xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision 0ecc27f28d202a3356a8601e6762b601ea822c4c)
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 int mrvl_logtype;
88 
89 struct mrvl_ifnames {
90 	const char *names[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
91 	int idx;
92 };
93 
94 /*
95  * To use buffer harvesting based on loopback port shadow queue structure
96  * was introduced for buffers information bookkeeping.
97  *
98  * Before sending the packet, related buffer information (pp2_buff_inf) is
99  * stored in shadow queue. After packet is transmitted no longer used
100  * packet buffer is released back to it's original hardware pool,
101  * on condition it originated from interface.
102  * In case it  was generated by application itself i.e: mbuf->port field is
103  * 0xff then its released to software mempool.
104  */
105 struct mrvl_shadow_txq {
106 	int head;           /* write index - used when sending buffers */
107 	int tail;           /* read index - used when releasing buffers */
108 	u16 size;           /* queue occupied size */
109 	u16 num_to_release; /* number of descriptors sent, that can be
110 			     * released
111 			     */
112 	struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
113 };
114 
115 struct mrvl_rxq {
116 	struct mrvl_priv *priv;
117 	struct rte_mempool *mp;
118 	int queue_id;
119 	int port_id;
120 	int cksum_enabled;
121 	uint64_t bytes_recv;
122 	uint64_t drop_mac;
123 };
124 
125 struct mrvl_txq {
126 	struct mrvl_priv *priv;
127 	int queue_id;
128 	int port_id;
129 	uint64_t bytes_sent;
130 	struct mrvl_shadow_txq shadow_txqs[RTE_MAX_LCORE];
131 	int tx_deferred_start;
132 };
133 
134 static int mrvl_lcore_first;
135 static int mrvl_lcore_last;
136 static int mrvl_dev_num;
137 
138 static int mrvl_fill_bpool(struct mrvl_rxq *rxq, int num);
139 static inline void mrvl_free_sent_buffers(struct pp2_ppio *ppio,
140 			struct pp2_hif *hif, unsigned int core_id,
141 			struct mrvl_shadow_txq *sq, int qid, int force);
142 
143 static uint16_t mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
144 				  uint16_t nb_pkts);
145 static uint16_t mrvl_tx_sg_pkt_burst(void *txq,	struct rte_mbuf **tx_pkts,
146 				     uint16_t nb_pkts);
147 static int rte_pmd_mrvl_remove(struct rte_vdev_device *vdev);
148 static void mrvl_deinit_pp2(void);
149 static void mrvl_deinit_hifs(void);
150 
151 
152 #define MRVL_XSTATS_TBL_ENTRY(name) { \
153 	#name, offsetof(struct pp2_ppio_statistics, name),	\
154 	sizeof(((struct pp2_ppio_statistics *)0)->name)		\
155 }
156 
157 /* Table with xstats data */
158 static struct {
159 	const char *name;
160 	unsigned int offset;
161 	unsigned int size;
162 } mrvl_xstats_tbl[] = {
163 	MRVL_XSTATS_TBL_ENTRY(rx_bytes),
164 	MRVL_XSTATS_TBL_ENTRY(rx_packets),
165 	MRVL_XSTATS_TBL_ENTRY(rx_unicast_packets),
166 	MRVL_XSTATS_TBL_ENTRY(rx_errors),
167 	MRVL_XSTATS_TBL_ENTRY(rx_fullq_dropped),
168 	MRVL_XSTATS_TBL_ENTRY(rx_bm_dropped),
169 	MRVL_XSTATS_TBL_ENTRY(rx_early_dropped),
170 	MRVL_XSTATS_TBL_ENTRY(rx_fifo_dropped),
171 	MRVL_XSTATS_TBL_ENTRY(rx_cls_dropped),
172 	MRVL_XSTATS_TBL_ENTRY(tx_bytes),
173 	MRVL_XSTATS_TBL_ENTRY(tx_packets),
174 	MRVL_XSTATS_TBL_ENTRY(tx_unicast_packets),
175 	MRVL_XSTATS_TBL_ENTRY(tx_errors)
176 };
177 
178 static inline void
179 mrvl_fill_shadowq(struct mrvl_shadow_txq *sq, struct rte_mbuf *buf)
180 {
181 	sq->ent[sq->head].buff.cookie = (uint64_t)buf;
182 	sq->ent[sq->head].buff.addr = buf ?
183 		rte_mbuf_data_iova_default(buf) : 0;
184 
185 	sq->ent[sq->head].bpool =
186 		(unlikely(!buf || buf->port >= RTE_MAX_ETHPORTS ||
187 		 buf->refcnt > 1)) ? NULL :
188 		 mrvl_port_to_bpool_lookup[buf->port];
189 
190 	sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
191 	sq->size++;
192 }
193 
194 static inline void
195 mrvl_fill_desc(struct pp2_ppio_desc *desc, struct rte_mbuf *buf)
196 {
197 	pp2_ppio_outq_desc_reset(desc);
198 	pp2_ppio_outq_desc_set_phys_addr(desc, rte_pktmbuf_iova(buf));
199 	pp2_ppio_outq_desc_set_pkt_offset(desc, 0);
200 	pp2_ppio_outq_desc_set_pkt_len(desc, rte_pktmbuf_data_len(buf));
201 }
202 
203 static inline int
204 mrvl_get_bpool_size(int pp2_id, int pool_id)
205 {
206 	int i;
207 	int size = 0;
208 
209 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
210 		size += mrvl_port_bpool_size[pp2_id][pool_id][i];
211 
212 	return size;
213 }
214 
215 static inline int
216 mrvl_reserve_bit(int *bitmap, int max)
217 {
218 	int n = sizeof(*bitmap) * 8 - __builtin_clz(*bitmap);
219 
220 	if (n >= max)
221 		return -1;
222 
223 	*bitmap |= 1 << n;
224 
225 	return n;
226 }
227 
228 static int
229 mrvl_init_hif(int core_id)
230 {
231 	struct pp2_hif_params params;
232 	char match[MRVL_MATCH_LEN];
233 	int ret;
234 
235 	ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
236 	if (ret < 0) {
237 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
238 		return ret;
239 	}
240 
241 	snprintf(match, sizeof(match), "hif-%d", ret);
242 	memset(&params, 0, sizeof(params));
243 	params.match = match;
244 	params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
245 	ret = pp2_hif_init(&params, &hifs[core_id]);
246 	if (ret) {
247 		MRVL_LOG(ERR, "Failed to initialize hif %d", core_id);
248 		return ret;
249 	}
250 
251 	return 0;
252 }
253 
254 static inline struct pp2_hif*
255 mrvl_get_hif(struct mrvl_priv *priv, int core_id)
256 {
257 	int ret;
258 
259 	if (likely(hifs[core_id] != NULL))
260 		return hifs[core_id];
261 
262 	rte_spinlock_lock(&priv->lock);
263 
264 	ret = mrvl_init_hif(core_id);
265 	if (ret < 0) {
266 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
267 		goto out;
268 	}
269 
270 	if (core_id < mrvl_lcore_first)
271 		mrvl_lcore_first = core_id;
272 
273 	if (core_id > mrvl_lcore_last)
274 		mrvl_lcore_last = core_id;
275 out:
276 	rte_spinlock_unlock(&priv->lock);
277 
278 	return hifs[core_id];
279 }
280 
281 /**
282  * Set tx burst function according to offload flag
283  *
284  * @param dev
285  *   Pointer to Ethernet device structure.
286  */
287 static void
288 mrvl_set_tx_function(struct rte_eth_dev *dev)
289 {
290 	struct mrvl_priv *priv = dev->data->dev_private;
291 
292 	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
293 	if (priv->multiseg) {
294 		RTE_LOG(INFO, PMD, "Using multi-segment tx callback\n");
295 		dev->tx_pkt_burst = mrvl_tx_sg_pkt_burst;
296 	} else {
297 		RTE_LOG(INFO, PMD, "Using single-segment tx callback\n");
298 		dev->tx_pkt_burst = mrvl_tx_pkt_burst;
299 	}
300 }
301 
302 /**
303  * Configure rss based on dpdk rss configuration.
304  *
305  * @param priv
306  *   Pointer to private structure.
307  * @param rss_conf
308  *   Pointer to RSS configuration.
309  *
310  * @return
311  *   0 on success, negative error value otherwise.
312  */
313 static int
314 mrvl_configure_rss(struct mrvl_priv *priv, struct rte_eth_rss_conf *rss_conf)
315 {
316 	if (rss_conf->rss_key)
317 		MRVL_LOG(WARNING, "Changing hash key is not supported");
318 
319 	if (rss_conf->rss_hf == 0) {
320 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
321 	} else if (rss_conf->rss_hf & ETH_RSS_IPV4) {
322 		priv->ppio_params.inqs_params.hash_type =
323 			PP2_PPIO_HASH_T_2_TUPLE;
324 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP) {
325 		priv->ppio_params.inqs_params.hash_type =
326 			PP2_PPIO_HASH_T_5_TUPLE;
327 		priv->rss_hf_tcp = 1;
328 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) {
329 		priv->ppio_params.inqs_params.hash_type =
330 			PP2_PPIO_HASH_T_5_TUPLE;
331 		priv->rss_hf_tcp = 0;
332 	} else {
333 		return -EINVAL;
334 	}
335 
336 	return 0;
337 }
338 
339 /**
340  * Ethernet device configuration.
341  *
342  * Prepare the driver for a given number of TX and RX queues and
343  * configure RSS.
344  *
345  * @param dev
346  *   Pointer to Ethernet device structure.
347  *
348  * @return
349  *   0 on success, negative error value otherwise.
350  */
351 static int
352 mrvl_dev_configure(struct rte_eth_dev *dev)
353 {
354 	struct mrvl_priv *priv = dev->data->dev_private;
355 	int ret;
356 
357 	if (priv->ppio) {
358 		MRVL_LOG(INFO, "Device reconfiguration is not supported");
359 		return -EINVAL;
360 	}
361 
362 	if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_NONE &&
363 	    dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
364 		MRVL_LOG(INFO, "Unsupported rx multi queue mode %d",
365 			dev->data->dev_conf.rxmode.mq_mode);
366 		return -EINVAL;
367 	}
368 
369 	if (dev->data->dev_conf.rxmode.split_hdr_size) {
370 		MRVL_LOG(INFO, "Split headers not supported");
371 		return -EINVAL;
372 	}
373 
374 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
375 		dev->data->mtu = dev->data->dev_conf.rxmode.max_rx_pkt_len -
376 				 MRVL_PP2_ETH_HDRS_LEN;
377 
378 	if (dev->data->dev_conf.txmode.offloads & DEV_TX_OFFLOAD_MULTI_SEGS)
379 		priv->multiseg = 1;
380 
381 	ret = mrvl_configure_rxqs(priv, dev->data->port_id,
382 				  dev->data->nb_rx_queues);
383 	if (ret < 0)
384 		return ret;
385 
386 	ret = mrvl_configure_txqs(priv, dev->data->port_id,
387 				  dev->data->nb_tx_queues);
388 	if (ret < 0)
389 		return ret;
390 
391 	priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
392 	priv->ppio_params.maintain_stats = 1;
393 	priv->nb_rx_queues = dev->data->nb_rx_queues;
394 
395 	ret = mrvl_tm_init(dev);
396 	if (ret < 0)
397 		return ret;
398 
399 	if (dev->data->nb_rx_queues == 1 &&
400 	    dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
401 		MRVL_LOG(WARNING, "Disabling hash for 1 rx queue");
402 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
403 
404 		return 0;
405 	}
406 
407 	return mrvl_configure_rss(priv,
408 				  &dev->data->dev_conf.rx_adv_conf.rss_conf);
409 }
410 
411 /**
412  * DPDK callback to change the MTU.
