xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision f5057be340e44f3edc0fe90fa875eb89a4c49b4f)
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 int
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 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
865 		return 0;
866 
867 	mrvl_flush_rx_queues(dev);
868 	mrvl_flush_tx_shadow_queues(dev);
869 	mrvl_flow_deinit(dev);
870 	mrvl_mtr_deinit(dev);
871 
872 	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
873 		struct pp2_ppio_tc_params *tc_params =
874 			&priv->ppio_params.inqs_params.tcs_params[i];
875 
876 		if (tc_params->inqs_params) {
877 			rte_free(tc_params->inqs_params);
878 			tc_params->inqs_params = NULL;
879 		}
880 	}
881 
882 	if (priv->cls_tbl) {
883 		pp2_cls_tbl_deinit(priv->cls_tbl);
884 		priv->cls_tbl = NULL;
885 	}
886 
887 	if (priv->qos_tbl) {
888 		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
889 		priv->qos_tbl = NULL;
890 	}
891 
892 	mrvl_flush_bpool(dev);
893 	mrvl_tm_deinit(dev);
894 
895 	if (priv->ppio) {
896 		pp2_ppio_deinit(priv->ppio);
897 		priv->ppio = NULL;
898 	}
899 
900 	/* policer must be released after ppio deinitialization */
901 	if (priv->default_policer) {
902 		pp2_cls_plcr_deinit(priv->default_policer);
903 		priv->default_policer = NULL;
904 	}
905 
906 
907 	if (priv->bpool) {
908 		pp2_bpool_deinit(priv->bpool);
909 		used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
910 		priv->bpool = NULL;
911 	}
912 
913 	mrvl_dev_num--;
914 
915 	if (mrvl_dev_num == 0) {
916 		MRVL_LOG(INFO, "Perform MUSDK deinit");
917 		mrvl_deinit_hifs();
918 		mrvl_deinit_pp2();
919 		rte_mvep_deinit(MVEP_MOD_T_PP2);
920 	}
921 
922 	return 0;
923 }
924 
925 /**
926  * DPDK callback to retrieve physical link information.
927  *
928  * @param dev
929  *   Pointer to Ethernet device structure.
930  * @param wait_to_complete
931  *   Wait for request completion (ignored).
932  *
933  * @return
934  *   0 on success, negative error value otherwise.
935  */
936 static int
937 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
938 {
939 	/*
940 	 * TODO
941 	 * once MUSDK provides necessary API use it here
942 	 */
943 	struct mrvl_priv *priv = dev->data->dev_private;
944 	struct ethtool_cmd edata;
945 	struct ifreq req;
946 	int ret, fd, link_up;
947 
948 	if (!priv->ppio)
949 		return -EPERM;
950 
951 	edata.cmd = ETHTOOL_GSET;
952 
953 	strcpy(req.ifr_name, dev->data->name);
954 	req.ifr_data = (void *)&edata;
955 
956 	fd = socket(AF_INET, SOCK_DGRAM, 0);
957 	if (fd == -1)
958 		return -EFAULT;
959 
960 	ret = ioctl(fd, SIOCETHTOOL, &req);
961 	if (ret == -1) {
962 		close(fd);
963 		return -EFAULT;
964 	}
965 
966 	close(fd);
967 
968 	switch (ethtool_cmd_speed(&edata)) {
969 	case SPEED_10:
970 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
971 		break;
972 	case SPEED_100:
973 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
974 		break;
975 	case SPEED_1000:
976 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
977 		break;
978 	case SPEED_10000:
979 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
980 		break;
981 	default:
982 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
983 	}
984 
985 	dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
986 							 ETH_LINK_HALF_DUPLEX;
987 	dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
988 							   ETH_LINK_FIXED;
989 	pp2_ppio_get_link_state(priv->ppio, &link_up);
990 	dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
991 
992 	return 0;
993 }
994 
995 /**
996  * DPDK callback to enable promiscuous mode.
997  *
998  * @param dev
999  *   Pointer to Ethernet device structure.
1000  *
1001  * @return
1002  *   0 on success, negative error value otherwise.
1003  */
1004 static int
1005 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
1006 {
1007 	struct mrvl_priv *priv = dev->data->dev_private;
1008 	int ret;
1009 
1010 	if (!priv->ppio)
1011 		return 0;
1012 
1013 	if (priv->isolated)
1014 		return 0;
1015 
1016 	ret = pp2_ppio_set_promisc(priv->ppio, 1);
1017 	if (ret) {
1018 		MRVL_LOG(ERR, "Failed to enable promiscuous mode");
1019 		return -EAGAIN;
1020 	}
1021 
1022 	return 0;
1023 }
1024 
1025 /**
1026  * DPDK callback to enable allmulti mode.
1027  *
1028  * @param dev
1029  *   Pointer to Ethernet device structure.
1030  *
1031  * @return
1032  *   0 on success, negative error value otherwise.
1033  */
1034 static int
1035 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
1036 {
1037 	struct mrvl_priv *priv = dev->data->dev_private;
1038 	int ret;
1039 
1040 	if (!priv->ppio)
1041 		return 0;
1042 
1043 	if (priv->isolated)
1044 		return 0;
1045 
1046 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
1047 	if (ret) {
1048 		MRVL_LOG(ERR, "Failed enable all-multicast mode");
1049 		return -EAGAIN;
1050 	}
1051 
1052 	return 0;
1053 }
1054 
1055 /**
1056  * DPDK callback to disable promiscuous mode.
1057  *
1058  * @param dev
1059  *   Pointer to Ethernet device structure.
1060  *
1061  * @return
1062  *   0 on success, negative error value otherwise.
1063  */
1064 static int
1065 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
1066 {
1067 	struct mrvl_priv *priv = dev->data->dev_private;
1068 	int ret;
1069 
1070 	if (!priv->ppio)
1071 		return 0;
1072 
1073 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
1074 	if (ret) {
1075 		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
1076 		return -EAGAIN;
1077 	}
1078 
1079 	return 0;
1080 }
1081 
1082 /**
1083  * DPDK callback to disable allmulticast mode.
1084  *
1085  * @param dev
1086  *   Pointer to Ethernet device structure.
1087  *
1088  * @return
1089  *   0 on success, negative error value otherwise.
1090  */
1091 static int
1092 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
1093 {
1094 	struct mrvl_priv *priv = dev->data->dev_private;
1095 	int ret;
1096 
1097 	if (!priv->ppio)
1098 		return 0;
1099 
1100 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
1101 	if (ret) {
1102 		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
1103 		return -EAGAIN;
1104 	}
1105 
1106 	return 0;
1107 }
1108 
1109 /**
1110  * DPDK callback to remove a MAC address.
1111  *
1112  * @param dev
1113  *   Pointer to Ethernet device structure.
1114  * @param index
1115  *   MAC address index.
1116  */
1117 static void
1118 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1119 {
1120 	struct mrvl_priv *priv = dev->data->dev_private;
1121 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1122 	int ret;
1123 
1124 	if (!priv->ppio)
1125 		return;
1126 
1127 	if (priv->isolated)
1128 		return;
1129 
1130 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
1131 				       dev->data->mac_addrs[index].addr_bytes);
1132 	if (ret) {
1133 		rte_ether_format_addr(buf, sizeof(buf),
1134 				  &dev->data->mac_addrs[index]);
1135 		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
1136 	}
1137 }
1138 
1139 /**
1140  * DPDK callback to add a MAC address.
1141  *
1142  * @param dev
1143  *   Pointer to Ethernet device structure.
1144  * @param mac_addr
1145  *   MAC address to register.
1146  * @param index
1147  *   MAC address index.
1148  * @param vmdq
1149  *   VMDq pool index to associate address with (unused).
1150  *
1151  * @return
1152  *   0 on success, negative error value otherwise.
1153  */
1154 static int
1155 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1156 		  uint32_t index, uint32_t vmdq __rte_unused)
1157 {
1158 	struct mrvl_priv *priv = dev->data->dev_private;
1159 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1160 	int ret;
1161 
1162 	if (priv->isolated)
1163 		return -ENOTSUP;
1164 
1165 	if (index == 0)
1166 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1167 		return -1;
1168 
1169 	if (!priv->ppio)
1170 		return 0;
1171 
1172 	/*
1173 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1174 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1175 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1176 	 * 21 respectively.
