xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision ff0b8b10dc479a893d610f92a66e43f3858b55ef)
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 #define MRVL_MAC_ADDRS_MAX 25
45 #define MRVL_MATCH_LEN 16
46 #define MRVL_PKT_EFFEC_OFFS (MRVL_PKT_OFFS + MV_MH_SIZE)
47 /* Maximum allowable packet size */
48 #define MRVL_PKT_SIZE_MAX (10240 - MV_MH_SIZE)
49 
50 #define MRVL_IFACE_NAME_ARG "iface"
51 #define MRVL_CFG_ARG "cfg"
52 
53 #define MRVL_BURST_SIZE 64
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_JUMBO_FRAME | \
63 			  DEV_RX_OFFLOAD_CHECKSUM)
64 
65 /** Port Tx offloads capabilities */
66 #define MRVL_TX_OFFLOADS (DEV_TX_OFFLOAD_IPV4_CKSUM | \
67 			  DEV_TX_OFFLOAD_UDP_CKSUM | \
68 			  DEV_TX_OFFLOAD_TCP_CKSUM | \
69 			  DEV_TX_OFFLOAD_MULTI_SEGS)
70 
71 static const char * const valid_args[] = {
72 	MRVL_IFACE_NAME_ARG,
73 	MRVL_CFG_ARG,
74 	NULL
75 };
76 
77 static int used_hifs = MRVL_MUSDK_HIFS_RESERVED;
78 static struct pp2_hif *hifs[RTE_MAX_LCORE];
79 static int used_bpools[PP2_NUM_PKT_PROC] = {
80 	[0 ... PP2_NUM_PKT_PROC - 1] = MRVL_MUSDK_BPOOLS_RESERVED
81 };
82 
83 static struct pp2_bpool *mrvl_port_to_bpool_lookup[RTE_MAX_ETHPORTS];
84 static int mrvl_port_bpool_size[PP2_NUM_PKT_PROC][PP2_BPOOL_NUM_POOLS][RTE_MAX_LCORE];
85 static uint64_t cookie_addr_high = MRVL_COOKIE_ADDR_INVALID;
86 
87 struct mrvl_ifnames {
88 	const char *names[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
89 	int idx;
90 };
91 
92 /*
93  * To use buffer harvesting based on loopback port shadow queue structure
94  * was introduced for buffers information bookkeeping.
95  *
96  * Before sending the packet, related buffer information (pp2_buff_inf) is
97  * stored in shadow queue. After packet is transmitted no longer used
98  * packet buffer is released back to it's original hardware pool,
99  * on condition it originated from interface.
100  * In case it  was generated by application itself i.e: mbuf->port field is
101  * 0xff then its released to software mempool.
102  */
103 struct mrvl_shadow_txq {
104 	int head;           /* write index - used when sending buffers */
105 	int tail;           /* read index - used when releasing buffers */
106 	u16 size;           /* queue occupied size */
107 	u16 num_to_release; /* number of descriptors sent, that can be
108 			     * released
109 			     */
110 	struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
111 };
112 
113 struct mrvl_rxq {
114 	struct mrvl_priv *priv;
115 	struct rte_mempool *mp;
116 	int queue_id;
117 	int port_id;
118 	int cksum_enabled;
119 	uint64_t bytes_recv;
120 	uint64_t drop_mac;
121 };
122 
123 struct mrvl_txq {
124 	struct mrvl_priv *priv;
125 	int queue_id;
126 	int port_id;
127 	uint64_t bytes_sent;
128 	struct mrvl_shadow_txq shadow_txqs[RTE_MAX_LCORE];
129 	int tx_deferred_start;
130 };
131 
132 static int mrvl_lcore_first;
133 static int mrvl_lcore_last;
134 static int mrvl_dev_num;
135 
136 static int mrvl_fill_bpool(struct mrvl_rxq *rxq, int num);
137 static inline void mrvl_free_sent_buffers(struct pp2_ppio *ppio,
138 			struct pp2_hif *hif, unsigned int core_id,
139 			struct mrvl_shadow_txq *sq, int qid, int force);
140 
141 static uint16_t mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
142 				  uint16_t nb_pkts);
143 static uint16_t mrvl_tx_sg_pkt_burst(void *txq,	struct rte_mbuf **tx_pkts,
144 				     uint16_t nb_pkts);
145 static int rte_pmd_mrvl_remove(struct rte_vdev_device *vdev);
146 static void mrvl_deinit_pp2(void);
147 static void mrvl_deinit_hifs(void);
148 
149 static int
150 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
151 		  uint32_t index, uint32_t vmdq __rte_unused);
152 static int
153 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr);
154 static int
155 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on);
156 static int mrvl_promiscuous_enable(struct rte_eth_dev *dev);
157 static int mrvl_allmulticast_enable(struct rte_eth_dev *dev);
158 
159 #define MRVL_XSTATS_TBL_ENTRY(name) { \
160 	#name, offsetof(struct pp2_ppio_statistics, name),	\
161 	sizeof(((struct pp2_ppio_statistics *)0)->name)		\
162 }
163 
164 /* Table with xstats data */
165 static struct {
166 	const char *name;
167 	unsigned int offset;
168 	unsigned int size;
169 } mrvl_xstats_tbl[] = {
170 	MRVL_XSTATS_TBL_ENTRY(rx_bytes),
171 	MRVL_XSTATS_TBL_ENTRY(rx_packets),
172 	MRVL_XSTATS_TBL_ENTRY(rx_unicast_packets),
173 	MRVL_XSTATS_TBL_ENTRY(rx_errors),
174 	MRVL_XSTATS_TBL_ENTRY(rx_fullq_dropped),
175 	MRVL_XSTATS_TBL_ENTRY(rx_bm_dropped),
176 	MRVL_XSTATS_TBL_ENTRY(rx_early_dropped),
177 	MRVL_XSTATS_TBL_ENTRY(rx_fifo_dropped),
178 	MRVL_XSTATS_TBL_ENTRY(rx_cls_dropped),
179 	MRVL_XSTATS_TBL_ENTRY(tx_bytes),
180 	MRVL_XSTATS_TBL_ENTRY(tx_packets),
181 	MRVL_XSTATS_TBL_ENTRY(tx_unicast_packets),
182 	MRVL_XSTATS_TBL_ENTRY(tx_errors)
183 };
184 
185 static inline void
186 mrvl_fill_shadowq(struct mrvl_shadow_txq *sq, struct rte_mbuf *buf)
187 {
188 	sq->ent[sq->head].buff.cookie = (uint64_t)buf;
189 	sq->ent[sq->head].buff.addr = buf ?
190 		rte_mbuf_data_iova_default(buf) : 0;
191 
192 	sq->ent[sq->head].bpool =
193 		(unlikely(!buf || buf->port >= RTE_MAX_ETHPORTS ||
194 		 buf->refcnt > 1)) ? NULL :
195 		 mrvl_port_to_bpool_lookup[buf->port];
196 
197 	sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
198 	sq->size++;
199 }
200 
201 static inline void
202 mrvl_fill_desc(struct pp2_ppio_desc *desc, struct rte_mbuf *buf)
203 {
204 	pp2_ppio_outq_desc_reset(desc);
205 	pp2_ppio_outq_desc_set_phys_addr(desc, rte_pktmbuf_iova(buf));
206 	pp2_ppio_outq_desc_set_pkt_offset(desc, 0);
207 	pp2_ppio_outq_desc_set_pkt_len(desc, rte_pktmbuf_data_len(buf));
208 }
209 
210 static inline int
211 mrvl_get_bpool_size(int pp2_id, int pool_id)
212 {
213 	int i;
214 	int size = 0;
215 
216 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
217 		size += mrvl_port_bpool_size[pp2_id][pool_id][i];
218 
219 	return size;
220 }
221 
222 static inline int
223 mrvl_reserve_bit(int *bitmap, int max)
224 {
225 	int n = sizeof(*bitmap) * 8 - __builtin_clz(*bitmap);
226 
227 	if (n >= max)
228 		return -1;
229 
230 	*bitmap |= 1 << n;
231 
232 	return n;
233 }
234 
235 static int
236 mrvl_init_hif(int core_id)
237 {
238 	struct pp2_hif_params params;
239 	char match[MRVL_MATCH_LEN];
240 	int ret;
241 
242 	ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
243 	if (ret < 0) {
244 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
245 		return ret;
246 	}
247 
248 	snprintf(match, sizeof(match), "hif-%d", ret);
249 	memset(&params, 0, sizeof(params));
250 	params.match = match;
251 	params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
252 	ret = pp2_hif_init(&params, &hifs[core_id]);
253 	if (ret) {
254 		MRVL_LOG(ERR, "Failed to initialize hif %d", core_id);
255 		return ret;
256 	}
257 
258 	return 0;
259 }
260 
261 static inline struct pp2_hif*
262 mrvl_get_hif(struct mrvl_priv *priv, int core_id)
263 {
264 	int ret;
265 
266 	if (likely(hifs[core_id] != NULL))
267 		return hifs[core_id];
268 
269 	rte_spinlock_lock(&priv->lock);
270 
271 	ret = mrvl_init_hif(core_id);
272 	if (ret < 0) {
273 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
274 		goto out;
275 	}
276 
277 	if (core_id < mrvl_lcore_first)
278 		mrvl_lcore_first = core_id;
279 
280 	if (core_id > mrvl_lcore_last)
281 		mrvl_lcore_last = core_id;
282 out:
283 	rte_spinlock_unlock(&priv->lock);
284 
285 	return hifs[core_id];
286 }
287 
288 /**
289  * Set tx burst function according to offload flag
290  *
291  * @param dev
292  *   Pointer to Ethernet device structure.
293  */
294 static void
295 mrvl_set_tx_function(struct rte_eth_dev *dev)
296 {
297 	struct mrvl_priv *priv = dev->data->dev_private;
298 
299 	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
300 	if (priv->multiseg) {
301 		RTE_LOG(INFO, PMD, "Using multi-segment tx callback\n");
302 		dev->tx_pkt_burst = mrvl_tx_sg_pkt_burst;
303 	} else {
304 		RTE_LOG(INFO, PMD, "Using single-segment tx callback\n");
305 		dev->tx_pkt_burst = mrvl_tx_pkt_burst;
306 	}
307 }
308 
309 /**
310  * Configure rss based on dpdk rss configuration.
311  *
312  * @param priv
313  *   Pointer to private structure.
314  * @param rss_conf
315  *   Pointer to RSS configuration.
316  *
317  * @return
318  *   0 on success, negative error value otherwise.
319  */
320 static int
321 mrvl_configure_rss(struct mrvl_priv *priv, struct rte_eth_rss_conf *rss_conf)
322 {
323 	if (rss_conf->rss_key)
324 		MRVL_LOG(WARNING, "Changing hash key is not supported");
325 
326 	if (rss_conf->rss_hf == 0) {
327 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
328 	} else if (rss_conf->rss_hf & ETH_RSS_IPV4) {
329 		priv->ppio_params.inqs_params.hash_type =
330 			PP2_PPIO_HASH_T_2_TUPLE;
331 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP) {
332 		priv->ppio_params.inqs_params.hash_type =
333 			PP2_PPIO_HASH_T_5_TUPLE;
334 		priv->rss_hf_tcp = 1;
335 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) {
336 		priv->ppio_params.inqs_params.hash_type =
337 			PP2_PPIO_HASH_T_5_TUPLE;
338 		priv->rss_hf_tcp = 0;
339 	} else {
340 		return -EINVAL;
341 	}
342 
343 	return 0;
344 }
345 
346 /**
347  * Ethernet device configuration.
348  *
349  * Prepare the driver for a given number of TX and RX queues and
350  * configure RSS.
351  *
352  * @param dev
353  *   Pointer to Ethernet device structure.
354  *
355  * @return
356  *   0 on success, negative error value otherwise.
