xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision 10b71caecbe1cddcbb65c050ca775fba575e88db)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2017 Marvell International Ltd.
3  * Copyright(c) 2017 Semihalf.
4  * All rights reserved.
5  */
6 
7 #include <rte_string_fns.h>
8 #include <rte_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 0x01
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 
150 #define MRVL_XSTATS_TBL_ENTRY(name) { \
151 	#name, offsetof(struct pp2_ppio_statistics, name),	\
152 	sizeof(((struct pp2_ppio_statistics *)0)->name)		\
153 }
154 
155 /* Table with xstats data */
156 static struct {
157 	const char *name;
158 	unsigned int offset;
159 	unsigned int size;
160 } mrvl_xstats_tbl[] = {
161 	MRVL_XSTATS_TBL_ENTRY(rx_bytes),
162 	MRVL_XSTATS_TBL_ENTRY(rx_packets),
163 	MRVL_XSTATS_TBL_ENTRY(rx_unicast_packets),
164 	MRVL_XSTATS_TBL_ENTRY(rx_errors),
165 	MRVL_XSTATS_TBL_ENTRY(rx_fullq_dropped),
166 	MRVL_XSTATS_TBL_ENTRY(rx_bm_dropped),
167 	MRVL_XSTATS_TBL_ENTRY(rx_early_dropped),
168 	MRVL_XSTATS_TBL_ENTRY(rx_fifo_dropped),
169 	MRVL_XSTATS_TBL_ENTRY(rx_cls_dropped),
170 	MRVL_XSTATS_TBL_ENTRY(tx_bytes),
171 	MRVL_XSTATS_TBL_ENTRY(tx_packets),
172 	MRVL_XSTATS_TBL_ENTRY(tx_unicast_packets),
173 	MRVL_XSTATS_TBL_ENTRY(tx_errors)
174 };
175 
176 static inline void
177 mrvl_fill_shadowq(struct mrvl_shadow_txq *sq, struct rte_mbuf *buf)
178 {
179 	sq->ent[sq->head].buff.cookie = (uint64_t)buf;
180 	sq->ent[sq->head].buff.addr = buf ?
181 		rte_mbuf_data_iova_default(buf) : 0;
182 
183 	sq->ent[sq->head].bpool =
184 		(unlikely(!buf || buf->port >= RTE_MAX_ETHPORTS ||
185 		 buf->refcnt > 1)) ? NULL :
186 		 mrvl_port_to_bpool_lookup[buf->port];
187 
188 	sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
189 	sq->size++;
190 }
191 
192 static inline void
193 mrvl_fill_desc(struct pp2_ppio_desc *desc, struct rte_mbuf *buf)
194 {
195 	pp2_ppio_outq_desc_reset(desc);
196 	pp2_ppio_outq_desc_set_phys_addr(desc, rte_pktmbuf_iova(buf));
197 	pp2_ppio_outq_desc_set_pkt_offset(desc, 0);
198 	pp2_ppio_outq_desc_set_pkt_len(desc, rte_pktmbuf_data_len(buf));
199 }
200 
201 static inline int
202 mrvl_get_bpool_size(int pp2_id, int pool_id)
203 {
204 	int i;
205 	int size = 0;
206 
207 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
208 		size += mrvl_port_bpool_size[pp2_id][pool_id][i];
209 
210 	return size;
211 }
212 
213 static inline int
214 mrvl_reserve_bit(int *bitmap, int max)
215 {
216 	int n = sizeof(*bitmap) * 8 - __builtin_clz(*bitmap);
217 
218 	if (n >= max)
219 		return -1;
220 
221 	*bitmap |= 1 << n;
222 
223 	return n;
224 }
225 
226 static int
227 mrvl_init_hif(int core_id)
228 {
229 	struct pp2_hif_params params;
230 	char match[MRVL_MATCH_LEN];
231 	int ret;
232 
233 	ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
234 	if (ret < 0) {
235 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
236 		return ret;
237 	}
238 
239 	snprintf(match, sizeof(match), "hif-%d", ret);
240 	memset(&params, 0, sizeof(params));
241 	params.match = match;
242 	params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
243 	ret = pp2_hif_init(&params, &hifs[core_id]);
244 	if (ret) {
245 		MRVL_LOG(ERR, "Failed to initialize hif %d", core_id);
246 		return ret;
247 	}
248 
249 	return 0;
250 }
251 
252 static inline struct pp2_hif*
253 mrvl_get_hif(struct mrvl_priv *priv, int core_id)
254 {
255 	int ret;
256 
257 	if (likely(hifs[core_id] != NULL))
258 		return hifs[core_id];
259 
260 	rte_spinlock_lock(&priv->lock);
261 
262 	ret = mrvl_init_hif(core_id);
263 	if (ret < 0) {
264 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
265 		goto out;
266 	}
267 
268 	if (core_id < mrvl_lcore_first)
269 		mrvl_lcore_first = core_id;
270 
271 	if (core_id > mrvl_lcore_last)
272 		mrvl_lcore_last = core_id;
273 out:
274 	rte_spinlock_unlock(&priv->lock);
275 
276 	return hifs[core_id];
277 }
278 
279 /**
280  * Set tx burst function according to offload flag
281  *
282  * @param dev
283  *   Pointer to Ethernet device structure.
284  */
285 static void
286 mrvl_set_tx_function(struct rte_eth_dev *dev)
287 {
288 	struct mrvl_priv *priv = dev->data->dev_private;
289 
290 	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
291 	if (priv->multiseg) {
292 		RTE_LOG(INFO, PMD, "Using multi-segment tx callback\n");
293 		dev->tx_pkt_burst = mrvl_tx_sg_pkt_burst;
294 	} else {
295 		RTE_LOG(INFO, PMD, "Using single-segment tx callback\n");
296 		dev->tx_pkt_burst = mrvl_tx_pkt_burst;
297 	}
298 }
299 
300 /**
301  * Configure rss based on dpdk rss configuration.
302  *
303  * @param priv
304  *   Pointer to private structure.
305  * @param rss_conf
306  *   Pointer to RSS configuration.
307  *
308  * @return
309  *   0 on success, negative error value otherwise.
310  */
311 static int
312 mrvl_configure_rss(struct mrvl_priv *priv, struct rte_eth_rss_conf *rss_conf)
313 {
314 	if (rss_conf->rss_key)
315 		MRVL_LOG(WARNING, "Changing hash key is not supported");
316 
317 	if (rss_conf->rss_hf == 0) {
318 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
319 	} else if (rss_conf->rss_hf & ETH_RSS_IPV4) {
320 		priv->ppio_params.inqs_params.hash_type =
321 			PP2_PPIO_HASH_T_2_TUPLE;
322 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP) {
323 		priv->ppio_params.inqs_params.hash_type =
324 			PP2_PPIO_HASH_T_5_TUPLE;
325 		priv->rss_hf_tcp = 1;
326 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) {
327 		priv->ppio_params.inqs_params.hash_type =
328 			PP2_PPIO_HASH_T_5_TUPLE;
329 		priv->rss_hf_tcp = 0;
330 	} else {
331 		return -EINVAL;
332 	}
333 
334 	return 0;
335 }
336 
337 /**
338  * Ethernet device configuration.
339  *
340  * Prepare the driver for a given number of TX and RX queues and
341  * configure RSS.
342  *
343  * @param dev
344  *   Pointer to Ethernet device structure.
345  *
346  * @return
347  *   0 on success, negative error value otherwise.
348  */
349 static int
350 mrvl_dev_configure(struct rte_eth_dev *dev)
351 {
352 	struct mrvl_priv *priv = dev->data->dev_private;
353 	int ret;
354 
355 	if (priv->ppio) {
356 		MRVL_LOG(INFO, "Device reconfiguration is not supported");
357 		return -EINVAL;
358 	}
359 
360 	if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_NONE &&
361 	    dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
362 		MRVL_LOG(INFO, "Unsupported rx multi queue mode %d",
363 			dev->data->dev_conf.rxmode.mq_mode);
364 		return -EINVAL;
365 	}
366 
367 	if (dev->data->dev_conf.rxmode.split_hdr_size) {
368 		MRVL_LOG(INFO, "Split headers not supported");
369 		return -EINVAL;
370 	}
371 
372 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
373 		dev->data->mtu = dev->data->dev_conf.rxmode.max_rx_pkt_len -
374 				 MRVL_PP2_ETH_HDRS_LEN;
375 
376 	if (dev->data->dev_conf.txmode.offloads & DEV_TX_OFFLOAD_MULTI_SEGS)
377 		priv->multiseg = 1;
378 
379 	ret = mrvl_configure_rxqs(priv, dev->data->port_id,
380 				  dev->data->nb_rx_queues);
381 	if (ret < 0)
382 		return ret;
383 
384 	ret = mrvl_configure_txqs(priv, dev->data->port_id,
385 				  dev->data->nb_tx_queues);
386 	if (ret < 0)
387 		return ret;
388 
389 	priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
390 	priv->ppio_params.maintain_stats = 1;
391 	priv->nb_rx_queues = dev->data->nb_rx_queues;
392 
393 	ret = mrvl_tm_init(dev);
394 	if (ret < 0)
395 		return ret;
396 
397 	if (dev->data->nb_rx_queues == 1 &&
398 	    dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
399 		MRVL_LOG(WARNING, "Disabling hash for 1 rx queue");
400 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
401 
402 		return 0;
403 	}
404 
405 	return mrvl_configure_rss(priv,
406 				  &dev->data->dev_conf.rx_adv_conf.rss_conf);
407 }
408 
409 /**
410  * DPDK callback to change the MTU.
411  *
412  * Setting the MTU affects hardware MRU (packets larger than the MRU
413  * will be dropped).
414  *
415  * @param dev
416  *   Pointer to Ethernet device structure.
417  * @param mtu
418  *   New MTU.
419  *
420  * @return
421  *   0 on success, negative error value otherwise.
422  */
423 static int
424 mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
425 {
426 	struct mrvl_priv *priv = dev->data->dev_private;
427 	uint16_t mru;
428 	uint16_t mbuf_data_size = 0; /* SW buffer size */
429 	int ret;
430 
431 	mru = MRVL_PP2_MTU_TO_MRU(mtu);
432 	/*
433 	 * min_rx_buf_size is equal to mbuf data size
434 	 * if pmd didn't set it differently
435 	 */
436 	mbuf_data_size = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
437 	/* Prevent PMD from:
438 	 * - setting mru greater than the mbuf size resulting in
439 	 * hw and sw buffer size mismatch
440 	 * - setting mtu that requires the support of scattered packets
441 	 * when this feature has not been enabled/supported so far
442 	 * (TODO check scattered_rx flag here once scattered RX is supported).