1177 	 *
1178 	 * If more than uc_filter_max uc addresses were added to filter list
1179 	 * then NIC will switch to promiscuous mode automatically.
1180 	 *
1181 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1182 	 * were added to filter list then NIC will switch to all-multicast mode
1183 	 * automatically.
1184 	 */
1185 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1186 	if (ret) {
1187 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1188 		MRVL_LOG(ERR, "Failed to add mac %s", buf);
1189 		return -1;
1190 	}
1191 
1192 	return 0;
1193 }
1194 
1195 /**
1196  * DPDK callback to set the primary MAC address.
1197  *
1198  * @param dev
1199  *   Pointer to Ethernet device structure.
1200  * @param mac_addr
1201  *   MAC address to register.
1202  *
1203  * @return
1204  *   0 on success, negative error value otherwise.
1205  */
1206 static int
1207 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1208 {
1209 	struct mrvl_priv *priv = dev->data->dev_private;
1210 	int ret;
1211 
1212 	if (!priv->ppio)
1213 		return 0;
1214 
1215 	if (priv->isolated)
1216 		return -ENOTSUP;
1217 
1218 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1219 	if (ret) {
1220 		char buf[RTE_ETHER_ADDR_FMT_SIZE];
1221 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1222 		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1223 	}
1224 
1225 	return ret;
1226 }
1227 
1228 /**
1229  * DPDK callback to get device statistics.
1230  *
1231  * @param dev
1232  *   Pointer to Ethernet device structure.
1233  * @param stats
1234  *   Stats structure output buffer.
1235  *
1236  * @return
1237  *   0 on success, negative error value otherwise.
1238  */
1239 static int
1240 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1241 {
1242 	struct mrvl_priv *priv = dev->data->dev_private;
1243 	struct pp2_ppio_statistics ppio_stats;
1244 	uint64_t drop_mac = 0;
1245 	unsigned int i, idx, ret;
1246 
1247 	if (!priv->ppio)
1248 		return -EPERM;
1249 
1250 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1251 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1252 		struct pp2_ppio_inq_statistics rx_stats;
1253 
1254 		if (!rxq)
1255 			continue;
1256 
1257 		idx = rxq->queue_id;
1258 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1259 			MRVL_LOG(ERR,
1260 				"rx queue %d stats out of range (0 - %d)",
1261 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1262 			continue;
1263 		}
1264 
1265 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1266 						  priv->rxq_map[idx].tc,
1267 						  priv->rxq_map[idx].inq,
1268 						  &rx_stats, 0);
1269 		if (unlikely(ret)) {
1270 			MRVL_LOG(ERR,
1271 				"Failed to update rx queue %d stats", idx);
1272 			break;
1273 		}
1274 
1275 		stats->q_ibytes[idx] = rxq->bytes_recv;
1276 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1277 		stats->q_errors[idx] = rx_stats.drop_early +
1278 				       rx_stats.drop_fullq +
1279 				       rx_stats.drop_bm +
1280 				       rxq->drop_mac;
1281 		stats->ibytes += rxq->bytes_recv;
1282 		drop_mac += rxq->drop_mac;
1283 	}
1284 
1285 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1286 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1287 		struct pp2_ppio_outq_statistics tx_stats;
1288 
1289 		if (!txq)
1290 			continue;
1291 
1292 		idx = txq->queue_id;
1293 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1294 			MRVL_LOG(ERR,
1295 				"tx queue %d stats out of range (0 - %d)",
1296 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1297 		}
1298 
1299 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1300 						   &tx_stats, 0);
1301 		if (unlikely(ret)) {
1302 			MRVL_LOG(ERR,
1303 				"Failed to update tx queue %d stats", idx);
1304 			break;
1305 		}
1306 
1307 		stats->q_opackets[idx] = tx_stats.deq_desc;
1308 		stats->q_obytes[idx] = txq->bytes_sent;
1309 		stats->obytes += txq->bytes_sent;
1310 	}
1311 
1312 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1313 	if (unlikely(ret)) {
1314 		MRVL_LOG(ERR, "Failed to update port statistics");
1315 		return ret;
1316 	}
1317 
1318 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1319 	stats->opackets += ppio_stats.tx_packets;
1320 	stats->imissed += ppio_stats.rx_fullq_dropped +
1321 			  ppio_stats.rx_bm_dropped +
1322 			  ppio_stats.rx_early_dropped +
1323 			  ppio_stats.rx_fifo_dropped +
1324 			  ppio_stats.rx_cls_dropped;
1325 	stats->ierrors = drop_mac;
1326 
1327 	return 0;
1328 }
1329 
1330 /**
1331  * DPDK callback to clear device statistics.
1332  *
1333  * @param dev
1334  *   Pointer to Ethernet device structure.
1335  *
1336  * @return
1337  *   0 on success, negative error value otherwise.
1338  */
1339 static int
1340 mrvl_stats_reset(struct rte_eth_dev *dev)
1341 {
1342 	struct mrvl_priv *priv = dev->data->dev_private;
1343 	int i;
1344 
1345 	if (!priv->ppio)
1346 		return 0;
1347 
1348 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1349 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1350 
1351 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1352 					    priv->rxq_map[i].inq, NULL, 1);
1353 		rxq->bytes_recv = 0;
1354 		rxq->drop_mac = 0;
1355 	}
1356 
1357 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1358 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1359 
1360 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1361 		txq->bytes_sent = 0;
1362 	}
1363 
1364 	return pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1365 }
1366 
1367 /**
1368  * DPDK callback to get extended statistics.
1369  *
1370  * @param dev
1371  *   Pointer to Ethernet device structure.
1372  * @param stats
1373  *   Pointer to xstats table.
1374  * @param n
1375  *   Number of entries in xstats table.
1376  * @return
1377  *   Negative value on error, number of read xstats otherwise.
1378  */
1379 static int
1380 mrvl_xstats_get(struct rte_eth_dev *dev,
1381 		struct rte_eth_xstat *stats, unsigned int n)
1382 {
1383 	struct mrvl_priv *priv = dev->data->dev_private;
1384 	struct pp2_ppio_statistics ppio_stats;
1385 	unsigned int i;
1386 
1387 	if (!stats)
1388 		return 0;
1389 
1390 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1391 	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1392 		uint64_t val;
1393 
1394 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1395 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1396 					    mrvl_xstats_tbl[i].offset);
1397 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1398 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1399 					    mrvl_xstats_tbl[i].offset);
1400 		else
1401 			return -EINVAL;
1402 
1403 		stats[i].id = i;
1404 		stats[i].value = val;
1405 	}
1406 
1407 	return n;
1408 }
1409 
1410 /**
1411  * DPDK callback to reset extended statistics.
1412  *
1413  * @param dev
1414  *   Pointer to Ethernet device structure.
1415  *
1416  * @return
1417  *   0 on success, negative error value otherwise.
1418  */
1419 static int
1420 mrvl_xstats_reset(struct rte_eth_dev *dev)
1421 {
1422 	return mrvl_stats_reset(dev);
1423 }
1424 
1425 /**
1426  * DPDK callback to get extended statistics names.
1427  *
1428  * @param dev (unused)
1429  *   Pointer to Ethernet device structure.
1430  * @param xstats_names
1431  *   Pointer to xstats names table.
1432  * @param size
1433  *   Size of the xstats names table.
1434  * @return
1435  *   Number of read names.
1436  */
1437 static int
1438 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1439 		      struct rte_eth_xstat_name *xstats_names,
1440 		      unsigned int size)
1441 {
1442 	unsigned int i;
1443 
1444 	if (!xstats_names)
1445 		return RTE_DIM(mrvl_xstats_tbl);
1446 
1447 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1448 		strlcpy(xstats_names[i].name, mrvl_xstats_tbl[i].name,
1449 			RTE_ETH_XSTATS_NAME_SIZE);
1450 
1451 	return size;
1452 }
1453 
1454 /**
1455  * DPDK callback to get information about the device.
1456  *
1457  * @param dev
1458  *   Pointer to Ethernet device structure (unused).
1459  * @param info
1460  *   Info structure output buffer.