413  *
414  * Setting the MTU affects hardware MRU (packets larger than the MRU
415  * will be dropped).
416  *
417  * @param dev
418  *   Pointer to Ethernet device structure.
419  * @param mtu
420  *   New MTU.
421  *
422  * @return
423  *   0 on success, negative error value otherwise.
424  */
425 static int
426 mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
427 {
428 	struct mrvl_priv *priv = dev->data->dev_private;
429 	uint16_t mru;
430 	uint16_t mbuf_data_size = 0; /* SW buffer size */
431 	int ret;
432 
433 	mru = MRVL_PP2_MTU_TO_MRU(mtu);
434 	/*
435 	 * min_rx_buf_size is equal to mbuf data size
436 	 * if pmd didn't set it differently
437 	 */
438 	mbuf_data_size = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
439 	/* Prevent PMD from:
440 	 * - setting mru greater than the mbuf size resulting in
441 	 * hw and sw buffer size mismatch
442 	 * - setting mtu that requires the support of scattered packets
443 	 * when this feature has not been enabled/supported so far
444 	 * (TODO check scattered_rx flag here once scattered RX is supported).
445 	 */
446 	if (mru + MRVL_PKT_OFFS > mbuf_data_size) {
447 		mru = mbuf_data_size - MRVL_PKT_OFFS;
448 		mtu = MRVL_PP2_MRU_TO_MTU(mru);
449 		MRVL_LOG(WARNING, "MTU too big, max MTU possible limitted "
450 			"by current mbuf size: %u. Set MTU to %u, MRU to %u",
451 			mbuf_data_size, mtu, mru);
452 	}
453 
454 	if (mtu < RTE_ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX) {
455 		MRVL_LOG(ERR, "Invalid MTU [%u] or MRU [%u]", mtu, mru);
456 		return -EINVAL;
457 	}
458 
459 	dev->data->mtu = mtu;
460 	dev->data->dev_conf.rxmode.max_rx_pkt_len = mru - MV_MH_SIZE;
461 
462 	if (!priv->ppio)
463 		return 0;
464 
465 	ret = pp2_ppio_set_mru(priv->ppio, mru);
466 	if (ret) {
467 		MRVL_LOG(ERR, "Failed to change MRU");
468 		return ret;
469 	}
470 
471 	ret = pp2_ppio_set_mtu(priv->ppio, mtu);
472 	if (ret) {
473 		MRVL_LOG(ERR, "Failed to change MTU");
474 		return ret;
475 	}
476 
477 	return 0;
478 }
479 
480 /**
481  * DPDK callback to bring the link up.
482  *
483  * @param dev
484  *   Pointer to Ethernet device structure.
485  *
486  * @return
487  *   0 on success, negative error value otherwise.
488  */
489 static int
490 mrvl_dev_set_link_up(struct rte_eth_dev *dev)
491 {
492 	struct mrvl_priv *priv = dev->data->dev_private;
493 	int ret;
494 
495 	if (!priv->ppio)
496 		return -EPERM;
497 
498 	ret = pp2_ppio_enable(priv->ppio);
499 	if (ret)
500 		return ret;
501 
502 	/*
503 	 * mtu/mru can be updated if pp2_ppio_enable() was called at least once
504 	 * as pp2_ppio_enable() changes port->t_mode from default 0 to
505 	 * PP2_TRAFFIC_INGRESS_EGRESS.
506 	 *
507 	 * Set mtu to default DPDK value here.
508 	 */
509 	ret = mrvl_mtu_set(dev, dev->data->mtu);
510 	if (ret)
511 		pp2_ppio_disable(priv->ppio);
512 
513 	return ret;
514 }
515 
516 /**
517  * DPDK callback to bring the link down.
518  *
519  * @param dev
520  *   Pointer to Ethernet device structure.
521  *
522  * @return
523  *   0 on success, negative error value otherwise.
524  */
525 static int
526 mrvl_dev_set_link_down(struct rte_eth_dev *dev)
527 {
528 	struct mrvl_priv *priv = dev->data->dev_private;
529 
530 	if (!priv->ppio)
531 		return -EPERM;
532 
533 	return pp2_ppio_disable(priv->ppio);
534 }
535 
536 /**
537  * DPDK callback to start tx queue.
538  *
539  * @param dev
540  *   Pointer to Ethernet device structure.
541  * @param queue_id
542  *   Transmit queue index.
543  *
544  * @return
545  *   0 on success, negative error value otherwise.
546  */
547 static int
548 mrvl_tx_queue_start(struct rte_eth_dev *dev, uint16_t queue_id)
549 {
550 	struct mrvl_priv *priv = dev->data->dev_private;
551 	int ret;
552 
553 	if (!priv)
554 		return -EPERM;
555 
556 	/* passing 1 enables given tx queue */
557 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 1);
558 	if (ret) {
559 		MRVL_LOG(ERR, "Failed to start txq %d", queue_id);
560 		return ret;
561 	}
562 
563 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
564 
565 	return 0;
566 }
567 
568 /**
569  * DPDK callback to stop tx queue.
570  *
571  * @param dev
572  *   Pointer to Ethernet device structure.
573  * @param queue_id
574  *   Transmit queue index.
575  *
576  * @return
577  *   0 on success, negative error value otherwise.
578  */
579 static int
580 mrvl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t queue_id)
581 {
582 	struct mrvl_priv *priv = dev->data->dev_private;
583 	int ret;
584 
585 	if (!priv->ppio)
586 		return -EPERM;
587 
588 	/* passing 0 disables given tx queue */
589 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 0);
590 	if (ret) {
591 		MRVL_LOG(ERR, "Failed to stop txq %d", queue_id);
592 		return ret;
593 	}
594 
595 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
596 
597 	return 0;
598 }
599 
600 /**
601  * DPDK callback to start the device.
602  *
603  * @param dev
604  *   Pointer to Ethernet device structure.
605  *
606  * @return
607  *   0 on success, negative errno value on failure.
608  */
609 static int
610 mrvl_dev_start(struct rte_eth_dev *dev)
611 {
612 	struct mrvl_priv *priv = dev->data->dev_private;
613 	char match[MRVL_MATCH_LEN];
614 	int ret = 0, i, def_init_size;
615 
616 	if (priv->ppio)
617 		return mrvl_dev_set_link_up(dev);
618 
619 	snprintf(match, sizeof(match), "ppio-%d:%d",
620 		 priv->pp_id, priv->ppio_id);
621 	priv->ppio_params.match = match;
622 
623 	/*
624 	 * Calculate the minimum bpool size for refill feature as follows:
625 	 * 2 default burst sizes multiply by number of rx queues.
626 	 * If the bpool size will be below this value, new buffers will
627 	 * be added to the pool.
628 	 */
629 	priv->bpool_min_size = priv->nb_rx_queues * MRVL_BURST_SIZE * 2;
630 
631 	/* In case initial bpool size configured in queues setup is
632 	 * smaller than minimum size add more buffers
633 	 */
634 	def_init_size = priv->bpool_min_size + MRVL_BURST_SIZE * 2;
635 	if (priv->bpool_init_size < def_init_size) {
636 		int buffs_to_add = def_init_size - priv->bpool_init_size;
637 
638 		priv->bpool_init_size += buffs_to_add;
639 		ret = mrvl_fill_bpool(dev->data->rx_queues[0], buffs_to_add);
640 		if (ret)
641 			MRVL_LOG(ERR, "Failed to add buffers to bpool");
642 	}
643 
644 	/*
645 	 * Calculate the maximum bpool size for refill feature as follows:
646 	 * maximum number of descriptors in rx queue multiply by number
647 	 * of rx queues plus minimum bpool size.
648 	 * In case the bpool size will exceed this value, superfluous buffers
649 	 * will be removed
650 	 */
651 	priv->bpool_max_size = (priv->nb_rx_queues * MRVL_PP2_RXD_MAX) +
652 				priv->bpool_min_size;
653 
654 	ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
655 	if (ret) {
656 		MRVL_LOG(ERR, "Failed to init ppio");
657 		return ret;
658 	}
659 
660 	/*
661 	 * In case there are some some stale uc/mc mac addresses flush them
662 	 * here. It cannot be done during mrvl_dev_close() as port information
663 	 * is already gone at that point (due to pp2_ppio_deinit() in
664 	 * mrvl_dev_stop()).
665 	 */
666 	if (!priv->uc_mc_flushed) {
667 		ret = pp2_ppio_flush_mac_addrs(priv->ppio, 1, 1);
668 		if (ret) {
669 			MRVL_LOG(ERR,
670 				"Failed to flush uc/mc filter list");
671 			goto out;
672 		}
673 		priv->uc_mc_flushed = 1;
674 	}
675 
676 	if (!priv->vlan_flushed) {
677 		ret = pp2_ppio_flush_vlan(priv->ppio);
678 		if (ret) {
679 			MRVL_LOG(ERR, "Failed to flush vlan list");
680 			/*
681 			 * TODO
682 			 * once pp2_ppio_flush_vlan() is supported jump to out
683 			 * goto out;
684 			 */
685 		}
686 		priv->vlan_flushed = 1;
687 	}
688 	ret = mrvl_mtu_set(dev, dev->data->mtu);
689 	if (ret)
690 		MRVL_LOG(ERR, "Failed to set MTU to %d", dev->data->mtu);
691 
692 	/* For default QoS config, don't start classifier. */
693 	if (mrvl_qos_cfg  &&
694 	    mrvl_qos_cfg->port[dev->data->port_id].use_global_defaults == 0) {
695 		ret = mrvl_start_qos_mapping(priv);
696 		if (ret) {
697 			MRVL_LOG(ERR, "Failed to setup QoS mapping");
698 			goto out;
699 		}
700 	}
701 
702 	ret = mrvl_dev_set_link_up(dev);
703 	if (ret) {
704 		MRVL_LOG(ERR, "Failed to set link up");
705 		goto out;
706 	}
707 
708 	/* start tx queues */
709 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
710 		struct mrvl_txq *txq = dev->data->tx_queues[i];
711 
712 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
713 
714 		if (!txq->tx_deferred_start)
715 			continue;
716 
717 		/*
718 		 * All txqs are started by default. Stop them
719 		 * so that tx_deferred_start works as expected.
720 		 */
721 		ret = mrvl_tx_queue_stop(dev, i);
722 		if (ret)
723 			goto out;
724 	}
725 
726 	mrvl_flow_init(dev);
727 	mrvl_mtr_init(dev);
728 	mrvl_set_tx_function(dev);
729 
730 	return 0;
731 out:
732 	MRVL_LOG(ERR, "Failed to start device");
733 	pp2_ppio_deinit(priv->ppio);
734 	return ret;
735 }
736 
737 /**
738  * Flush receive queues.
739  *
740  * @param dev
741  *   Pointer to Ethernet device structure.
742  */
743 static void
744 mrvl_flush_rx_queues(struct rte_eth_dev *dev)
745 {
746 	int i;
747 
748 	MRVL_LOG(INFO, "Flushing rx queues");
749 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
750 		int ret, num;
751 
752 		do {
753 			struct mrvl_rxq *q = dev->data->rx_queues[i];
754 			struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];
755 
756 			num = MRVL_PP2_RXD_MAX;
757 			ret = pp2_ppio_recv(q->priv->ppio,
758 					    q->priv->rxq_map[q->queue_id].tc,
759 					    q->priv->rxq_map[q->queue_id].inq,
760 					    descs, (uint16_t *)&num);
761 		} while (ret == 0 && num);
762 	}
763 }
764 
765 /**
766  * Flush transmit shadow queues.