443 	 */
444 	if (mru + MRVL_PKT_OFFS > mbuf_data_size) {
445 		mru = mbuf_data_size - MRVL_PKT_OFFS;
446 		mtu = MRVL_PP2_MRU_TO_MTU(mru);
447 		MRVL_LOG(WARNING, "MTU too big, max MTU possible limitted "
448 			"by current mbuf size: %u. Set MTU to %u, MRU to %u",
449 			mbuf_data_size, mtu, mru);
450 	}
451 
452 	if (mtu < RTE_ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX) {
453 		MRVL_LOG(ERR, "Invalid MTU [%u] or MRU [%u]", mtu, mru);
454 		return -EINVAL;
455 	}
456 
457 	dev->data->mtu = mtu;
458 	dev->data->dev_conf.rxmode.max_rx_pkt_len = mru - MV_MH_SIZE;
459 
460 	if (!priv->ppio)
461 		return 0;
462 
463 	ret = pp2_ppio_set_mru(priv->ppio, mru);
464 	if (ret) {
465 		MRVL_LOG(ERR, "Failed to change MRU");
466 		return ret;
467 	}
468 
469 	ret = pp2_ppio_set_mtu(priv->ppio, mtu);
470 	if (ret) {
471 		MRVL_LOG(ERR, "Failed to change MTU");
472 		return ret;
473 	}
474 
475 	return 0;
476 }
477 
478 /**
479  * DPDK callback to bring the link up.
480  *
481  * @param dev
482  *   Pointer to Ethernet device structure.
483  *
484  * @return
485  *   0 on success, negative error value otherwise.
486  */
487 static int
488 mrvl_dev_set_link_up(struct rte_eth_dev *dev)
489 {
490 	struct mrvl_priv *priv = dev->data->dev_private;
491 	int ret;
492 
493 	if (!priv->ppio)
494 		return -EPERM;
495 
496 	ret = pp2_ppio_enable(priv->ppio);
497 	if (ret)
498 		return ret;
499 
500 	/*
501 	 * mtu/mru can be updated if pp2_ppio_enable() was called at least once
502 	 * as pp2_ppio_enable() changes port->t_mode from default 0 to
503 	 * PP2_TRAFFIC_INGRESS_EGRESS.
504 	 *
505 	 * Set mtu to default DPDK value here.
506 	 */
507 	ret = mrvl_mtu_set(dev, dev->data->mtu);
508 	if (ret)
509 		pp2_ppio_disable(priv->ppio);
510 
511 	return ret;
512 }
513 
514 /**
515  * DPDK callback to bring the link down.
516  *
517  * @param dev
518  *   Pointer to Ethernet device structure.
519  *
520  * @return
521  *   0 on success, negative error value otherwise.
522  */
523 static int
524 mrvl_dev_set_link_down(struct rte_eth_dev *dev)
525 {
526 	struct mrvl_priv *priv = dev->data->dev_private;
527 
528 	if (!priv->ppio)
529 		return -EPERM;
530 
531 	return pp2_ppio_disable(priv->ppio);
532 }
533 
534 /**
535  * DPDK callback to start tx queue.
536  *
537  * @param dev
538  *   Pointer to Ethernet device structure.
539  * @param queue_id
540  *   Transmit queue index.
541  *
542  * @return
543  *   0 on success, negative error value otherwise.
544  */
545 static int
546 mrvl_tx_queue_start(struct rte_eth_dev *dev, uint16_t queue_id)
547 {
548 	struct mrvl_priv *priv = dev->data->dev_private;
549 	int ret;
550 
551 	if (!priv)
552 		return -EPERM;
553 
554 	/* passing 1 enables given tx queue */
555 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 1);
556 	if (ret) {
557 		MRVL_LOG(ERR, "Failed to start txq %d", queue_id);
558 		return ret;
559 	}
560 
561 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
562 
563 	return 0;
564 }
565 
566 /**
567  * DPDK callback to stop tx queue.
568  *
569  * @param dev
570  *   Pointer to Ethernet device structure.
571  * @param queue_id
572  *   Transmit queue index.
573  *
574  * @return
575  *   0 on success, negative error value otherwise.
576  */
577 static int
578 mrvl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t queue_id)
579 {
580 	struct mrvl_priv *priv = dev->data->dev_private;
581 	int ret;
582 
583 	if (!priv->ppio)
584 		return -EPERM;
585 
586 	/* passing 0 disables given tx queue */
587 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 0);
588 	if (ret) {
589 		MRVL_LOG(ERR, "Failed to stop txq %d", queue_id);
590 		return ret;
591 	}
592 
593 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
594 
595 	return 0;
596 }
597 
598 /**
599  * DPDK callback to start the device.
600  *
601  * @param dev
602  *   Pointer to Ethernet device structure.
603  *
604  * @return
605  *   0 on success, negative errno value on failure.
606  */
607 static int
608 mrvl_dev_start(struct rte_eth_dev *dev)
609 {
610 	struct mrvl_priv *priv = dev->data->dev_private;
611 	char match[MRVL_MATCH_LEN];
612 	int ret = 0, i, def_init_size;
613 
614 	if (priv->ppio)
615 		return mrvl_dev_set_link_up(dev);
616 
617 	snprintf(match, sizeof(match), "ppio-%d:%d",
618 		 priv->pp_id, priv->ppio_id);
619 	priv->ppio_params.match = match;
620 
621 	/*
622 	 * Calculate the minimum bpool size for refill feature as follows:
623 	 * 2 default burst sizes multiply by number of rx queues.
624 	 * If the bpool size will be below this value, new buffers will
625 	 * be added to the pool.
626 	 */
627 	priv->bpool_min_size = priv->nb_rx_queues * MRVL_BURST_SIZE * 2;
628 
629 	/* In case initial bpool size configured in queues setup is
630 	 * smaller than minimum size add more buffers
631 	 */
632 	def_init_size = priv->bpool_min_size + MRVL_BURST_SIZE * 2;
633 	if (priv->bpool_init_size < def_init_size) {
634 		int buffs_to_add = def_init_size - priv->bpool_init_size;
635 
636 		priv->bpool_init_size += buffs_to_add;
637 		ret = mrvl_fill_bpool(dev->data->rx_queues[0], buffs_to_add);
638 		if (ret)
639 			MRVL_LOG(ERR, "Failed to add buffers to bpool");
640 	}
641 
642 	/*
643 	 * Calculate the maximum bpool size for refill feature as follows:
644 	 * maximum number of descriptors in rx queue multiply by number
645 	 * of rx queues plus minimum bpool size.
646 	 * In case the bpool size will exceed this value, superfluous buffers
647 	 * will be removed
648 	 */
649 	priv->bpool_max_size = (priv->nb_rx_queues * MRVL_PP2_RXD_MAX) +
650 				priv->bpool_min_size;
651 
652 	ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
653 	if (ret) {
654 		MRVL_LOG(ERR, "Failed to init ppio");
655 		return ret;
656 	}
657 
658 	/*
659 	 * In case there are some some stale uc/mc mac addresses flush them
660 	 * here. It cannot be done during mrvl_dev_close() as port information
661 	 * is already gone at that point (due to pp2_ppio_deinit() in
662 	 * mrvl_dev_stop()).
663 	 */
664 	if (!priv->uc_mc_flushed) {
665 		ret = pp2_ppio_flush_mac_addrs(priv->ppio, 1, 1);
666 		if (ret) {
667 			MRVL_LOG(ERR,
668 				"Failed to flush uc/mc filter list");
669 			goto out;
670 		}
671 		priv->uc_mc_flushed = 1;
672 	}
673 
674 	if (!priv->vlan_flushed) {
675 		ret = pp2_ppio_flush_vlan(priv->ppio);
676 		if (ret) {
677 			MRVL_LOG(ERR, "Failed to flush vlan list");
678 			/*
679 			 * TODO
680 			 * once pp2_ppio_flush_vlan() is supported jump to out
681 			 * goto out;
682 			 */
683 		}
684 		priv->vlan_flushed = 1;
685 	}
686 	ret = mrvl_mtu_set(dev, dev->data->mtu);
687 	if (ret)
688 		MRVL_LOG(ERR, "Failed to set MTU to %d", dev->data->mtu);
689 
690 	/* For default QoS config, don't start classifier. */
691 	if (mrvl_qos_cfg  &&
692 	    mrvl_qos_cfg->port[dev->data->port_id].use_global_defaults == 0) {
693 		ret = mrvl_start_qos_mapping(priv);
694 		if (ret) {
695 			MRVL_LOG(ERR, "Failed to setup QoS mapping");
696 			goto out;
697 		}
698 	}
699 
700 	ret = mrvl_dev_set_link_up(dev);
701 	if (ret) {
702 		MRVL_LOG(ERR, "Failed to set link up");
703 		goto out;
704 	}
705 
706 	/* start tx queues */
707 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
708 		struct mrvl_txq *txq = dev->data->tx_queues[i];
709 
710 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
711 
712 		if (!txq->tx_deferred_start)
713 			continue;
714 
715 		/*
716 		 * All txqs are started by default. Stop them
717 		 * so that tx_deferred_start works as expected.
718 		 */
719 		ret = mrvl_tx_queue_stop(dev, i);
720 		if (ret)
721 			goto out;
722 	}
723 
724 	mrvl_flow_init(dev);
725 	mrvl_mtr_init(dev);
726 	mrvl_set_tx_function(dev);
727 
728 	return 0;
729 out:
730 	MRVL_LOG(ERR, "Failed to start device");
731 	pp2_ppio_deinit(priv->ppio);
732 	return ret;
733 }
734 
735 /**
736  * Flush receive queues.
737  *
738  * @param dev
739  *   Pointer to Ethernet device structure.
740  */
741 static void
742 mrvl_flush_rx_queues(struct rte_eth_dev *dev)
743 {
744 	int i;
745 
746 	MRVL_LOG(INFO, "Flushing rx queues");
747 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
748 		int ret, num;
749 
750 		do {
751 			struct mrvl_rxq *q = dev->data->rx_queues[i];
752 			struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];
753 
754 			num = MRVL_PP2_RXD_MAX;
755 			ret = pp2_ppio_recv(q->priv->ppio,
756 					    q->priv->rxq_map[q->queue_id].tc,
757 					    q->priv->rxq_map[q->queue_id].inq,
758 					    descs, (uint16_t *)&num);
759 		} while (ret == 0 && num);
760 	}
761 }
762 
763 /**
764  * Flush transmit shadow queues.
765  *
766  * @param dev
767  *   Pointer to Ethernet device structure.
768  */
769 static void
770 mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
771 {
772 	int i, j;
773 	struct mrvl_txq *txq;
774 
775 	MRVL_LOG(INFO, "Flushing tx shadow queues");
776 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
777 		txq = (struct mrvl_txq *)dev->data->tx_queues[i];
778 
779 		for (j = 0; j < RTE_MAX_LCORE; j++) {
780 			struct mrvl_shadow_txq *sq;
781 
782 			if (!hifs[j])
783 				continue;
784 
785 			sq = &txq->shadow_txqs[j];
786 			mrvl_free_sent_buffers(txq->priv->ppio,
787 				hifs[j], j, sq, txq->queue_id, 1);
788 			while (sq->tail != sq->head) {
789 				uint64_t addr = cookie_addr_high |
790 					sq->ent[sq->tail].buff.cookie;
791 				rte_pktmbuf_free(
792 					(struct rte_mbuf *)addr);
793 				sq->tail = (sq->tail + 1) &
794 					    MRVL_PP2_TX_SHADOWQ_MASK;
795 			}
796 			memset(sq, 0, sizeof(*sq));
797 		}
798 	}
799 }
800 
801 /**
802  * Flush hardware bpool (buffer-pool).
803  *
804  * @param dev
805  *   Pointer to Ethernet device structure.