357  */
358 static int
359 mrvl_dev_configure(struct rte_eth_dev *dev)
360 {
361 	struct mrvl_priv *priv = dev->data->dev_private;
362 	int ret;
363 
364 	if (priv->ppio) {
365 		MRVL_LOG(INFO, "Device reconfiguration is not supported");
366 		return -EINVAL;
367 	}
368 
369 	if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_NONE &&
370 	    dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
371 		MRVL_LOG(INFO, "Unsupported rx multi queue mode %d",
372 			dev->data->dev_conf.rxmode.mq_mode);
373 		return -EINVAL;
374 	}
375 
376 	if (dev->data->dev_conf.rxmode.split_hdr_size) {
377 		MRVL_LOG(INFO, "Split headers not supported");
378 		return -EINVAL;
379 	}
380 
381 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
382 		dev->data->mtu = dev->data->dev_conf.rxmode.max_rx_pkt_len -
383 				 MRVL_PP2_ETH_HDRS_LEN;
384 
385 	if (dev->data->dev_conf.txmode.offloads & DEV_TX_OFFLOAD_MULTI_SEGS)
386 		priv->multiseg = 1;
387 
388 	ret = mrvl_configure_rxqs(priv, dev->data->port_id,
389 				  dev->data->nb_rx_queues);
390 	if (ret < 0)
391 		return ret;
392 
393 	ret = mrvl_configure_txqs(priv, dev->data->port_id,
394 				  dev->data->nb_tx_queues);
395 	if (ret < 0)
396 		return ret;
397 
398 	priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
399 	priv->ppio_params.maintain_stats = 1;
400 	priv->nb_rx_queues = dev->data->nb_rx_queues;
401 
402 	ret = mrvl_tm_init(dev);
403 	if (ret < 0)
404 		return ret;
405 
406 	if (dev->data->nb_rx_queues == 1 &&
407 	    dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
408 		MRVL_LOG(WARNING, "Disabling hash for 1 rx queue");
409 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
410 
411 		return 0;
412 	}
413 
414 	return mrvl_configure_rss(priv,
415 			&dev->data->dev_conf.rx_adv_conf.rss_conf);
416 }
417 
418 /**
419  * DPDK callback to change the MTU.
420  *
421  * Setting the MTU affects hardware MRU (packets larger than the MRU
422  * will be dropped).
423  *
424  * @param dev
425  *   Pointer to Ethernet device structure.
426  * @param mtu
427  *   New MTU.
428  *
429  * @return
430  *   0 on success, negative error value otherwise.
431  */
432 static int
433 mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
434 {
435 	struct mrvl_priv *priv = dev->data->dev_private;
436 	uint16_t mru;
437 	uint16_t mbuf_data_size = 0; /* SW buffer size */
438 	int ret;
439 
440 	mru = MRVL_PP2_MTU_TO_MRU(mtu);
441 	/*
442 	 * min_rx_buf_size is equal to mbuf data size
443 	 * if pmd didn't set it differently
444 	 */
445 	mbuf_data_size = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
446 	/* Prevent PMD from:
447 	 * - setting mru greater than the mbuf size resulting in
448 	 * hw and sw buffer size mismatch
449 	 * - setting mtu that requires the support of scattered packets
450 	 * when this feature has not been enabled/supported so far
451 	 * (TODO check scattered_rx flag here once scattered RX is supported).
452 	 */
453 	if (mru - RTE_ETHER_CRC_LEN + MRVL_PKT_OFFS > mbuf_data_size) {
454 		mru = mbuf_data_size + RTE_ETHER_CRC_LEN - MRVL_PKT_OFFS;
455 		mtu = MRVL_PP2_MRU_TO_MTU(mru);
456 		MRVL_LOG(WARNING, "MTU too big, max MTU possible limitted "
457 			"by current mbuf size: %u. Set MTU to %u, MRU to %u",
458 			mbuf_data_size, mtu, mru);
459 	}
460 
461 	if (mtu < RTE_ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX) {
462 		MRVL_LOG(ERR, "Invalid MTU [%u] or MRU [%u]", mtu, mru);
463 		return -EINVAL;
464 	}
465 
466 	dev->data->mtu = mtu;
467 	dev->data->dev_conf.rxmode.max_rx_pkt_len = mru - MV_MH_SIZE;
468 
469 	if (!priv->ppio)
470 		return 0;
471 
472 	ret = pp2_ppio_set_mru(priv->ppio, mru);
473 	if (ret) {
474 		MRVL_LOG(ERR, "Failed to change MRU");
475 		return ret;
476 	}
477 
478 	ret = pp2_ppio_set_mtu(priv->ppio, mtu);
479 	if (ret) {
480 		MRVL_LOG(ERR, "Failed to change MTU");
481 		return ret;
482 	}
483 
484 	return 0;
485 }
486 
487 /**
488  * DPDK callback to bring the link up.
489  *
490  * @param dev
491  *   Pointer to Ethernet device structure.
492  *
493  * @return
494  *   0 on success, negative error value otherwise.
495  */
496 static int
497 mrvl_dev_set_link_up(struct rte_eth_dev *dev)
498 {
499 	struct mrvl_priv *priv = dev->data->dev_private;
500 	int ret;
501 
502 	if (!priv->ppio) {
503 		dev->data->dev_link.link_status = ETH_LINK_UP;
504 		return 0;
505 	}
506 
507 	ret = pp2_ppio_enable(priv->ppio);
508 	if (ret)
509 		return ret;
510 
511 	/*
512 	 * mtu/mru can be updated if pp2_ppio_enable() was called at least once
513 	 * as pp2_ppio_enable() changes port->t_mode from default 0 to
514 	 * PP2_TRAFFIC_INGRESS_EGRESS.
515 	 *
516 	 * Set mtu to default DPDK value here.
517 	 */
518 	ret = mrvl_mtu_set(dev, dev->data->mtu);
519 	if (ret) {
520 		pp2_ppio_disable(priv->ppio);
521 		return ret;
522 	}
523 
524 	dev->data->dev_link.link_status = ETH_LINK_UP;
525 	return 0;
526 }
527 
528 /**
529  * DPDK callback to bring the link down.
530  *
531  * @param dev
532  *   Pointer to Ethernet device structure.
533  *
534  * @return
535  *   0 on success, negative error value otherwise.
536  */
537 static int
538 mrvl_dev_set_link_down(struct rte_eth_dev *dev)
539 {
540 	struct mrvl_priv *priv = dev->data->dev_private;
541 	int ret;
542 
543 	if (!priv->ppio) {
544 		dev->data->dev_link.link_status = ETH_LINK_DOWN;
545 		return 0;
546 	}
547 	ret = pp2_ppio_disable(priv->ppio);
548 	if (ret)
549 		return ret;
550 
551 	dev->data->dev_link.link_status = ETH_LINK_DOWN;
552 	return 0;
553 }
554 
555 /**
556  * DPDK callback to start tx queue.
557  *
558  * @param dev
559  *   Pointer to Ethernet device structure.
560  * @param queue_id
561  *   Transmit queue index.
562  *
563  * @return
564  *   0 on success, negative error value otherwise.
565  */
566 static int
567 mrvl_tx_queue_start(struct rte_eth_dev *dev, uint16_t queue_id)
568 {
569 	struct mrvl_priv *priv = dev->data->dev_private;
570 	int ret;
571 
572 	if (!priv)
573 		return -EPERM;
574 
575 	/* passing 1 enables given tx queue */
576 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 1);
577 	if (ret) {
578 		MRVL_LOG(ERR, "Failed to start txq %d", queue_id);
579 		return ret;
580 	}
581 
582 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
583 
584 	return 0;
585 }
586 
587 /**
588  * DPDK callback to stop tx queue.
589  *
590  * @param dev
591  *   Pointer to Ethernet device structure.
592  * @param queue_id
593  *   Transmit queue index.
594  *
595  * @return
596  *   0 on success, negative error value otherwise.
597  */
598 static int
599 mrvl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t queue_id)
600 {
601 	struct mrvl_priv *priv = dev->data->dev_private;
602 	int ret;
603 
604 	if (!priv->ppio)
605 		return -EPERM;
606 
607 	/* passing 0 disables given tx queue */
608 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 0);
609 	if (ret) {
610 		MRVL_LOG(ERR, "Failed to stop txq %d", queue_id);
611 		return ret;
612 	}
613 
614 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
615 
616 	return 0;
617 }
618 
619 /**
620  * Populate VLAN Filter configuration.
621  *
622  * @param dev
623  *   Pointer to Ethernet device structure.
624  * @param on
625  *   Toggle filter.
626  *
627  * @return
628  *   0 on success, negative error value otherwise.
629  */
630 static int mrvl_populate_vlan_table(struct rte_eth_dev *dev, int on)
631 {
632 	uint32_t j;
633 	int ret;
634 	struct rte_vlan_filter_conf *vfc;
635 
636 	vfc = &dev->data->vlan_filter_conf;
637 	for (j = 0; j < RTE_DIM(vfc->ids); j++) {
638 		uint64_t vlan;
639 		uint64_t vbit;
640 		uint64_t ids = vfc->ids[j];
641 
642 		if (ids == 0)
643 			continue;
644 
645 		while (ids) {
646 			vlan = 64 * j;
647 			/* count trailing zeroes */
648 			vbit = ~ids & (ids - 1);
649 			/* clear least significant bit set */
650 			ids ^= (ids ^ (ids - 1)) ^ vbit;
651 			for (; vbit; vlan++)
652 				vbit >>= 1;
653 			ret = mrvl_vlan_filter_set(dev, vlan, on);
654 			if (ret) {
655 				MRVL_LOG(ERR, "Failed to setup VLAN filter\n");
656 				return ret;
657 			}
658 		}
659 	}
660 
661 	return 0;
662 }
663 
664 /**
665  * DPDK callback to start the device.
666  *
667  * @param dev
668  *   Pointer to Ethernet device structure.
669  *
670  * @return
671  *   0 on success, negative errno value on failure.
672  */
673 static int
674 mrvl_dev_start(struct rte_eth_dev *dev)
675 {
676 	struct mrvl_priv *priv = dev->data->dev_private;
677 	char match[MRVL_MATCH_LEN];
678 	int ret = 0, i, def_init_size;
679 	struct rte_ether_addr *mac_addr;
680 
681 	if (priv->ppio)
682 		return mrvl_dev_set_link_up(dev);
683 
684 	snprintf(match, sizeof(match), "ppio-%d:%d",
685 		 priv->pp_id, priv->ppio_id);
686 	priv->ppio_params.match = match;
687 
688 	/*
689 	 * Calculate the minimum bpool size for refill feature as follows:
690 	 * 2 default burst sizes multiply by number of rx queues.
691 	 * If the bpool size will be below this value, new buffers will
692 	 * be added to the pool.
693 	 */
694 	priv->bpool_min_size = priv->nb_rx_queues * MRVL_BURST_SIZE * 2;
695 
696 	/* In case initial bpool size configured in queues setup is
697 	 * smaller than minimum size add more buffers
698 	 */
699 	def_init_size = priv->bpool_min_size + MRVL_BURST_SIZE * 2;
700 	if (priv->bpool_init_size < def_init_size) {
701 		int buffs_to_add = def_init_size - priv->bpool_init_size;
702 
703 		priv->bpool_init_size += buffs_to_add;
704 		ret = mrvl_fill_bpool(dev->data->rx_queues[0], buffs_to_add);
705 		if (ret)
706 			MRVL_LOG(ERR, "Failed to add buffers to bpool");
707 	}
708 
709 	/*
710 	 * Calculate the maximum bpool size for refill feature as follows:
711 	 * maximum number of descriptors in rx queue multiply by number
712 	 * of rx queues plus minimum bpool size.
713 	 * In case the bpool size will exceed this value, superfluous buffers
714 	 * will be removed
715 	 */
716 	priv->bpool_max_size = (priv->nb_rx_queues * MRVL_PP2_RXD_MAX) +
717 				priv->bpool_min_size;
718 
719 	ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
720 	if (ret) {
721 		MRVL_LOG(ERR, "Failed to init ppio");
722 		return ret;
723 	}
724 
725 	/*
726 	 * In case there are some some stale uc/mc mac addresses flush them
727 	 * here. It cannot be done during mrvl_dev_close() as port information
728 	 * is already gone at that point (due to pp2_ppio_deinit() in
729 	 * mrvl_dev_stop()).