767  *
768  * @param dev
769  *   Pointer to Ethernet device structure.
770  */
771 static void
772 mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
773 {
774 	int i, j;
775 	struct mrvl_txq *txq;
776 
777 	MRVL_LOG(INFO, "Flushing tx shadow queues");
778 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
779 		txq = (struct mrvl_txq *)dev->data->tx_queues[i];
780 
781 		for (j = 0; j < RTE_MAX_LCORE; j++) {
782 			struct mrvl_shadow_txq *sq;
783 
784 			if (!hifs[j])
785 				continue;
786 
787 			sq = &txq->shadow_txqs[j];
788 			mrvl_free_sent_buffers(txq->priv->ppio,
789 				hifs[j], j, sq, txq->queue_id, 1);
790 			while (sq->tail != sq->head) {
791 				uint64_t addr = cookie_addr_high |
792 					sq->ent[sq->tail].buff.cookie;
793 				rte_pktmbuf_free(
794 					(struct rte_mbuf *)addr);
795 				sq->tail = (sq->tail + 1) &
796 					    MRVL_PP2_TX_SHADOWQ_MASK;
797 			}
798 			memset(sq, 0, sizeof(*sq));
799 		}
800 	}
801 }
802 
803 /**
804  * Flush hardware bpool (buffer-pool).
805  *
806  * @param dev
807  *   Pointer to Ethernet device structure.
808  */
809 static void
810 mrvl_flush_bpool(struct rte_eth_dev *dev)
811 {
812 	struct mrvl_priv *priv = dev->data->dev_private;
813 	struct pp2_hif *hif;
814 	uint32_t num;
815 	int ret;
816 	unsigned int core_id = rte_lcore_id();
817 
818 	if (core_id == LCORE_ID_ANY)
819 		core_id = 0;
820 
821 	hif = mrvl_get_hif(priv, core_id);
822 
823 	ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
824 	if (ret) {
825 		MRVL_LOG(ERR, "Failed to get bpool buffers number");
826 		return;
827 	}
828 
829 	while (num--) {
830 		struct pp2_buff_inf inf;
831 		uint64_t addr;
832 
833 		ret = pp2_bpool_get_buff(hif, priv->bpool, &inf);
834 		if (ret)
835 			break;
836 
837 		addr = cookie_addr_high | inf.cookie;
838 		rte_pktmbuf_free((struct rte_mbuf *)addr);
839 	}
840 }
841 
842 /**
843  * DPDK callback to stop the device.
844  *
845  * @param dev
846  *   Pointer to Ethernet device structure.
847  */
848 static void
849 mrvl_dev_stop(struct rte_eth_dev *dev)
850 {
851 	mrvl_dev_set_link_down(dev);
852 }
853 
854 /**
855  * DPDK callback to close the device.
856  *
857  * @param dev
858  *   Pointer to Ethernet device structure.
859  */
860 static void
861 mrvl_dev_close(struct rte_eth_dev *dev)
862 {
863 	struct mrvl_priv *priv = dev->data->dev_private;
864 	size_t i;
865 
866 	mrvl_flush_rx_queues(dev);
867 	mrvl_flush_tx_shadow_queues(dev);
868 	mrvl_flow_deinit(dev);
869 	mrvl_mtr_deinit(dev);
870 
871 	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
872 		struct pp2_ppio_tc_params *tc_params =
873 			&priv->ppio_params.inqs_params.tcs_params[i];
874 
875 		if (tc_params->inqs_params) {
876 			rte_free(tc_params->inqs_params);
877 			tc_params->inqs_params = NULL;
878 		}
879 	}
880 
881 	if (priv->cls_tbl) {
882 		pp2_cls_tbl_deinit(priv->cls_tbl);
883 		priv->cls_tbl = NULL;
884 	}
885 
886 	if (priv->qos_tbl) {
887 		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
888 		priv->qos_tbl = NULL;
889 	}
890 
891 	mrvl_flush_bpool(dev);
892 	mrvl_tm_deinit(dev);
893 
894 	if (priv->ppio) {
895 		pp2_ppio_deinit(priv->ppio);
896 		priv->ppio = NULL;
897 	}
898 
899 	/* policer must be released after ppio deinitialization */
900 	if (priv->default_policer) {
901 		pp2_cls_plcr_deinit(priv->default_policer);
902 		priv->default_policer = NULL;
903 	}
904 
905 
906 	if (priv->bpool) {
907 		pp2_bpool_deinit(priv->bpool);
908 		used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
909 		priv->bpool = NULL;
910 	}
911 
912 	mrvl_dev_num--;
913 
914 	if (mrvl_dev_num == 0) {
915 		MRVL_LOG(INFO, "Perform MUSDK deinit");
916 		mrvl_deinit_hifs();
917 		mrvl_deinit_pp2();
918 		rte_mvep_deinit(MVEP_MOD_T_PP2);
919 	}
920 }
921 
922 /**
923  * DPDK callback to retrieve physical link information.
924  *
925  * @param dev
926  *   Pointer to Ethernet device structure.
927  * @param wait_to_complete
928  *   Wait for request completion (ignored).
929  *
930  * @return
931  *   0 on success, negative error value otherwise.
932  */
933 static int
934 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
935 {
936 	/*
937 	 * TODO
938 	 * once MUSDK provides necessary API use it here
939 	 */
940 	struct mrvl_priv *priv = dev->data->dev_private;
941 	struct ethtool_cmd edata;
942 	struct ifreq req;
943 	int ret, fd, link_up;
944 
945 	if (!priv->ppio)
946 		return -EPERM;
947 
948 	edata.cmd = ETHTOOL_GSET;
949 
950 	strcpy(req.ifr_name, dev->data->name);
951 	req.ifr_data = (void *)&edata;
952 
953 	fd = socket(AF_INET, SOCK_DGRAM, 0);
954 	if (fd == -1)
955 		return -EFAULT;
956 
957 	ret = ioctl(fd, SIOCETHTOOL, &req);
958 	if (ret == -1) {
959 		close(fd);
960 		return -EFAULT;
961 	}
962 
963 	close(fd);
964 
965 	switch (ethtool_cmd_speed(&edata)) {
966 	case SPEED_10:
967 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
968 		break;
969 	case SPEED_100:
970 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
971 		break;
972 	case SPEED_1000:
973 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
974 		break;
975 	case SPEED_10000:
976 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
977 		break;
978 	default:
979 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
980 	}
981 
982 	dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
983 							 ETH_LINK_HALF_DUPLEX;
984 	dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
985 							   ETH_LINK_FIXED;
986 	pp2_ppio_get_link_state(priv->ppio, &link_up);
987 	dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
988 
989 	return 0;
990 }
991 
992 /**
993  * DPDK callback to enable promiscuous mode.
994  *
995  * @param dev
996  *   Pointer to Ethernet device structure.
997  *
998  * @return
999  *   0 on success, negative error value otherwise.
1000  */
1001 static int
1002 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
1003 {
1004 	struct mrvl_priv *priv = dev->data->dev_private;
1005 	int ret;
1006 
1007 	if (!priv->ppio)
1008 		return 0;
1009 
1010 	if (priv->isolated)
1011 		return 0;
1012 
1013 	ret = pp2_ppio_set_promisc(priv->ppio, 1);
1014 	if (ret) {
1015 		MRVL_LOG(ERR, "Failed to enable promiscuous mode");
1016 		return -EAGAIN;
1017 	}
1018 
1019 	return 0;
1020 }
1021 
1022 /**
1023  * DPDK callback to enable allmulti mode.
1024  *
1025  * @param dev
1026  *   Pointer to Ethernet device structure.
1027  *
1028  * @return
1029  *   0 on success, negative error value otherwise.
1030  */
1031 static int
1032 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
1033 {
1034 	struct mrvl_priv *priv = dev->data->dev_private;
1035 	int ret;
1036 
1037 	if (!priv->ppio)
1038 		return 0;
1039 
1040 	if (priv->isolated)
1041 		return 0;
1042 
1043 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
1044 	if (ret) {
1045 		MRVL_LOG(ERR, "Failed enable all-multicast mode");
1046 		return -EAGAIN;
1047 	}
1048 
1049 	return 0;
1050 }
1051 
1052 /**
1053  * DPDK callback to disable promiscuous mode.
1054  *
1055  * @param dev
1056  *   Pointer to Ethernet device structure.
1057  *
1058  * @return
1059  *   0 on success, negative error value otherwise.
1060  */
1061 static int
1062 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
1063 {
1064 	struct mrvl_priv *priv = dev->data->dev_private;
1065 	int ret;
1066 
1067 	if (!priv->ppio)
1068 		return 0;
1069 
1070 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
1071 	if (ret) {
1072 		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
1073 		return -EAGAIN;
1074 	}
1075 
1076 	return 0;
1077 }
1078 
1079 /**
1080  * DPDK callback to disable allmulticast mode.
1081  *
1082  * @param dev
1083  *   Pointer to Ethernet device structure.
1084  *
1085  * @return
1086  *   0 on success, negative error value otherwise.
1087  */
1088 static int
1089 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
1090 {
1091 	struct mrvl_priv *priv = dev->data->dev_private;
1092 	int ret;
1093 
1094 	if (!priv->ppio)
1095 		return 0;
1096 
1097 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
1098 	if (ret) {
1099 		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
1100 		return -EAGAIN;
1101 	}
1102 
1103 	return 0;
1104 }
1105 
1106 /**
1107  * DPDK callback to remove a MAC address.
1108  *
1109  * @param dev
1110  *   Pointer to Ethernet device structure.
1111  * @param index
1112  *   MAC address index.
1113  */
1114 static void
1115 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1116 {
1117 	struct mrvl_priv *priv = dev->data->dev_private;
1118 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1119 	int ret;
1120 
1121 	if (!priv->ppio)
1122 		return;
1123 
1124 	if (priv->isolated)
1125 		return;
1126 
1127 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
1128 				       dev->data->mac_addrs[index].addr_bytes);
1129 	if (ret) {
1130 		rte_ether_format_addr(buf, sizeof(buf),
1131 				  &dev->data->mac_addrs[index]);
1132 		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
1133 	}
1134 }
1135 
1136 /**
1137  * DPDK callback to add a MAC address.
1138  *
1139  * @param dev
1140  *   Pointer to Ethernet device structure.
1141  * @param mac_addr
1142  *   MAC address to register.
1143  * @param index
1144  *   MAC address index.
1145  * @param vmdq
1146  *   VMDq pool index to associate address with (unused).
1147  *
1148  * @return
1149  *   0 on success, negative error value otherwise.
1150  */
1151 static int
1152 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1153 		  uint32_t index, uint32_t vmdq __rte_unused)
1154 {
1155 	struct mrvl_priv *priv = dev->data->dev_private;
1156 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1157 	int ret;
1158 
1159 	if (priv->isolated)
1160 		return -ENOTSUP;
1161 
1162 	if (index == 0)
1163 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1164 		return -1;
1165 
1166 	if (!priv->ppio)
1167 		return 0;
1168 
1169 	/*
1170 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1171 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1172 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1173 	 * 21 respectively.
1174 	 *
1175 	 * If more than uc_filter_max uc addresses were added to filter list
1176 	 * then NIC will switch to promiscuous mode automatically.
1177 	 *
1178 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1179 	 * were added to filter list then NIC will switch to all-multicast mode
1180 	 * automatically.