806  */
807 static void
808 mrvl_flush_bpool(struct rte_eth_dev *dev)
809 {
810 	struct mrvl_priv *priv = dev->data->dev_private;
811 	struct pp2_hif *hif;
812 	uint32_t num;
813 	int ret;
814 	unsigned int core_id = rte_lcore_id();
815 
816 	if (core_id == LCORE_ID_ANY)
817 		core_id = rte_get_master_lcore();
818 
819 	hif = mrvl_get_hif(priv, core_id);
820 
821 	ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
822 	if (ret) {
823 		MRVL_LOG(ERR, "Failed to get bpool buffers number");
824 		return;
825 	}
826 
827 	while (num--) {
828 		struct pp2_buff_inf inf;
829 		uint64_t addr;
830 
831 		ret = pp2_bpool_get_buff(hif, priv->bpool, &inf);
832 		if (ret)
833 			break;
834 
835 		addr = cookie_addr_high | inf.cookie;
836 		rte_pktmbuf_free((struct rte_mbuf *)addr);
837 	}
838 }
839 
840 /**
841  * DPDK callback to stop the device.
842  *
843  * @param dev
844  *   Pointer to Ethernet device structure.
845  */
846 static void
847 mrvl_dev_stop(struct rte_eth_dev *dev)
848 {
849 	mrvl_dev_set_link_down(dev);
850 }
851 
852 /**
853  * DPDK callback to close the device.
854  *
855  * @param dev
856  *   Pointer to Ethernet device structure.
857  */
858 static void
859 mrvl_dev_close(struct rte_eth_dev *dev)
860 {
861 	struct mrvl_priv *priv = dev->data->dev_private;
862 	size_t i;
863 
864 	mrvl_flush_rx_queues(dev);
865 	mrvl_flush_tx_shadow_queues(dev);
866 	mrvl_flow_deinit(dev);
867 	mrvl_mtr_deinit(dev);
868 
869 	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
870 		struct pp2_ppio_tc_params *tc_params =
871 			&priv->ppio_params.inqs_params.tcs_params[i];
872 
873 		if (tc_params->inqs_params) {
874 			rte_free(tc_params->inqs_params);
875 			tc_params->inqs_params = NULL;
876 		}
877 	}
878 
879 	if (priv->cls_tbl) {
880 		pp2_cls_tbl_deinit(priv->cls_tbl);
881 		priv->cls_tbl = NULL;
882 	}
883 
884 	if (priv->qos_tbl) {
885 		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
886 		priv->qos_tbl = NULL;
887 	}
888 
889 	mrvl_flush_bpool(dev);
890 	mrvl_tm_deinit(dev);
891 
892 	if (priv->ppio) {
893 		pp2_ppio_deinit(priv->ppio);
894 		priv->ppio = NULL;
895 	}
896 
897 	/* policer must be released after ppio deinitialization */
898 	if (priv->default_policer) {
899 		pp2_cls_plcr_deinit(priv->default_policer);
900 		priv->default_policer = NULL;
901 	}
902 
903 
904 	if (priv->bpool) {
905 		pp2_bpool_deinit(priv->bpool);
906 		used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
907 		priv->bpool = NULL;
908 	}
909 
910 	mrvl_dev_num--;
911 
912 	if (mrvl_dev_num == 0) {
913 		MRVL_LOG(INFO, "Perform MUSDK deinit");
914 		mrvl_deinit_hifs();
915 		mrvl_deinit_pp2();
916 		rte_mvep_deinit(MVEP_MOD_T_PP2);
917 	}
918 }
919 
920 /**
921  * DPDK callback to retrieve physical link information.
922  *
923  * @param dev
924  *   Pointer to Ethernet device structure.
925  * @param wait_to_complete
926  *   Wait for request completion (ignored).
927  *
928  * @return
929  *   0 on success, negative error value otherwise.
930  */
931 static int
932 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
933 {
934 	/*
935 	 * TODO
936 	 * once MUSDK provides necessary API use it here
937 	 */
938 	struct mrvl_priv *priv = dev->data->dev_private;
939 	struct ethtool_cmd edata;
940 	struct ifreq req;
941 	int ret, fd, link_up;
942 
943 	if (!priv->ppio)
944 		return -EPERM;
945 
946 	edata.cmd = ETHTOOL_GSET;
947 
948 	strcpy(req.ifr_name, dev->data->name);
949 	req.ifr_data = (void *)&edata;
950 
951 	fd = socket(AF_INET, SOCK_DGRAM, 0);
952 	if (fd == -1)
953 		return -EFAULT;
954 
955 	ret = ioctl(fd, SIOCETHTOOL, &req);
956 	if (ret == -1) {
957 		close(fd);
958 		return -EFAULT;
959 	}
960 
961 	close(fd);
962 
963 	switch (ethtool_cmd_speed(&edata)) {
964 	case SPEED_10:
965 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
966 		break;
967 	case SPEED_100:
968 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
969 		break;
970 	case SPEED_1000:
971 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
972 		break;
973 	case SPEED_10000:
974 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
975 		break;
976 	default:
977 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
978 	}
979 
980 	dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
981 							 ETH_LINK_HALF_DUPLEX;
982 	dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
983 							   ETH_LINK_FIXED;
984 	pp2_ppio_get_link_state(priv->ppio, &link_up);
985 	dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
986 
987 	return 0;
988 }
989 
990 /**
991  * DPDK callback to enable promiscuous mode.
992  *
993  * @param dev
994  *   Pointer to Ethernet device structure.
995  *
996  * @return
997  *   0 on success, negative error value otherwise.
998  */
999 static int
1000 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
1001 {
1002 	struct mrvl_priv *priv = dev->data->dev_private;
1003 	int ret;
1004 
1005 	if (!priv->ppio)
1006 		return 0;
1007 
1008 	if (priv->isolated)
1009 		return 0;
1010 
1011 	ret = pp2_ppio_set_promisc(priv->ppio, 1);
1012 	if (ret) {
1013 		MRVL_LOG(ERR, "Failed to enable promiscuous mode");
1014 		return -EAGAIN;
1015 	}
1016 
1017 	return 0;
1018 }
1019 
1020 /**
1021  * DPDK callback to enable allmulti mode.
1022  *
1023  * @param dev
1024  *   Pointer to Ethernet device structure.
1025  *
1026  * @return
1027  *   0 on success, negative error value otherwise.
1028  */
1029 static int
1030 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
1031 {
1032 	struct mrvl_priv *priv = dev->data->dev_private;
1033 	int ret;
1034 
1035 	if (!priv->ppio)
1036 		return 0;
1037 
1038 	if (priv->isolated)
1039 		return 0;
1040 
1041 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
1042 	if (ret) {
1043 		MRVL_LOG(ERR, "Failed enable all-multicast mode");
1044 		return -EAGAIN;
1045 	}
1046 
1047 	return 0;
1048 }
1049 
1050 /**
1051  * DPDK callback to disable promiscuous mode.
1052  *
1053  * @param dev
1054  *   Pointer to Ethernet device structure.
1055  *
1056  * @return
1057  *   0 on success, negative error value otherwise.
1058  */
1059 static int
1060 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
1061 {
1062 	struct mrvl_priv *priv = dev->data->dev_private;
1063 	int ret;
1064 
1065 	if (!priv->ppio)
1066 		return 0;
1067 
1068 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
1069 	if (ret) {
1070 		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
1071 		return -EAGAIN;
1072 	}
1073 
1074 	return 0;
1075 }
1076 
1077 /**
1078  * DPDK callback to disable allmulticast mode.
1079  *
1080  * @param dev
1081  *   Pointer to Ethernet device structure.
1082  *
1083  * @return
1084  *   0 on success, negative error value otherwise.
1085  */
1086 static int
1087 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
1088 {
1089 	struct mrvl_priv *priv = dev->data->dev_private;
1090 	int ret;
1091 
1092 	if (!priv->ppio)
1093 		return 0;
1094 
1095 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
1096 	if (ret) {
1097 		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
1098 		return -EAGAIN;
1099 	}
1100 
1101 	return 0;
1102 }
1103 
1104 /**
1105  * DPDK callback to remove a MAC address.
1106  *
1107  * @param dev
1108  *   Pointer to Ethernet device structure.
1109  * @param index
1110  *   MAC address index.
1111  */
1112 static void
1113 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1114 {
1115 	struct mrvl_priv *priv = dev->data->dev_private;
1116 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1117 	int ret;
1118 
1119 	if (!priv->ppio)
1120 		return;
1121 
1122 	if (priv->isolated)
1123 		return;
1124 
1125 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
1126 				       dev->data->mac_addrs[index].addr_bytes);
1127 	if (ret) {
1128 		rte_ether_format_addr(buf, sizeof(buf),
1129 				  &dev->data->mac_addrs[index]);
1130 		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
1131 	}
1132 }
1133 
1134 /**
1135  * DPDK callback to add a MAC address.
1136  *
1137  * @param dev
1138  *   Pointer to Ethernet device structure.
1139  * @param mac_addr
1140  *   MAC address to register.
1141  * @param index
1142  *   MAC address index.
1143  * @param vmdq
1144  *   VMDq pool index to associate address with (unused).
1145  *
1146  * @return
1147  *   0 on success, negative error value otherwise.
1148  */
1149 static int
1150 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1151 		  uint32_t index, uint32_t vmdq __rte_unused)
1152 {
1153 	struct mrvl_priv *priv = dev->data->dev_private;
1154 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1155 	int ret;
1156 
1157 	if (priv->isolated)
1158 		return -ENOTSUP;
1159 
1160 	if (index == 0)
1161 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1162 		return -1;
1163 
1164 	if (!priv->ppio)
1165 		return 0;
1166 
1167 	/*
1168 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1169 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1170 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1171 	 * 21 respectively.
1172 	 *
1173 	 * If more than uc_filter_max uc addresses were added to filter list
1174 	 * then NIC will switch to promiscuous mode automatically.
1175 	 *
1176 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1177 	 * were added to filter list then NIC will switch to all-multicast mode
1178 	 * automatically.
1179 	 */
1180 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1181 	if (ret) {
1182 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1183 		MRVL_LOG(ERR, "Failed to add mac %s", buf);
1184 		return -1;
1185 	}
1186 
1187 	return 0;
1188 }
1189 
1190 /**
1191  * DPDK callback to set the primary MAC address.
1192  *
1193  * @param dev
1194  *   Pointer to Ethernet device structure.
1195  * @param mac_addr
1196  *   MAC address to register.
1197  *
1198  * @return
1199  *   0 on success, negative error value otherwise.
1200  */
1201 static int
1202 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1203 {
1204 	struct mrvl_priv *priv = dev->data->dev_private;
1205 	int ret;
1206 
1207 	if (!priv->ppio)
1208 		return 0;
1209 
1210 	if (priv->isolated)
1211 		return -ENOTSUP;
1212 
1213 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1214 	if (ret) {
1215 		char buf[RTE_ETHER_ADDR_FMT_SIZE];
1216 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1217 		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1218 	}
1219 
1220 	return ret;
1221 }
1222 
1223 /**
1224  * DPDK callback to get device statistics.
1225  *
1226  * @param dev
1227  *   Pointer to Ethernet device structure.
1228  * @param stats
1229  *   Stats structure output buffer.
1230  *
1231  * @return
1232  *   0 on success, negative error value otherwise.