730 	 */
731 	if (!priv->uc_mc_flushed) {
732 		ret = pp2_ppio_flush_mac_addrs(priv->ppio, 1, 1);
733 		if (ret) {
734 			MRVL_LOG(ERR,
735 				"Failed to flush uc/mc filter list");
736 			goto out;
737 		}
738 		priv->uc_mc_flushed = 1;
739 	}
740 
741 	ret = mrvl_mtu_set(dev, dev->data->mtu);
742 	if (ret)
743 		MRVL_LOG(ERR, "Failed to set MTU to %d", dev->data->mtu);
744 
745 	if (!rte_is_zero_ether_addr(&dev->data->mac_addrs[0]))
746 		mrvl_mac_addr_set(dev, &dev->data->mac_addrs[0]);
747 
748 	for (i = 1; i < MRVL_MAC_ADDRS_MAX; i++) {
749 		mac_addr = &dev->data->mac_addrs[i];
750 
751 		/* skip zero address */
752 		if (rte_is_zero_ether_addr(mac_addr))
753 			continue;
754 
755 		mrvl_mac_addr_add(dev, mac_addr, i, 0);
756 	}
757 
758 	if (dev->data->all_multicast == 1)
759 		mrvl_allmulticast_enable(dev);
760 
761 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_VLAN_FILTER) {
762 		ret = mrvl_populate_vlan_table(dev, 1);
763 		if (ret) {
764 			MRVL_LOG(ERR, "Failed to populate VLAN table");
765 			goto out;
766 		}
767 	}
768 
769 	/* For default QoS config, don't start classifier. */
770 	if (mrvl_qos_cfg  &&
771 	    mrvl_qos_cfg->port[dev->data->port_id].use_global_defaults == 0) {
772 		ret = mrvl_start_qos_mapping(priv);
773 		if (ret) {
774 			MRVL_LOG(ERR, "Failed to setup QoS mapping");
775 			goto out;
776 		}
777 	}
778 
779 	ret = pp2_ppio_set_loopback(priv->ppio, dev->data->dev_conf.lpbk_mode);
780 	if (ret) {
781 		MRVL_LOG(ERR, "Failed to set loopback");
782 		goto out;
783 	}
784 
785 	if (dev->data->promiscuous == 1)
786 		mrvl_promiscuous_enable(dev);
787 
788 	if (dev->data->dev_link.link_status == ETH_LINK_UP) {
789 		ret = mrvl_dev_set_link_up(dev);
790 		if (ret) {
791 			MRVL_LOG(ERR, "Failed to set link up");
792 			dev->data->dev_link.link_status = ETH_LINK_DOWN;
793 			goto out;
794 		}
795 	}
796 
797 	/* start tx queues */
798 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
799 		struct mrvl_txq *txq = dev->data->tx_queues[i];
800 
801 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
802 
803 		if (!txq->tx_deferred_start)
804 			continue;
805 
806 		/*
807 		 * All txqs are started by default. Stop them
808 		 * so that tx_deferred_start works as expected.
809 		 */
810 		ret = mrvl_tx_queue_stop(dev, i);
811 		if (ret)
812 			goto out;
813 	}
814 
815 	mrvl_flow_init(dev);
816 	mrvl_mtr_init(dev);
817 	mrvl_set_tx_function(dev);
818 
819 	return 0;
820 out:
821 	MRVL_LOG(ERR, "Failed to start device");
822 	pp2_ppio_deinit(priv->ppio);
823 	return ret;
824 }
825 
826 /**
827  * Flush receive queues.
828  *
829  * @param dev
830  *   Pointer to Ethernet device structure.
831  */
832 static void
833 mrvl_flush_rx_queues(struct rte_eth_dev *dev)
834 {
835 	int i;
836 
837 	MRVL_LOG(INFO, "Flushing rx queues");
838 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
839 		int ret, num;
840 
841 		do {
842 			struct mrvl_rxq *q = dev->data->rx_queues[i];
843 			struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];
844 
845 			num = MRVL_PP2_RXD_MAX;
846 			ret = pp2_ppio_recv(q->priv->ppio,
847 					    q->priv->rxq_map[q->queue_id].tc,
848 					    q->priv->rxq_map[q->queue_id].inq,
849 					    descs, (uint16_t *)&num);
850 		} while (ret == 0 && num);
851 	}
852 }
853 
854 /**
855  * Flush transmit shadow queues.
856  *
857  * @param dev
858  *   Pointer to Ethernet device structure.
859  */
860 static void
861 mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
862 {
863 	int i, j;
864 	struct mrvl_txq *txq;
865 
866 	MRVL_LOG(INFO, "Flushing tx shadow queues");
867 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
868 		txq = (struct mrvl_txq *)dev->data->tx_queues[i];
869 
870 		for (j = 0; j < RTE_MAX_LCORE; j++) {
871 			struct mrvl_shadow_txq *sq;
872 
873 			if (!hifs[j])
874 				continue;
875 
876 			sq = &txq->shadow_txqs[j];
877 			mrvl_free_sent_buffers(txq->priv->ppio,
878 				hifs[j], j, sq, txq->queue_id, 1);
879 			while (sq->tail != sq->head) {
880 				uint64_t addr = cookie_addr_high |
881 					sq->ent[sq->tail].buff.cookie;
882 				rte_pktmbuf_free(
883 					(struct rte_mbuf *)addr);
884 				sq->tail = (sq->tail + 1) &
885 					    MRVL_PP2_TX_SHADOWQ_MASK;
886 			}
887 			memset(sq, 0, sizeof(*sq));
888 		}
889 	}
890 }
891 
892 /**
893  * Flush hardware bpool (buffer-pool).
894  *
895  * @param dev
896  *   Pointer to Ethernet device structure.
897  */
898 static void
899 mrvl_flush_bpool(struct rte_eth_dev *dev)
900 {
901 	struct mrvl_priv *priv = dev->data->dev_private;
902 	struct pp2_hif *hif;
903 	uint32_t num;
904 	int ret;
905 	unsigned int core_id = rte_lcore_id();
906 
907 	if (core_id == LCORE_ID_ANY)
908 		core_id = rte_get_main_lcore();
909 
910 	hif = mrvl_get_hif(priv, core_id);
911 
912 	ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
913 	if (ret) {
914 		MRVL_LOG(ERR, "Failed to get bpool buffers number");
915 		return;
916 	}
917 
918 	while (num--) {
919 		struct pp2_buff_inf inf;
920 		uint64_t addr;
921 
922 		ret = pp2_bpool_get_buff(hif, priv->bpool, &inf);
923 		if (ret)
924 			break;
925 
926 		addr = cookie_addr_high | inf.cookie;
927 		rte_pktmbuf_free((struct rte_mbuf *)addr);
928 	}
929 }
930 
931 /**
932  * DPDK callback to stop the device.
933  *
934  * @param dev
935  *   Pointer to Ethernet device structure.
936  */
937 static int
938 mrvl_dev_stop(struct rte_eth_dev *dev)
939 {
940 	return mrvl_dev_set_link_down(dev);
941 }
942 
943 /**
944  * DPDK callback to close the device.
945  *
946  * @param dev
947  *   Pointer to Ethernet device structure.
948  */
949 static int
950 mrvl_dev_close(struct rte_eth_dev *dev)
951 {
952 	struct mrvl_priv *priv = dev->data->dev_private;
953 	size_t i;
954 
955 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
956 		return 0;
957 
958 	mrvl_flush_rx_queues(dev);
959 	mrvl_flush_tx_shadow_queues(dev);
960 	mrvl_flow_deinit(dev);
961 	mrvl_mtr_deinit(dev);
962 
963 	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
964 		struct pp2_ppio_tc_params *tc_params =
965 			&priv->ppio_params.inqs_params.tcs_params[i];
966 
967 		if (tc_params->inqs_params) {
968 			rte_free(tc_params->inqs_params);
969 			tc_params->inqs_params = NULL;
970 		}
971 	}
972 
973 	if (priv->cls_tbl) {
974 		pp2_cls_tbl_deinit(priv->cls_tbl);
975 		priv->cls_tbl = NULL;
976 	}
977 
978 	if (priv->qos_tbl) {
979 		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
980 		priv->qos_tbl = NULL;
981 	}
982 
983 	mrvl_flush_bpool(dev);
984 	mrvl_tm_deinit(dev);
985 
986 	if (priv->ppio) {
987 		pp2_ppio_deinit(priv->ppio);
988 		priv->ppio = NULL;
989 	}
990 
991 	/* policer must be released after ppio deinitialization */
992 	if (priv->default_policer) {
993 		pp2_cls_plcr_deinit(priv->default_policer);
994 		priv->default_policer = NULL;
995 	}
996 
997 
998 	if (priv->bpool) {
999 		pp2_bpool_deinit(priv->bpool);
1000 		used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
1001 		priv->bpool = NULL;
1002 	}
1003 
1004 	mrvl_dev_num--;
1005 
1006 	if (mrvl_dev_num == 0) {
1007 		MRVL_LOG(INFO, "Perform MUSDK deinit");
1008 		mrvl_deinit_hifs();
1009 		mrvl_deinit_pp2();
1010 		rte_mvep_deinit(MVEP_MOD_T_PP2);
1011 	}
1012 
1013 	return 0;
1014 }
1015 
1016 /**
1017  * DPDK callback to retrieve physical link information.
1018  *
1019  * @param dev
1020  *   Pointer to Ethernet device structure.
1021  * @param wait_to_complete
1022  *   Wait for request completion (ignored).
1023  *
1024  * @return
1025  *   0 on success, negative error value otherwise.
1026  */
1027 static int
1028 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1029 {
1030 	/*
1031 	 * TODO
1032 	 * once MUSDK provides necessary API use it here
1033 	 */
1034 	struct mrvl_priv *priv = dev->data->dev_private;
1035 	struct ethtool_cmd edata;
1036 	struct ifreq req;
1037 	int ret, fd, link_up;
1038 
1039 	if (!priv->ppio)
1040 		return -EPERM;
1041 
1042 	edata.cmd = ETHTOOL_GSET;
1043 
1044 	strcpy(req.ifr_name, dev->data->name);
1045 	req.ifr_data = (void *)&edata;
1046 
1047 	fd = socket(AF_INET, SOCK_DGRAM, 0);
1048 	if (fd == -1)
1049 		return -EFAULT;
1050 
1051 	ret = ioctl(fd, SIOCETHTOOL, &req);
1052 	if (ret == -1) {
1053 		close(fd);
1054 		return -EFAULT;
1055 	}
1056 
1057 	close(fd);
1058 
1059 	switch (ethtool_cmd_speed(&edata)) {
1060 	case SPEED_10:
1061 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
1062 		break;
1063 	case SPEED_100:
1064 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
1065 		break;
1066 	case SPEED_1000:
1067 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
1068 		break;
1069 	case SPEED_10000:
1070 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
1071 		break;
1072 	default:
1073 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
1074 	}
1075 
1076 	dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
1077 							 ETH_LINK_HALF_DUPLEX;
1078 	dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
1079 							   ETH_LINK_FIXED;
1080 	pp2_ppio_get_link_state(priv->ppio, &link_up);
1081 	dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
1082 
1083 	return 0;
1084 }
1085 
1086 /**
1087  * DPDK callback to enable promiscuous mode.
1088  *
1089  * @param dev
1090  *   Pointer to Ethernet device structure.
1091  *
1092  * @return
1093  *   0 on success, negative error value otherwise.
1094  */
1095 static int
1096 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
1097 {
1098 	struct mrvl_priv *priv = dev->data->dev_private;
1099 	int ret;
1100 
1101 	if (priv->isolated)
1102 		return -ENOTSUP;
1103 
1104 	if (!priv->ppio)
1105 		return 0;
1106 
1107 	ret = pp2_ppio_set_promisc(priv->ppio, 1);
1108 	if (ret) {
1109 		MRVL_LOG(ERR, "Failed to enable promiscuous mode");
1110 		return -EAGAIN;
1111 	}
1112 
1113 	return 0;
1114 }
1115 
1116 /**
1117  * DPDK callback to enable allmulti mode.
1118  *
1119  * @param dev
1120  *   Pointer to Ethernet device structure.
1121  *
1122  * @return
1123  *   0 on success, negative error value otherwise.
1124  */
1125 static int
1126 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
1127 {
1128 	struct mrvl_priv *priv = dev->data->dev_private;
1129 	int ret;
1130 
1131 	if (priv->isolated)
1132 		return -ENOTSUP;
1133 
1134 	if (!priv->ppio)
1135 		return 0;
1136 
1137 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
1138 	if (ret) {
1139 		MRVL_LOG(ERR, "Failed enable all-multicast mode");
1140 		return -EAGAIN;
1141 	}
1142 
1143 	return 0;
1144 }
1145 
1146 /**
1147  * DPDK callback to disable promiscuous mode.