1181 	 */
1182 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1183 	if (ret) {
1184 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1185 		MRVL_LOG(ERR, "Failed to add mac %s", buf);
1186 		return -1;
1187 	}
1188 
1189 	return 0;
1190 }
1191 
1192 /**
1193  * DPDK callback to set the primary MAC address.
1194  *
1195  * @param dev
1196  *   Pointer to Ethernet device structure.
1197  * @param mac_addr
1198  *   MAC address to register.
1199  *
1200  * @return
1201  *   0 on success, negative error value otherwise.
1202  */
1203 static int
1204 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1205 {
1206 	struct mrvl_priv *priv = dev->data->dev_private;
1207 	int ret;
1208 
1209 	if (!priv->ppio)
1210 		return 0;
1211 
1212 	if (priv->isolated)
1213 		return -ENOTSUP;
1214 
1215 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1216 	if (ret) {
1217 		char buf[RTE_ETHER_ADDR_FMT_SIZE];
1218 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1219 		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1220 	}
1221 
1222 	return ret;
1223 }
1224 
1225 /**
1226  * DPDK callback to get device statistics.
1227  *
1228  * @param dev
1229  *   Pointer to Ethernet device structure.
1230  * @param stats
1231  *   Stats structure output buffer.
1232  *
1233  * @return
1234  *   0 on success, negative error value otherwise.
1235  */
1236 static int
1237 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1238 {
1239 	struct mrvl_priv *priv = dev->data->dev_private;
1240 	struct pp2_ppio_statistics ppio_stats;
1241 	uint64_t drop_mac = 0;
1242 	unsigned int i, idx, ret;
1243 
1244 	if (!priv->ppio)
1245 		return -EPERM;
1246 
1247 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1248 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1249 		struct pp2_ppio_inq_statistics rx_stats;
1250 
1251 		if (!rxq)
1252 			continue;
1253 
1254 		idx = rxq->queue_id;
1255 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1256 			MRVL_LOG(ERR,
1257 				"rx queue %d stats out of range (0 - %d)",
1258 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1259 			continue;
1260 		}
1261 
1262 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1263 						  priv->rxq_map[idx].tc,
1264 						  priv->rxq_map[idx].inq,
1265 						  &rx_stats, 0);
1266 		if (unlikely(ret)) {
1267 			MRVL_LOG(ERR,
1268 				"Failed to update rx queue %d stats", idx);
1269 			break;
1270 		}
1271 
1272 		stats->q_ibytes[idx] = rxq->bytes_recv;
1273 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1274 		stats->q_errors[idx] = rx_stats.drop_early +
1275 				       rx_stats.drop_fullq +
1276 				       rx_stats.drop_bm +
1277 				       rxq->drop_mac;
1278 		stats->ibytes += rxq->bytes_recv;
1279 		drop_mac += rxq->drop_mac;
1280 	}
1281 
1282 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1283 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1284 		struct pp2_ppio_outq_statistics tx_stats;
1285 
1286 		if (!txq)
1287 			continue;
1288 
1289 		idx = txq->queue_id;
1290 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1291 			MRVL_LOG(ERR,
1292 				"tx queue %d stats out of range (0 - %d)",
1293 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1294 		}
1295 
1296 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1297 						   &tx_stats, 0);
1298 		if (unlikely(ret)) {
1299 			MRVL_LOG(ERR,
1300 				"Failed to update tx queue %d stats", idx);
1301 			break;
1302 		}
1303 
1304 		stats->q_opackets[idx] = tx_stats.deq_desc;
1305 		stats->q_obytes[idx] = txq->bytes_sent;
1306 		stats->obytes += txq->bytes_sent;
1307 	}
1308 
1309 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1310 	if (unlikely(ret)) {
1311 		MRVL_LOG(ERR, "Failed to update port statistics");
1312 		return ret;
1313 	}
1314 
1315 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1316 	stats->opackets += ppio_stats.tx_packets;
1317 	stats->imissed += ppio_stats.rx_fullq_dropped +
1318 			  ppio_stats.rx_bm_dropped +
1319 			  ppio_stats.rx_early_dropped +
1320 			  ppio_stats.rx_fifo_dropped +
1321 			  ppio_stats.rx_cls_dropped;
1322 	stats->ierrors = drop_mac;
1323 
1324 	return 0;
1325 }
1326 
1327 /**
1328  * DPDK callback to clear device statistics.
1329  *
1330  * @param dev
1331  *   Pointer to Ethernet device structure.
1332  *
1333  * @return
1334  *   0 on success, negative error value otherwise.
1335  */
1336 static int
1337 mrvl_stats_reset(struct rte_eth_dev *dev)
1338 {
1339 	struct mrvl_priv *priv = dev->data->dev_private;
1340 	int i;
1341 
1342 	if (!priv->ppio)
1343 		return 0;
1344 
1345 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1346 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1347 
1348 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1349 					    priv->rxq_map[i].inq, NULL, 1);
1350 		rxq->bytes_recv = 0;
1351 		rxq->drop_mac = 0;
1352 	}
1353 
1354 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1355 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1356 
1357 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1358 		txq->bytes_sent = 0;
1359 	}
1360 
1361 	return pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1362 }
1363 
1364 /**
1365  * DPDK callback to get extended statistics.
1366  *
1367  * @param dev
1368  *   Pointer to Ethernet device structure.
1369  * @param stats
1370  *   Pointer to xstats table.
1371  * @param n
1372  *   Number of entries in xstats table.
1373  * @return
1374  *   Negative value on error, number of read xstats otherwise.
1375  */
1376 static int
1377 mrvl_xstats_get(struct rte_eth_dev *dev,
1378 		struct rte_eth_xstat *stats, unsigned int n)
1379 {
1380 	struct mrvl_priv *priv = dev->data->dev_private;
1381 	struct pp2_ppio_statistics ppio_stats;
1382 	unsigned int i;
1383 
1384 	if (!stats)
1385 		return 0;
1386 
1387 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1388 	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1389 		uint64_t val;
1390 
1391 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1392 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1393 					    mrvl_xstats_tbl[i].offset);
1394 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1395 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1396 					    mrvl_xstats_tbl[i].offset);
1397 		else
1398 			return -EINVAL;
1399 
1400 		stats[i].id = i;
1401 		stats[i].value = val;
1402 	}
1403 
1404 	return n;
1405 }
1406 
1407 /**
1408  * DPDK callback to reset extended statistics.
1409  *
1410  * @param dev
1411  *   Pointer to Ethernet device structure.
1412  *
1413  * @return
1414  *   0 on success, negative error value otherwise.
1415  */
1416 static int
1417 mrvl_xstats_reset(struct rte_eth_dev *dev)
1418 {
1419 	return mrvl_stats_reset(dev);
1420 }
1421 
1422 /**
1423  * DPDK callback to get extended statistics names.
1424  *
1425  * @param dev (unused)
1426  *   Pointer to Ethernet device structure.
1427  * @param xstats_names
1428  *   Pointer to xstats names table.
1429  * @param size
1430  *   Size of the xstats names table.
1431  * @return
1432  *   Number of read names.
1433  */
1434 static int
1435 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1436 		      struct rte_eth_xstat_name *xstats_names,
1437 		      unsigned int size)
1438 {
1439 	unsigned int i;
1440 
1441 	if (!xstats_names)
1442 		return RTE_DIM(mrvl_xstats_tbl);
1443 
1444 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1445 		strlcpy(xstats_names[i].name, mrvl_xstats_tbl[i].name,
1446 			RTE_ETH_XSTATS_NAME_SIZE);
1447 
1448 	return size;
1449 }
1450 
1451 /**
1452  * DPDK callback to get information about the device.
1453  *
1454  * @param dev
1455  *   Pointer to Ethernet device structure (unused).
1456  * @param info
1457  *   Info structure output buffer.
1458  */
1459 static int
1460 mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
1461 		   struct rte_eth_dev_info *info)
1462 {
1463 	info->speed_capa = ETH_LINK_SPEED_10M |
1464 			   ETH_LINK_SPEED_100M |
1465 			   ETH_LINK_SPEED_1G |
1466 			   ETH_LINK_SPEED_10G;
1467 
1468 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1469 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1470 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1471 
1472 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1473 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1474 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1475 
1476 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1477 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1478 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1479 
1480 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1481 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1482 
1483 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1484 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1485 
1486 	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1487 				       ETH_RSS_NONFRAG_IPV4_TCP |
1488 				       ETH_RSS_NONFRAG_IPV4_UDP;
1489 
1490 	/* By default packets are dropped if no descriptors are available */
1491 	info->default_rxconf.rx_drop_en = 1;
1492 
1493 	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1494 
1495 	return 0;
1496 }
1497 
1498 /**
1499  * Return supported packet types.
1500  *
1501  * @param dev
1502  *   Pointer to Ethernet device structure (unused).
1503  *
1504  * @return
1505  *   Const pointer to the table with supported packet types.
1506  */
1507 static const uint32_t *
1508 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1509 {
1510 	static const uint32_t ptypes[] = {
1511 		RTE_PTYPE_L2_ETHER,
1512 		RTE_PTYPE_L2_ETHER_VLAN,
1513 		RTE_PTYPE_L2_ETHER_QINQ,
1514 		RTE_PTYPE_L3_IPV4,
1515 		RTE_PTYPE_L3_IPV4_EXT,
1516 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1517 		RTE_PTYPE_L3_IPV6,
1518 		RTE_PTYPE_L3_IPV6_EXT,
1519 		RTE_PTYPE_L2_ETHER_ARP,
1520 		RTE_PTYPE_L4_TCP,
1521 		RTE_PTYPE_L4_UDP
1522 	};
1523 
1524 	return ptypes;
1525 }
1526 
1527 /**
1528  * DPDK callback to get information about specific receive queue.
1529  *
1530  * @param dev
1531  *   Pointer to Ethernet device structure.
1532  * @param rx_queue_id
1533  *   Receive queue index.
1534  * @param qinfo
1535  *   Receive queue information structure.
1536  */
1537 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1538 			      struct rte_eth_rxq_info *qinfo)
1539 {
1540 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1541 	struct mrvl_priv *priv = dev->data->dev_private;
1542 	int inq = priv->rxq_map[rx_queue_id].inq;
1543 	int tc = priv->rxq_map[rx_queue_id].tc;
1544 	struct pp2_ppio_tc_params *tc_params =
1545 		&priv->ppio_params.inqs_params.tcs_params[tc];
1546 
1547 	qinfo->mp = q->mp;
1548 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1549 }
1550 
1551 /**
1552  * DPDK callback to get information about specific transmit queue.
1553  *
1554  * @param dev
1555  *   Pointer to Ethernet device structure.
1556  * @param tx_queue_id
1557  *   Transmit queue index.
1558  * @param qinfo
1559  *   Transmit queue information structure.
1560  */
1561 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1562 			      struct rte_eth_txq_info *qinfo)
1563 {
1564 	struct mrvl_priv *priv = dev->data->dev_private;
1565 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1566 
1567 	qinfo->nb_desc =
1568 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1569 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1570 }
1571 
1572 /**
1573  * DPDK callback to Configure a VLAN filter.
1574  *
1575  * @param dev
1576  *   Pointer to Ethernet device structure.
1577  * @param vlan_id
1578  *   VLAN ID to filter.
1579  * @param on
1580  *   Toggle filter.
1581  *
1582  * @return
1583  *   0 on success, negative error value otherwise.
