMEM;
3123 
3124 	priv = mrvl_priv_create(name);
3125 	if (!priv) {
3126 		ret = -ENOMEM;
3127 		goto out_free;
3128 	}
3129 	eth_dev->data->dev_private = priv;
3130 
3131 	eth_dev->data->mac_addrs =
3132 		rte_zmalloc("mac_addrs",
3133 			    RTE_ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
3134 	if (!eth_dev->data->mac_addrs) {
3135 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
3136 		ret = -ENOMEM;
3137 		goto out_free;
3138 	}
3139 
3140 	memset(&req, 0, sizeof(req));
3141 	strcpy(req.ifr_name, name);
3142 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
3143 	if (ret)
3144 		goto out_free;
3145 
3146 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
3147 	       req.ifr_addr.sa_data, RTE_ETHER_ADDR_LEN);
3148 
3149 	eth_dev->device = &vdev->device;
3150 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
3151 	mrvl_set_tx_function(eth_dev);
3152 	eth_dev->dev_ops = &mrvl_ops;
3153 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
3154 
3155 	eth_dev->data->dev_link.link_status = ETH_LINK_UP;
3156 
3157 	rte_eth_dev_probing_finish(eth_dev);
3158 	return 0;
3159 out_free:
3160 	rte_eth_dev_release_port(eth_dev);
3161 
3162 	return ret;
3163 }
3164 
3165 /**
3166  * Callback used by rte_kvargs_process() during argument parsing.
3167  *
3168  * @param key
3169  *   Pointer to the parsed key (unused).
3170  * @param value
3171  *   Pointer to the parsed value.
3172  * @param extra_args
3173  *   Pointer to the extra arguments which contains address of the
3174  *   table of pointers to parsed interface names.
3175  *
3176  * @return
3177  *   Always 0.
3178  */
3179 static int
3180 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
3181 		 void *extra_args)
3182 {
3183 	struct mrvl_ifnames *ifnames = extra_args;
3184 
3185 	ifnames->names[ifnames->idx++] = value;
3186 
3187 	return 0;
3188 }
3189 
3190 /**
3191  * DPDK callback to register the virtual device.
3192  *
3193  * @param vdev
3194  *   Pointer to the virtual device.
3195  *
3196  * @return
3197  *   0 on success, negative error value otherwise.
3198  */
3199 static int
3200 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
3201 {
3202 	struct rte_kvargs *kvlist;
3203 	struct mrvl_ifnames ifnames;
3204 	int ret = -EINVAL;
3205 	uint32_t i, ifnum, cfgnum;
3206 	const char *params;
3207 
3208 	params = rte_vdev_device_args(vdev);
3209 	if (!params)
3210 		return -EINVAL;
3211 
3212 	kvlist = rte_kvargs_parse(params, valid_args);
3213 	if (!kvlist)
3214 		return -EINVAL;
3215 
3216 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
3217 	if (ifnum > RTE_DIM(ifnames.names))
3218 		goto out_free_kvlist;
3219 
3220 	ifnames.idx = 0;
3221 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
3222 			   mrvl_get_ifnames, &ifnames);
3223 
3224 
3225 	/*
3226 	 * The below system initialization should be done only once,
3227 	 * on the first provided configuration file
3228 	 */
3229 	if (!mrvl_cfg) {
3230 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
3231 		MRVL_LOG(INFO, "Parsing config file!");
3232 		if (cfgnum > 1) {
3233 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
3234 			goto out_free_kvlist;
3235 		} else if (cfgnum == 1) {
3236 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
3237 					   mrvl_get_cfg, &mrvl_cfg);
3238 		}
3239 	}
3240 
3241 	if (mrvl_dev_num)
3242 		goto init_devices;
3243 
3244 	MRVL_LOG(INFO, "Perform MUSDK initializations");
3245 
3246 	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
3247 	if (ret)
3248 		goto out_free_kvlist;
3249 
3250 	ret = mrvl_init_pp2();
3251 	if (ret) {
3252 		MRVL_LOG(ERR, "Failed to init PP!");
3253 		rte_mvep_deinit(MVEP_MOD_T_PP2);
3254 		goto out_free_kvlist;
3255 	}
3256 
3257 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
3258 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
3259 
3260 	mrvl_lcore_first = RTE_MAX_LCORE;
3261 	mrvl_lcore_last = 0;
3262 
3263 init_devices:
3264 	for (i = 0; i < ifnum; i++) {
3265 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
3266 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
3267 		if (ret)
3268 			goto out_cleanup;
3269 		mrvl_dev_num++;
3270 	}
3271 
3272 	rte_kvargs_free(kvlist);
3273 
3274 	return 0;
3275 out_cleanup:
3276 	rte_pmd_mrvl_remove(vdev);
3277 
3278 out_free_kvlist:
3279 	rte_kvargs_free(kvlist);
3280 
3281 	return ret;
3282 }
3283 
3284 /**
3285  * DPDK callback to remove virtual device.
3286  *
3287  * @param vdev
3288  *   Pointer to the removed virtual device.
3289  *
3290  * @return
3291  *   0 on success, negative error value otherwise.
3292  */
3293 static int
3294 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
3295 {
3296 	uint16_t port_id;
3297 	int ret = 0;
3298 
3299 	RTE_ETH_FOREACH_DEV(port_id) {
3300 		if (rte_eth_devices[port_id].device != &vdev->device)
3301 			continue;
3302 		ret |= rte_eth_dev_close(port_id);
3303 	}
3304 
3305 	return ret == 0 ? 0 : -EIO;
3306 }
3307 
3308 static struct rte_vdev_driver pmd_mrvl_drv = {
3309 	.probe = rte_pmd_mrvl_probe,
3310 	.remove = rte_pmd_mrvl_remove,
3311 };
3312 
3313 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
3314 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
3315 RTE_LOG_REGISTER_DEFAULT(mrvl_logtype, NOTICE);
3316