1148  *
1149  * @param dev
1150  *   Pointer to Ethernet device structure.
1151  *
1152  * @return
1153  *   0 on success, negative error value otherwise.
1154  */
1155 static int
1156 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
1157 {
1158 	struct mrvl_priv *priv = dev->data->dev_private;
1159 	int ret;
1160 
1161 	if (priv->isolated)
1162 		return -ENOTSUP;
1163 
1164 	if (!priv->ppio)
1165 		return 0;
1166 
1167 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
1168 	if (ret) {
1169 		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
1170 		return -EAGAIN;
1171 	}
1172 
1173 	return 0;
1174 }
1175 
1176 /**
1177  * DPDK callback to disable allmulticast mode.
1178  *
1179  * @param dev
1180  *   Pointer to Ethernet device structure.
1181  *
1182  * @return
1183  *   0 on success, negative error value otherwise.
1184  */
1185 static int
1186 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
1187 {
1188 	struct mrvl_priv *priv = dev->data->dev_private;
1189 	int ret;
1190 
1191 	if (priv->isolated)
1192 		return -ENOTSUP;
1193 
1194 	if (!priv->ppio)
1195 		return 0;
1196 
1197 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
1198 	if (ret) {
1199 		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
1200 		return -EAGAIN;
1201 	}
1202 
1203 	return 0;
1204 }
1205 
1206 /**
1207  * DPDK callback to remove a MAC address.
1208  *
1209  * @param dev
1210  *   Pointer to Ethernet device structure.
1211  * @param index
1212  *   MAC address index.
1213  */
1214 static void
1215 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1216 {
1217 	struct mrvl_priv *priv = dev->data->dev_private;
1218 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1219 	int ret;
1220 
1221 	if (priv->isolated)
1222 		return;
1223 
1224 	if (!priv->ppio)
1225 		return;
1226 
1227 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
1228 				       dev->data->mac_addrs[index].addr_bytes);
1229 	if (ret) {
1230 		rte_ether_format_addr(buf, sizeof(buf),
1231 				  &dev->data->mac_addrs[index]);
1232 		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
1233 	}
1234 }
1235 
1236 /**
1237  * DPDK callback to add a MAC address.
1238  *
1239  * @param dev
1240  *   Pointer to Ethernet device structure.
1241  * @param mac_addr
1242  *   MAC address to register.
1243  * @param index
1244  *   MAC address index.
1245  * @param vmdq
1246  *   VMDq pool index to associate address with (unused).
1247  *
1248  * @return
1249  *   0 on success, negative error value otherwise.
1250  */
1251 static int
1252 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1253 		  uint32_t index, uint32_t vmdq __rte_unused)
1254 {
1255 	struct mrvl_priv *priv = dev->data->dev_private;
1256 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1257 	int ret;
1258 
1259 	if (priv->isolated)
1260 		return -ENOTSUP;
1261 
1262 	if (!priv->ppio)
1263 		return 0;
1264 
1265 	if (index == 0)
1266 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1267 		return -1;
1268 
1269 	/*
1270 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1271 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1272 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1273 	 * 21 respectively.
1274 	 *
1275 	 * If more than uc_filter_max uc addresses were added to filter list
1276 	 * then NIC will switch to promiscuous mode automatically.
1277 	 *
1278 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1279 	 * were added to filter list then NIC will switch to all-multicast mode
1280 	 * automatically.
1281 	 */
1282 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1283 	if (ret) {
1284 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1285 		MRVL_LOG(ERR, "Failed to add mac %s", buf);
1286 		return -1;
1287 	}
1288 
1289 	return 0;
1290 }
1291 
1292 /**
1293  * DPDK callback to set the primary MAC address.
1294  *
1295  * @param dev
1296  *   Pointer to Ethernet device structure.
1297  * @param mac_addr
1298  *   MAC address to register.
1299  *
1300  * @return
1301  *   0 on success, negative error value otherwise.
1302  */
1303 static int
1304 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1305 {
1306 	struct mrvl_priv *priv = dev->data->dev_private;
1307 	int ret;
1308 
1309 	if (priv->isolated)
1310 		return -ENOTSUP;
1311 
1312 	if (!priv->ppio)
1313 		return 0;
1314 
1315 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1316 	if (ret) {
1317 		char buf[RTE_ETHER_ADDR_FMT_SIZE];
1318 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1319 		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1320 	}
1321 
1322 	return ret;
1323 }
1324 
1325 /**
1326  * DPDK callback to get device statistics.
1327  *
1328  * @param dev
1329  *   Pointer to Ethernet device structure.
1330  * @param stats
1331  *   Stats structure output buffer.
1332  *
1333  * @return
1334  *   0 on success, negative error value otherwise.
1335  */
1336 static int
1337 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1338 {
1339 	struct mrvl_priv *priv = dev->data->dev_private;
1340 	struct pp2_ppio_statistics ppio_stats;
1341 	uint64_t drop_mac = 0;
1342 	unsigned int i, idx, ret;
1343 
1344 	if (!priv->ppio)
1345 		return -EPERM;
1346 
1347 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1348 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1349 		struct pp2_ppio_inq_statistics rx_stats;
1350 
1351 		if (!rxq)
1352 			continue;
1353 
1354 		idx = rxq->queue_id;
1355 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1356 			MRVL_LOG(ERR,
1357 				"rx queue %d stats out of range (0 - %d)",
1358 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1359 			continue;
1360 		}
1361 
1362 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1363 						  priv->rxq_map[idx].tc,
1364 						  priv->rxq_map[idx].inq,
1365 						  &rx_stats, 0);
1366 		if (unlikely(ret)) {
1367 			MRVL_LOG(ERR,
1368 				"Failed to update rx queue %d stats", idx);
1369 			break;
1370 		}
1371 
1372 		stats->q_ibytes[idx] = rxq->bytes_recv;
1373 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1374 		stats->q_errors[idx] = rx_stats.drop_early +
1375 				       rx_stats.drop_fullq +
1376 				       rx_stats.drop_bm +
1377 				       rxq->drop_mac;
1378 		stats->ibytes += rxq->bytes_recv;
1379 		drop_mac += rxq->drop_mac;
1380 	}
1381 
1382 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1383 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1384 		struct pp2_ppio_outq_statistics tx_stats;
1385 
1386 		if (!txq)
1387 			continue;
1388 
1389 		idx = txq->queue_id;
1390 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1391 			MRVL_LOG(ERR,
1392 				"tx queue %d stats out of range (0 - %d)",
1393 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1394 		}
1395 
1396 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1397 						   &tx_stats, 0);
1398 		if (unlikely(ret)) {
1399 			MRVL_LOG(ERR,
1400 				"Failed to update tx queue %d stats", idx);
1401 			break;
1402 		}
1403 
1404 		stats->q_opackets[idx] = tx_stats.deq_desc;
1405 		stats->q_obytes[idx] = txq->bytes_sent;
1406 		stats->obytes += txq->bytes_sent;
1407 	}
1408 
1409 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1410 	if (unlikely(ret)) {
1411 		MRVL_LOG(ERR, "Failed to update port statistics");
1412 		return ret;
1413 	}
1414 
1415 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1416 	stats->opackets += ppio_stats.tx_packets;
1417 	stats->imissed += ppio_stats.rx_fullq_dropped +
1418 			  ppio_stats.rx_bm_dropped +
1419 			  ppio_stats.rx_early_dropped +
1420 			  ppio_stats.rx_fifo_dropped +
1421 			  ppio_stats.rx_cls_dropped;
1422 	stats->ierrors = drop_mac;
1423 
1424 	return 0;
1425 }
1426 
1427 /**
1428  * DPDK callback to clear device statistics.
1429  *
1430  * @param dev
1431  *   Pointer to Ethernet device structure.
1432  *
1433  * @return
1434  *   0 on success, negative error value otherwise.
1435  */
1436 static int
1437 mrvl_stats_reset(struct rte_eth_dev *dev)
1438 {
1439 	struct mrvl_priv *priv = dev->data->dev_private;
1440 	int i;
1441 
1442 	if (!priv->ppio)
1443 		return 0;
1444 
1445 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1446 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1447 
1448 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1449 					    priv->rxq_map[i].inq, NULL, 1);
1450 		rxq->bytes_recv = 0;
1451 		rxq->drop_mac = 0;
1452 	}
1453 
1454 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1455 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1456 
1457 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1458 		txq->bytes_sent = 0;
1459 	}
1460 
1461 	return pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1462 }
1463 
1464 /**
1465  * DPDK callback to get extended statistics.
1466  *
1467  * @param dev
1468  *   Pointer to Ethernet device structure.
1469  * @param stats
1470  *   Pointer to xstats table.
1471  * @param n
1472  *   Number of entries in xstats table.
1473  * @return
1474  *   Negative value on error, number of read xstats otherwise.
1475  */
1476 static int
1477 mrvl_xstats_get(struct rte_eth_dev *dev,
1478 		struct rte_eth_xstat *stats, unsigned int n)
1479 {
1480 	struct mrvl_priv *priv = dev->data->dev_private;
1481 	struct pp2_ppio_statistics ppio_stats;
1482 	unsigned int i;
1483 
1484 	if (!stats)
1485 		return 0;
1486 
1487 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1488 	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1489 		uint64_t val;
1490 
1491 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1492 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1493 					    mrvl_xstats_tbl[i].offset);
1494 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1495 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1496 					    mrvl_xstats_tbl[i].offset);
1497 		else
1498 			return -EINVAL;
1499 
1500 		stats[i].id = i;
1501 		stats[i].value = val;
1502 	}
1503 
1504 	return n;
1505 }
1506 
1507 /**
1508  * DPDK callback to reset extended statistics.
1509  *
1510  * @param dev
1511  *   Pointer to Ethernet device structure.
1512  *
1513  * @return
1514  *   0 on success, negative error value otherwise.
1515  */
1516 static int
1517 mrvl_xstats_reset(struct rte_eth_dev *dev)
1518 {
1519 	return mrvl_stats_reset(dev);
1520 }
1521 
1522 /**
1523  * DPDK callback to get extended statistics names.
1524  *
1525  * @param dev (unused)
1526  *   Pointer to Ethernet device structure.
1527  * @param xstats_names
1528  *   Pointer to xstats names table.
1529  * @param size
1530  *   Size of the xstats names table.
1531  * @return
1532  *   Number of read names.
1533  */
1534 static int
1535 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1536 		      struct rte_eth_xstat_name *xstats_names,
1537 		      unsigned int size)
1538 {
1539 	unsigned int i;
1540 
1541 	if (!xstats_names)
1542 		return RTE_DIM(mrvl_xstats_tbl);
1543 
1544 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1545 		strlcpy(xstats_names[i].name, mrvl_xstats_tbl[i].name,
1546 			RTE_ETH_XSTATS_NAME_SIZE);
1547 
1548 	return size;
1549 }
1550 
1551 /**
1552  * DPDK callback to get information about the device.
1553  *
1554  * @param dev
1555  *   Pointer to Ethernet device structure (unused).
1556  * @param info
1557  *   Info structure output buffer.
1558  */
1559 static int
1560 mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
1561 		   struct rte_eth_dev_info *info)
1562 {
1563 	info->speed_capa = ETH_LINK_SPEED_10M |
1564 			   ETH_LINK_SPEED_100M |
1565 			   ETH_LINK_SPEED_1G |
1566 			   ETH_LINK_SPEED_10G;
1567 
1568 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1569 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1570 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1571 
1572 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1573 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1574 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1575 
1576 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1577 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1578 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1579 
1580 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1581 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1582 
1583 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1584 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1585 
1586 	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1587 				       ETH_RSS_NONFRAG_IPV4_TCP |
1588 				       ETH_RSS_NONFRAG_IPV4_UDP;
1589 
1590 	/* By default packets are dropped if no descriptors are available */
1591 	info->default_rxconf.rx_drop_en = 1;
1592 
1593 	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1594 
1595 	return 0;
1596 }
1597 
1598 /**
1599  * Return supported packet types.