1461  */
1462 static int
1463 mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
1464 		   struct rte_eth_dev_info *info)
1465 {
1466 	info->speed_capa = ETH_LINK_SPEED_10M |
1467 			   ETH_LINK_SPEED_100M |
1468 			   ETH_LINK_SPEED_1G |
1469 			   ETH_LINK_SPEED_10G;
1470 
1471 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1472 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1473 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1474 
1475 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1476 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1477 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1478 
1479 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1480 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1481 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1482 
1483 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1484 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1485 
1486 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1487 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1488 
1489 	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1490 				       ETH_RSS_NONFRAG_IPV4_TCP |
1491 				       ETH_RSS_NONFRAG_IPV4_UDP;
1492 
1493 	/* By default packets are dropped if no descriptors are available */
1494 	info->default_rxconf.rx_drop_en = 1;
1495 
1496 	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1497 
1498 	return 0;
1499 }
1500 
1501 /**
1502  * Return supported packet types.
1503  *
1504  * @param dev
1505  *   Pointer to Ethernet device structure (unused).
1506  *
1507  * @return
1508  *   Const pointer to the table with supported packet types.
1509  */
1510 static const uint32_t *
1511 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1512 {
1513 	static const uint32_t ptypes[] = {
1514 		RTE_PTYPE_L2_ETHER,
1515 		RTE_PTYPE_L2_ETHER_VLAN,
1516 		RTE_PTYPE_L2_ETHER_QINQ,
1517 		RTE_PTYPE_L3_IPV4,
1518 		RTE_PTYPE_L3_IPV4_EXT,
1519 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1520 		RTE_PTYPE_L3_IPV6,
1521 		RTE_PTYPE_L3_IPV6_EXT,
1522 		RTE_PTYPE_L2_ETHER_ARP,
1523 		RTE_PTYPE_L4_TCP,
1524 		RTE_PTYPE_L4_UDP
1525 	};
1526 
1527 	return ptypes;
1528 }
1529 
1530 /**
1531  * DPDK callback to get information about specific receive queue.
1532  *
1533  * @param dev
1534  *   Pointer to Ethernet device structure.
1535  * @param rx_queue_id
1536  *   Receive queue index.
1537  * @param qinfo
1538  *   Receive queue information structure.
1539  */
1540 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1541 			      struct rte_eth_rxq_info *qinfo)
1542 {
1543 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1544 	struct mrvl_priv *priv = dev->data->dev_private;
1545 	int inq = priv->rxq_map[rx_queue_id].inq;
1546 	int tc = priv->rxq_map[rx_queue_id].tc;
1547 	struct pp2_ppio_tc_params *tc_params =
1548 		&priv->ppio_params.inqs_params.tcs_params[tc];
1549 
1550 	qinfo->mp = q->mp;
1551 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1552 }
1553 
1554 /**
1555  * DPDK callback to get information about specific transmit queue.
1556  *
1557  * @param dev
1558  *   Pointer to Ethernet device structure.
1559  * @param tx_queue_id
1560  *   Transmit queue index.
1561  * @param qinfo
1562  *   Transmit queue information structure.
1563  */
1564 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1565 			      struct rte_eth_txq_info *qinfo)
1566 {
1567 	struct mrvl_priv *priv = dev->data->dev_private;
1568 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1569 
1570 	qinfo->nb_desc =
1571 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1572 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1573 }
1574 
1575 /**
1576  * DPDK callback to Configure a VLAN filter.
1577  *
1578  * @param dev
1579  *   Pointer to Ethernet device structure.
1580  * @param vlan_id
1581  *   VLAN ID to filter.
1582  * @param on
1583  *   Toggle filter.
1584  *
1585  * @return
1586  *   0 on success, negative error value otherwise.
1587  */
1588 static int
1589 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1590 {
1591 	struct mrvl_priv *priv = dev->data->dev_private;
1592 
1593 	if (!priv->ppio)
1594 		return -EPERM;
1595 
1596 	if (priv->isolated)
1597 		return -ENOTSUP;
1598 
1599 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1600 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1601 }
1602 
1603 /**
1604  * Release buffers to hardware bpool (buffer-pool)
1605  *
1606  * @param rxq
1607  *   Receive queue pointer.
1608  * @param num
1609  *   Number of buffers to release to bpool.
1610  *
1611  * @return
1612  *   0 on success, negative error value otherwise.
1613  */
1614 static int
1615 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1616 {
1617 	struct buff_release_entry entries[MRVL_PP2_RXD_MAX];
1618 	struct rte_mbuf *mbufs[MRVL_PP2_RXD_MAX];
1619 	int i, ret;
1620 	unsigned int core_id;
1621 	struct pp2_hif *hif;
1622 	struct pp2_bpool *bpool;
1623 
1624 	core_id = rte_lcore_id();
1625 	if (core_id == LCORE_ID_ANY)
1626 		core_id = rte_get_master_lcore();
1627 
1628 	hif = mrvl_get_hif(rxq->priv, core_id);
1629 	if (!hif)
1630 		return -1;
1631 
1632 	bpool = rxq->priv->bpool;
1633 
1634 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1635 	if (ret)
1636 		return ret;
1637 
1638 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1639 		cookie_addr_high =
1640 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1641 
1642 	for (i = 0; i < num; i++) {
1643 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1644 			!= cookie_addr_high) {
1645 			MRVL_LOG(ERR,
1646 				"mbuf virtual addr high 0x%lx out of range",
1647 				(uint64_t)mbufs[i] >> 32);
1648 			goto out;
1649 		}
1650 
1651 		entries[i].buff.addr =
1652 			rte_mbuf_data_iova_default(mbufs[i]);
1653 		entries[i].buff.cookie = (uint64_t)mbufs[i];
1654 		entries[i].bpool = bpool;
1655 	}
1656 
1657 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1658 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1659 
1660 	if (i != num)
1661 		goto out;
1662 
1663 	return 0;
1664 out:
1665 	for (; i < num; i++)
1666 		rte_pktmbuf_free(mbufs[i]);
1667 
1668 	return -1;
1669 }
1670 
1671 /**
1672  * DPDK callback to configure the receive queue.
1673  *
1674  * @param dev
1675  *   Pointer to Ethernet device structure.
1676  * @param idx
1677  *   RX queue index.
1678  * @param desc
1679  *   Number of descriptors to configure in queue.
1680  * @param socket
1681  *   NUMA socket on which memory must be allocated.
1682  * @param conf
1683  *   Thresholds parameters.
1684  * @param mp
1685  *   Memory pool for buffer allocations.
1686  *
1687  * @return
1688  *   0 on success, negative error value otherwise.
1689  */
1690 static int
1691 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1692 		    unsigned int socket,
1693 		    const struct rte_eth_rxconf *conf,
1694 		    struct rte_mempool *mp)
1695 {
1696 	struct mrvl_priv *priv = dev->data->dev_private;
1697 	struct mrvl_rxq *rxq;
1698 	uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
1699 	uint32_t max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1700 	int ret, tc, inq;
1701 	uint64_t offloads;
1702 
1703 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
1704 
1705 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1706 		/*
1707 		 * Unknown TC mapping, mapping will not have a correct queue.
1708 		 */
1709 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
1710 			idx, priv->ppio_id);
1711 		return -EFAULT;
1712 	}
1713 
1714 	frame_size = buf_size - RTE_PKTMBUF_HEADROOM - MRVL_PKT_EFFEC_OFFS;
1715 	if (frame_size < max_rx_pkt_len) {
1716 		MRVL_LOG(WARNING,
1717 			"Mbuf size must be increased to %u bytes to hold up "
1718 			"to %u bytes of data.",
1719 			buf_size + max_rx_pkt_len - frame_size,
1720 			max_rx_pkt_len);
1721 		dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
1722 		MRVL_LOG(INFO, "Setting max rx pkt len to %u",
1723 			dev->data->dev_conf.rxmode.max_rx_pkt_len);
1724 	}
1725 
1726 	if (dev->data->rx_queues[idx]) {
1727 		rte_free(dev->data->rx_queues[idx]);
1728 		dev->data->rx_queues[idx] = NULL;
1729 	}
1730 
1731 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
1732 	if (!rxq)
1733 		return -ENOMEM;
1734 
1735 	rxq->priv = priv;
1736 	rxq->mp = mp;
1737 	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
1738 	rxq->queue_id = idx;
1739 	rxq->port_id = dev->data->port_id;
1740 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
1741 
1742 	tc = priv->rxq_map[rxq->queue_id].tc,
1743 	inq = priv->rxq_map[rxq->queue_id].inq;
1744 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
1745 		desc;
1746 
1747 	ret = mrvl_fill_bpool(rxq, desc);
1748 	if (ret) {
1749 		rte_free(rxq);
1750 		return ret;
1751 	}
1752 
1753 	priv->bpool_init_size += desc;
1754 
1755 	dev->data->rx_queues[idx] = rxq;
1756 
1757 	return 0;
1758 }
1759 
1760 /**
1761  * DPDK callback to release the receive queue.