1584  */
1585 static int
1586 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1587 {
1588 	struct mrvl_priv *priv = dev->data->dev_private;
1589 
1590 	if (!priv->ppio)
1591 		return -EPERM;
1592 
1593 	if (priv->isolated)
1594 		return -ENOTSUP;
1595 
1596 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1597 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1598 }
1599 
1600 /**
1601  * Release buffers to hardware bpool (buffer-pool)
1602  *
1603  * @param rxq
1604  *   Receive queue pointer.
1605  * @param num
1606  *   Number of buffers to release to bpool.
1607  *
1608  * @return
1609  *   0 on success, negative error value otherwise.
1610  */
1611 static int
1612 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1613 {
1614 	struct buff_release_entry entries[MRVL_PP2_RXD_MAX];
1615 	struct rte_mbuf *mbufs[MRVL_PP2_RXD_MAX];
1616 	int i, ret;
1617 	unsigned int core_id;
1618 	struct pp2_hif *hif;
1619 	struct pp2_bpool *bpool;
1620 
1621 	core_id = rte_lcore_id();
1622 	if (core_id == LCORE_ID_ANY)
1623 		core_id = 0;
1624 
1625 	hif = mrvl_get_hif(rxq->priv, core_id);
1626 	if (!hif)
1627 		return -1;
1628 
1629 	bpool = rxq->priv->bpool;
1630 
1631 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1632 	if (ret)
1633 		return ret;
1634 
1635 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1636 		cookie_addr_high =
1637 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1638 
1639 	for (i = 0; i < num; i++) {
1640 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1641 			!= cookie_addr_high) {
1642 			MRVL_LOG(ERR,
1643 				"mbuf virtual addr high 0x%lx out of range",
1644 				(uint64_t)mbufs[i] >> 32);
1645 			goto out;
1646 		}
1647 
1648 		entries[i].buff.addr =
1649 			rte_mbuf_data_iova_default(mbufs[i]);
1650 		entries[i].buff.cookie = (uint64_t)mbufs[i];
1651 		entries[i].bpool = bpool;
1652 	}
1653 
1654 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1655 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1656 
1657 	if (i != num)
1658 		goto out;
1659 
1660 	return 0;
1661 out:
1662 	for (; i < num; i++)
1663 		rte_pktmbuf_free(mbufs[i]);
1664 
1665 	return -1;
1666 }
1667 
1668 /**
1669  * DPDK callback to configure the receive queue.
1670  *
1671  * @param dev
1672  *   Pointer to Ethernet device structure.
1673  * @param idx
1674  *   RX queue index.
1675  * @param desc
1676  *   Number of descriptors to configure in queue.
1677  * @param socket
1678  *   NUMA socket on which memory must be allocated.
1679  * @param conf
1680  *   Thresholds parameters.
1681  * @param mp
1682  *   Memory pool for buffer allocations.
1683  *
1684  * @return
1685  *   0 on success, negative error value otherwise.
1686  */
1687 static int
1688 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1689 		    unsigned int socket,
1690 		    const struct rte_eth_rxconf *conf,
1691 		    struct rte_mempool *mp)
1692 {
1693 	struct mrvl_priv *priv = dev->data->dev_private;
1694 	struct mrvl_rxq *rxq;
1695 	uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
1696 	uint32_t max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1697 	int ret, tc, inq;
1698 	uint64_t offloads;
1699 
1700 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
1701 
1702 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1703 		/*
1704 		 * Unknown TC mapping, mapping will not have a correct queue.
1705 		 */
1706 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
1707 			idx, priv->ppio_id);
1708 		return -EFAULT;
1709 	}
1710 
1711 	frame_size = buf_size - RTE_PKTMBUF_HEADROOM - MRVL_PKT_EFFEC_OFFS;
1712 	if (frame_size < max_rx_pkt_len) {
1713 		MRVL_LOG(WARNING,
1714 			"Mbuf size must be increased to %u bytes to hold up "
1715 			"to %u bytes of data.",
1716 			buf_size + max_rx_pkt_len - frame_size,
1717 			max_rx_pkt_len);
1718 		dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
1719 		MRVL_LOG(INFO, "Setting max rx pkt len to %u",
1720 			dev->data->dev_conf.rxmode.max_rx_pkt_len);
1721 	}
1722 
1723 	if (dev->data->rx_queues[idx]) {
1724 		rte_free(dev->data->rx_queues[idx]);
1725 		dev->data->rx_queues[idx] = NULL;
1726 	}
1727 
1728 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
1729 	if (!rxq)
1730 		return -ENOMEM;
1731 
1732 	rxq->priv = priv;
1733 	rxq->mp = mp;
1734 	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
1735 	rxq->queue_id = idx;
1736 	rxq->port_id = dev->data->port_id;
1737 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
1738 
1739 	tc = priv->rxq_map[rxq->queue_id].tc,
1740 	inq = priv->rxq_map[rxq->queue_id].inq;
1741 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
1742 		desc;
1743 
1744 	ret = mrvl_fill_bpool(rxq, desc);
1745 	if (ret) {
1746 		rte_free(rxq);
1747 		return ret;
1748 	}
1749 
1750 	priv->bpool_init_size += desc;
1751 
1752 	dev->data->rx_queues[idx] = rxq;
1753 
1754 	return 0;
1755 }
1756 
1757 /**
1758  * DPDK callback to release the receive queue.
1759  *
1760  * @param rxq
1761  *   Generic receive queue pointer.
1762  */
1763 static void
1764 mrvl_rx_queue_release(void *rxq)
1765 {
1766 	struct mrvl_rxq *q = rxq;
1767 	struct pp2_ppio_tc_params *tc_params;
1768 	int i, num, tc, inq;
1769 	struct pp2_hif *hif;
1770 	unsigned int core_id = rte_lcore_id();
1771 
1772 	if (core_id == LCORE_ID_ANY)
1773 		core_id = 0;
1774 
1775 	if (!q)
1776 		return;
1777 
1778 	hif = mrvl_get_hif(q->priv, core_id);
1779 
1780 	if (!hif)
1781 		return;
1782 
1783 	tc = q->priv->rxq_map[q->queue_id].tc;
1784 	inq = q->priv->rxq_map[q->queue_id].inq;
1785 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
1786 	num = tc_params->inqs_params[inq].size;
1787 	for (i = 0; i < num; i++) {
1788 		struct pp2_buff_inf inf;
1789 		uint64_t addr;
1790 
1791 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
1792 		addr = cookie_addr_high | inf.cookie;
1793 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1794 	}
1795 
1796 	rte_free(q);
1797 }
1798 
1799 /**
1800  * DPDK callback to configure the transmit queue.
1801  *
1802  * @param dev
1803  *   Pointer to Ethernet device structure.
1804  * @param idx
1805  *   Transmit queue index.
1806  * @param desc
1807  *   Number of descriptors to configure in the queue.
1808  * @param socket
1809  *   NUMA socket on which memory must be allocated.
1810  * @param conf
1811  *   Tx queue configuration parameters.
1812  *
1813  * @return
1814  *   0 on success, negative error value otherwise.
1815  */
1816 static int
1817 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1818 		    unsigned int socket,
1819 		    const struct rte_eth_txconf *conf)
1820 {
1821 	struct mrvl_priv *priv = dev->data->dev_private;
1822 	struct mrvl_txq *txq;
1823 
1824 	if (dev->data->tx_queues[idx]) {
1825 		rte_free(dev->data->tx_queues[idx]);
1826 		dev->data->tx_queues[idx] = NULL;
1827 	}
1828 
1829 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
1830 	if (!txq)
1831 		return -ENOMEM;
1832 
1833 	txq->priv = priv;
1834 	txq->queue_id = idx;
1835 	txq->port_id = dev->data->port_id;
1836 	txq->tx_deferred_start = conf->tx_deferred_start;
1837 	dev->data->tx_queues[idx] = txq;
1838 
1839 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
1840 
1841 	return 0;
1842 }
1843 
1844 /**
1845  * DPDK callback to release the transmit queue.
1846  *
1847  * @param txq
1848  *   Generic transmit queue pointer.
1849  */
1850 static void
1851 mrvl_tx_queue_release(void *txq)
1852 {
1853 	struct mrvl_txq *q = txq;
1854 
1855 	if (!q)
1856 		return;
1857 
1858 	rte_free(q);
1859 }
1860 
1861 /**
1862  * DPDK callback to get flow control configuration.
1863  *
1864  * @param dev
1865  *  Pointer to Ethernet device structure.
1866  * @param fc_conf
1867  *  Pointer to the flow control configuration.
1868  *
1869  * @return
1870  *  0 on success, negative error value otherwise.
1871  */
1872 static int
1873 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1874 {
1875 	struct mrvl_priv *priv = dev->data->dev_private;
1876 	int ret, en;
1877 
1878 	if (!priv)
1879 		return -EPERM;
1880 
1881 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
1882 	if (ret) {
1883 		MRVL_LOG(ERR, "Failed to read rx pause state");
1884 		return ret;
1885 	}
1886 
1887 	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
1888 
1889 	return 0;
1890 }
1891 
1892 /**
1893  * DPDK callback to set flow control configuration.
1894  *
1895  * @param dev
1896  *  Pointer to Ethernet device structure.
1897  * @param fc_conf
1898  *  Pointer to the flow control configuration.
1899  *
1900  * @return
1901  *  0 on success, negative error value otherwise.
1902  */
1903 static int
1904 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1905 {
1906 	struct mrvl_priv *priv = dev->data->dev_private;
1907 
1908 	if (!priv)
1909 		return -EPERM;
1910 
1911 	if (fc_conf->high_water ||
1912 	    fc_conf->low_water ||
1913 	    fc_conf->pause_time ||
1914 	    fc_conf->mac_ctrl_frame_fwd ||
1915 	    fc_conf->autoneg) {
1916 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
1917 
1918 		return -EINVAL;
1919 	}
1920 
1921 	if (fc_conf->mode == RTE_FC_NONE ||
1922 	    fc_conf->mode == RTE_FC_RX_PAUSE) {
1923 		int ret, en;
1924 
1925 		en = fc_conf->mode == RTE_FC_NONE ? 0 : 1;
1926 		ret = pp2_ppio_set_rx_pause(priv->ppio, en);
1927 		if (ret)
1928 			MRVL_LOG(ERR,
1929 				"Failed to change flowctrl on RX side");
1930 
1931 		return ret;
1932 	}
1933 
1934 	return 0;
1935 }
1936 
1937 /**
1938  * Update RSS hash configuration
1939  *
1940  * @param dev
1941  *   Pointer to Ethernet device structure.
1942  * @param rss_conf
1943  *   Pointer to RSS configuration.
1944  *
1945  * @return
1946  *   0 on success, negative error value otherwise.
1947  */
1948 static int
1949 mrvl_rss_hash_update(struct rte_eth_dev *dev,
1950 		     struct rte_eth_rss_conf *rss_conf)
1951 {
1952 	struct mrvl_priv *priv = dev->data->dev_private;
1953 
1954 	if (priv->isolated)
1955 		return -ENOTSUP;
1956 
1957 	return mrvl_configure_rss(priv, rss_conf);
1958 }
1959 
1960 /**
1961  * DPDK callback to get RSS hash configuration.
1962  *
1963  * @param dev
1964  *   Pointer to Ethernet device structure.
1965  * @rss_conf
1966  *   Pointer to RSS configuration.
1967  *
1968  * @return
1969  *   Always 0.