1233  */
1234 static int
1235 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1236 {
1237 	struct mrvl_priv *priv = dev->data->dev_private;
1238 	struct pp2_ppio_statistics ppio_stats;
1239 	uint64_t drop_mac = 0;
1240 	unsigned int i, idx, ret;
1241 
1242 	if (!priv->ppio)
1243 		return -EPERM;
1244 
1245 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1246 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1247 		struct pp2_ppio_inq_statistics rx_stats;
1248 
1249 		if (!rxq)
1250 			continue;
1251 
1252 		idx = rxq->queue_id;
1253 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1254 			MRVL_LOG(ERR,
1255 				"rx queue %d stats out of range (0 - %d)",
1256 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1257 			continue;
1258 		}
1259 
1260 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1261 						  priv->rxq_map[idx].tc,
1262 						  priv->rxq_map[idx].inq,
1263 						  &rx_stats, 0);
1264 		if (unlikely(ret)) {
1265 			MRVL_LOG(ERR,
1266 				"Failed to update rx queue %d stats", idx);
1267 			break;
1268 		}
1269 
1270 		stats->q_ibytes[idx] = rxq->bytes_recv;
1271 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1272 		stats->q_errors[idx] = rx_stats.drop_early +
1273 				       rx_stats.drop_fullq +
1274 				       rx_stats.drop_bm +
1275 				       rxq->drop_mac;
1276 		stats->ibytes += rxq->bytes_recv;
1277 		drop_mac += rxq->drop_mac;
1278 	}
1279 
1280 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1281 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1282 		struct pp2_ppio_outq_statistics tx_stats;
1283 
1284 		if (!txq)
1285 			continue;
1286 
1287 		idx = txq->queue_id;
1288 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1289 			MRVL_LOG(ERR,
1290 				"tx queue %d stats out of range (0 - %d)",
1291 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1292 		}
1293 
1294 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1295 						   &tx_stats, 0);
1296 		if (unlikely(ret)) {
1297 			MRVL_LOG(ERR,
1298 				"Failed to update tx queue %d stats", idx);
1299 			break;
1300 		}
1301 
1302 		stats->q_opackets[idx] = tx_stats.deq_desc;
1303 		stats->q_obytes[idx] = txq->bytes_sent;
1304 		stats->obytes += txq->bytes_sent;
1305 	}
1306 
1307 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1308 	if (unlikely(ret)) {
1309 		MRVL_LOG(ERR, "Failed to update port statistics");
1310 		return ret;
1311 	}
1312 
1313 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1314 	stats->opackets += ppio_stats.tx_packets;
1315 	stats->imissed += ppio_stats.rx_fullq_dropped +
1316 			  ppio_stats.rx_bm_dropped +
1317 			  ppio_stats.rx_early_dropped +
1318 			  ppio_stats.rx_fifo_dropped +
1319 			  ppio_stats.rx_cls_dropped;
1320 	stats->ierrors = drop_mac;
1321 
1322 	return 0;
1323 }
1324 
1325 /**
1326  * DPDK callback to clear device statistics.
1327  *
1328  * @param dev
1329  *   Pointer to Ethernet device structure.
1330  *
1331  * @return
1332  *   0 on success, negative error value otherwise.
1333  */
1334 static int
1335 mrvl_stats_reset(struct rte_eth_dev *dev)
1336 {
1337 	struct mrvl_priv *priv = dev->data->dev_private;
1338 	int i;
1339 
1340 	if (!priv->ppio)
1341 		return 0;
1342 
1343 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1344 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1345 
1346 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1347 					    priv->rxq_map[i].inq, NULL, 1);
1348 		rxq->bytes_recv = 0;
1349 		rxq->drop_mac = 0;
1350 	}
1351 
1352 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1353 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1354 
1355 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1356 		txq->bytes_sent = 0;
1357 	}
1358 
1359 	return pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1360 }
1361 
1362 /**
1363  * DPDK callback to get extended statistics.
1364  *
1365  * @param dev
1366  *   Pointer to Ethernet device structure.
1367  * @param stats
1368  *   Pointer to xstats table.
1369  * @param n
1370  *   Number of entries in xstats table.
1371  * @return
1372  *   Negative value on error, number of read xstats otherwise.
1373  */
1374 static int
1375 mrvl_xstats_get(struct rte_eth_dev *dev,
1376 		struct rte_eth_xstat *stats, unsigned int n)
1377 {
1378 	struct mrvl_priv *priv = dev->data->dev_private;
1379 	struct pp2_ppio_statistics ppio_stats;
1380 	unsigned int i;
1381 
1382 	if (!stats)
1383 		return 0;
1384 
1385 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1386 	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1387 		uint64_t val;
1388 
1389 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1390 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1391 					    mrvl_xstats_tbl[i].offset);
1392 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1393 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1394 					    mrvl_xstats_tbl[i].offset);
1395 		else
1396 			return -EINVAL;
1397 
1398 		stats[i].id = i;
1399 		stats[i].value = val;
1400 	}
1401 
1402 	return n;
1403 }
1404 
1405 /**
1406  * DPDK callback to reset extended statistics.
1407  *
1408  * @param dev
1409  *   Pointer to Ethernet device structure.
1410  *
1411  * @return
1412  *   0 on success, negative error value otherwise.
1413  */
1414 static int
1415 mrvl_xstats_reset(struct rte_eth_dev *dev)
1416 {
1417 	return mrvl_stats_reset(dev);
1418 }
1419 
1420 /**
1421  * DPDK callback to get extended statistics names.
1422  *
1423  * @param dev (unused)
1424  *   Pointer to Ethernet device structure.
1425  * @param xstats_names
1426  *   Pointer to xstats names table.
1427  * @param size
1428  *   Size of the xstats names table.
1429  * @return
1430  *   Number of read names.
1431  */
1432 static int
1433 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1434 		      struct rte_eth_xstat_name *xstats_names,
1435 		      unsigned int size)
1436 {
1437 	unsigned int i;
1438 
1439 	if (!xstats_names)
1440 		return RTE_DIM(mrvl_xstats_tbl);
1441 
1442 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1443 		strlcpy(xstats_names[i].name, mrvl_xstats_tbl[i].name,
1444 			RTE_ETH_XSTATS_NAME_SIZE);
1445 
1446 	return size;
1447 }
1448 
1449 /**
1450  * DPDK callback to get information about the device.
1451  *
1452  * @param dev
1453  *   Pointer to Ethernet device structure (unused).
1454  * @param info
1455  *   Info structure output buffer.
1456  */
1457 static int
1458 mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
1459 		   struct rte_eth_dev_info *info)
1460 {
1461 	info->speed_capa = ETH_LINK_SPEED_10M |
1462 			   ETH_LINK_SPEED_100M |
1463 			   ETH_LINK_SPEED_1G |
1464 			   ETH_LINK_SPEED_10G;
1465 
1466 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1467 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1468 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1469 
1470 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1471 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1472 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1473 
1474 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1475 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1476 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1477 
1478 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1479 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1480 
1481 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1482 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1483 
1484 	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1485 				       ETH_RSS_NONFRAG_IPV4_TCP |
1486 				       ETH_RSS_NONFRAG_IPV4_UDP;
1487 
1488 	/* By default packets are dropped if no descriptors are available */
1489 	info->default_rxconf.rx_drop_en = 1;
1490 
1491 	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1492 
1493 	return 0;
1494 }
1495 
1496 /**
1497  * Return supported packet types.
1498  *
1499  * @param dev
1500  *   Pointer to Ethernet device structure (unused).
1501  *
1502  * @return
1503  *   Const pointer to the table with supported packet types.
1504  */
1505 static const uint32_t *
1506 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1507 {
1508 	static const uint32_t ptypes[] = {
1509 		RTE_PTYPE_L2_ETHER,
1510 		RTE_PTYPE_L2_ETHER_VLAN,
1511 		RTE_PTYPE_L2_ETHER_QINQ,
1512 		RTE_PTYPE_L3_IPV4,
1513 		RTE_PTYPE_L3_IPV4_EXT,
1514 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1515 		RTE_PTYPE_L3_IPV6,
1516 		RTE_PTYPE_L3_IPV6_EXT,
1517 		RTE_PTYPE_L2_ETHER_ARP,
1518 		RTE_PTYPE_L4_TCP,
1519 		RTE_PTYPE_L4_UDP
1520 	};
1521 
1522 	return ptypes;
1523 }
1524 
1525 /**
1526  * DPDK callback to get information about specific receive queue.
1527  *
1528  * @param dev
1529  *   Pointer to Ethernet device structure.
1530  * @param rx_queue_id
1531  *   Receive queue index.
1532  * @param qinfo
1533  *   Receive queue information structure.
1534  */
1535 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1536 			      struct rte_eth_rxq_info *qinfo)
1537 {
1538 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1539 	struct mrvl_priv *priv = dev->data->dev_private;
1540 	int inq = priv->rxq_map[rx_queue_id].inq;
1541 	int tc = priv->rxq_map[rx_queue_id].tc;
1542 	struct pp2_ppio_tc_params *tc_params =
1543 		&priv->ppio_params.inqs_params.tcs_params[tc];
1544 
1545 	qinfo->mp = q->mp;
1546 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1547 }
1548 
1549 /**
1550  * DPDK callback to get information about specific transmit queue.
1551  *
1552  * @param dev
1553  *   Pointer to Ethernet device structure.
1554  * @param tx_queue_id
1555  *   Transmit queue index.
1556  * @param qinfo
1557  *   Transmit queue information structure.
1558  */
1559 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1560 			      struct rte_eth_txq_info *qinfo)
1561 {
1562 	struct mrvl_priv *priv = dev->data->dev_private;
1563 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1564 
1565 	qinfo->nb_desc =
1566 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1567 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1568 }
1569 
1570 /**
1571  * DPDK callback to Configure a VLAN filter.
1572  *
1573  * @param dev
1574  *   Pointer to Ethernet device structure.
1575  * @param vlan_id
1576  *   VLAN ID to filter.
1577  * @param on
1578  *   Toggle filter.
1579  *
1580  * @return
1581  *   0 on success, negative error value otherwise.
1582  */
1583 static int
1584 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1585 {
1586 	struct mrvl_priv *priv = dev->data->dev_private;
1587 
1588 	if (!priv->ppio)
1589 		return -EPERM;
1590 
1591 	if (priv->isolated)
1592 		return -ENOTSUP;
1593 
1594 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1595 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1596 }
1597 
1598 /**
1599  * Release buffers to hardware bpool (buffer-pool)
1600  *
1601  * @param rxq
1602  *   Receive queue pointer.
1603  * @param num
1604  *   Number of buffers to release to bpool.
1605  *
1606  * @return
1607  *   0 on success, negative error value otherwise.