1762  *
1763  * @param rxq
1764  *   Generic receive queue pointer.
1765  */
1766 static void
1767 mrvl_rx_queue_release(void *rxq)
1768 {
1769 	struct mrvl_rxq *q = rxq;
1770 	struct pp2_ppio_tc_params *tc_params;
1771 	int i, num, tc, inq;
1772 	struct pp2_hif *hif;
1773 	unsigned int core_id = rte_lcore_id();
1774 
1775 	if (core_id == LCORE_ID_ANY)
1776 		core_id = rte_get_master_lcore();
1777 
1778 	if (!q)
1779 		return;
1780 
1781 	hif = mrvl_get_hif(q->priv, core_id);
1782 
1783 	if (!hif)
1784 		return;
1785 
1786 	tc = q->priv->rxq_map[q->queue_id].tc;
1787 	inq = q->priv->rxq_map[q->queue_id].inq;
1788 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
1789 	num = tc_params->inqs_params[inq].size;
1790 	for (i = 0; i < num; i++) {
1791 		struct pp2_buff_inf inf;
1792 		uint64_t addr;
1793 
1794 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
1795 		addr = cookie_addr_high | inf.cookie;
1796 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1797 	}
1798 
1799 	rte_free(q);
1800 }
1801 
1802 /**
1803  * DPDK callback to configure the transmit queue.
1804  *
1805  * @param dev
1806  *   Pointer to Ethernet device structure.
1807  * @param idx
1808  *   Transmit queue index.
1809  * @param desc
1810  *   Number of descriptors to configure in the queue.
1811  * @param socket
1812  *   NUMA socket on which memory must be allocated.
1813  * @param conf
1814  *   Tx queue configuration parameters.
1815  *
1816  * @return
1817  *   0 on success, negative error value otherwise.
1818  */
1819 static int
1820 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1821 		    unsigned int socket,
1822 		    const struct rte_eth_txconf *conf)
1823 {
1824 	struct mrvl_priv *priv = dev->data->dev_private;
1825 	struct mrvl_txq *txq;
1826 
1827 	if (dev->data->tx_queues[idx]) {
1828 		rte_free(dev->data->tx_queues[idx]);
1829 		dev->data->tx_queues[idx] = NULL;
1830 	}
1831 
1832 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
1833 	if (!txq)
1834 		return -ENOMEM;
1835 
1836 	txq->priv = priv;
1837 	txq->queue_id = idx;
1838 	txq->port_id = dev->data->port_id;
1839 	txq->tx_deferred_start = conf->tx_deferred_start;
1840 	dev->data->tx_queues[idx] = txq;
1841 
1842 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
1843 
1844 	return 0;
1845 }
1846 
1847 /**
1848  * DPDK callback to release the transmit queue.
1849  *
1850  * @param txq
1851  *   Generic transmit queue pointer.
1852  */
1853 static void
1854 mrvl_tx_queue_release(void *txq)
1855 {
1856 	struct mrvl_txq *q = txq;
1857 
1858 	if (!q)
1859 		return;
1860 
1861 	rte_free(q);
1862 }
1863 
1864 /**
1865  * DPDK callback to get flow control configuration.
1866  *
1867  * @param dev
1868  *  Pointer to Ethernet device structure.
1869  * @param fc_conf
1870  *  Pointer to the flow control configuration.
1871  *
1872  * @return
1873  *  0 on success, negative error value otherwise.
1874  */
1875 static int
1876 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1877 {
1878 	struct mrvl_priv *priv = dev->data->dev_private;
1879 	int ret, en;
1880 
1881 	if (!priv)
1882 		return -EPERM;
1883 
1884 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
1885 	if (ret) {
1886 		MRVL_LOG(ERR, "Failed to read rx pause state");
1887 		return ret;
1888 	}
1889 
1890 	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
1891 
1892 	return 0;
1893 }
1894 
1895 /**
1896  * DPDK callback to set flow control configuration.
1897  *
1898  * @param dev
1899  *  Pointer to Ethernet device structure.
1900  * @param fc_conf
1901  *  Pointer to the flow control configuration.
1902  *
1903  * @return
1904  *  0 on success, negative error value otherwise.
1905  */
1906 static int
1907 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1908 {
1909 	struct mrvl_priv *priv = dev->data->dev_private;
1910 
1911 	if (!priv)
1912 		return -EPERM;
1913 
1914 	if (fc_conf->high_water ||
1915 	    fc_conf->low_water ||
1916 	    fc_conf->pause_time ||
1917 	    fc_conf->mac_ctrl_frame_fwd ||
1918 	    fc_conf->autoneg) {
1919 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
1920 
1921 		return -EINVAL;
1922 	}
1923 
1924 	if (fc_conf->mode == RTE_FC_NONE ||
1925 	    fc_conf->mode == RTE_FC_RX_PAUSE) {
1926 		int ret, en;
1927 
1928 		en = fc_conf->mode == RTE_FC_NONE ? 0 : 1;
1929 		ret = pp2_ppio_set_rx_pause(priv->ppio, en);
1930 		if (ret)
1931 			MRVL_LOG(ERR,
1932 				"Failed to change flowctrl on RX side");
1933 
1934 		return ret;
1935 	}
1936 
1937 	return 0;
1938 }
1939 
1940 /**
1941  * Update RSS hash configuration
1942  *
1943  * @param dev
1944  *   Pointer to Ethernet device structure.
1945  * @param rss_conf
1946  *   Pointer to RSS configuration.
1947  *
1948  * @return
1949  *   0 on success, negative error value otherwise.
1950  */
1951 static int
1952 mrvl_rss_hash_update(struct rte_eth_dev *dev,
1953 		     struct rte_eth_rss_conf *rss_conf)
1954 {
1955 	struct mrvl_priv *priv = dev->data->dev_private;
1956 
1957 	if (priv->isolated)
1958 		return -ENOTSUP;
1959 
1960 	return mrvl_configure_rss(priv, rss_conf);
1961 }
1962 
1963 /**
1964  * DPDK callback to get RSS hash configuration.
1965  *
1966  * @param dev
1967  *   Pointer to Ethernet device structure.
1968  * @rss_conf
1969  *   Pointer to RSS configuration.
1970  *
1971  * @return
1972  *   Always 0.
1973  */
1974 static int
1975 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
1976 		       struct rte_eth_rss_conf *rss_conf)
1977 {
1978 	struct mrvl_priv *priv = dev->data->dev_private;
1979 	enum pp2_ppio_hash_type hash_type =
1980 		priv->ppio_params.inqs_params.hash_type;
1981 
1982 	rss_conf->rss_key = NULL;
1983 
1984 	if (hash_type == PP2_PPIO_HASH_T_NONE)
1985 		rss_conf->rss_hf = 0;
1986 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
1987 		rss_conf->rss_hf = ETH_RSS_IPV4;
1988 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
1989 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
1990 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
1991 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
1992 
1993 	return 0;
1994 }
1995 
1996 /**
1997  * DPDK callback to get rte_flow callbacks.
1998  *
1999  * @param dev
2000  *   Pointer to the device structure.
2001  * @param filer_type
2002  *   Flow filter type.
2003  * @param filter_op
2004  *   Flow filter operation.
2005  * @param arg
2006  *   Pointer to pass the flow ops.
2007  *
2008  * @return
2009  *   0 on success, negative error value otherwise.
2010  */
2011 static int
2012 mrvl_eth_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
2013 		     enum rte_filter_type filter_type,
2014 		     enum rte_filter_op filter_op, void *arg)
2015 {
2016 	switch (filter_type) {
2017 	case RTE_ETH_FILTER_GENERIC:
2018 		if (filter_op != RTE_ETH_FILTER_GET)
2019 			return -EINVAL;
2020 		*(const void **)arg = &mrvl_flow_ops;
2021 		return 0;
2022 	default:
2023 		MRVL_LOG(WARNING, "Filter type (%d) not supported",
2024 				filter_type);
2025 		return -EINVAL;
2026 	}
2027 }
2028 
2029 /**
2030  * DPDK callback to get rte_mtr callbacks.