1970  */
1971 static int
1972 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
1973 		       struct rte_eth_rss_conf *rss_conf)
1974 {
1975 	struct mrvl_priv *priv = dev->data->dev_private;
1976 	enum pp2_ppio_hash_type hash_type =
1977 		priv->ppio_params.inqs_params.hash_type;
1978 
1979 	rss_conf->rss_key = NULL;
1980 
1981 	if (hash_type == PP2_PPIO_HASH_T_NONE)
1982 		rss_conf->rss_hf = 0;
1983 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
1984 		rss_conf->rss_hf = ETH_RSS_IPV4;
1985 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
1986 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
1987 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
1988 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
1989 
1990 	return 0;
1991 }
1992 
1993 /**
1994  * DPDK callback to get rte_flow callbacks.
1995  *
1996  * @param dev
1997  *   Pointer to the device structure.
1998  * @param filer_type
1999  *   Flow filter type.
2000  * @param filter_op
2001  *   Flow filter operation.
2002  * @param arg
2003  *   Pointer to pass the flow ops.
2004  *
2005  * @return
2006  *   0 on success, negative error value otherwise.
2007  */
2008 static int
2009 mrvl_eth_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
2010 		     enum rte_filter_type filter_type,
2011 		     enum rte_filter_op filter_op, void *arg)
2012 {
2013 	switch (filter_type) {
2014 	case RTE_ETH_FILTER_GENERIC:
2015 		if (filter_op != RTE_ETH_FILTER_GET)
2016 			return -EINVAL;
2017 		*(const void **)arg = &mrvl_flow_ops;
2018 		return 0;
2019 	default:
2020 		MRVL_LOG(WARNING, "Filter type (%d) not supported",
2021 				filter_type);
2022 		return -EINVAL;
2023 	}
2024 }
2025 
2026 /**
2027  * DPDK callback to get rte_mtr callbacks.
2028  *
2029  * @param dev
2030  *   Pointer to the device structure.
2031  * @param ops
2032  *   Pointer to pass the mtr ops.
2033  *
2034  * @return
2035  *   Always 0.
2036  */
2037 static int
2038 mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2039 {
2040 	*(const void **)ops = &mrvl_mtr_ops;
2041 
2042 	return 0;
2043 }
2044 
2045 /**
2046  * DPDK callback to get rte_tm callbacks.
2047  *
2048  * @param dev
2049  *   Pointer to the device structure.
2050  * @param ops
2051  *   Pointer to pass the tm ops.
2052  *
2053  * @return
2054  *   Always 0.
2055  */
2056 static int
2057 mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2058 {
2059 	*(const void **)ops = &mrvl_tm_ops;
2060 
2061 	return 0;
2062 }
2063 
2064 static const struct eth_dev_ops mrvl_ops = {
2065 	.dev_configure = mrvl_dev_configure,
2066 	.dev_start = mrvl_dev_start,
2067 	.dev_stop = mrvl_dev_stop,
2068 	.dev_set_link_up = mrvl_dev_set_link_up,
2069 	.dev_set_link_down = mrvl_dev_set_link_down,
2070 	.dev_close = mrvl_dev_close,
2071 	.link_update = mrvl_link_update,
2072 	.promiscuous_enable = mrvl_promiscuous_enable,
2073 	.allmulticast_enable = mrvl_allmulticast_enable,
2074 	.promiscuous_disable = mrvl_promiscuous_disable,
2075 	.allmulticast_disable = mrvl_allmulticast_disable,
2076 	.mac_addr_remove = mrvl_mac_addr_remove,
2077 	.mac_addr_add = mrvl_mac_addr_add,
2078 	.mac_addr_set = mrvl_mac_addr_set,
2079 	.mtu_set = mrvl_mtu_set,
2080 	.stats_get = mrvl_stats_get,
2081 	.stats_reset = mrvl_stats_reset,
2082 	.xstats_get = mrvl_xstats_get,
2083 	.xstats_reset = mrvl_xstats_reset,
2084 	.xstats_get_names = mrvl_xstats_get_names,
2085 	.dev_infos_get = mrvl_dev_infos_get,
2086 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
2087 	.rxq_info_get = mrvl_rxq_info_get,
2088 	.txq_info_get = mrvl_txq_info_get,
2089 	.vlan_filter_set = mrvl_vlan_filter_set,
2090 	.tx_queue_start = mrvl_tx_queue_start,
2091 	.tx_queue_stop = mrvl_tx_queue_stop,
2092 	.rx_queue_setup = mrvl_rx_queue_setup,
2093 	.rx_queue_release = mrvl_rx_queue_release,
2094 	.tx_queue_setup = mrvl_tx_queue_setup,
2095 	.tx_queue_release = mrvl_tx_queue_release,
2096 	.flow_ctrl_get = mrvl_flow_ctrl_get,
2097 	.flow_ctrl_set = mrvl_flow_ctrl_set,
2098 	.rss_hash_update = mrvl_rss_hash_update,
2099 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
2100 	.filter_ctrl = mrvl_eth_filter_ctrl,
2101 	.mtr_ops_get = mrvl_mtr_ops_get,
2102 	.tm_ops_get = mrvl_tm_ops_get,
2103 };
2104 
2105 /**
2106  * Return packet type information and l3/l4 offsets.
2107  *
2108  * @param desc
2109  *   Pointer to the received packet descriptor.
2110  * @param l3_offset
2111  *   l3 packet offset.
2112  * @param l4_offset
2113  *   l4 packet offset.
2114  *
2115  * @return
2116  *   Packet type information.
2117  */
2118 static inline uint64_t
2119 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
2120 				    uint8_t *l3_offset, uint8_t *l4_offset)
2121 {
2122 	enum pp2_inq_l3_type l3_type;
2123 	enum pp2_inq_l4_type l4_type;
2124 	enum pp2_inq_vlan_tag vlan_tag;
2125 	uint64_t packet_type;
2126 
2127 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
2128 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
2129 	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
2130 
2131 	packet_type = RTE_PTYPE_L2_ETHER;
2132 
2133 	switch (vlan_tag) {
2134 	case PP2_INQ_VLAN_TAG_SINGLE:
2135 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
2136 		break;
2137 	case PP2_INQ_VLAN_TAG_DOUBLE:
2138 	case PP2_INQ_VLAN_TAG_TRIPLE:
2139 		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
2140 		break;
2141 	default:
2142 		break;
2143 	}
2144 
2145 	switch (l3_type) {
2146 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
2147 		packet_type |= RTE_PTYPE_L3_IPV4;
2148 		break;
2149 	case PP2_INQ_L3_TYPE_IPV4_OK:
2150 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
2151 		break;
2152 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
2153 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
2154 		break;
2155 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
2156 		packet_type |= RTE_PTYPE_L3_IPV6;
2157 		break;
2158 	case PP2_INQ_L3_TYPE_IPV6_EXT:
2159 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
2160 		break;
2161 	case PP2_INQ_L3_TYPE_ARP:
2162 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
2163 		/*
2164 		 * In case of ARP l4_offset is set to wrong value.
2165 		 * Set it to proper one so that later on mbuf->l3_len can be
2166 		 * calculated subtracting l4_offset and l3_offset.
2167 		 */
2168 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
2169 		break;
2170 	default:
2171 		MRVL_LOG(DEBUG, "Failed to recognise l3 packet type");
2172 		break;
2173 	}
2174 
2175 	switch (l4_type) {
2176 	case PP2_INQ_L4_TYPE_TCP:
2177 		packet_type |= RTE_PTYPE_L4_TCP;
2178 		break;
2179 	case PP2_INQ_L4_TYPE_UDP:
2180 		packet_type |= RTE_PTYPE_L4_UDP;
2181 		break;
2182 	default:
2183 		MRVL_LOG(DEBUG, "Failed to recognise l4 packet type");
2184 		break;
2185 	}
2186 
2187 	return packet_type;
2188 }
2189 
2190 /**
2191  * Get offload information from the received packet descriptor.
2192  *
2193  * @param desc
2194  *   Pointer to the received packet descriptor.
2195  *
2196  * @return
2197  *   Mbuf offload flags.
2198  */
2199 static inline uint64_t
2200 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc)
2201 {
2202 	uint64_t flags;
2203 	enum pp2_inq_desc_status status;
2204 
2205 	status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2206 	if (unlikely(status != PP2_DESC_ERR_OK))
2207 		flags = PKT_RX_IP_CKSUM_BAD;
2208 	else
2209 		flags = PKT_RX_IP_CKSUM_GOOD;
2210 
2211 	status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2212 	if (unlikely(status != PP2_DESC_ERR_OK))
2213 		flags |= PKT_RX_L4_CKSUM_BAD;
2214 	else
2215 		flags |= PKT_RX_L4_CKSUM_GOOD;
2216 
2217 	return flags;
2218 }
2219 
2220 /**
2221  * DPDK callback for receive.
2222  *
2223  * @param rxq
2224  *   Generic pointer to the receive queue.
2225  * @param rx_pkts
2226  *   Array to store received packets.
2227  * @param nb_pkts
2228  *   Maximum number of packets in array.
2229  *
2230  * @return
2231  *   Number of packets successfully received.
2232  */
2233 static uint16_t
2234 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2235 {
2236 	struct mrvl_rxq *q = rxq;
2237 	struct pp2_ppio_desc descs[nb_pkts];
2238 	struct pp2_bpool *bpool;
2239 	int i, ret, rx_done = 0;
2240 	int num;
2241 	struct pp2_hif *hif;
2242 	unsigned int core_id = rte_lcore_id();
2243 
2244 	hif = mrvl_get_hif(q->priv, core_id);
2245 
2246 	if (unlikely(!q->priv->ppio || !hif))
2247 		return 0;
2248 
2249 	bpool = q->priv->bpool;
2250 
2251 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2252 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2253 	if (unlikely(ret < 0)) {
2254 		MRVL_LOG(ERR, "Failed to receive packets");
2255 		return 0;
2256 	}
2257 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2258 
2259 	for (i = 0; i < nb_pkts; i++) {
2260 		struct rte_mbuf *mbuf;
2261 		uint8_t l3_offset, l4_offset;
2262 		enum pp2_inq_desc_status status;
2263 		uint64_t addr;
2264 
2265 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2266 			struct pp2_ppio_desc *pref_desc;
2267 			u64 pref_addr;
2268 
2269 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2270 			pref_addr = cookie_addr_high |
2271 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2272 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2273 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2274 		}
2275 
2276 		addr = cookie_addr_high |
2277 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2278 		mbuf = (struct rte_mbuf *)addr;
2279 		rte_pktmbuf_reset(mbuf);
2280 
2281 		/* drop packet in case of mac, overrun or resource error */
2282 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2283 		if (unlikely(status != PP2_DESC_ERR_OK)) {
2284 			struct pp2_buff_inf binf = {
2285 				.addr = rte_mbuf_data_iova_default(mbuf),
2286 				.cookie = (uint64_t)mbuf,
2287 			};
2288 
2289 			pp2_bpool_put_buff(hif, bpool, &binf);
2290 			mrvl_port_bpool_size
2291 				[bpool->pp2_id][bpool->id][core_id]++;
2292 			q->drop_mac++;
2293 			continue;
2294 		}
2295 
2296 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2297 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2298 		mbuf->data_len = mbuf->pkt_len;
2299 		mbuf->port = q->port_id;
2300 		mbuf->packet_type =
2301 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2302 							    &l3_offset,
2303 							    &l4_offset);
2304 		mbuf->l2_len = l3_offset;
2305 		mbuf->l3_len = l4_offset - l3_offset;
2306 
2307 		if (likely(q->cksum_enabled))
2308 			mbuf->ol_flags = mrvl_desc_to_ol_flags(&descs[i]);
2309 
2310 		rx_pkts[rx_done++] = mbuf;
2311 		q->bytes_recv += mbuf->pkt_len;
2312 	}
2313 
2314 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2315 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2316 
2317 		if (unlikely(num <= q->priv->bpool_min_size ||
2318 			     (!rx_done && num < q->priv->bpool_init_size))) {
2319 			ret = mrvl_fill_bpool(q, MRVL_BURST_SIZE);
2320 			if (ret)
2321 				MRVL_LOG(ERR, "Failed to fill bpool");
2322 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2323 			int i;
2324 			int pkt_to_remove = num - q->priv->bpool_init_size;
2325 			struct rte_mbuf *mbuf;
2326 			struct pp2_buff_inf buff;
2327 
2328 			MRVL_LOG(DEBUG,
2329 				"port-%d:%d: bpool %d oversize - remove %d buffers (pool size: %d -> %d)",
2330 				bpool->pp2_id, q->priv->ppio->port_id,
2331 				bpool->id, pkt_to_remove, num,
2332 				q->priv->bpool_init_size);
2333 
2334 			for (i = 0; i < pkt_to_remove; i++) {
2335 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2336 				if (ret)
2337 					break;
2338 				mbuf = (struct rte_mbuf *)
2339 					(cookie_addr_high | buff.cookie);
2340 				rte_pktmbuf_free(mbuf);
2341 			}
2342 			mrvl_port_bpool_size
2343 				[bpool->pp2_id][bpool->id][core_id] -= i;
2344 		}
2345 		rte_spinlock_unlock(&q->priv->lock);
2346 	}
2347 
2348 	return rx_done;
2349 }
2350 
2351 /**
2352  * Prepare offload information.