1600  *
1601  * @param dev
1602  *   Pointer to Ethernet device structure (unused).
1603  *
1604  * @return
1605  *   Const pointer to the table with supported packet types.
1606  */
1607 static const uint32_t *
1608 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1609 {
1610 	static const uint32_t ptypes[] = {
1611 		RTE_PTYPE_L2_ETHER,
1612 		RTE_PTYPE_L2_ETHER_VLAN,
1613 		RTE_PTYPE_L2_ETHER_QINQ,
1614 		RTE_PTYPE_L3_IPV4,
1615 		RTE_PTYPE_L3_IPV4_EXT,
1616 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1617 		RTE_PTYPE_L3_IPV6,
1618 		RTE_PTYPE_L3_IPV6_EXT,
1619 		RTE_PTYPE_L2_ETHER_ARP,
1620 		RTE_PTYPE_L4_TCP,
1621 		RTE_PTYPE_L4_UDP
1622 	};
1623 
1624 	return ptypes;
1625 }
1626 
1627 /**
1628  * DPDK callback to get information about specific receive queue.
1629  *
1630  * @param dev
1631  *   Pointer to Ethernet device structure.
1632  * @param rx_queue_id
1633  *   Receive queue index.
1634  * @param qinfo
1635  *   Receive queue information structure.
1636  */
1637 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1638 			      struct rte_eth_rxq_info *qinfo)
1639 {
1640 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1641 	struct mrvl_priv *priv = dev->data->dev_private;
1642 	int inq = priv->rxq_map[rx_queue_id].inq;
1643 	int tc = priv->rxq_map[rx_queue_id].tc;
1644 	struct pp2_ppio_tc_params *tc_params =
1645 		&priv->ppio_params.inqs_params.tcs_params[tc];
1646 
1647 	qinfo->mp = q->mp;
1648 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1649 }
1650 
1651 /**
1652  * DPDK callback to get information about specific transmit queue.
1653  *
1654  * @param dev
1655  *   Pointer to Ethernet device structure.
1656  * @param tx_queue_id
1657  *   Transmit queue index.
1658  * @param qinfo
1659  *   Transmit queue information structure.
1660  */
1661 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1662 			      struct rte_eth_txq_info *qinfo)
1663 {
1664 	struct mrvl_priv *priv = dev->data->dev_private;
1665 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1666 
1667 	qinfo->nb_desc =
1668 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1669 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1670 }
1671 
1672 /**
1673  * DPDK callback to Configure a VLAN filter.
1674  *
1675  * @param dev
1676  *   Pointer to Ethernet device structure.
1677  * @param vlan_id
1678  *   VLAN ID to filter.
1679  * @param on
1680  *   Toggle filter.
1681  *
1682  * @return
1683  *   0 on success, negative error value otherwise.
1684  */
1685 static int
1686 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1687 {
1688 	struct mrvl_priv *priv = dev->data->dev_private;
1689 
1690 	if (priv->isolated)
1691 		return -ENOTSUP;
1692 
1693 	if (!priv->ppio)
1694 		return 0;
1695 
1696 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1697 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1698 }
1699 
1700 /**
1701  * DPDK callback to Configure VLAN offload.
1702  *
1703  * @param dev
1704  *   Pointer to Ethernet device structure.
1705  * @param mask
1706  *   VLAN offload mask.
1707  *
1708  * @return
1709  *   0 on success, negative error value otherwise.
1710  */
1711 static int mrvl_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1712 {
1713 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
1714 	int ret;
1715 
1716 	if (mask & ETH_VLAN_STRIP_MASK)
1717 		MRVL_LOG(ERR, "VLAN stripping is not supported\n");
1718 
1719 	if (mask & ETH_VLAN_FILTER_MASK) {
1720 		if (rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1721 			ret = mrvl_populate_vlan_table(dev, 1);
1722 		else
1723 			ret = mrvl_populate_vlan_table(dev, 0);
1724 
1725 		if (ret)
1726 			return ret;
1727 	}
1728 
1729 	if (mask & ETH_VLAN_EXTEND_MASK)
1730 		MRVL_LOG(ERR, "Extend VLAN not supported\n");
1731 
1732 	return 0;
1733 }
1734 
1735 /**
1736  * Release buffers to hardware bpool (buffer-pool)
1737  *
1738  * @param rxq
1739  *   Receive queue pointer.
1740  * @param num
1741  *   Number of buffers to release to bpool.
1742  *
1743  * @return
1744  *   0 on success, negative error value otherwise.
1745  */
1746 static int
1747 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1748 {
1749 	struct buff_release_entry entries[num];
1750 	struct rte_mbuf *mbufs[num];
1751 	int i, ret;
1752 	unsigned int core_id;
1753 	struct pp2_hif *hif;
1754 	struct pp2_bpool *bpool;
1755 
1756 	core_id = rte_lcore_id();
1757 	if (core_id == LCORE_ID_ANY)
1758 		core_id = rte_get_main_lcore();
1759 
1760 	hif = mrvl_get_hif(rxq->priv, core_id);
1761 	if (!hif)
1762 		return -1;
1763 
1764 	bpool = rxq->priv->bpool;
1765 
1766 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1767 	if (ret)
1768 		return ret;
1769 
1770 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1771 		cookie_addr_high =
1772 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1773 
1774 	for (i = 0; i < num; i++) {
1775 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1776 			!= cookie_addr_high) {
1777 			MRVL_LOG(ERR,
1778 				"mbuf virtual addr high is out of range "
1779 				"0x%x instead of 0x%x\n",
1780 				(uint32_t)((uint64_t)mbufs[i] >> 32),
1781 				(uint32_t)(cookie_addr_high >> 32));
1782 			goto out;
1783 		}
1784 
1785 		entries[i].buff.addr =
1786 			rte_mbuf_data_iova_default(mbufs[i]);
1787 		entries[i].buff.cookie = (uintptr_t)mbufs[i];
1788 		entries[i].bpool = bpool;
1789 	}
1790 
1791 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1792 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1793 
1794 	if (i != num)
1795 		goto out;
1796 
1797 	return 0;
1798 out:
1799 	for (; i < num; i++)
1800 		rte_pktmbuf_free(mbufs[i]);
1801 
1802 	return -1;
1803 }
1804 
1805 /**
1806  * DPDK callback to configure the receive queue.
1807  *
1808  * @param dev
1809  *   Pointer to Ethernet device structure.
1810  * @param idx
1811  *   RX queue index.
1812  * @param desc
1813  *   Number of descriptors to configure in queue.
1814  * @param socket
1815  *   NUMA socket on which memory must be allocated.
1816  * @param conf
1817  *   Thresholds parameters.
1818  * @param mp
1819  *   Memory pool for buffer allocations.
1820  *
1821  * @return
1822  *   0 on success, negative error value otherwise.
1823  */
1824 static int
1825 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1826 		    unsigned int socket,
1827 		    const struct rte_eth_rxconf *conf,
1828 		    struct rte_mempool *mp)
1829 {
1830 	struct mrvl_priv *priv = dev->data->dev_private;
1831 	struct mrvl_rxq *rxq;
1832 	uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
1833 	uint32_t max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1834 	int ret, tc, inq;
1835 	uint64_t offloads;
1836 
1837 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
1838 
1839 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1840 		/*
1841 		 * Unknown TC mapping, mapping will not have a correct queue.
1842 		 */
1843 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
1844 			idx, priv->ppio_id);
1845 		return -EFAULT;
1846 	}
1847 
1848 	frame_size = buf_size - RTE_PKTMBUF_HEADROOM -
1849 		     MRVL_PKT_EFFEC_OFFS + RTE_ETHER_CRC_LEN;
1850 	if (frame_size < max_rx_pkt_len) {
1851 		MRVL_LOG(WARNING,
1852 			"Mbuf size must be increased to %u bytes to hold up "
1853 			"to %u bytes of data.",
1854 			buf_size + max_rx_pkt_len - frame_size,
1855 			max_rx_pkt_len);
1856 		dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
1857 		MRVL_LOG(INFO, "Setting max rx pkt len to %u",
1858 			dev->data->dev_conf.rxmode.max_rx_pkt_len);
1859 	}
1860 
1861 	if (dev->data->rx_queues[idx]) {
1862 		rte_free(dev->data->rx_queues[idx]);
1863 		dev->data->rx_queues[idx] = NULL;
1864 	}
1865 
1866 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
1867 	if (!rxq)
1868 		return -ENOMEM;
1869 
1870 	rxq->priv = priv;
1871 	rxq->mp = mp;
1872 	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
1873 	rxq->queue_id = idx;
1874 	rxq->port_id = dev->data->port_id;
1875 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
1876 
1877 	tc = priv->rxq_map[rxq->queue_id].tc,
1878 	inq = priv->rxq_map[rxq->queue_id].inq;
1879 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
1880 		desc;
1881 
1882 	ret = mrvl_fill_bpool(rxq, desc);
1883 	if (ret) {
1884 		rte_free(rxq);
1885 		return ret;
1886 	}
1887 
1888 	priv->bpool_init_size += desc;
1889 
1890 	dev->data->rx_queues[idx] = rxq;
1891 
1892 	return 0;
1893 }
1894 
1895 /**
1896  * DPDK callback to release the receive queue.
1897  *
1898  * @param rxq
1899  *   Generic receive queue pointer.
1900  */
1901 static void
1902 mrvl_rx_queue_release(void *rxq)
1903 {
1904 	struct mrvl_rxq *q = rxq;
1905 	struct pp2_ppio_tc_params *tc_params;
1906 	int i, num, tc, inq;
1907 	struct pp2_hif *hif;
1908 	unsigned int core_id = rte_lcore_id();
1909 
1910 	if (core_id == LCORE_ID_ANY)
1911 		core_id = rte_get_main_lcore();
1912 
1913 	if (!q)
1914 		return;
1915 
1916 	hif = mrvl_get_hif(q->priv, core_id);
1917 
1918 	if (!hif)
1919 		return;
1920 
1921 	tc = q->priv->rxq_map[q->queue_id].tc;
1922 	inq = q->priv->rxq_map[q->queue_id].inq;
1923 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
1924 	num = tc_params->inqs_params[inq].size;
1925 	for (i = 0; i < num; i++) {
1926 		struct pp2_buff_inf inf;
1927 		uint64_t addr;
1928 
1929 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
1930 		addr = cookie_addr_high | inf.cookie;
1931 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1932 	}
1933 
1934 	rte_free(q);
1935 }
1936 
1937 /**
1938  * DPDK callback to configure the transmit queue.
1939  *
1940  * @param dev
1941  *   Pointer to Ethernet device structure.
1942  * @param idx
1943  *   Transmit queue index.
1944  * @param desc
1945  *   Number of descriptors to configure in the queue.
1946  * @param socket
1947  *   NUMA socket on which memory must be allocated.
1948  * @param conf
1949  *   Tx queue configuration parameters.
1950  *
1951  * @return
1952  *   0 on success, negative error value otherwise.
1953  */
1954 static int
1955 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1956 		    unsigned int socket,
1957 		    const struct rte_eth_txconf *conf)
1958 {
1959 	struct mrvl_priv *priv = dev->data->dev_private;
1960 	struct mrvl_txq *txq;
1961 
1962 	if (dev->data->tx_queues[idx]) {
1963 		rte_free(dev->data->tx_queues[idx]);
1964 		dev->data->tx_queues[idx] = NULL;
1965 	}
1966 
1967 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
1968 	if (!txq)
1969 		return -ENOMEM;
1970 
1971 	txq->priv = priv;
1972 	txq->queue_id = idx;
1973 	txq->port_id = dev->data->port_id;
1974 	txq->tx_deferred_start = conf->tx_deferred_start;
1975 	dev->data->tx_queues[idx] = txq;
1976 
1977 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
1978 
1979 	return 0;
1980 }
1981 
1982 /**
1983  * DPDK callback to release the transmit queue.
1984  *
1985  * @param txq
1986  *   Generic transmit queue pointer.
1987  */
1988 static void
1989 mrvl_tx_queue_release(void *txq)
1990 {
1991 	struct mrvl_txq *q = txq;
1992 
1993 	if (!q)
1994 		return;
1995 
1996 	rte_free(q);
1997 }
1998 
1999 /**
2000  * DPDK callback to get flow control configuration.