2031  *
2032  * @param dev
2033  *   Pointer to the device structure.
2034  * @param ops
2035  *   Pointer to pass the mtr ops.
2036  *
2037  * @return
2038  *   Always 0.
2039  */
2040 static int
2041 mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2042 {
2043 	*(const void **)ops = &mrvl_mtr_ops;
2044 
2045 	return 0;
2046 }
2047 
2048 /**
2049  * DPDK callback to get rte_tm callbacks.
2050  *
2051  * @param dev
2052  *   Pointer to the device structure.
2053  * @param ops
2054  *   Pointer to pass the tm ops.
2055  *
2056  * @return
2057  *   Always 0.
2058  */
2059 static int
2060 mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2061 {
2062 	*(const void **)ops = &mrvl_tm_ops;
2063 
2064 	return 0;
2065 }
2066 
2067 static const struct eth_dev_ops mrvl_ops = {
2068 	.dev_configure = mrvl_dev_configure,
2069 	.dev_start = mrvl_dev_start,
2070 	.dev_stop = mrvl_dev_stop,
2071 	.dev_set_link_up = mrvl_dev_set_link_up,
2072 	.dev_set_link_down = mrvl_dev_set_link_down,
2073 	.dev_close = mrvl_dev_close,
2074 	.link_update = mrvl_link_update,
2075 	.promiscuous_enable = mrvl_promiscuous_enable,
2076 	.allmulticast_enable = mrvl_allmulticast_enable,
2077 	.promiscuous_disable = mrvl_promiscuous_disable,
2078 	.allmulticast_disable = mrvl_allmulticast_disable,
2079 	.mac_addr_remove = mrvl_mac_addr_remove,
2080 	.mac_addr_add = mrvl_mac_addr_add,
2081 	.mac_addr_set = mrvl_mac_addr_set,
2082 	.mtu_set = mrvl_mtu_set,
2083 	.stats_get = mrvl_stats_get,
2084 	.stats_reset = mrvl_stats_reset,
2085 	.xstats_get = mrvl_xstats_get,
2086 	.xstats_reset = mrvl_xstats_reset,
2087 	.xstats_get_names = mrvl_xstats_get_names,
2088 	.dev_infos_get = mrvl_dev_infos_get,
2089 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
2090 	.rxq_info_get = mrvl_rxq_info_get,
2091 	.txq_info_get = mrvl_txq_info_get,
2092 	.vlan_filter_set = mrvl_vlan_filter_set,
2093 	.tx_queue_start = mrvl_tx_queue_start,
2094 	.tx_queue_stop = mrvl_tx_queue_stop,
2095 	.rx_queue_setup = mrvl_rx_queue_setup,
2096 	.rx_queue_release = mrvl_rx_queue_release,
2097 	.tx_queue_setup = mrvl_tx_queue_setup,
2098 	.tx_queue_release = mrvl_tx_queue_release,
2099 	.flow_ctrl_get = mrvl_flow_ctrl_get,
2100 	.flow_ctrl_set = mrvl_flow_ctrl_set,
2101 	.rss_hash_update = mrvl_rss_hash_update,
2102 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
2103 	.filter_ctrl = mrvl_eth_filter_ctrl,
2104 	.mtr_ops_get = mrvl_mtr_ops_get,
2105 	.tm_ops_get = mrvl_tm_ops_get,
2106 };
2107 
2108 /**
2109  * Return packet type information and l3/l4 offsets.
2110  *
2111  * @param desc
2112  *   Pointer to the received packet descriptor.
2113  * @param l3_offset
2114  *   l3 packet offset.
2115  * @param l4_offset
2116  *   l4 packet offset.
2117  *
2118  * @return
2119  *   Packet type information.
2120  */
2121 static inline uint64_t
2122 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
2123 				    uint8_t *l3_offset, uint8_t *l4_offset)
2124 {
2125 	enum pp2_inq_l3_type l3_type;
2126 	enum pp2_inq_l4_type l4_type;
2127 	enum pp2_inq_vlan_tag vlan_tag;
2128 	uint64_t packet_type;
2129 
2130 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
2131 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
2132 	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
2133 
2134 	packet_type = RTE_PTYPE_L2_ETHER;
2135 
2136 	switch (vlan_tag) {
2137 	case PP2_INQ_VLAN_TAG_SINGLE:
2138 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
2139 		break;
2140 	case PP2_INQ_VLAN_TAG_DOUBLE:
2141 	case PP2_INQ_VLAN_TAG_TRIPLE:
2142 		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
2143 		break;
2144 	default:
2145 		break;
2146 	}
2147 
2148 	switch (l3_type) {
2149 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
2150 		packet_type |= RTE_PTYPE_L3_IPV4;
2151 		break;
2152 	case PP2_INQ_L3_TYPE_IPV4_OK:
2153 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
2154 		break;
2155 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
2156 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
2157 		break;
2158 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
2159 		packet_type |= RTE_PTYPE_L3_IPV6;
2160 		break;
2161 	case PP2_INQ_L3_TYPE_IPV6_EXT:
2162 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
2163 		break;
2164 	case PP2_INQ_L3_TYPE_ARP:
2165 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
2166 		/*
2167 		 * In case of ARP l4_offset is set to wrong value.
2168 		 * Set it to proper one so that later on mbuf->l3_len can be
2169 		 * calculated subtracting l4_offset and l3_offset.
2170 		 */
2171 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
2172 		break;
2173 	default:
2174 		MRVL_LOG(DEBUG, "Failed to recognise l3 packet type");
2175 		break;
2176 	}
2177 
2178 	switch (l4_type) {
2179 	case PP2_INQ_L4_TYPE_TCP:
2180 		packet_type |= RTE_PTYPE_L4_TCP;
2181 		break;
2182 	case PP2_INQ_L4_TYPE_UDP:
2183 		packet_type |= RTE_PTYPE_L4_UDP;
2184 		break;
2185 	default:
2186 		MRVL_LOG(DEBUG, "Failed to recognise l4 packet type");
2187 		break;
2188 	}
2189 
2190 	return packet_type;
2191 }
2192 
2193 /**
2194  * Get offload information from the received packet descriptor.
2195  *
2196  * @param desc
2197  *   Pointer to the received packet descriptor.
2198  *
2199  * @return
2200  *   Mbuf offload flags.
2201  */
2202 static inline uint64_t
2203 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc)
2204 {
2205 	uint64_t flags;
2206 	enum pp2_inq_desc_status status;
2207 
2208 	status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2209 	if (unlikely(status != PP2_DESC_ERR_OK))
2210 		flags = PKT_RX_IP_CKSUM_BAD;
2211 	else
2212 		flags = PKT_RX_IP_CKSUM_GOOD;
2213 
2214 	status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2215 	if (unlikely(status != PP2_DESC_ERR_OK))
2216 		flags |= PKT_RX_L4_CKSUM_BAD;
2217 	else
2218 		flags |= PKT_RX_L4_CKSUM_GOOD;
2219 
2220 	return flags;
2221 }
2222 
2223 /**
2224  * DPDK callback for receive.
2225  *
2226  * @param rxq
2227  *   Generic pointer to the receive queue.
2228  * @param rx_pkts
2229  *   Array to store received packets.
2230  * @param nb_pkts
2231  *   Maximum number of packets in array.
2232  *
2233  * @return
2234  *   Number of packets successfully received.