1608  */
1609 static int
1610 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1611 {
1612 	struct buff_release_entry entries[MRVL_PP2_RXD_MAX];
1613 	struct rte_mbuf *mbufs[MRVL_PP2_RXD_MAX];
1614 	int i, ret;
1615 	unsigned int core_id;
1616 	struct pp2_hif *hif;
1617 	struct pp2_bpool *bpool;
1618 
1619 	core_id = rte_lcore_id();
1620 	if (core_id == LCORE_ID_ANY)
1621 		core_id = rte_get_master_lcore();
1622 
1623 	hif = mrvl_get_hif(rxq->priv, core_id);
1624 	if (!hif)
1625 		return -1;
1626 
1627 	bpool = rxq->priv->bpool;
1628 
1629 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1630 	if (ret)
1631 		return ret;
1632 
1633 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1634 		cookie_addr_high =
1635 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1636 
1637 	for (i = 0; i < num; i++) {
1638 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1639 			!= cookie_addr_high) {
1640 			MRVL_LOG(ERR,
1641 				"mbuf virtual addr high 0x%lx out of range",
1642 				(uint64_t)mbufs[i] >> 32);
1643 			goto out;
1644 		}
1645 
1646 		entries[i].buff.addr =
1647 			rte_mbuf_data_iova_default(mbufs[i]);
1648 		entries[i].buff.cookie = (uint64_t)mbufs[i];
1649 		entries[i].bpool = bpool;
1650 	}
1651 
1652 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1653 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1654 
1655 	if (i != num)
1656 		goto out;
1657 
1658 	return 0;
1659 out:
1660 	for (; i < num; i++)
1661 		rte_pktmbuf_free(mbufs[i]);
1662 
1663 	return -1;
1664 }
1665 
1666 /**
1667  * DPDK callback to configure the receive queue.
1668  *
1669  * @param dev
1670  *   Pointer to Ethernet device structure.
1671  * @param idx
1672  *   RX queue index.
1673  * @param desc
1674  *   Number of descriptors to configure in queue.
1675  * @param socket
1676  *   NUMA socket on which memory must be allocated.
1677  * @param conf
1678  *   Thresholds parameters.
1679  * @param mp
1680  *   Memory pool for buffer allocations.
1681  *
1682  * @return
1683  *   0 on success, negative error value otherwise.
1684  */
1685 static int
1686 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1687 		    unsigned int socket,
1688 		    const struct rte_eth_rxconf *conf,
1689 		    struct rte_mempool *mp)
1690 {
1691 	struct mrvl_priv *priv = dev->data->dev_private;
1692 	struct mrvl_rxq *rxq;
1693 	uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
1694 	uint32_t max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1695 	int ret, tc, inq;
1696 	uint64_t offloads;
1697 
1698 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
1699 
1700 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1701 		/*
1702 		 * Unknown TC mapping, mapping will not have a correct queue.
1703 		 */
1704 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
1705 			idx, priv->ppio_id);
1706 		return -EFAULT;
1707 	}
1708 
1709 	frame_size = buf_size - RTE_PKTMBUF_HEADROOM - MRVL_PKT_EFFEC_OFFS;
1710 	if (frame_size < max_rx_pkt_len) {
1711 		MRVL_LOG(WARNING,
1712 			"Mbuf size must be increased to %u bytes to hold up "
1713 			"to %u bytes of data.",
1714 			buf_size + max_rx_pkt_len - frame_size,
1715 			max_rx_pkt_len);
1716 		dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
1717 		MRVL_LOG(INFO, "Setting max rx pkt len to %u",
1718 			dev->data->dev_conf.rxmode.max_rx_pkt_len);
1719 	}
1720 
1721 	if (dev->data->rx_queues[idx]) {
1722 		rte_free(dev->data->rx_queues[idx]);
1723 		dev->data->rx_queues[idx] = NULL;
1724 	}
1725 
1726 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
1727 	if (!rxq)
1728 		return -ENOMEM;
1729 
1730 	rxq->priv = priv;
1731 	rxq->mp = mp;
1732 	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
1733 	rxq->queue_id = idx;
1734 	rxq->port_id = dev->data->port_id;
1735 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
1736 
1737 	tc = priv->rxq_map[rxq->queue_id].tc,
1738 	inq = priv->rxq_map[rxq->queue_id].inq;
1739 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
1740 		desc;
1741 
1742 	ret = mrvl_fill_bpool(rxq, desc);
1743 	if (ret) {
1744 		rte_free(rxq);
1745 		return ret;
1746 	}
1747 
1748 	priv->bpool_init_size += desc;
1749 
1750 	dev->data->rx_queues[idx] = rxq;
1751 
1752 	return 0;
1753 }
1754 
1755 /**
1756  * DPDK callback to release the receive queue.
1757  *
1758  * @param rxq
1759  *   Generic receive queue pointer.
1760  */
1761 static void
1762 mrvl_rx_queue_release(void *rxq)
1763 {
1764 	struct mrvl_rxq *q = rxq;
1765 	struct pp2_ppio_tc_params *tc_params;
1766 	int i, num, tc, inq;
1767 	struct pp2_hif *hif;
1768 	unsigned int core_id = rte_lcore_id();
1769 
1770 	if (core_id == LCORE_ID_ANY)
1771 		core_id = rte_get_master_lcore();
1772 
1773 	if (!q)
1774 		return;
1775 
1776 	hif = mrvl_get_hif(q->priv, core_id);
1777 
1778 	if (!hif)
1779 		return;
1780 
1781 	tc = q->priv->rxq_map[q->queue_id].tc;
1782 	inq = q->priv->rxq_map[q->queue_id].inq;
1783 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
1784 	num = tc_params->inqs_params[inq].size;
1785 	for (i = 0; i < num; i++) {
1786 		struct pp2_buff_inf inf;
1787 		uint64_t addr;
1788 
1789 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
1790 		addr = cookie_addr_high | inf.cookie;
1791 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1792 	}
1793 
1794 	rte_free(q);
1795 }
1796 
1797 /**
1798  * DPDK callback to configure the transmit queue.
1799  *
1800  * @param dev
1801  *   Pointer to Ethernet device structure.
1802  * @param idx
1803  *   Transmit queue index.
1804  * @param desc
1805  *   Number of descriptors to configure in the queue.
1806  * @param socket
1807  *   NUMA socket on which memory must be allocated.
1808  * @param conf
1809  *   Tx queue configuration parameters.
1810  *
1811  * @return
1812  *   0 on success, negative error value otherwise.
1813  */
1814 static int
1815 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1816 		    unsigned int socket,
1817 		    const struct rte_eth_txconf *conf)
1818 {
1819 	struct mrvl_priv *priv = dev->data->dev_private;
1820 	struct mrvl_txq *txq;
1821 
1822 	if (dev->data->tx_queues[idx]) {
1823 		rte_free(dev->data->tx_queues[idx]);
1824 		dev->data->tx_queues[idx] = NULL;
1825 	}
1826 
1827 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
1828 	if (!txq)
1829 		return -ENOMEM;
1830 
1831 	txq->priv = priv;
1832 	txq->queue_id = idx;
1833 	txq->port_id = dev->data->port_id;
1834 	txq->tx_deferred_start = conf->tx_deferred_start;
1835 	dev->data->tx_queues[idx] = txq;
1836 
1837 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
1838 
1839 	return 0;
1840 }
1841 
1842 /**
1843  * DPDK callback to release the transmit queue.
1844  *
1845  * @param txq
1846  *   Generic transmit queue pointer.
1847  */
1848 static void
1849 mrvl_tx_queue_release(void *txq)
1850 {
1851 	struct mrvl_txq *q = txq;
1852 
1853 	if (!q)
1854 		return;
1855 
1856 	rte_free(q);
1857 }
1858 
1859 /**
1860  * DPDK callback to get flow control configuration.
1861  *
1862  * @param dev
1863  *  Pointer to Ethernet device structure.
1864  * @param fc_conf
1865  *  Pointer to the flow control configuration.
1866  *
1867  * @return
1868  *  0 on success, negative error value otherwise.
1869  */
1870 static int
1871 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1872 {
1873 	struct mrvl_priv *priv = dev->data->dev_private;
1874 	int ret, en;
1875 
1876 	if (!priv)
1877 		return -EPERM;
1878 
1879 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
1880 	if (ret) {
1881 		MRVL_LOG(ERR, "Failed to read rx pause state");
1882 		return ret;
1883 	}
1884 
1885 	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
1886 
1887 	return 0;
1888 }
1889 
1890 /**
1891  * DPDK callback to set flow control configuration.
1892  *
1893  * @param dev
1894  *  Pointer to Ethernet device structure.
1895  * @param fc_conf
1896  *  Pointer to the flow control configuration.
1897  *
1898  * @return
1899  *  0 on success, negative error value otherwise.
1900  */
1901 static int
1902 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1903 {
1904 	struct mrvl_priv *priv = dev->data->dev_private;
1905 
1906 	if (!priv)
1907 		return -EPERM;
1908 
1909 	if (fc_conf->high_water ||
1910 	    fc_conf->low_water ||
1911 	    fc_conf->pause_time ||
1912 	    fc_conf->mac_ctrl_frame_fwd ||
1913 	    fc_conf->autoneg) {
1914 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
1915 
1916 		return -EINVAL;
1917 	}
1918 
1919 	if (fc_conf->mode == RTE_FC_NONE ||
1920 	    fc_conf->mode == RTE_FC_RX_PAUSE) {
1921 		int ret, en;
1922 
1923 		en = fc_conf->mode == RTE_FC_NONE ? 0 : 1;
1924 		ret = pp2_ppio_set_rx_pause(priv->ppio, en);
1925 		if (ret)
1926 			MRVL_LOG(ERR,
1927 				"Failed to change flowctrl on RX side");
1928 
1929 		return ret;
1930 	}
1931 
1932 	return 0;
1933 }
1934 
1935 /**
1936  * Update RSS hash configuration
1937  *
1938  * @param dev
1939  *   Pointer to Ethernet device structure.
1940  * @param rss_conf
1941  *   Pointer to RSS configuration.
1942  *
1943  * @return
1944  *   0 on success, negative error value otherwise.
1945  */
1946 static int
1947 mrvl_rss_hash_update(struct rte_eth_dev *dev,
1948 		     struct rte_eth_rss_conf *rss_conf)
1949 {
1950 	struct mrvl_priv *priv = dev->data->dev_private;
1951 
1952 	if (priv->isolated)
1953 		return -ENOTSUP;
1954 
1955 	return mrvl_configure_rss(priv, rss_conf);
1956 }
1957 
1958 /**
1959  * DPDK callback to get RSS hash configuration.
1960  *
1961  * @param dev
1962  *   Pointer to Ethernet device structure.
1963  * @rss_conf
1964  *   Pointer to RSS configuration.
1965  *
1966  * @return
1967  *   Always 0.
1968  */
1969 static int
1970 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
1971 		       struct rte_eth_rss_conf *rss_conf)
1972 {
1973 	struct mrvl_priv *priv = dev->data->dev_private;
1974 	enum pp2_ppio_hash_type hash_type =
1975 		priv->ppio_params.inqs_params.hash_type;
1976 
1977 	rss_conf->rss_key = NULL;
1978 
1979 	if (hash_type == PP2_PPIO_HASH_T_NONE)
1980 		rss_conf->rss_hf = 0;
1981 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
1982 		rss_conf->rss_hf = ETH_RSS_IPV4;
1983 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
1984 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
1985 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
1986 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
1987 
1988 	return 0;
1989 }
1990 
1991 /**
1992  * DPDK callback to get rte_flow callbacks.
1993  *
1994  * @param dev
1995  *   Pointer to the device structure.
1996  * @param filer_type
1997  *   Flow filter type.
1998  * @param filter_op
1999  *   Flow filter operation.
2000  * @param arg
2001  *   Pointer to pass the flow ops.