2001  *
2002  * @param dev
2003  *  Pointer to Ethernet device structure.
2004  * @param fc_conf
2005  *  Pointer to the flow control configuration.
2006  *
2007  * @return
2008  *  0 on success, negative error value otherwise.
2009  */
2010 static int
2011 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2012 {
2013 	struct mrvl_priv *priv = dev->data->dev_private;
2014 	int ret, en;
2015 
2016 	if (!priv)
2017 		return -EPERM;
2018 
2019 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
2020 	if (ret) {
2021 		MRVL_LOG(ERR, "Failed to read rx pause state");
2022 		return ret;
2023 	}
2024 
2025 	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
2026 
2027 	return 0;
2028 }
2029 
2030 /**
2031  * DPDK callback to set flow control configuration.
2032  *
2033  * @param dev
2034  *  Pointer to Ethernet device structure.
2035  * @param fc_conf
2036  *  Pointer to the flow control configuration.
2037  *
2038  * @return
2039  *  0 on success, negative error value otherwise.
2040  */
2041 static int
2042 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2043 {
2044 	struct mrvl_priv *priv = dev->data->dev_private;
2045 
2046 	if (!priv)
2047 		return -EPERM;
2048 
2049 	if (fc_conf->high_water ||
2050 	    fc_conf->low_water ||
2051 	    fc_conf->pause_time ||
2052 	    fc_conf->mac_ctrl_frame_fwd ||
2053 	    fc_conf->autoneg) {
2054 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
2055 
2056 		return -EINVAL;
2057 	}
2058 
2059 	if (fc_conf->mode == RTE_FC_NONE ||
2060 	    fc_conf->mode == RTE_FC_RX_PAUSE) {
2061 		int ret, en;
2062 
2063 		en = fc_conf->mode == RTE_FC_NONE ? 0 : 1;
2064 		ret = pp2_ppio_set_rx_pause(priv->ppio, en);
2065 		if (ret)
2066 			MRVL_LOG(ERR,
2067 				"Failed to change flowctrl on RX side");
2068 
2069 		return ret;
2070 	}
2071 
2072 	return 0;
2073 }
2074 
2075 /**
2076  * Update RSS hash configuration
2077  *
2078  * @param dev
2079  *   Pointer to Ethernet device structure.
2080  * @param rss_conf
2081  *   Pointer to RSS configuration.
2082  *
2083  * @return
2084  *   0 on success, negative error value otherwise.
2085  */
2086 static int
2087 mrvl_rss_hash_update(struct rte_eth_dev *dev,
2088 		     struct rte_eth_rss_conf *rss_conf)
2089 {
2090 	struct mrvl_priv *priv = dev->data->dev_private;
2091 
2092 	if (priv->isolated)
2093 		return -ENOTSUP;
2094 
2095 	return mrvl_configure_rss(priv, rss_conf);
2096 }
2097 
2098 /**
2099  * DPDK callback to get RSS hash configuration.
2100  *
2101  * @param dev
2102  *   Pointer to Ethernet device structure.
2103  * @rss_conf
2104  *   Pointer to RSS configuration.
2105  *
2106  * @return
2107  *   Always 0.
2108  */
2109 static int
2110 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
2111 		       struct rte_eth_rss_conf *rss_conf)
2112 {
2113 	struct mrvl_priv *priv = dev->data->dev_private;
2114 	enum pp2_ppio_hash_type hash_type =
2115 		priv->ppio_params.inqs_params.hash_type;
2116 
2117 	rss_conf->rss_key = NULL;
2118 
2119 	if (hash_type == PP2_PPIO_HASH_T_NONE)
2120 		rss_conf->rss_hf = 0;
2121 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
2122 		rss_conf->rss_hf = ETH_RSS_IPV4;
2123 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
2124 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
2125 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
2126 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
2127 
2128 	return 0;
2129 }
2130 
2131 /**
2132  * DPDK callback to get rte_flow callbacks.
2133  *
2134  * @param dev
2135  *   Pointer to the device structure.
2136  * @param filer_type
2137  *   Flow filter type.
2138  * @param filter_op
2139  *   Flow filter operation.
2140  * @param arg
2141  *   Pointer to pass the flow ops.
2142  *
2143  * @return
2144  *   0 on success, negative error value otherwise.
2145  */
2146 static int
2147 mrvl_eth_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
2148 		     enum rte_filter_type filter_type,
2149 		     enum rte_filter_op filter_op, void *arg)
2150 {
2151 	switch (filter_type) {
2152 	case RTE_ETH_FILTER_GENERIC:
2153 		if (filter_op != RTE_ETH_FILTER_GET)
2154 			return -EINVAL;
2155 		*(const void **)arg = &mrvl_flow_ops;
2156 		return 0;
2157 	default:
2158 		MRVL_LOG(WARNING, "Filter type (%d) not supported",
2159 				filter_type);
2160 		return -EINVAL;
2161 	}
2162 }
2163 
2164 /**
2165  * DPDK callback to get rte_mtr callbacks.
2166  *
2167  * @param dev
2168  *   Pointer to the device structure.
2169  * @param ops
2170  *   Pointer to pass the mtr ops.
2171  *
2172  * @return
2173  *   Always 0.
2174  */
2175 static int
2176 mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2177 {
2178 	*(const void **)ops = &mrvl_mtr_ops;
2179 
2180 	return 0;
2181 }
2182 
2183 /**
2184  * DPDK callback to get rte_tm callbacks.
2185  *
2186  * @param dev
2187  *   Pointer to the device structure.
2188  * @param ops
2189  *   Pointer to pass the tm ops.
2190  *
2191  * @return
2192  *   Always 0.
2193  */
2194 static int
2195 mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2196 {
2197 	*(const void **)ops = &mrvl_tm_ops;
2198 
2199 	return 0;
2200 }
2201 
2202 static const struct eth_dev_ops mrvl_ops = {
2203 	.dev_configure = mrvl_dev_configure,
2204 	.dev_start = mrvl_dev_start,
2205 	.dev_stop = mrvl_dev_stop,
2206 	.dev_set_link_up = mrvl_dev_set_link_up,
2207 	.dev_set_link_down = mrvl_dev_set_link_down,
2208 	.dev_close = mrvl_dev_close,
2209 	.link_update = mrvl_link_update,
2210 	.promiscuous_enable = mrvl_promiscuous_enable,
2211 	.allmulticast_enable = mrvl_allmulticast_enable,
2212 	.promiscuous_disable = mrvl_promiscuous_disable,
2213 	.allmulticast_disable = mrvl_allmulticast_disable,
2214 	.mac_addr_remove = mrvl_mac_addr_remove,
2215 	.mac_addr_add = mrvl_mac_addr_add,
2216 	.mac_addr_set = mrvl_mac_addr_set,
2217 	.mtu_set = mrvl_mtu_set,
2218 	.stats_get = mrvl_stats_get,
2219 	.stats_reset = mrvl_stats_reset,
2220 	.xstats_get = mrvl_xstats_get,
2221 	.xstats_reset = mrvl_xstats_reset,
2222 	.xstats_get_names = mrvl_xstats_get_names,
2223 	.dev_infos_get = mrvl_dev_infos_get,
2224 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
2225 	.rxq_info_get = mrvl_rxq_info_get,
2226 	.txq_info_get = mrvl_txq_info_get,
2227 	.vlan_filter_set = mrvl_vlan_filter_set,
2228 	.vlan_offload_set = mrvl_vlan_offload_set,
2229 	.tx_queue_start = mrvl_tx_queue_start,
2230 	.tx_queue_stop = mrvl_tx_queue_stop,
2231 	.rx_queue_setup = mrvl_rx_queue_setup,
2232 	.rx_queue_release = mrvl_rx_queue_release,
2233 	.tx_queue_setup = mrvl_tx_queue_setup,
2234 	.tx_queue_release = mrvl_tx_queue_release,
2235 	.flow_ctrl_get = mrvl_flow_ctrl_get,
2236 	.flow_ctrl_set = mrvl_flow_ctrl_set,
2237 	.rss_hash_update = mrvl_rss_hash_update,
2238 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
2239 	.filter_ctrl = mrvl_eth_filter_ctrl,
2240 	.mtr_ops_get = mrvl_mtr_ops_get,
2241 	.tm_ops_get = mrvl_tm_ops_get,
2242 };
2243 
2244 /**
2245  * Return packet type information and l3/l4 offsets.
2246  *
2247  * @param desc
2248  *   Pointer to the received packet descriptor.
2249  * @param l3_offset
2250  *   l3 packet offset.
2251  * @param l4_offset
2252  *   l4 packet offset.
2253  *
2254  * @return
2255  *   Packet type information.
2256  */
2257 static inline uint64_t
2258 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
2259 				    uint8_t *l3_offset, uint8_t *l4_offset)
2260 {
2261 	enum pp2_inq_l3_type l3_type;
2262 	enum pp2_inq_l4_type l4_type;
2263 	enum pp2_inq_vlan_tag vlan_tag;
2264 	uint64_t packet_type;
2265 
2266 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
2267 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
2268 	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
2269 
2270 	packet_type = RTE_PTYPE_L2_ETHER;
2271 
2272 	switch (vlan_tag) {
2273 	case PP2_INQ_VLAN_TAG_SINGLE:
2274 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
2275 		break;
2276 	case PP2_INQ_VLAN_TAG_DOUBLE:
2277 	case PP2_INQ_VLAN_TAG_TRIPLE:
2278 		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
2279 		break;
2280 	default:
2281 		break;
2282 	}
2283 
2284 	switch (l3_type) {
2285 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
2286 		packet_type |= RTE_PTYPE_L3_IPV4;
2287 		break;
2288 	case PP2_INQ_L3_TYPE_IPV4_OK:
2289 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
2290 		break;
2291 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
2292 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
2293 		break;
2294 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
2295 		packet_type |= RTE_PTYPE_L3_IPV6;
2296 		break;
2297 	case PP2_INQ_L3_TYPE_IPV6_EXT:
2298 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
2299 		break;
2300 	case PP2_INQ_L3_TYPE_ARP:
2301 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
2302 		/*
2303 		 * In case of ARP l4_offset is set to wrong value.
2304 		 * Set it to proper one so that later on mbuf->l3_len can be
2305 		 * calculated subtracting l4_offset and l3_offset.
2306 		 */
2307 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
2308 		break;
2309 	default:
2310 		break;
2311 	}
2312 
2313 	switch (l4_type) {
2314 	case PP2_INQ_L4_TYPE_TCP:
2315 		packet_type |= RTE_PTYPE_L4_TCP;
2316 		break;
2317 	case PP2_INQ_L4_TYPE_UDP:
2318 		packet_type |= RTE_PTYPE_L4_UDP;
2319 		break;
2320 	default:
2321 		break;
2322 	}
2323 
2324 	return packet_type;
2325 }
2326 
2327 /**
2328  * Get offload information from the received packet descriptor.
2329  *
2330  * @param desc
2331  *   Pointer to the received packet descriptor.
2332  *
2333  * @return
2334  *   Mbuf offload flags.
2335  */
2336 static inline uint64_t
2337 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc)
2338 {
2339 	uint64_t flags;
2340 	enum pp2_inq_desc_status status;
2341 
2342 	status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2343 	if (unlikely(status != PP2_DESC_ERR_OK))
2344 		flags = PKT_RX_IP_CKSUM_BAD;
2345 	else
2346 		flags = PKT_RX_IP_CKSUM_GOOD;
2347 
2348 	status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2349 	if (unlikely(status != PP2_DESC_ERR_OK))
2350 		flags |= PKT_RX_L4_CKSUM_BAD;
2351 	else
2352 		flags |= PKT_RX_L4_CKSUM_GOOD;
2353 
2354 	return flags;
2355 }
2356 
2357 /**
2358  * DPDK callback for receive.
2359  *
2360  * @param rxq
2361  *   Generic pointer to the receive queue.
2362  * @param rx_pkts
2363  *   Array to store received packets.
2364  * @param nb_pkts
2365  *   Maximum number of packets in array.
2366  *
2367  * @return
2368  *   Number of packets successfully received.