2235  */
2236 static uint16_t
2237 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2238 {
2239 	struct mrvl_rxq *q = rxq;
2240 	struct pp2_ppio_desc descs[nb_pkts];
2241 	struct pp2_bpool *bpool;
2242 	int i, ret, rx_done = 0;
2243 	int num;
2244 	struct pp2_hif *hif;
2245 	unsigned int core_id = rte_lcore_id();
2246 
2247 	hif = mrvl_get_hif(q->priv, core_id);
2248 
2249 	if (unlikely(!q->priv->ppio || !hif))
2250 		return 0;
2251 
2252 	bpool = q->priv->bpool;
2253 
2254 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2255 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2256 	if (unlikely(ret < 0)) {
2257 		MRVL_LOG(ERR, "Failed to receive packets");
2258 		return 0;
2259 	}
2260 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2261 
2262 	for (i = 0; i < nb_pkts; i++) {
2263 		struct rte_mbuf *mbuf;
2264 		uint8_t l3_offset, l4_offset;
2265 		enum pp2_inq_desc_status status;
2266 		uint64_t addr;
2267 
2268 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2269 			struct pp2_ppio_desc *pref_desc;
2270 			u64 pref_addr;
2271 
2272 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2273 			pref_addr = cookie_addr_high |
2274 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2275 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2276 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2277 		}
2278 
2279 		addr = cookie_addr_high |
2280 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2281 		mbuf = (struct rte_mbuf *)addr;
2282 		rte_pktmbuf_reset(mbuf);
2283 
2284 		/* drop packet in case of mac, overrun or resource error */
2285 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2286 		if (unlikely(status != PP2_DESC_ERR_OK)) {
2287 			struct pp2_buff_inf binf = {
2288 				.addr = rte_mbuf_data_iova_default(mbuf),
2289 				.cookie = (uint64_t)mbuf,
2290 			};
2291 
2292 			pp2_bpool_put_buff(hif, bpool, &binf);
2293 			mrvl_port_bpool_size
2294 				[bpool->pp2_id][bpool->id][core_id]++;
2295 			q->drop_mac++;
2296 			continue;
2297 		}
2298 
2299 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2300 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2301 		mbuf->data_len = mbuf->pkt_len;
2302 		mbuf->port = q->port_id;
2303 		mbuf->packet_type =
2304 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2305 							    &l3_offset,
2306 							    &l4_offset);
2307 		mbuf->l2_len = l3_offset;
2308 		mbuf->l3_len = l4_offset - l3_offset;
2309 
2310 		if (likely(q->cksum_enabled))
2311 			mbuf->ol_flags = mrvl_desc_to_ol_flags(&descs[i]);
2312 
2313 		rx_pkts[rx_done++] = mbuf;
2314 		q->bytes_recv += mbuf->pkt_len;
2315 	}
2316 
2317 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2318 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2319 
2320 		if (unlikely(num <= q->priv->bpool_min_size ||
2321 			     (!rx_done && num < q->priv->bpool_init_size))) {
2322 			ret = mrvl_fill_bpool(q, MRVL_BURST_SIZE);
2323 			if (ret)
2324 				MRVL_LOG(ERR, "Failed to fill bpool");
2325 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2326 			int i;
2327 			int pkt_to_remove = num - q->priv->bpool_init_size;
2328 			struct rte_mbuf *mbuf;
2329 			struct pp2_buff_inf buff;
2330 
2331 			MRVL_LOG(DEBUG,
2332 				"port-%d:%d: bpool %d oversize - remove %d buffers (pool size: %d -> %d)",
2333 				bpool->pp2_id, q->priv->ppio->port_id,
2334 				bpool->id, pkt_to_remove, num,
2335 				q->priv->bpool_init_size);
2336 
2337 			for (i = 0; i < pkt_to_remove; i++) {
2338 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2339 				if (ret)
2340 					break;
2341 				mbuf = (struct rte_mbuf *)
2342 					(cookie_addr_high | buff.cookie);
2343 				rte_pktmbuf_free(mbuf);
2344 			}
2345 			mrvl_port_bpool_size
2346 				[bpool->pp2_id][bpool->id][core_id] -= i;
2347 		}
2348 		rte_spinlock_unlock(&q->priv->lock);
2349 	}
2350 
2351 	return rx_done;
2352 }
2353 
2354 /**
2355  * Prepare offload information.
2356  *
2357  * @param ol_flags
2358  *   Offload flags.
2359  * @param packet_type
2360  *   Packet type bitfield.
2361  * @param l3_type
2362  *   Pointer to the pp2_ouq_l3_type structure.
2363  * @param l4_type
2364  *   Pointer to the pp2_outq_l4_type structure.
2365  * @param gen_l3_cksum
2366  *   Will be set to 1 in case l3 checksum is computed.
2367  * @param l4_cksum
2368  *   Will be set to 1 in case l4 checksum is computed.
2369  *
2370  * @return
2371  *   0 on success, negative error value otherwise.
2372  */
2373 static inline int
2374 mrvl_prepare_proto_info(uint64_t ol_flags, uint32_t packet_type,
2375 			enum pp2_outq_l3_type *l3_type,
2376 			enum pp2_outq_l4_type *l4_type,
2377 			int *gen_l3_cksum,
2378 			int *gen_l4_cksum)
2379 {
2380 	/*
2381 	 * Based on ol_flags prepare information
2382 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2383 	 * for offloading.
2384 	 */
2385 	if (ol_flags & PKT_TX_IPV4) {
2386 		*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2387 		*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2388 	} else if (ol_flags & PKT_TX_IPV6) {
2389 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2390 		/* no checksum for ipv6 header */
2391 		*gen_l3_cksum = 0;
2392 	} else {
2393 		/* if something different then stop processing */
2394 		return -1;
2395 	}
2396 
2397 	ol_flags &= PKT_TX_L4_MASK;
2398 	if ((packet_type & RTE_PTYPE_L4_TCP) &&
2399 	    ol_flags == PKT_TX_TCP_CKSUM) {
2400 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2401 		*gen_l4_cksum = 1;
2402 	} else if ((packet_type & RTE_PTYPE_L4_UDP) &&
2403 		   ol_flags == PKT_TX_UDP_CKSUM) {
2404 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2405 		*gen_l4_cksum = 1;
2406 	} else {
2407 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2408 		/* no checksum for other type */
2409 		*gen_l4_cksum = 0;
2410 	}
2411 
2412 	return 0;
2413 }
2414 
2415 /**
2416  * Release already sent buffers to bpool (buffer-pool).
2417  *
2418  * @param ppio
2419  *   Pointer to the port structure.
2420  * @param hif
2421  *   Pointer to the MUSDK hardware interface.
2422  * @param sq
2423  *   Pointer to the shadow queue.
2424  * @param qid
2425  *   Queue id number.
2426  * @param force
2427  *   Force releasing packets.
2428  */
2429 static inline void
2430 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2431 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2432 		       int qid, int force)
2433 {
2434 	struct buff_release_entry *entry;
2435 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2436 	int i;
2437 
2438 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2439 
2440 	sq->num_to_release += nb_done;
2441 
2442 	if (likely(!force &&
2443 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2444 		return;
2445 
2446 	nb_done = sq->num_to_release;
2447 	sq->num_to_release = 0;
2448 
2449 	for (i = 0; i < nb_done; i++) {
2450 		entry = &sq->ent[sq->tail + num];
2451 		if (unlikely(!entry->buff.addr)) {
2452 			MRVL_LOG(ERR,
2453 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2454 				sq->tail, (u64)entry->buff.cookie,
2455 				(u64)entry->buff.addr);
2456 			skip_bufs = 1;
2457 			goto skip;
2458 		}
2459 
2460 		if (unlikely(!entry->bpool)) {
2461 			struct rte_mbuf *mbuf;
2462 
2463 			mbuf = (struct rte_mbuf *)
2464 			       (cookie_addr_high | entry->buff.cookie);
2465 			rte_pktmbuf_free(mbuf);
2466 			skip_bufs = 1;
2467 			goto skip;
2468 		}
2469 
2470 		mrvl_port_bpool_size
2471 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2472 		num++;
2473 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2474 			goto skip;
2475 		continue;
2476 skip:
2477 		if (likely(num))
2478 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2479 		num += skip_bufs;
2480 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2481 		sq->size -= num;
2482 		num = 0;
2483 		skip_bufs = 0;
2484 	}
2485 
2486 	if (likely(num)) {
2487 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2488 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2489 		sq->size -= num;
2490 	}
2491 }
2492 
2493 /**
2494  * DPDK callback for transmit.
2495  *
2496  * @param txq
2497  *   Generic pointer transmit queue.
2498  * @param tx_pkts
2499  *   Packets to transmit.
2500  * @param nb_pkts
2501  *   Number of packets in array.
2502  *
2503  * @return
2504  *   Number of packets successfully transmitted.