2002  *
2003  * @return
2004  *   0 on success, negative error value otherwise.
2005  */
2006 static int
2007 mrvl_eth_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
2008 		     enum rte_filter_type filter_type,
2009 		     enum rte_filter_op filter_op, void *arg)
2010 {
2011 	switch (filter_type) {
2012 	case RTE_ETH_FILTER_GENERIC:
2013 		if (filter_op != RTE_ETH_FILTER_GET)
2014 			return -EINVAL;
2015 		*(const void **)arg = &mrvl_flow_ops;
2016 		return 0;
2017 	default:
2018 		MRVL_LOG(WARNING, "Filter type (%d) not supported",
2019 				filter_type);
2020 		return -EINVAL;
2021 	}
2022 }
2023 
2024 /**
2025  * DPDK callback to get rte_mtr callbacks.
2026  *
2027  * @param dev
2028  *   Pointer to the device structure.
2029  * @param ops
2030  *   Pointer to pass the mtr ops.
2031  *
2032  * @return
2033  *   Always 0.
2034  */
2035 static int
2036 mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2037 {
2038 	*(const void **)ops = &mrvl_mtr_ops;
2039 
2040 	return 0;
2041 }
2042 
2043 /**
2044  * DPDK callback to get rte_tm callbacks.
2045  *
2046  * @param dev
2047  *   Pointer to the device structure.
2048  * @param ops
2049  *   Pointer to pass the tm ops.
2050  *
2051  * @return
2052  *   Always 0.
2053  */
2054 static int
2055 mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2056 {
2057 	*(const void **)ops = &mrvl_tm_ops;
2058 
2059 	return 0;
2060 }
2061 
2062 static const struct eth_dev_ops mrvl_ops = {
2063 	.dev_configure = mrvl_dev_configure,
2064 	.dev_start = mrvl_dev_start,
2065 	.dev_stop = mrvl_dev_stop,
2066 	.dev_set_link_up = mrvl_dev_set_link_up,
2067 	.dev_set_link_down = mrvl_dev_set_link_down,
2068 	.dev_close = mrvl_dev_close,
2069 	.link_update = mrvl_link_update,
2070 	.promiscuous_enable = mrvl_promiscuous_enable,
2071 	.allmulticast_enable = mrvl_allmulticast_enable,
2072 	.promiscuous_disable = mrvl_promiscuous_disable,
2073 	.allmulticast_disable = mrvl_allmulticast_disable,
2074 	.mac_addr_remove = mrvl_mac_addr_remove,
2075 	.mac_addr_add = mrvl_mac_addr_add,
2076 	.mac_addr_set = mrvl_mac_addr_set,
2077 	.mtu_set = mrvl_mtu_set,
2078 	.stats_get = mrvl_stats_get,
2079 	.stats_reset = mrvl_stats_reset,
2080 	.xstats_get = mrvl_xstats_get,
2081 	.xstats_reset = mrvl_xstats_reset,
2082 	.xstats_get_names = mrvl_xstats_get_names,
2083 	.dev_infos_get = mrvl_dev_infos_get,
2084 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
2085 	.rxq_info_get = mrvl_rxq_info_get,
2086 	.txq_info_get = mrvl_txq_info_get,
2087 	.vlan_filter_set = mrvl_vlan_filter_set,
2088 	.tx_queue_start = mrvl_tx_queue_start,
2089 	.tx_queue_stop = mrvl_tx_queue_stop,
2090 	.rx_queue_setup = mrvl_rx_queue_setup,
2091 	.rx_queue_release = mrvl_rx_queue_release,
2092 	.tx_queue_setup = mrvl_tx_queue_setup,
2093 	.tx_queue_release = mrvl_tx_queue_release,
2094 	.flow_ctrl_get = mrvl_flow_ctrl_get,
2095 	.flow_ctrl_set = mrvl_flow_ctrl_set,
2096 	.rss_hash_update = mrvl_rss_hash_update,
2097 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
2098 	.filter_ctrl = mrvl_eth_filter_ctrl,
2099 	.mtr_ops_get = mrvl_mtr_ops_get,
2100 	.tm_ops_get = mrvl_tm_ops_get,
2101 };
2102 
2103 /**
2104  * Return packet type information and l3/l4 offsets.
2105  *
2106  * @param desc
2107  *   Pointer to the received packet descriptor.
2108  * @param l3_offset
2109  *   l3 packet offset.
2110  * @param l4_offset
2111  *   l4 packet offset.
2112  *
2113  * @return
2114  *   Packet type information.
2115  */
2116 static inline uint64_t
2117 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
2118 				    uint8_t *l3_offset, uint8_t *l4_offset)
2119 {
2120 	enum pp2_inq_l3_type l3_type;
2121 	enum pp2_inq_l4_type l4_type;
2122 	enum pp2_inq_vlan_tag vlan_tag;
2123 	uint64_t packet_type;
2124 
2125 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
2126 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
2127 	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
2128 
2129 	packet_type = RTE_PTYPE_L2_ETHER;
2130 
2131 	switch (vlan_tag) {
2132 	case PP2_INQ_VLAN_TAG_SINGLE:
2133 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
2134 		break;
2135 	case PP2_INQ_VLAN_TAG_DOUBLE:
2136 	case PP2_INQ_VLAN_TAG_TRIPLE:
2137 		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
2138 		break;
2139 	default:
2140 		break;
2141 	}
2142 
2143 	switch (l3_type) {
2144 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
2145 		packet_type |= RTE_PTYPE_L3_IPV4;
2146 		break;
2147 	case PP2_INQ_L3_TYPE_IPV4_OK:
2148 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
2149 		break;
2150 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
2151 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
2152 		break;
2153 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
2154 		packet_type |= RTE_PTYPE_L3_IPV6;
2155 		break;
2156 	case PP2_INQ_L3_TYPE_IPV6_EXT:
2157 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
2158 		break;
2159 	case PP2_INQ_L3_TYPE_ARP:
2160 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
2161 		/*
2162 		 * In case of ARP l4_offset is set to wrong value.
2163 		 * Set it to proper one so that later on mbuf->l3_len can be
2164 		 * calculated subtracting l4_offset and l3_offset.
2165 		 */
2166 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
2167 		break;
2168 	default:
2169 		MRVL_LOG(DEBUG, "Failed to recognise l3 packet type");
2170 		break;
2171 	}
2172 
2173 	switch (l4_type) {
2174 	case PP2_INQ_L4_TYPE_TCP:
2175 		packet_type |= RTE_PTYPE_L4_TCP;
2176 		break;
2177 	case PP2_INQ_L4_TYPE_UDP:
2178 		packet_type |= RTE_PTYPE_L4_UDP;
2179 		break;
2180 	default:
2181 		MRVL_LOG(DEBUG, "Failed to recognise l4 packet type");
2182 		break;
2183 	}
2184 
2185 	return packet_type;
2186 }
2187 
2188 /**
2189  * Get offload information from the received packet descriptor.
2190  *
2191  * @param desc
2192  *   Pointer to the received packet descriptor.
2193  *
2194  * @return
2195  *   Mbuf offload flags.
2196  */
2197 static inline uint64_t
2198 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc)
2199 {
2200 	uint64_t flags;
2201 	enum pp2_inq_desc_status status;
2202 
2203 	status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2204 	if (unlikely(status != PP2_DESC_ERR_OK))
2205 		flags = PKT_RX_IP_CKSUM_BAD;
2206 	else
2207 		flags = PKT_RX_IP_CKSUM_GOOD;
2208 
2209 	status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2210 	if (unlikely(status != PP2_DESC_ERR_OK))
2211 		flags |= PKT_RX_L4_CKSUM_BAD;
2212 	else
2213 		flags |= PKT_RX_L4_CKSUM_GOOD;
2214 
2215 	return flags;
2216 }
2217 
2218 /**
2219  * DPDK callback for receive.
2220  *
2221  * @param rxq
2222  *   Generic pointer to the receive queue.
2223  * @param rx_pkts
2224  *   Array to store received packets.
2225  * @param nb_pkts
2226  *   Maximum number of packets in array.
2227  *
2228  * @return
2229  *   Number of packets successfully received.
2230  */
2231 static uint16_t
2232 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2233 {
2234 	struct mrvl_rxq *q = rxq;
2235 	struct pp2_ppio_desc descs[nb_pkts];
2236 	struct pp2_bpool *bpool;
2237 	int i, ret, rx_done = 0;
2238 	int num;
2239 	struct pp2_hif *hif;
2240 	unsigned int core_id = rte_lcore_id();
2241 
2242 	hif = mrvl_get_hif(q->priv, core_id);
2243 
2244 	if (unlikely(!q->priv->ppio || !hif))
2245 		return 0;
2246 
2247 	bpool = q->priv->bpool;
2248 
2249 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2250 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2251 	if (unlikely(ret < 0)) {
2252 		MRVL_LOG(ERR, "Failed to receive packets");
2253 		return 0;
2254 	}
2255 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2256 
2257 	for (i = 0; i < nb_pkts; i++) {
2258 		struct rte_mbuf *mbuf;
2259 		uint8_t l3_offset, l4_offset;
2260 		enum pp2_inq_desc_status status;
2261 		uint64_t addr;
2262 
2263 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2264 			struct pp2_ppio_desc *pref_desc;
2265 			u64 pref_addr;
2266 
2267 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2268 			pref_addr = cookie_addr_high |
2269 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2270 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2271 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2272 		}
2273 
2274 		addr = cookie_addr_high |
2275 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2276 		mbuf = (struct rte_mbuf *)addr;
2277 		rte_pktmbuf_reset(mbuf);
2278 
2279 		/* drop packet in case of mac, overrun or resource error */
2280 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2281 		if (unlikely(status != PP2_DESC_ERR_OK)) {
2282 			struct pp2_buff_inf binf = {
2283 				.addr = rte_mbuf_data_iova_default(mbuf),
2284 				.cookie = (uint64_t)mbuf,
2285 			};
2286 
2287 			pp2_bpool_put_buff(hif, bpool, &binf);
2288 			mrvl_port_bpool_size
2289 				[bpool->pp2_id][bpool->id][core_id]++;
2290 			q->drop_mac++;
2291 			continue;
2292 		}
2293 
2294 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2295 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2296 		mbuf->data_len = mbuf->pkt_len;
2297 		mbuf->port = q->port_id;
2298 		mbuf->packet_type =
2299 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2300 							    &l3_offset,
2301 							    &l4_offset);
2302 		mbuf->l2_len = l3_offset;
2303 		mbuf->l3_len = l4_offset - l3_offset;
2304 
2305 		if (likely(q->cksum_enabled))
2306 			mbuf->ol_flags = mrvl_desc_to_ol_flags(&descs[i]);
2307 
2308 		rx_pkts[rx_done++] = mbuf;
2309 		q->bytes_recv += mbuf->pkt_len;
2310 	}
2311 
2312 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2313 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2314 
2315 		if (unlikely(num <= q->priv->bpool_min_size ||
2316 			     (!rx_done && num < q->priv->bpool_init_size))) {
2317 			ret = mrvl_fill_bpool(q, MRVL_BURST_SIZE);
2318 			if (ret)
2319 				MRVL_LOG(ERR, "Failed to fill bpool");
2320 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2321 			int i;
2322 			int pkt_to_remove = num - q->priv->bpool_init_size;
2323 			struct rte_mbuf *mbuf;
2324 			struct pp2_buff_inf buff;
2325 
2326 			MRVL_LOG(DEBUG,
2327 				"port-%d:%d: bpool %d oversize - remove %d buffers (pool size: %d -> %d)",
2328 				bpool->pp2_id, q->priv->ppio->port_id,
2329 				bpool->id, pkt_to_remove, num,
2330 				q->priv->bpool_init_size);
2331 
2332 			for (i = 0; i < pkt_to_remove; i++) {
2333 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2334 				if (ret)
2335 					break;
2336 				mbuf = (struct rte_mbuf *)
2337 					(cookie_addr_high | buff.cookie);
2338 				rte_pktmbuf_free(mbuf);
2339 			}
2340 			mrvl_port_bpool_size
2341 				[bpool->pp2_id][bpool->id][core_id] -= i;
2342 		}
2343 		rte_spinlock_unlock(&q->priv->lock);
2344 	}
2345 
2346 	return rx_done;
2347 }
2348 
2349 /**
2350  * Prepare offload information.