2369  */
2370 static uint16_t
2371 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2372 {
2373 	struct mrvl_rxq *q = rxq;
2374 	struct pp2_ppio_desc descs[nb_pkts];
2375 	struct pp2_bpool *bpool;
2376 	int i, ret, rx_done = 0;
2377 	int num;
2378 	struct pp2_hif *hif;
2379 	unsigned int core_id = rte_lcore_id();
2380 
2381 	hif = mrvl_get_hif(q->priv, core_id);
2382 
2383 	if (unlikely(!q->priv->ppio || !hif))
2384 		return 0;
2385 
2386 	bpool = q->priv->bpool;
2387 
2388 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2389 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2390 	if (unlikely(ret < 0))
2391 		return 0;
2392 
2393 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2394 
2395 	for (i = 0; i < nb_pkts; i++) {
2396 		struct rte_mbuf *mbuf;
2397 		uint8_t l3_offset, l4_offset;
2398 		enum pp2_inq_desc_status status;
2399 		uint64_t addr;
2400 
2401 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2402 			struct pp2_ppio_desc *pref_desc;
2403 			u64 pref_addr;
2404 
2405 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2406 			pref_addr = cookie_addr_high |
2407 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2408 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2409 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2410 		}
2411 
2412 		addr = cookie_addr_high |
2413 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2414 		mbuf = (struct rte_mbuf *)addr;
2415 		rte_pktmbuf_reset(mbuf);
2416 
2417 		/* drop packet in case of mac, overrun or resource error */
2418 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2419 		if (unlikely(status != PP2_DESC_ERR_OK)) {
2420 			struct pp2_buff_inf binf = {
2421 				.addr = rte_mbuf_data_iova_default(mbuf),
2422 				.cookie = (uint64_t)mbuf,
2423 			};
2424 
2425 			pp2_bpool_put_buff(hif, bpool, &binf);
2426 			mrvl_port_bpool_size
2427 				[bpool->pp2_id][bpool->id][core_id]++;
2428 			q->drop_mac++;
2429 			continue;
2430 		}
2431 
2432 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2433 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2434 		mbuf->data_len = mbuf->pkt_len;
2435 		mbuf->port = q->port_id;
2436 		mbuf->packet_type =
2437 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2438 							    &l3_offset,
2439 							    &l4_offset);
2440 		mbuf->l2_len = l3_offset;
2441 		mbuf->l3_len = l4_offset - l3_offset;
2442 
2443 		if (likely(q->cksum_enabled))
2444 			mbuf->ol_flags = mrvl_desc_to_ol_flags(&descs[i]);
2445 
2446 		rx_pkts[rx_done++] = mbuf;
2447 		q->bytes_recv += mbuf->pkt_len;
2448 	}
2449 
2450 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2451 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2452 
2453 		if (unlikely(num <= q->priv->bpool_min_size ||
2454 			     (!rx_done && num < q->priv->bpool_init_size))) {
2455 			mrvl_fill_bpool(q, MRVL_BURST_SIZE);
2456 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2457 			int i;
2458 			int pkt_to_remove = num - q->priv->bpool_init_size;
2459 			struct rte_mbuf *mbuf;
2460 			struct pp2_buff_inf buff;
2461 
2462 			for (i = 0; i < pkt_to_remove; i++) {
2463 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2464 				if (ret)
2465 					break;
2466 				mbuf = (struct rte_mbuf *)
2467 					(cookie_addr_high | buff.cookie);
2468 				rte_pktmbuf_free(mbuf);
2469 			}
2470 			mrvl_port_bpool_size
2471 				[bpool->pp2_id][bpool->id][core_id] -= i;
2472 		}
2473 		rte_spinlock_unlock(&q->priv->lock);
2474 	}
2475 
2476 	return rx_done;
2477 }
2478 
2479 /**
2480  * Prepare offload information.
2481  *
2482  * @param ol_flags
2483  *   Offload flags.
2484  * @param packet_type
2485  *   Packet type bitfield.
2486  * @param l3_type
2487  *   Pointer to the pp2_ouq_l3_type structure.
2488  * @param l4_type
2489  *   Pointer to the pp2_outq_l4_type structure.
2490  * @param gen_l3_cksum
2491  *   Will be set to 1 in case l3 checksum is computed.
2492  * @param l4_cksum
2493  *   Will be set to 1 in case l4 checksum is computed.
2494  *
2495  * @return
2496  *   0 on success, negative error value otherwise.
2497  */
2498 static inline int
2499 mrvl_prepare_proto_info(uint64_t ol_flags, uint32_t packet_type,
2500 			enum pp2_outq_l3_type *l3_type,
2501 			enum pp2_outq_l4_type *l4_type,
2502 			int *gen_l3_cksum,
2503 			int *gen_l4_cksum)
2504 {
2505 	/*
2506 	 * Based on ol_flags prepare information
2507 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2508 	 * for offloading.
2509 	 */
2510 	if (ol_flags & PKT_TX_IPV4) {
2511 		*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2512 		*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2513 	} else if (ol_flags & PKT_TX_IPV6) {
2514 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2515 		/* no checksum for ipv6 header */
2516 		*gen_l3_cksum = 0;
2517 	} else {
2518 		/* if something different then stop processing */
2519 		return -1;
2520 	}
2521 
2522 	ol_flags &= PKT_TX_L4_MASK;
2523 	if ((packet_type & RTE_PTYPE_L4_TCP) &&
2524 	    ol_flags == PKT_TX_TCP_CKSUM) {
2525 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2526 		*gen_l4_cksum = 1;
2527 	} else if ((packet_type & RTE_PTYPE_L4_UDP) &&
2528 		   ol_flags == PKT_TX_UDP_CKSUM) {
2529 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2530 		*gen_l4_cksum = 1;
2531 	} else {
2532 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2533 		/* no checksum for other type */
2534 		*gen_l4_cksum = 0;
2535 	}
2536 
2537 	return 0;
2538 }
2539 
2540 /**
2541  * Release already sent buffers to bpool (buffer-pool).
2542  *
2543  * @param ppio
2544  *   Pointer to the port structure.
2545  * @param hif
2546  *   Pointer to the MUSDK hardware interface.
2547  * @param sq
2548  *   Pointer to the shadow queue.
2549  * @param qid
2550  *   Queue id number.
2551  * @param force
2552  *   Force releasing packets.
2553  */
2554 static inline void
2555 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2556 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2557 		       int qid, int force)
2558 {
2559 	struct buff_release_entry *entry;
2560 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2561 	int i;
2562 
2563 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2564 
2565 	sq->num_to_release += nb_done;
2566 
2567 	if (likely(!force &&
2568 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2569 		return;
2570 
2571 	nb_done = sq->num_to_release;
2572 	sq->num_to_release = 0;
2573 
2574 	for (i = 0; i < nb_done; i++) {
2575 		entry = &sq->ent[sq->tail + num];
2576 		if (unlikely(!entry->buff.addr)) {
2577 			MRVL_LOG(ERR,
2578 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2579 				sq->tail, (u64)entry->buff.cookie,
2580 				(u64)entry->buff.addr);
2581 			skip_bufs = 1;
2582 			goto skip;
2583 		}
2584 
2585 		if (unlikely(!entry->bpool)) {
2586 			struct rte_mbuf *mbuf;
2587 
2588 			mbuf = (struct rte_mbuf *)entry->buff.cookie;
2589 			rte_pktmbuf_free(mbuf);
2590 			skip_bufs = 1;
2591 			goto skip;
2592 		}
2593 
2594 		mrvl_port_bpool_size
2595 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2596 		num++;
2597 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2598 			goto skip;
2599 		continue;
2600 skip:
2601 		if (likely(num))
2602 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2603 		num += skip_bufs;
2604 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2605 		sq->size -= num;
2606 		num = 0;
2607 		skip_bufs = 0;
2608 	}
2609 
2610 	if (likely(num)) {
2611 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2612 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2613 		sq->size -= num;
2614 	}
2615 }
2616 
2617 /**
2618  * DPDK callback for transmit.
2619  *
2620  * @param txq
2621  *   Generic pointer transmit queue.
2622  * @param tx_pkts
2623  *   Packets to transmit.
2624  * @param nb_pkts
2625  *   Number of packets in array.
2626  *
2627  * @return
2628  *   Number of packets successfully transmitted.
2629  */
2630 static uint16_t
2631 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2632 {
2633 	struct mrvl_txq *q = txq;
2634 	struct mrvl_shadow_txq *sq;
2635 	struct pp2_hif *hif;
2636 	struct pp2_ppio_desc descs[nb_pkts];
2637 	unsigned int core_id = rte_lcore_id();
2638 	int i, ret, bytes_sent = 0;
2639 	uint16_t num, sq_free_size;
2640 	uint64_t addr;
2641 
2642 	hif = mrvl_get_hif(q->priv, core_id);
2643 	sq = &q->shadow_txqs[core_id];
2644 
2645 	if (unlikely(!q->priv->ppio || !hif))
2646 		return 0;
2647 
2648 	if (sq->size)
2649 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2650 				       sq, q->queue_id, 0);
2651 
2652 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2653 	if (unlikely(nb_pkts > sq_free_size))
2654 		nb_pkts = sq_free_size;
2655 
2656 	for (i = 0; i < nb_pkts; i++) {
2657 		struct rte_mbuf *mbuf = tx_pkts[i];
2658 		int gen_l3_cksum, gen_l4_cksum;
2659 		enum pp2_outq_l3_type l3_type;
2660 		enum pp2_outq_l4_type l4_type;
2661 
2662 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2663 			struct rte_mbuf *pref_pkt_hdr;
2664 
2665 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2666 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2667 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2668 		}
2669 
2670 		mrvl_fill_shadowq(sq, mbuf);
2671 		mrvl_fill_desc(&descs[i], mbuf);
2672 
2673 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2674 		/*
2675 		 * in case unsupported ol_flags were passed
2676 		 * do not update descriptor offload information
2677 		 */
2678 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2679 					      &l3_type, &l4_type, &gen_l3_cksum,
2680 					      &gen_l4_cksum);
2681 		if (unlikely(ret))
2682 			continue;
2683 
2684 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2685 						  mbuf->l2_len,
2686 						  mbuf->l2_len + mbuf->l3_len,
2687 						  gen_l3_cksum, gen_l4_cksum);
2688 	}
2689 
2690 	num = nb_pkts;
2691 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2692 	/* number of packets that were not sent */
2693 	if (unlikely(num > nb_pkts)) {
2694 		for (i = nb_pkts; i < num; i++) {
2695 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2696 				MRVL_PP2_TX_SHADOWQ_MASK;
2697 			addr = sq->ent[sq->head].buff.cookie;
2698 			bytes_sent -=
2699 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2700 		}
2701 		sq->size -= num - nb_pkts;
2702 	}
2703 
2704 	q->bytes_sent += bytes_sent;
2705 
2706 	return nb_pkts;
2707 }
2708 
2709 /** DPDK callback for S/G transmit.
2710  *
2711  * @param txq
2712  *   Generic pointer transmit queue.
2713  * @param tx_pkts
2714  *   Packets to transmit.
2715  * @param nb_pkts
2716  *   Number of packets in array.
2717  *
2718  * @return
2719  *   Number of packets successfully transmitted.