2505  */
2506 static uint16_t
2507 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2508 {
2509 	struct mrvl_txq *q = txq;
2510 	struct mrvl_shadow_txq *sq;
2511 	struct pp2_hif *hif;
2512 	struct pp2_ppio_desc descs[nb_pkts];
2513 	unsigned int core_id = rte_lcore_id();
2514 	int i, ret, bytes_sent = 0;
2515 	uint16_t num, sq_free_size;
2516 	uint64_t addr;
2517 
2518 	hif = mrvl_get_hif(q->priv, core_id);
2519 	sq = &q->shadow_txqs[core_id];
2520 
2521 	if (unlikely(!q->priv->ppio || !hif))
2522 		return 0;
2523 
2524 	if (sq->size)
2525 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2526 				       sq, q->queue_id, 0);
2527 
2528 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2529 	if (unlikely(nb_pkts > sq_free_size)) {
2530 		MRVL_LOG(DEBUG,
2531 			"No room in shadow queue for %d packets! %d packets will be sent.",
2532 			nb_pkts, sq_free_size);
2533 		nb_pkts = sq_free_size;
2534 	}
2535 
2536 	for (i = 0; i < nb_pkts; i++) {
2537 		struct rte_mbuf *mbuf = tx_pkts[i];
2538 		int gen_l3_cksum, gen_l4_cksum;
2539 		enum pp2_outq_l3_type l3_type;
2540 		enum pp2_outq_l4_type l4_type;
2541 
2542 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2543 			struct rte_mbuf *pref_pkt_hdr;
2544 
2545 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2546 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2547 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2548 		}
2549 
2550 		mrvl_fill_shadowq(sq, mbuf);
2551 		mrvl_fill_desc(&descs[i], mbuf);
2552 
2553 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2554 		/*
2555 		 * in case unsupported ol_flags were passed
2556 		 * do not update descriptor offload information
2557 		 */
2558 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2559 					      &l3_type, &l4_type, &gen_l3_cksum,
2560 					      &gen_l4_cksum);
2561 		if (unlikely(ret))
2562 			continue;
2563 
2564 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2565 						  mbuf->l2_len,
2566 						  mbuf->l2_len + mbuf->l3_len,
2567 						  gen_l3_cksum, gen_l4_cksum);
2568 	}
2569 
2570 	num = nb_pkts;
2571 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2572 	/* number of packets that were not sent */
2573 	if (unlikely(num > nb_pkts)) {
2574 		for (i = nb_pkts; i < num; i++) {
2575 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2576 				MRVL_PP2_TX_SHADOWQ_MASK;
2577 			addr = cookie_addr_high | sq->ent[sq->head].buff.cookie;
2578 			bytes_sent -=
2579 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2580 		}
2581 		sq->size -= num - nb_pkts;
2582 	}
2583 
2584 	q->bytes_sent += bytes_sent;
2585 
2586 	return nb_pkts;
2587 }
2588 
2589 /** DPDK callback for S/G transmit.
2590  *
2591  * @param txq
2592  *   Generic pointer transmit queue.
2593  * @param tx_pkts
2594  *   Packets to transmit.
2595  * @param nb_pkts
2596  *   Number of packets in array.
2597  *
2598  * @return
2599  *   Number of packets successfully transmitted.
2600  */
2601 static uint16_t
2602 mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
2603 		     uint16_t nb_pkts)
2604 {
2605 	struct mrvl_txq *q = txq;
2606 	struct mrvl_shadow_txq *sq;
2607 	struct pp2_hif *hif;
2608 	struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
2609 	struct pp2_ppio_sg_pkts pkts;
2610 	uint8_t frags[nb_pkts];
2611 	unsigned int core_id = rte_lcore_id();
2612 	int i, j, ret, bytes_sent = 0;
2613 	int tail, tail_first;
2614 	uint16_t num, sq_free_size;
2615 	uint16_t nb_segs, total_descs = 0;
2616 	uint64_t addr;
2617 
2618 	hif = mrvl_get_hif(q->priv, core_id);
2619 	sq = &q->shadow_txqs[core_id];
2620 	pkts.frags = frags;
2621 	pkts.num = 0;
2622 
2623 	if (unlikely(!q->priv->ppio || !hif))
2624 		return 0;
2625 
2626 	if (sq->size)
2627 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2628 				       sq, q->queue_id, 0);
2629 
2630 	/* Save shadow queue free size */
2631 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2632 
2633 	tail = 0;
2634 	for (i = 0; i < nb_pkts; i++) {
2635 		struct rte_mbuf *mbuf = tx_pkts[i];
2636 		struct rte_mbuf *seg = NULL;
2637 		int gen_l3_cksum, gen_l4_cksum;
2638 		enum pp2_outq_l3_type l3_type;
2639 		enum pp2_outq_l4_type l4_type;
2640 
2641 		nb_segs = mbuf->nb_segs;
2642 		tail_first = tail;
2643 		total_descs += nb_segs;
2644 
2645 		/*
2646 		 * Check if total_descs does not exceed
2647 		 * shadow queue free size
2648 		 */
2649 		if (unlikely(total_descs > sq_free_size)) {
2650 			total_descs -= nb_segs;
2651 			RTE_LOG(DEBUG, PMD,
2652 				"No room in shadow queue for %d packets! "
2653 				"%d packets will be sent.\n",
2654 				nb_pkts, i);
2655 			break;
2656 		}
2657 
2658 		/* Check if nb_segs does not exceed the max nb of desc per
2659 		 * fragmented packet
2660 		 */
2661 		if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
2662 			total_descs -= nb_segs;
2663 			RTE_LOG(ERR, PMD,
2664 				"Too many segments. Packet won't be sent.\n");
2665 			break;
2666 		}
2667 
2668 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2669 			struct rte_mbuf *pref_pkt_hdr;
2670 
2671 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2672 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2673 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2674 		}
2675 
2676 		pkts.frags[pkts.num] = nb_segs;
2677 		pkts.num++;
2678 
2679 		seg = mbuf;
2680 		for (j = 0; j < nb_segs - 1; j++) {
2681 			/* For the subsequent segments, set shadow queue
2682 			 * buffer to NULL
2683 			 */
2684 			mrvl_fill_shadowq(sq, NULL);
2685 			mrvl_fill_desc(&descs[tail], seg);
2686 
2687 			tail++;
2688 			seg = seg->next;
2689 		}
2690 		/* Put first mbuf info in last shadow queue entry */
2691 		mrvl_fill_shadowq(sq, mbuf);
2692 		/* Update descriptor with last segment */
2693 		mrvl_fill_desc(&descs[tail++], seg);
2694 
2695 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2696 		/* In case unsupported ol_flags were passed
2697 		 * do not update descriptor offload information
2698 		 */
2699 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2700 					      &l3_type, &l4_type, &gen_l3_cksum,
2701 					      &gen_l4_cksum);
2702 		if (unlikely(ret))
2703 			continue;
2704 
2705 		pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
2706 						  l4_type, mbuf->l2_len,
2707 						  mbuf->l2_len + mbuf->l3_len,
2708 						  gen_l3_cksum, gen_l4_cksum);
2709 	}
2710 
2711 	num = total_descs;
2712 	pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
2713 			 &total_descs, &pkts);
2714 	/* number of packets that were not sent */
2715 	if (unlikely(num > total_descs)) {
2716 		for (i = total_descs; i < num; i++) {
2717 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2718 				MRVL_PP2_TX_SHADOWQ_MASK;
2719 
2720 			addr = sq->ent[sq->head].buff.cookie;
2721 			if (addr)
2722 				bytes_sent -=
2723 					rte_pktmbuf_pkt_len((struct rte_mbuf *)
2724 						(cookie_addr_high | addr));
2725 		}
2726 		sq->size -= num - total_descs;
2727 		nb_pkts = pkts.num;
2728 	}
2729 
2730 	q->bytes_sent += bytes_sent;
2731 
2732 	return nb_pkts;
2733 }
2734 
2735 /**
2736  * Initialize packet processor.
2737  *
2738  * @return
2739  *   0 on success, negative error value otherwise.
2740  */
2741 static int
2742 mrvl_init_pp2(void)
2743 {
2744 	struct pp2_init_params init_params;
2745 
2746 	memset(&init_params, 0, sizeof(init_params));
2747 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
2748 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
2749 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
2750 
2751 	return pp2_init(&init_params);
2752 }
2753 
2754 /**
2755  * Deinitialize packet processor.
2756  *
2757  * @return
2758  *   0 on success, negative error value otherwise.