2353  *
2354  * @param ol_flags
2355  *   Offload flags.
2356  * @param packet_type
2357  *   Packet type bitfield.
2358  * @param l3_type
2359  *   Pointer to the pp2_ouq_l3_type structure.
2360  * @param l4_type
2361  *   Pointer to the pp2_outq_l4_type structure.
2362  * @param gen_l3_cksum
2363  *   Will be set to 1 in case l3 checksum is computed.
2364  * @param l4_cksum
2365  *   Will be set to 1 in case l4 checksum is computed.
2366  *
2367  * @return
2368  *   0 on success, negative error value otherwise.
2369  */
2370 static inline int
2371 mrvl_prepare_proto_info(uint64_t ol_flags, uint32_t packet_type,
2372 			enum pp2_outq_l3_type *l3_type,
2373 			enum pp2_outq_l4_type *l4_type,
2374 			int *gen_l3_cksum,
2375 			int *gen_l4_cksum)
2376 {
2377 	/*
2378 	 * Based on ol_flags prepare information
2379 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2380 	 * for offloading.
2381 	 */
2382 	if (ol_flags & PKT_TX_IPV4) {
2383 		*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2384 		*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2385 	} else if (ol_flags & PKT_TX_IPV6) {
2386 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2387 		/* no checksum for ipv6 header */
2388 		*gen_l3_cksum = 0;
2389 	} else {
2390 		/* if something different then stop processing */
2391 		return -1;
2392 	}
2393 
2394 	ol_flags &= PKT_TX_L4_MASK;
2395 	if ((packet_type & RTE_PTYPE_L4_TCP) &&
2396 	    ol_flags == PKT_TX_TCP_CKSUM) {
2397 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2398 		*gen_l4_cksum = 1;
2399 	} else if ((packet_type & RTE_PTYPE_L4_UDP) &&
2400 		   ol_flags == PKT_TX_UDP_CKSUM) {
2401 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2402 		*gen_l4_cksum = 1;
2403 	} else {
2404 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2405 		/* no checksum for other type */
2406 		*gen_l4_cksum = 0;
2407 	}
2408 
2409 	return 0;
2410 }
2411 
2412 /**
2413  * Release already sent buffers to bpool (buffer-pool).
2414  *
2415  * @param ppio
2416  *   Pointer to the port structure.
2417  * @param hif
2418  *   Pointer to the MUSDK hardware interface.
2419  * @param sq
2420  *   Pointer to the shadow queue.
2421  * @param qid
2422  *   Queue id number.
2423  * @param force
2424  *   Force releasing packets.
2425  */
2426 static inline void
2427 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2428 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2429 		       int qid, int force)
2430 {
2431 	struct buff_release_entry *entry;
2432 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2433 	int i;
2434 
2435 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2436 
2437 	sq->num_to_release += nb_done;
2438 
2439 	if (likely(!force &&
2440 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2441 		return;
2442 
2443 	nb_done = sq->num_to_release;
2444 	sq->num_to_release = 0;
2445 
2446 	for (i = 0; i < nb_done; i++) {
2447 		entry = &sq->ent[sq->tail + num];
2448 		if (unlikely(!entry->buff.addr)) {
2449 			MRVL_LOG(ERR,
2450 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2451 				sq->tail, (u64)entry->buff.cookie,
2452 				(u64)entry->buff.addr);
2453 			skip_bufs = 1;
2454 			goto skip;
2455 		}
2456 
2457 		if (unlikely(!entry->bpool)) {
2458 			struct rte_mbuf *mbuf;
2459 
2460 			mbuf = (struct rte_mbuf *)
2461 			       (cookie_addr_high | entry->buff.cookie);
2462 			rte_pktmbuf_free(mbuf);
2463 			skip_bufs = 1;
2464 			goto skip;
2465 		}
2466 
2467 		mrvl_port_bpool_size
2468 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2469 		num++;
2470 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2471 			goto skip;
2472 		continue;
2473 skip:
2474 		if (likely(num))
2475 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2476 		num += skip_bufs;
2477 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2478 		sq->size -= num;
2479 		num = 0;
2480 		skip_bufs = 0;
2481 	}
2482 
2483 	if (likely(num)) {
2484 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2485 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2486 		sq->size -= num;
2487 	}
2488 }
2489 
2490 /**
2491  * DPDK callback for transmit.
2492  *
2493  * @param txq
2494  *   Generic pointer transmit queue.
2495  * @param tx_pkts
2496  *   Packets to transmit.
2497  * @param nb_pkts
2498  *   Number of packets in array.
2499  *
2500  * @return
2501  *   Number of packets successfully transmitted.
2502  */
2503 static uint16_t
2504 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2505 {
2506 	struct mrvl_txq *q = txq;
2507 	struct mrvl_shadow_txq *sq;
2508 	struct pp2_hif *hif;
2509 	struct pp2_ppio_desc descs[nb_pkts];
2510 	unsigned int core_id = rte_lcore_id();
2511 	int i, ret, bytes_sent = 0;
2512 	uint16_t num, sq_free_size;
2513 	uint64_t addr;
2514 
2515 	hif = mrvl_get_hif(q->priv, core_id);
2516 	sq = &q->shadow_txqs[core_id];
2517 
2518 	if (unlikely(!q->priv->ppio || !hif))
2519 		return 0;
2520 
2521 	if (sq->size)
2522 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2523 				       sq, q->queue_id, 0);
2524 
2525 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2526 	if (unlikely(nb_pkts > sq_free_size)) {
2527 		MRVL_LOG(DEBUG,
2528 			"No room in shadow queue for %d packets! %d packets will be sent.",
2529 			nb_pkts, sq_free_size);
2530 		nb_pkts = sq_free_size;
2531 	}
2532 
2533 	for (i = 0; i < nb_pkts; i++) {
2534 		struct rte_mbuf *mbuf = tx_pkts[i];
2535 		int gen_l3_cksum, gen_l4_cksum;
2536 		enum pp2_outq_l3_type l3_type;
2537 		enum pp2_outq_l4_type l4_type;
2538 
2539 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2540 			struct rte_mbuf *pref_pkt_hdr;
2541 
2542 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2543 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2544 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2545 		}
2546 
2547 		mrvl_fill_shadowq(sq, mbuf);
2548 		mrvl_fill_desc(&descs[i], mbuf);
2549 
2550 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2551 		/*
2552 		 * in case unsupported ol_flags were passed
2553 		 * do not update descriptor offload information
2554 		 */
2555 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2556 					      &l3_type, &l4_type, &gen_l3_cksum,
2557 					      &gen_l4_cksum);
2558 		if (unlikely(ret))
2559 			continue;
2560 
2561 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2562 						  mbuf->l2_len,
2563 						  mbuf->l2_len + mbuf->l3_len,
2564 						  gen_l3_cksum, gen_l4_cksum);
2565 	}
2566 
2567 	num = nb_pkts;
2568 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2569 	/* number of packets that were not sent */
2570 	if (unlikely(num > nb_pkts)) {
2571 		for (i = nb_pkts; i < num; i++) {
2572 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2573 				MRVL_PP2_TX_SHADOWQ_MASK;
2574 			addr = cookie_addr_high | sq->ent[sq->head].buff.cookie;
2575 			bytes_sent -=
2576 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2577 		}
2578 		sq->size -= num - nb_pkts;
2579 	}
2580 
2581 	q->bytes_sent += bytes_sent;
2582 
2583 	return nb_pkts;
2584 }
2585 
2586 /** DPDK callback for S/G transmit.
2587  *
2588  * @param txq
2589  *   Generic pointer transmit queue.
2590  * @param tx_pkts
2591  *   Packets to transmit.
2592  * @param nb_pkts
2593  *   Number of packets in array.
2594  *
2595  * @return
2596  *   Number of packets successfully transmitted.