2351  *
2352  * @param ol_flags
2353  *   Offload flags.
2354  * @param packet_type
2355  *   Packet type bitfield.
2356  * @param l3_type
2357  *   Pointer to the pp2_ouq_l3_type structure.
2358  * @param l4_type
2359  *   Pointer to the pp2_outq_l4_type structure.
2360  * @param gen_l3_cksum
2361  *   Will be set to 1 in case l3 checksum is computed.
2362  * @param l4_cksum
2363  *   Will be set to 1 in case l4 checksum is computed.
2364  *
2365  * @return
2366  *   0 on success, negative error value otherwise.
2367  */
2368 static inline int
2369 mrvl_prepare_proto_info(uint64_t ol_flags, uint32_t packet_type,
2370 			enum pp2_outq_l3_type *l3_type,
2371 			enum pp2_outq_l4_type *l4_type,
2372 			int *gen_l3_cksum,
2373 			int *gen_l4_cksum)
2374 {
2375 	/*
2376 	 * Based on ol_flags prepare information
2377 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2378 	 * for offloading.
2379 	 */
2380 	if (ol_flags & PKT_TX_IPV4) {
2381 		*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2382 		*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2383 	} else if (ol_flags & PKT_TX_IPV6) {
2384 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2385 		/* no checksum for ipv6 header */
2386 		*gen_l3_cksum = 0;
2387 	} else {
2388 		/* if something different then stop processing */
2389 		return -1;
2390 	}
2391 
2392 	ol_flags &= PKT_TX_L4_MASK;
2393 	if ((packet_type & RTE_PTYPE_L4_TCP) &&
2394 	    ol_flags == PKT_TX_TCP_CKSUM) {
2395 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2396 		*gen_l4_cksum = 1;
2397 	} else if ((packet_type & RTE_PTYPE_L4_UDP) &&
2398 		   ol_flags == PKT_TX_UDP_CKSUM) {
2399 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2400 		*gen_l4_cksum = 1;
2401 	} else {
2402 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2403 		/* no checksum for other type */
2404 		*gen_l4_cksum = 0;
2405 	}
2406 
2407 	return 0;
2408 }
2409 
2410 /**
2411  * Release already sent buffers to bpool (buffer-pool).
2412  *
2413  * @param ppio
2414  *   Pointer to the port structure.
2415  * @param hif
2416  *   Pointer to the MUSDK hardware interface.
2417  * @param sq
2418  *   Pointer to the shadow queue.
2419  * @param qid
2420  *   Queue id number.
2421  * @param force
2422  *   Force releasing packets.
2423  */
2424 static inline void
2425 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2426 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2427 		       int qid, int force)
2428 {
2429 	struct buff_release_entry *entry;
2430 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2431 	int i;
2432 
2433 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2434 
2435 	sq->num_to_release += nb_done;
2436 
2437 	if (likely(!force &&
2438 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2439 		return;
2440 
2441 	nb_done = sq->num_to_release;
2442 	sq->num_to_release = 0;
2443 
2444 	for (i = 0; i < nb_done; i++) {
2445 		entry = &sq->ent[sq->tail + num];
2446 		if (unlikely(!entry->buff.addr)) {
2447 			MRVL_LOG(ERR,
2448 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2449 				sq->tail, (u64)entry->buff.cookie,
2450 				(u64)entry->buff.addr);
2451 			skip_bufs = 1;
2452 			goto skip;
2453 		}
2454 
2455 		if (unlikely(!entry->bpool)) {
2456 			struct rte_mbuf *mbuf;
2457 
2458 			mbuf = (struct rte_mbuf *)
2459 			       (cookie_addr_high | entry->buff.cookie);
2460 			rte_pktmbuf_free(mbuf);
2461 			skip_bufs = 1;
2462 			goto skip;
2463 		}
2464 
2465 		mrvl_port_bpool_size
2466 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2467 		num++;
2468 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2469 			goto skip;
2470 		continue;
2471 skip:
2472 		if (likely(num))
2473 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2474 		num += skip_bufs;
2475 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2476 		sq->size -= num;
2477 		num = 0;
2478 		skip_bufs = 0;
2479 	}
2480 
2481 	if (likely(num)) {
2482 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2483 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2484 		sq->size -= num;
2485 	}
2486 }
2487 
2488 /**
2489  * DPDK callback for transmit.
2490  *
2491  * @param txq
2492  *   Generic pointer transmit queue.
2493  * @param tx_pkts
2494  *   Packets to transmit.
2495  * @param nb_pkts
2496  *   Number of packets in array.
2497  *
2498  * @return
2499  *   Number of packets successfully transmitted.
2500  */
2501 static uint16_t
2502 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2503 {
2504 	struct mrvl_txq *q = txq;
2505 	struct mrvl_shadow_txq *sq;
2506 	struct pp2_hif *hif;
2507 	struct pp2_ppio_desc descs[nb_pkts];
2508 	unsigned int core_id = rte_lcore_id();
2509 	int i, ret, bytes_sent = 0;
2510 	uint16_t num, sq_free_size;
2511 	uint64_t addr;
2512 
2513 	hif = mrvl_get_hif(q->priv, core_id);
2514 	sq = &q->shadow_txqs[core_id];
2515 
2516 	if (unlikely(!q->priv->ppio || !hif))
2517 		return 0;
2518 
2519 	if (sq->size)
2520 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2521 				       sq, q->queue_id, 0);
2522 
2523 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2524 	if (unlikely(nb_pkts > sq_free_size)) {
2525 		MRVL_LOG(DEBUG,
2526 			"No room in shadow queue for %d packets! %d packets will be sent.",
2527 			nb_pkts, sq_free_size);
2528 		nb_pkts = sq_free_size;
2529 	}
2530 
2531 	for (i = 0; i < nb_pkts; i++) {
2532 		struct rte_mbuf *mbuf = tx_pkts[i];
2533 		int gen_l3_cksum, gen_l4_cksum;
2534 		enum pp2_outq_l3_type l3_type;
2535 		enum pp2_outq_l4_type l4_type;
2536 
2537 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2538 			struct rte_mbuf *pref_pkt_hdr;
2539 
2540 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2541 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2542 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2543 		}
2544 
2545 		mrvl_fill_shadowq(sq, mbuf);
2546 		mrvl_fill_desc(&descs[i], mbuf);
2547 
2548 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2549 		/*
2550 		 * in case unsupported ol_flags were passed
2551 		 * do not update descriptor offload information
2552 		 */
2553 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2554 					      &l3_type, &l4_type, &gen_l3_cksum,
2555 					      &gen_l4_cksum);
2556 		if (unlikely(ret))
2557 			continue;
2558 
2559 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2560 						  mbuf->l2_len,
2561 						  mbuf->l2_len + mbuf->l3_len,
2562 						  gen_l3_cksum, gen_l4_cksum);
2563 	}
2564 
2565 	num = nb_pkts;
2566 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2567 	/* number of packets that were not sent */
2568 	if (unlikely(num > nb_pkts)) {
2569 		for (i = nb_pkts; i < num; i++) {
2570 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2571 				MRVL_PP2_TX_SHADOWQ_MASK;
2572 			addr = cookie_addr_high | sq->ent[sq->head].buff.cookie;
2573 			bytes_sent -=
2574 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2575 		}
2576 		sq->size -= num - nb_pkts;
2577 	}
2578 
2579 	q->bytes_sent += bytes_sent;
2580 
2581 	return nb_pkts;
2582 }
2583 
2584 /** DPDK callback for S/G transmit.
2585  *
2586  * @param txq
2587  *   Generic pointer transmit queue.
2588  * @param tx_pkts
2589  *   Packets to transmit.
2590  * @param nb_pkts
2591  *   Number of packets in array.
2592  *
2593  * @return
2594  *   Number of packets successfully transmitted.
2595  */
2596 static uint16_t
2597 mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
2598 		     uint16_t nb_pkts)
2599 {
2600 	struct mrvl_txq *q = txq;
2601 	struct mrvl_shadow_txq *sq;
2602 	struct pp2_hif *hif;
2603 	struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
2604 	struct pp2_ppio_sg_pkts pkts;
2605 	uint8_t frags[nb_pkts];
2606 	unsigned int core_id = rte_lcore_id();
2607 	int i, j, ret, bytes_sent = 0;
2608 	int tail, tail_first;
2609 	uint16_t num, sq_free_size;
2610 	uint16_t nb_segs, total_descs = 0;
2611 	uint64_t addr;
2612 
2613 	hif = mrvl_get_hif(q->priv, core_id);
2614 	sq = &q->shadow_txqs[core_id];
2615 	pkts.frags = frags;
2616 	pkts.num = 0;
2617 
2618 	if (unlikely(!q->priv->ppio || !hif))
2619 		return 0;
2620 
2621 	if (sq->size)
2622 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2623 				       sq, q->queue_id, 0);
2624 
2625 	/* Save shadow queue free size */
2626 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2627 
2628 	tail = 0;
2629 	for (i = 0; i < nb_pkts; i++) {
2630 		struct rte_mbuf *mbuf = tx_pkts[i];
2631 		struct rte_mbuf *seg = NULL;
2632 		int gen_l3_cksum, gen_l4_cksum;
2633 		enum pp2_outq_l3_type l3_type;
2634 		enum pp2_outq_l4_type l4_type;
2635 
2636 		nb_segs = mbuf->nb_segs;
2637 		tail_first = tail;
2638 		total_descs += nb_segs;
2639 
2640 		/*
2641 		 * Check if total_descs does not exceed
2642 		 * shadow queue free size
2643 		 */
2644 		if (unlikely(total_descs > sq_free_size)) {
2645 			total_descs -= nb_segs;
2646 			RTE_LOG(DEBUG, PMD,
2647 				"No room in shadow queue for %d packets! "
2648 				"%d packets will be sent.\n",
2649 				nb_pkts, i);
2650 			break;
2651 		}
2652 
2653 		/* Check if nb_segs does not exceed the max nb of desc per
2654 		 * fragmented packet
2655 		 */
2656 		if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
2657 			total_descs -= nb_segs;
2658 			RTE_LOG(ERR, PMD,
2659 				"Too many segments. Packet won't be sent.\n");
2660 			break;
2661 		}
2662 
2663 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2664 			struct rte_mbuf *pref_pkt_hdr;
2665 
2666 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2667 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2668 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2669 		}
2670 
2671 		pkts.frags[pkts.num] = nb_segs;
2672 		pkts.num++;
2673 
2674 		seg = mbuf;
2675 		for (j = 0; j < nb_segs - 1; j++) {
2676 			/* For the subsequent segments, set shadow queue
2677 			 * buffer to NULL
2678 			 */
2679 			mrvl_fill_shadowq(sq, NULL);
2680 			mrvl_fill_desc(&descs[tail], seg);
2681 
2682 			tail++;
2683 			seg = seg->next;
2684 		}
2685 		/* Put first mbuf info in last shadow queue entry */
2686 		mrvl_fill_shadowq(sq, mbuf);
2687 		/* Update descriptor with last segment */
2688 		mrvl_fill_desc(&descs[tail++], seg);
2689 
2690 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2691 		/* In case unsupported ol_flags were passed
2692 		 * do not update descriptor offload information
2693 		 */
2694 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2695 					      &l3_type, &l4_type, &gen_l3_cksum,
2696 					      &gen_l4_cksum);
2697 		if (unlikely(ret))
2698 			continue;
2699 
2700 		pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
2701 						  l4_type, mbuf->l2_len,
2702 						  mbuf->l2_len + mbuf->l3_len,
2703 						  gen_l3_cksum, gen_l4_cksum);
2704 	}
2705 
2706 	num = total_descs;
2707 	pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
2708 			 &total_descs, &pkts);
2709 	/* number of packets that were not sent */
2710 	if (unlikely(num > total_descs)) {
2711 		for (i = total_descs; i < num; i++) {
2712 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2713 				MRVL_PP2_TX_SHADOWQ_MASK;
2714 
2715 			addr = sq->ent[sq->head].buff.cookie;
2716 			if (addr)
2717 				bytes_sent -=
2718 					rte_pktmbuf_pkt_len((struct rte_mbuf *)
2719 						(cookie_addr_high | addr));
2720 		}
2721 		sq->size -= num - total_descs;
2722 		nb_pkts = pkts.num;
2723 	}
2724 
2725 	q->bytes_sent += bytes_sent;
2726 
2727 	return nb_pkts;
2728 }
2729 
2730 /**
2731  * Initialize packet processor.