2720  */
2721 static uint16_t
2722 mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
2723 		     uint16_t nb_pkts)
2724 {
2725 	struct mrvl_txq *q = txq;
2726 	struct mrvl_shadow_txq *sq;
2727 	struct pp2_hif *hif;
2728 	struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
2729 	struct pp2_ppio_sg_pkts pkts;
2730 	uint8_t frags[nb_pkts];
2731 	unsigned int core_id = rte_lcore_id();
2732 	int i, j, ret, bytes_sent = 0;
2733 	int tail, tail_first;
2734 	uint16_t num, sq_free_size;
2735 	uint16_t nb_segs, total_descs = 0;
2736 	uint64_t addr;
2737 
2738 	hif = mrvl_get_hif(q->priv, core_id);
2739 	sq = &q->shadow_txqs[core_id];
2740 	pkts.frags = frags;
2741 	pkts.num = 0;
2742 
2743 	if (unlikely(!q->priv->ppio || !hif))
2744 		return 0;
2745 
2746 	if (sq->size)
2747 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2748 				       sq, q->queue_id, 0);
2749 
2750 	/* Save shadow queue free size */
2751 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2752 
2753 	tail = 0;
2754 	for (i = 0; i < nb_pkts; i++) {
2755 		struct rte_mbuf *mbuf = tx_pkts[i];
2756 		struct rte_mbuf *seg = NULL;
2757 		int gen_l3_cksum, gen_l4_cksum;
2758 		enum pp2_outq_l3_type l3_type;
2759 		enum pp2_outq_l4_type l4_type;
2760 
2761 		nb_segs = mbuf->nb_segs;
2762 		tail_first = tail;
2763 		total_descs += nb_segs;
2764 
2765 		/*
2766 		 * Check if total_descs does not exceed
2767 		 * shadow queue free size
2768 		 */
2769 		if (unlikely(total_descs > sq_free_size)) {
2770 			total_descs -= nb_segs;
2771 			break;
2772 		}
2773 
2774 		/* Check if nb_segs does not exceed the max nb of desc per
2775 		 * fragmented packet
2776 		 */
2777 		if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
2778 			total_descs -= nb_segs;
2779 			RTE_LOG(ERR, PMD,
2780 				"Too many segments. Packet won't be sent.\n");
2781 			break;
2782 		}
2783 
2784 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2785 			struct rte_mbuf *pref_pkt_hdr;
2786 
2787 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2788 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2789 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2790 		}
2791 
2792 		pkts.frags[pkts.num] = nb_segs;
2793 		pkts.num++;
2794 
2795 		seg = mbuf;
2796 		for (j = 0; j < nb_segs - 1; j++) {
2797 			/* For the subsequent segments, set shadow queue
2798 			 * buffer to NULL
2799 			 */
2800 			mrvl_fill_shadowq(sq, NULL);
2801 			mrvl_fill_desc(&descs[tail], seg);
2802 
2803 			tail++;
2804 			seg = seg->next;
2805 		}
2806 		/* Put first mbuf info in last shadow queue entry */
2807 		mrvl_fill_shadowq(sq, mbuf);
2808 		/* Update descriptor with last segment */
2809 		mrvl_fill_desc(&descs[tail++], seg);
2810 
2811 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2812 		/* In case unsupported ol_flags were passed
2813 		 * do not update descriptor offload information
2814 		 */
2815 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2816 					      &l3_type, &l4_type, &gen_l3_cksum,
2817 					      &gen_l4_cksum);
2818 		if (unlikely(ret))
2819 			continue;
2820 
2821 		pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
2822 						  l4_type, mbuf->l2_len,
2823 						  mbuf->l2_len + mbuf->l3_len,
2824 						  gen_l3_cksum, gen_l4_cksum);
2825 	}
2826 
2827 	num = total_descs;
2828 	pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
2829 			 &total_descs, &pkts);
2830 	/* number of packets that were not sent */
2831 	if (unlikely(num > total_descs)) {
2832 		for (i = total_descs; i < num; i++) {
2833 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2834 				MRVL_PP2_TX_SHADOWQ_MASK;
2835 
2836 			addr = sq->ent[sq->head].buff.cookie;
2837 			if (addr)
2838 				bytes_sent -=
2839 					rte_pktmbuf_pkt_len((struct rte_mbuf *)
2840 						(cookie_addr_high | addr));
2841 		}
2842 		sq->size -= num - total_descs;
2843 		nb_pkts = pkts.num;
2844 	}
2845 
2846 	q->bytes_sent += bytes_sent;
2847 
2848 	return nb_pkts;
2849 }
2850 
2851 /**
2852  * Initialize packet processor.
2853  *
2854  * @return
2855  *   0 on success, negative error value otherwise.
2856  */
2857 static int
2858 mrvl_init_pp2(void)
2859 {
2860 	struct pp2_init_params init_params;
2861 
2862 	memset(&init_params, 0, sizeof(init_params));
2863 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
2864 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
2865 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
2866 
2867 	return pp2_init(&init_params);
2868 }
2869 
2870 /**
2871  * Deinitialize packet processor.
2872  *
2873  * @return
2874  *   0 on success, negative error value otherwise.
2875  */
2876 static void
2877 mrvl_deinit_pp2(void)
2878 {
2879 	pp2_deinit();
2880 }
2881 
2882 /**
2883  * Create private device structure.
2884  *
2885  * @param dev_name
2886  *   Pointer to the port name passed in the initialization parameters.
2887  *
2888  * @return
2889  *   Pointer to the newly allocated private device structure.
2890  */
2891 static struct mrvl_priv *
2892 mrvl_priv_create(const char *dev_name)
2893 {
2894 	struct pp2_bpool_params bpool_params;
2895 	char match[MRVL_MATCH_LEN];
2896 	struct mrvl_priv *priv;
2897 	int ret, bpool_bit;
2898 
2899 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
2900 	if (!priv)
2901 		return NULL;
2902 
2903 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
2904 				       &priv->pp_id, &priv->ppio_id);
2905 	if (ret)
2906 		goto out_free_priv;
2907 
2908 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
2909 				     PP2_BPOOL_NUM_POOLS);
2910 	if (bpool_bit < 0)
2911 		goto out_free_priv;
2912 	priv->bpool_bit = bpool_bit;
2913 
2914 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
2915 		 priv->bpool_bit);
2916 	memset(&bpool_params, 0, sizeof(bpool_params));
2917 	bpool_params.match = match;
2918 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
2919 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
2920 	if (ret)
2921 		goto out_clear_bpool_bit;
2922 
2923 	priv->ppio_params.type = PP2_PPIO_T_NIC;
2924 	rte_spinlock_init(&priv->lock);
2925 
2926 	return priv;
2927 out_clear_bpool_bit:
2928 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2929 out_free_priv:
2930 	rte_free(priv);
2931 	return NULL;
2932 }
2933 
2934 /**
2935  * Create device representing Ethernet port.
2936  *
2937  * @param name
2938  *   Pointer to the port's name.
2939  *
2940  * @return
2941  *   0 on success, negative error value otherwise.
2942  */
2943 static int
2944 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
2945 {
2946 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
2947 	struct rte_eth_dev *eth_dev;
2948 	struct mrvl_priv *priv;
2949 	struct ifreq req;
2950 
2951 	eth_dev = rte_eth_dev_allocate(name);
2952 	if (!eth_dev)
2953 		return -ENOMEM;
2954 
2955 	priv = mrvl_priv_create(name);
2956 	if (!priv) {
2957 		ret = -ENOMEM;
2958 		goto out_free;
2959 	}
2960 	eth_dev->data->dev_private = priv;
2961 
2962 	eth_dev->data->mac_addrs =
2963 		rte_zmalloc("mac_addrs",
2964 			    RTE_ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
2965 	if (!eth_dev->data->mac_addrs) {
2966 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
2967 		ret = -ENOMEM;
2968 		goto out_free;
2969 	}
2970 
2971 	memset(&req, 0, sizeof(req));
2972 	strcpy(req.ifr_name, name);
2973 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
2974 	if (ret)
2975 		goto out_free;
2976 
2977 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
2978 	       req.ifr_addr.sa_data, RTE_ETHER_ADDR_LEN);
2979 
2980 	eth_dev->device = &vdev->device;
2981 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
2982 	mrvl_set_tx_function(eth_dev);
2983 	eth_dev->dev_ops = &mrvl_ops;
2984 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2985 
2986 	eth_dev->data->dev_link.link_status = ETH_LINK_UP;
2987 
2988 	rte_eth_dev_probing_finish(eth_dev);
2989 	return 0;
2990 out_free:
2991 	rte_eth_dev_release_port(eth_dev);
2992 
2993 	return ret;
2994 }
2995 
2996 /**
2997  * Callback used by rte_kvargs_process() during argument parsing.
2998  *
2999  * @param key
3000  *   Pointer to the parsed key (unused).
3001  * @param value
3002  *   Pointer to the parsed value.
3003  * @param extra_args
3004  *   Pointer to the extra arguments which contains address of the
3005  *   table of pointers to parsed interface names.
3006  *
3007  * @return
3008  *   Always 0.
3009  */
3010 static int
3011 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
3012 		 void *extra_args)
3013 {
3014 	struct mrvl_ifnames *ifnames = extra_args;
3015 
3016 	ifnames->names[ifnames->idx++] = value;
3017 
3018 	return 0;
3019 }
3020 
3021 /**
3022  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
3023  */
3024 static void
3025 mrvl_deinit_hifs(void)
3026 {
3027 	int i;
3028 
3029 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
3030 		if (hifs[i])
3031 			pp2_hif_deinit(hifs[i]);
3032 	}
3033 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
3034 	memset(hifs, 0, sizeof(hifs));
3035 }
3036 
3037 /**
3038  * DPDK callback to register the virtual device.
3039  *
3040  * @param vdev
3041  *   Pointer to the virtual device.
3042  *
3043  * @return
3044  *   0 on success, negative error value otherwise.
3045  */
3046 static int
3047 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
3048 {
3049 	struct rte_kvargs *kvlist;
3050 	struct mrvl_ifnames ifnames;
3051 	int ret = -EINVAL;
3052 	uint32_t i, ifnum, cfgnum;
3053 	const char *params;
3054 
3055 	params = rte_vdev_device_args(vdev);
3056 	if (!params)
3057 		return -EINVAL;
3058 
3059 	kvlist = rte_kvargs_parse(params, valid_args);
3060 	if (!kvlist)
3061 		return -EINVAL;
3062 
3063 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
3064 	if (ifnum > RTE_DIM(ifnames.names))
3065 		goto out_free_kvlist;
3066 
3067 	ifnames.idx = 0;
3068 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
3069 			   mrvl_get_ifnames, &ifnames);
3070 
3071 
3072 	/*
3073 	 * The below system initialization should be done only once,
3074 	 * on the first provided configuration file
3075 	 */
3076 	if (!mrvl_qos_cfg) {
3077 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
3078 		MRVL_LOG(INFO, "Parsing config file!");
3079 		if (cfgnum > 1) {
3080 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
3081 			goto out_free_kvlist;
3082 		} else if (cfgnum == 1) {
3083 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
3084 					   mrvl_get_qoscfg, &mrvl_qos_cfg);
3085 		}
3086 	}
3087 
3088 	if (mrvl_dev_num)
3089 		goto init_devices;
3090 
3091 	MRVL_LOG(INFO, "Perform MUSDK initializations");
3092 
3093 	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
3094 	if (ret)
3095 		goto out_free_kvlist;
3096 
3097 	ret = mrvl_init_pp2();
3098 	if (ret) {
3099 		MRVL_LOG(ERR, "Failed to init PP!");
3100 		rte_mvep_deinit(MVEP_MOD_T_PP2);
3101 		goto out_free_kvlist;
3102 	}
3103 
3104 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
3105 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
3106 
3107 	mrvl_lcore_first = RTE_MAX_LCORE;
3108 	mrvl_lcore_last = 0;
3109 
3110 init_devices:
3111 	for (i = 0; i < ifnum; i++) {
3112 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
3113 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
3114 		if (ret)
3115 			goto out_cleanup;
3116 		mrvl_dev_num++;
3117 	}
3118 
3119 	rte_kvargs_free(kvlist);
3120 
3121 	return 0;
3122 out_cleanup:
3123 	rte_pmd_mrvl_remove(vdev);
3124 
3125 out_free_kvlist:
3126 	rte_kvargs_free(kvlist);
3127 
3128 	return ret;
3129 }
3130 
3131 /**
3132  * DPDK callback to remove virtual device.
3133  *
3134  * @param vdev
3135  *   Pointer to the removed virtual device.
3136  *
3137  * @return
3138  *   0 on success, negative error value otherwise.
3139  */
3140 static int
3141 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
3142 {
3143 	uint16_t port_id;
3144 	int ret = 0;
3145 
3146 	RTE_ETH_FOREACH_DEV(port_id) {
3147 		if (rte_eth_devices[port_id].device != &vdev->device)
3148 			continue;
3149 		ret |= rte_eth_dev_close(port_id);
3150 	}
3151 
3152 	return ret == 0 ? 0 : -EIO;
3153 }
3154 
3155 static struct rte_vdev_driver pmd_mrvl_drv = {
3156 	.probe = rte_pmd_mrvl_probe,
3157 	.remove = rte_pmd_mrvl_remove,
3158 };
3159 
3160 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
3161 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
3162 RTE_LOG_REGISTER(mrvl_logtype, pmd.net.mvpp2, NOTICE);
3163