2597  */
2598 static uint16_t
2599 mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
2600 		     uint16_t nb_pkts)
2601 {
2602 	struct mrvl_txq *q = txq;
2603 	struct mrvl_shadow_txq *sq;
2604 	struct pp2_hif *hif;
2605 	struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
2606 	struct pp2_ppio_sg_pkts pkts;
2607 	uint8_t frags[nb_pkts];
2608 	unsigned int core_id = rte_lcore_id();
2609 	int i, j, ret, bytes_sent = 0;
2610 	int tail, tail_first;
2611 	uint16_t num, sq_free_size;
2612 	uint16_t nb_segs, total_descs = 0;
2613 	uint64_t addr;
2614 
2615 	hif = mrvl_get_hif(q->priv, core_id);
2616 	sq = &q->shadow_txqs[core_id];
2617 	pkts.frags = frags;
2618 	pkts.num = 0;
2619 
2620 	if (unlikely(!q->priv->ppio || !hif))
2621 		return 0;
2622 
2623 	if (sq->size)
2624 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2625 				       sq, q->queue_id, 0);
2626 
2627 	/* Save shadow queue free size */
2628 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2629 
2630 	tail = 0;
2631 	for (i = 0; i < nb_pkts; i++) {
2632 		struct rte_mbuf *mbuf = tx_pkts[i];
2633 		struct rte_mbuf *seg = NULL;
2634 		int gen_l3_cksum, gen_l4_cksum;
2635 		enum pp2_outq_l3_type l3_type;
2636 		enum pp2_outq_l4_type l4_type;
2637 
2638 		nb_segs = mbuf->nb_segs;
2639 		tail_first = tail;
2640 		total_descs += nb_segs;
2641 
2642 		/*
2643 		 * Check if total_descs does not exceed
2644 		 * shadow queue free size
2645 		 */
2646 		if (unlikely(total_descs > sq_free_size)) {
2647 			total_descs -= nb_segs;
2648 			RTE_LOG(DEBUG, PMD,
2649 				"No room in shadow queue for %d packets! "
2650 				"%d packets will be sent.\n",
2651 				nb_pkts, i);
2652 			break;
2653 		}
2654 
2655 		/* Check if nb_segs does not exceed the max nb of desc per
2656 		 * fragmented packet
2657 		 */
2658 		if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
2659 			total_descs -= nb_segs;
2660 			RTE_LOG(ERR, PMD,
2661 				"Too many segments. Packet won't be sent.\n");
2662 			break;
2663 		}
2664 
2665 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2666 			struct rte_mbuf *pref_pkt_hdr;
2667 
2668 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2669 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2670 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2671 		}
2672 
2673 		pkts.frags[pkts.num] = nb_segs;
2674 		pkts.num++;
2675 
2676 		seg = mbuf;
2677 		for (j = 0; j < nb_segs - 1; j++) {
2678 			/* For the subsequent segments, set shadow queue
2679 			 * buffer to NULL
2680 			 */
2681 			mrvl_fill_shadowq(sq, NULL);
2682 			mrvl_fill_desc(&descs[tail], seg);
2683 
2684 			tail++;
2685 			seg = seg->next;
2686 		}
2687 		/* Put first mbuf info in last shadow queue entry */
2688 		mrvl_fill_shadowq(sq, mbuf);
2689 		/* Update descriptor with last segment */
2690 		mrvl_fill_desc(&descs[tail++], seg);
2691 
2692 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2693 		/* In case unsupported ol_flags were passed
2694 		 * do not update descriptor offload information
2695 		 */
2696 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2697 					      &l3_type, &l4_type, &gen_l3_cksum,
2698 					      &gen_l4_cksum);
2699 		if (unlikely(ret))
2700 			continue;
2701 
2702 		pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
2703 						  l4_type, mbuf->l2_len,
2704 						  mbuf->l2_len + mbuf->l3_len,
2705 						  gen_l3_cksum, gen_l4_cksum);
2706 	}
2707 
2708 	num = total_descs;
2709 	pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
2710 			 &total_descs, &pkts);
2711 	/* number of packets that were not sent */
2712 	if (unlikely(num > total_descs)) {
2713 		for (i = total_descs; i < num; i++) {
2714 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2715 				MRVL_PP2_TX_SHADOWQ_MASK;
2716 
2717 			addr = sq->ent[sq->head].buff.cookie;
2718 			if (addr)
2719 				bytes_sent -=
2720 					rte_pktmbuf_pkt_len((struct rte_mbuf *)
2721 						(cookie_addr_high | addr));
2722 		}
2723 		sq->size -= num - total_descs;
2724 		nb_pkts = pkts.num;
2725 	}
2726 
2727 	q->bytes_sent += bytes_sent;
2728 
2729 	return nb_pkts;
2730 }
2731 
2732 /**
2733  * Initialize packet processor.
2734  *
2735  * @return
2736  *   0 on success, negative error value otherwise.
2737  */
2738 static int
2739 mrvl_init_pp2(void)
2740 {
2741 	struct pp2_init_params init_params;
2742 
2743 	memset(&init_params, 0, sizeof(init_params));
2744 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
2745 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
2746 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
2747 
2748 	return pp2_init(&init_params);
2749 }
2750 
2751 /**
2752  * Deinitialize packet processor.
2753  *
2754  * @return
2755  *   0 on success, negative error value otherwise.
2756  */
2757 static void
2758 mrvl_deinit_pp2(void)
2759 {
2760 	pp2_deinit();
2761 }
2762 
2763 /**
2764  * Create private device structure.
2765  *
2766  * @param dev_name
2767  *   Pointer to the port name passed in the initialization parameters.
2768  *
2769  * @return
2770  *   Pointer to the newly allocated private device structure.
2771  */
2772 static struct mrvl_priv *
2773 mrvl_priv_create(const char *dev_name)
2774 {
2775 	struct pp2_bpool_params bpool_params;
2776 	char match[MRVL_MATCH_LEN];
2777 	struct mrvl_priv *priv;
2778 	int ret, bpool_bit;
2779 
2780 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
2781 	if (!priv)
2782 		return NULL;
2783 
2784 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
2785 				       &priv->pp_id, &priv->ppio_id);
2786 	if (ret)
2787 		goto out_free_priv;
2788 
2789 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
2790 				     PP2_BPOOL_NUM_POOLS);
2791 	if (bpool_bit < 0)
2792 		goto out_free_priv;
2793 	priv->bpool_bit = bpool_bit;
2794 
2795 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
2796 		 priv->bpool_bit);
2797 	memset(&bpool_params, 0, sizeof(bpool_params));
2798 	bpool_params.match = match;
2799 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
2800 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
2801 	if (ret)
2802 		goto out_clear_bpool_bit;
2803 
2804 	priv->ppio_params.type = PP2_PPIO_T_NIC;
2805 	rte_spinlock_init(&priv->lock);
2806 
2807 	return priv;
2808 out_clear_bpool_bit:
2809 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2810 out_free_priv:
2811 	rte_free(priv);
2812 	return NULL;
2813 }
2814 
2815 /**
2816  * Create device representing Ethernet port.
2817  *
2818  * @param name
2819  *   Pointer to the port's name.
2820  *
2821  * @return
2822  *   0 on success, negative error value otherwise.
2823  */
2824 static int
2825 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
2826 {
2827 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
2828 	struct rte_eth_dev *eth_dev;
2829 	struct mrvl_priv *priv;
2830 	struct ifreq req;
2831 
2832 	eth_dev = rte_eth_dev_allocate(name);
2833 	if (!eth_dev)
2834 		return -ENOMEM;
2835 
2836 	priv = mrvl_priv_create(name);
2837 	if (!priv) {
2838 		ret = -ENOMEM;
2839 		goto out_free;
2840 	}
2841 	eth_dev->data->dev_private = priv;
2842 
2843 	eth_dev->data->mac_addrs =
2844 		rte_zmalloc("mac_addrs",
2845 			    RTE_ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
2846 	if (!eth_dev->data->mac_addrs) {
2847 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
2848 		ret = -ENOMEM;
2849 		goto out_free;
2850 	}
2851 
2852 	memset(&req, 0, sizeof(req));
2853 	strcpy(req.ifr_name, name);
2854 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
2855 	if (ret)
2856 		goto out_free;
2857 
2858 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
2859 	       req.ifr_addr.sa_data, RTE_ETHER_ADDR_LEN);
2860 
2861 	eth_dev->data->kdrv = RTE_KDRV_NONE;
2862 	eth_dev->device = &vdev->device;
2863 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
2864 	mrvl_set_tx_function(eth_dev);
2865 	eth_dev->dev_ops = &mrvl_ops;
2866 
2867 	/* Flag to call rte_eth_dev_release_port() in rte_eth_dev_close(). */
2868 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
2869 
2870 	rte_eth_dev_probing_finish(eth_dev);
2871 	return 0;
2872 out_free:
2873 	rte_eth_dev_release_port(eth_dev);
2874 
2875 	return ret;
2876 }
2877 
2878 /**
2879  * Callback used by rte_kvargs_process() during argument parsing.
2880  *
2881  * @param key
2882  *   Pointer to the parsed key (unused).
2883  * @param value
2884  *   Pointer to the parsed value.
2885  * @param extra_args
2886  *   Pointer to the extra arguments which contains address of the
2887  *   table of pointers to parsed interface names.
2888  *
2889  * @return
2890  *   Always 0.
2891  */
2892 static int
2893 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
2894 		 void *extra_args)
2895 {
2896 	struct mrvl_ifnames *ifnames = extra_args;
2897 
2898 	ifnames->names[ifnames->idx++] = value;
2899 
2900 	return 0;
2901 }
2902 
2903 /**
2904  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
2905  */
2906 static void
2907 mrvl_deinit_hifs(void)
2908 {
2909 	int i;
2910 
2911 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
2912 		if (hifs[i])
2913 			pp2_hif_deinit(hifs[i]);
2914 	}
2915 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
2916 	memset(hifs, 0, sizeof(hifs));
2917 }
2918 
2919 /**
2920  * DPDK callback to register the virtual device.
2921  *
2922  * @param vdev
2923  *   Pointer to the virtual device.
2924  *
2925  * @return
2926  *   0 on success, negative error value otherwise.
2927  */
2928 static int
2929 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
2930 {
2931 	struct rte_kvargs *kvlist;
2932 	struct mrvl_ifnames ifnames;
2933 	int ret = -EINVAL;
2934 	uint32_t i, ifnum, cfgnum;
2935 	const char *params;
2936 
2937 	params = rte_vdev_device_args(vdev);
2938 	if (!params)
2939 		return -EINVAL;
2940 
2941 	kvlist = rte_kvargs_parse(params, valid_args);
2942 	if (!kvlist)
2943 		return -EINVAL;
2944 
2945 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
2946 	if (ifnum > RTE_DIM(ifnames.names))
2947 		goto out_free_kvlist;
2948 
2949 	ifnames.idx = 0;
2950 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
2951 			   mrvl_get_ifnames, &ifnames);
2952 
2953 
2954 	/*
2955 	 * The below system initialization should be done only once,
2956 	 * on the first provided configuration file
2957 	 */
2958 	if (!mrvl_qos_cfg) {
2959 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
2960 		MRVL_LOG(INFO, "Parsing config file!");
2961 		if (cfgnum > 1) {
2962 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
2963 			goto out_free_kvlist;
2964 		} else if (cfgnum == 1) {
2965 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
2966 					   mrvl_get_qoscfg, &mrvl_qos_cfg);
2967 		}
2968 	}
2969 
2970 	if (mrvl_dev_num)
2971 		goto init_devices;
2972 
2973 	MRVL_LOG(INFO, "Perform MUSDK initializations");
2974 
2975 	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
2976 	if (ret)
2977 		goto out_free_kvlist;
2978 
2979 	ret = mrvl_init_pp2();
2980 	if (ret) {
2981 		MRVL_LOG(ERR, "Failed to init PP!");
2982 		rte_mvep_deinit(MVEP_MOD_T_PP2);
2983 		goto out_free_kvlist;
2984 	}
2985 
2986 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
2987 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
2988 
2989 	mrvl_lcore_first = RTE_MAX_LCORE;
2990 	mrvl_lcore_last = 0;
2991 
2992 init_devices:
2993 	for (i = 0; i < ifnum; i++) {
2994 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
2995 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
2996 		if (ret)
2997 			goto out_cleanup;
2998 		mrvl_dev_num++;
2999 	}
3000 
3001 	rte_kvargs_free(kvlist);
3002 
3003 	return 0;
3004 out_cleanup:
3005 	rte_pmd_mrvl_remove(vdev);
3006 
3007 out_free_kvlist:
3008 	rte_kvargs_free(kvlist);
3009 
3010 	return ret;
3011 }
3012 
3013 /**
3014  * DPDK callback to remove virtual device.
3015  *
3016  * @param vdev
3017  *   Pointer to the removed virtual device.
3018  *
3019  * @return
3020  *   0 on success, negative error value otherwise.
3021  */
3022 static int
3023 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
3024 {
3025 	uint16_t port_id;
3026 
3027 	RTE_ETH_FOREACH_DEV(port_id) {
3028 		if (rte_eth_devices[port_id].device != &vdev->device)
3029 			continue;
3030 		rte_eth_dev_close(port_id);
3031 	}
3032 
3033 	return 0;
3034 }
3035 
3036 static struct rte_vdev_driver pmd_mrvl_drv = {
3037 	.probe = rte_pmd_mrvl_probe,
3038 	.remove = rte_pmd_mrvl_remove,
3039 };
3040 
3041 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
3042 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
3043 
3044 RTE_INIT(mrvl_init_log)
3045 {
3046 	mrvl_logtype = rte_log_register("pmd.net.mvpp2");
3047 	if (mrvl_logtype >= 0)
3048 		rte_log_set_level(mrvl_logtype, RTE_LOG_NOTICE);
3049 }
3050