2759  */
2760 static void
2761 mrvl_deinit_pp2(void)
2762 {
2763 	pp2_deinit();
2764 }
2765 
2766 /**
2767  * Create private device structure.
2768  *
2769  * @param dev_name
2770  *   Pointer to the port name passed in the initialization parameters.
2771  *
2772  * @return
2773  *   Pointer to the newly allocated private device structure.
2774  */
2775 static struct mrvl_priv *
2776 mrvl_priv_create(const char *dev_name)
2777 {
2778 	struct pp2_bpool_params bpool_params;
2779 	char match[MRVL_MATCH_LEN];
2780 	struct mrvl_priv *priv;
2781 	int ret, bpool_bit;
2782 
2783 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
2784 	if (!priv)
2785 		return NULL;
2786 
2787 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
2788 				       &priv->pp_id, &priv->ppio_id);
2789 	if (ret)
2790 		goto out_free_priv;
2791 
2792 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
2793 				     PP2_BPOOL_NUM_POOLS);
2794 	if (bpool_bit < 0)
2795 		goto out_free_priv;
2796 	priv->bpool_bit = bpool_bit;
2797 
2798 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
2799 		 priv->bpool_bit);
2800 	memset(&bpool_params, 0, sizeof(bpool_params));
2801 	bpool_params.match = match;
2802 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
2803 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
2804 	if (ret)
2805 		goto out_clear_bpool_bit;
2806 
2807 	priv->ppio_params.type = PP2_PPIO_T_NIC;
2808 	rte_spinlock_init(&priv->lock);
2809 
2810 	return priv;
2811 out_clear_bpool_bit:
2812 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2813 out_free_priv:
2814 	rte_free(priv);
2815 	return NULL;
2816 }
2817 
2818 /**
2819  * Create device representing Ethernet port.
2820  *
2821  * @param name
2822  *   Pointer to the port's name.
2823  *
2824  * @return
2825  *   0 on success, negative error value otherwise.
2826  */
2827 static int
2828 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
2829 {
2830 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
2831 	struct rte_eth_dev *eth_dev;
2832 	struct mrvl_priv *priv;
2833 	struct ifreq req;
2834 
2835 	eth_dev = rte_eth_dev_allocate(name);
2836 	if (!eth_dev)
2837 		return -ENOMEM;
2838 
2839 	priv = mrvl_priv_create(name);
2840 	if (!priv) {
2841 		ret = -ENOMEM;
2842 		goto out_free;
2843 	}
2844 	eth_dev->data->dev_private = priv;
2845 
2846 	eth_dev->data->mac_addrs =
2847 		rte_zmalloc("mac_addrs",
2848 			    RTE_ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
2849 	if (!eth_dev->data->mac_addrs) {
2850 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
2851 		ret = -ENOMEM;
2852 		goto out_free;
2853 	}
2854 
2855 	memset(&req, 0, sizeof(req));
2856 	strcpy(req.ifr_name, name);
2857 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
2858 	if (ret)
2859 		goto out_free;
2860 
2861 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
2862 	       req.ifr_addr.sa_data, RTE_ETHER_ADDR_LEN);
2863 
2864 	eth_dev->device = &vdev->device;
2865 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
2866 	mrvl_set_tx_function(eth_dev);
2867 	eth_dev->dev_ops = &mrvl_ops;
2868 
2869 	rte_eth_dev_probing_finish(eth_dev);
2870 	return 0;
2871 out_free:
2872 	rte_eth_dev_release_port(eth_dev);
2873 
2874 	return ret;
2875 }
2876 
2877 /**
2878  * Callback used by rte_kvargs_process() during argument parsing.
2879  *
2880  * @param key
2881  *   Pointer to the parsed key (unused).
2882  * @param value
2883  *   Pointer to the parsed value.
2884  * @param extra_args
2885  *   Pointer to the extra arguments which contains address of the
2886  *   table of pointers to parsed interface names.
2887  *
2888  * @return
2889  *   Always 0.
2890  */
2891 static int
2892 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
2893 		 void *extra_args)
2894 {
2895 	struct mrvl_ifnames *ifnames = extra_args;
2896 
2897 	ifnames->names[ifnames->idx++] = value;
2898 
2899 	return 0;
2900 }
2901 
2902 /**
2903  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
2904  */
2905 static void
2906 mrvl_deinit_hifs(void)
2907 {
2908 	int i;
2909 
2910 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
2911 		if (hifs[i])
2912 			pp2_hif_deinit(hifs[i]);
2913 	}
2914 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
2915 	memset(hifs, 0, sizeof(hifs));
2916 }
2917 
2918 /**
2919  * DPDK callback to register the virtual device.
2920  *
2921  * @param vdev
2922  *   Pointer to the virtual device.
2923  *
2924  * @return
2925  *   0 on success, negative error value otherwise.
2926  */
2927 static int
2928 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
2929 {
2930 	struct rte_kvargs *kvlist;
2931 	struct mrvl_ifnames ifnames;
2932 	int ret = -EINVAL;
2933 	uint32_t i, ifnum, cfgnum;
2934 	const char *params;
2935 
2936 	params = rte_vdev_device_args(vdev);
2937 	if (!params)
2938 		return -EINVAL;
2939 
2940 	kvlist = rte_kvargs_parse(params, valid_args);
2941 	if (!kvlist)
2942 		return -EINVAL;
2943 
2944 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
2945 	if (ifnum > RTE_DIM(ifnames.names))
2946 		goto out_free_kvlist;
2947 
2948 	ifnames.idx = 0;
2949 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
2950 			   mrvl_get_ifnames, &ifnames);
2951 
2952 
2953 	/*
2954 	 * The below system initialization should be done only once,
2955 	 * on the first provided configuration file
2956 	 */
2957 	if (!mrvl_qos_cfg) {
2958 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
2959 		MRVL_LOG(INFO, "Parsing config file!");
2960 		if (cfgnum > 1) {
2961 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
2962 			goto out_free_kvlist;
2963 		} else if (cfgnum == 1) {
2964 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
2965 					   mrvl_get_qoscfg, &mrvl_qos_cfg);
2966 		}
2967 	}
2968 
2969 	if (mrvl_dev_num)
2970 		goto init_devices;
2971 
2972 	MRVL_LOG(INFO, "Perform MUSDK initializations");
2973 
2974 	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
2975 	if (ret)
2976 		goto out_free_kvlist;
2977 
2978 	ret = mrvl_init_pp2();
2979 	if (ret) {
2980 		MRVL_LOG(ERR, "Failed to init PP!");
2981 		rte_mvep_deinit(MVEP_MOD_T_PP2);
2982 		goto out_free_kvlist;
2983 	}
2984 
2985 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
2986 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
2987 
2988 	mrvl_lcore_first = RTE_MAX_LCORE;
2989 	mrvl_lcore_last = 0;
2990 
2991 init_devices:
2992 	for (i = 0; i < ifnum; i++) {
2993 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
2994 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
2995 		if (ret)
2996 			goto out_cleanup;
2997 		mrvl_dev_num++;
2998 	}
2999 
3000 	rte_kvargs_free(kvlist);
3001 
3002 	return 0;
3003 out_cleanup:
3004 	rte_pmd_mrvl_remove(vdev);
3005 
3006 out_free_kvlist:
3007 	rte_kvargs_free(kvlist);
3008 
3009 	return ret;
3010 }
3011 
3012 /**
3013  * DPDK callback to remove virtual device.
3014  *
3015  * @param vdev
3016  *   Pointer to the removed virtual device.
3017  *
3018  * @return
3019  *   0 on success, negative error value otherwise.
3020  */
3021 static int
3022 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
3023 {
3024 	uint16_t port_id;
3025 
3026 	RTE_ETH_FOREACH_DEV(port_id) {
3027 		if (rte_eth_devices[port_id].device != &vdev->device)
3028 			continue;
3029 		rte_eth_dev_close(port_id);
3030 	}
3031 
3032 	return 0;
3033 }
3034 
3035 static struct rte_vdev_driver pmd_mrvl_drv = {
3036 	.probe = rte_pmd_mrvl_probe,
3037 	.remove = rte_pmd_mrvl_remove,
3038 };
3039 
3040 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
3041 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
3042 RTE_LOG_REGISTER(mrvl_logtype, pmd.net.mvpp2, NOTICE);
3043