2732  *
2733  * @return
2734  *   0 on success, negative error value otherwise.
2735  */
2736 static int
2737 mrvl_init_pp2(void)
2738 {
2739 	struct pp2_init_params init_params;
2740 
2741 	memset(&init_params, 0, sizeof(init_params));
2742 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
2743 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
2744 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
2745 
2746 	return pp2_init(&init_params);
2747 }
2748 
2749 /**
2750  * Deinitialize packet processor.
2751  *
2752  * @return
2753  *   0 on success, negative error value otherwise.
2754  */
2755 static void
2756 mrvl_deinit_pp2(void)
2757 {
2758 	pp2_deinit();
2759 }
2760 
2761 /**
2762  * Create private device structure.
2763  *
2764  * @param dev_name
2765  *   Pointer to the port name passed in the initialization parameters.
2766  *
2767  * @return
2768  *   Pointer to the newly allocated private device structure.
2769  */
2770 static struct mrvl_priv *
2771 mrvl_priv_create(const char *dev_name)
2772 {
2773 	struct pp2_bpool_params bpool_params;
2774 	char match[MRVL_MATCH_LEN];
2775 	struct mrvl_priv *priv;
2776 	int ret, bpool_bit;
2777 
2778 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
2779 	if (!priv)
2780 		return NULL;
2781 
2782 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
2783 				       &priv->pp_id, &priv->ppio_id);
2784 	if (ret)
2785 		goto out_free_priv;
2786 
2787 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
2788 				     PP2_BPOOL_NUM_POOLS);
2789 	if (bpool_bit < 0)
2790 		goto out_free_priv;
2791 	priv->bpool_bit = bpool_bit;
2792 
2793 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
2794 		 priv->bpool_bit);
2795 	memset(&bpool_params, 0, sizeof(bpool_params));
2796 	bpool_params.match = match;
2797 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
2798 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
2799 	if (ret)
2800 		goto out_clear_bpool_bit;
2801 
2802 	priv->ppio_params.type = PP2_PPIO_T_NIC;
2803 	rte_spinlock_init(&priv->lock);
2804 
2805 	return priv;
2806 out_clear_bpool_bit:
2807 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2808 out_free_priv:
2809 	rte_free(priv);
2810 	return NULL;
2811 }
2812 
2813 /**
2814  * Create device representing Ethernet port.
2815  *
2816  * @param name
2817  *   Pointer to the port's name.
2818  *
2819  * @return
2820  *   0 on success, negative error value otherwise.
2821  */
2822 static int
2823 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
2824 {
2825 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
2826 	struct rte_eth_dev *eth_dev;
2827 	struct mrvl_priv *priv;
2828 	struct ifreq req;
2829 
2830 	eth_dev = rte_eth_dev_allocate(name);
2831 	if (!eth_dev)
2832 		return -ENOMEM;
2833 
2834 	priv = mrvl_priv_create(name);
2835 	if (!priv) {
2836 		ret = -ENOMEM;
2837 		goto out_free;
2838 	}
2839 	eth_dev->data->dev_private = priv;
2840 
2841 	eth_dev->data->mac_addrs =
2842 		rte_zmalloc("mac_addrs",
2843 			    RTE_ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
2844 	if (!eth_dev->data->mac_addrs) {
2845 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
2846 		ret = -ENOMEM;
2847 		goto out_free;
2848 	}
2849 
2850 	memset(&req, 0, sizeof(req));
2851 	strcpy(req.ifr_name, name);
2852 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
2853 	if (ret)
2854 		goto out_free;
2855 
2856 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
2857 	       req.ifr_addr.sa_data, RTE_ETHER_ADDR_LEN);
2858 
2859 	eth_dev->data->kdrv = RTE_KDRV_NONE;
2860 	eth_dev->device = &vdev->device;
2861 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
2862 	mrvl_set_tx_function(eth_dev);
2863 	eth_dev->dev_ops = &mrvl_ops;
2864 
2865 	/* Flag to call rte_eth_dev_release_port() in rte_eth_dev_close(). */
2866 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
2867 
2868 	rte_eth_dev_probing_finish(eth_dev);
2869 	return 0;
2870 out_free:
2871 	rte_eth_dev_release_port(eth_dev);
2872 
2873 	return ret;
2874 }
2875 
2876 /**
2877  * Callback used by rte_kvargs_process() during argument parsing.
2878  *
2879  * @param key
2880  *   Pointer to the parsed key (unused).
2881  * @param value
2882  *   Pointer to the parsed value.
2883  * @param extra_args
2884  *   Pointer to the extra arguments which contains address of the
2885  *   table of pointers to parsed interface names.
2886  *
2887  * @return
2888  *   Always 0.
2889  */
2890 static int
2891 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
2892 		 void *extra_args)
2893 {
2894 	struct mrvl_ifnames *ifnames = extra_args;
2895 
2896 	ifnames->names[ifnames->idx++] = value;
2897 
2898 	return 0;
2899 }
2900 
2901 /**
2902  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
2903  */
2904 static void
2905 mrvl_deinit_hifs(void)
2906 {
2907 	int i;
2908 
2909 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
2910 		if (hifs[i])
2911 			pp2_hif_deinit(hifs[i]);
2912 	}
2913 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
2914 	memset(hifs, 0, sizeof(hifs));
2915 }
2916 
2917 /**
2918  * DPDK callback to register the virtual device.
2919  *
2920  * @param vdev
2921  *   Pointer to the virtual device.
2922  *
2923  * @return
2924  *   0 on success, negative error value otherwise.
2925  */
2926 static int
2927 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
2928 {
2929 	struct rte_kvargs *kvlist;
2930 	struct mrvl_ifnames ifnames;
2931 	int ret = -EINVAL;
2932 	uint32_t i, ifnum, cfgnum;
2933 	const char *params;
2934 
2935 	params = rte_vdev_device_args(vdev);
2936 	if (!params)
2937 		return -EINVAL;
2938 
2939 	kvlist = rte_kvargs_parse(params, valid_args);
2940 	if (!kvlist)
2941 		return -EINVAL;
2942 
2943 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
2944 	if (ifnum > RTE_DIM(ifnames.names))
2945 		goto out_free_kvlist;
2946 
2947 	ifnames.idx = 0;
2948 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
2949 			   mrvl_get_ifnames, &ifnames);
2950 
2951 
2952 	/*
2953 	 * The below system initialization should be done only once,
2954 	 * on the first provided configuration file
2955 	 */
2956 	if (!mrvl_qos_cfg) {
2957 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
2958 		MRVL_LOG(INFO, "Parsing config file!");
2959 		if (cfgnum > 1) {
2960 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
2961 			goto out_free_kvlist;
2962 		} else if (cfgnum == 1) {
2963 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
2964 					   mrvl_get_qoscfg, &mrvl_qos_cfg);
2965 		}
2966 	}
2967 
2968 	if (mrvl_dev_num)
2969 		goto init_devices;
2970 
2971 	MRVL_LOG(INFO, "Perform MUSDK initializations");
2972 
2973 	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
2974 	if (ret)
2975 		goto out_free_kvlist;
2976 
2977 	ret = mrvl_init_pp2();
2978 	if (ret) {
2979 		MRVL_LOG(ERR, "Failed to init PP!");
2980 		rte_mvep_deinit(MVEP_MOD_T_PP2);
2981 		goto out_free_kvlist;
2982 	}
2983 
2984 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
2985 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
2986 
2987 	mrvl_lcore_first = RTE_MAX_LCORE;
2988 	mrvl_lcore_last = 0;
2989 
2990 init_devices:
2991 	for (i = 0; i < ifnum; i++) {
2992 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
2993 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
2994 		if (ret)
2995 			goto out_cleanup;
2996 		mrvl_dev_num++;
2997 	}
2998 
2999 	rte_kvargs_free(kvlist);
3000 
3001 	return 0;
3002 out_cleanup:
3003 	rte_pmd_mrvl_remove(vdev);
3004 
3005 out_free_kvlist:
3006 	rte_kvargs_free(kvlist);
3007 
3008 	return ret;
3009 }
3010 
3011 /**
3012  * DPDK callback to remove virtual device.
3013  *
3014  * @param vdev
3015  *   Pointer to the removed virtual device.
3016  *
3017  * @return
3018  *   0 on success, negative error value otherwise.
3019  */
3020 static int
3021 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
3022 {
3023 	uint16_t port_id;
3024 
3025 	RTE_ETH_FOREACH_DEV(port_id) {
3026 		if (rte_eth_devices[port_id].device != &vdev->device)
3027 			continue;
3028 		rte_eth_dev_close(port_id);
3029 	}
3030 
3031 	return 0;
3032 }
3033 
3034 static struct rte_vdev_driver pmd_mrvl_drv = {
3035 	.probe = rte_pmd_mrvl_probe,
3036 	.remove = rte_pmd_mrvl_remove,
3037 };
3038 
3039 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
3040 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
3041 RTE_LOG_REGISTER(mrvl_logtype, pmd.net.mvpp2, NOTICE);
3042