xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision 1c9a958a8bfb7cb4e2082b97d01849d4ba91c13b)
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 <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 - RTE_ETHER_CRC_LEN + MRVL_PKT_OFFS > mbuf_data_size) {
445 		mru = mbuf_data_size + RTE_ETHER_CRC_LEN - 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 	ret = mrvl_mtu_set(dev, dev->data->mtu);
675 	if (ret)
676 		MRVL_LOG(ERR, "Failed to set MTU to %d", dev->data->mtu);
677 
678 	/* For default QoS config, don't start classifier. */
679 	if (mrvl_qos_cfg  &&
680 	    mrvl_qos_cfg->port[dev->data->port_id].use_global_defaults == 0) {
681 		ret = mrvl_start_qos_mapping(priv);
682 		if (ret) {
683 			MRVL_LOG(ERR, "Failed to setup QoS mapping");
684 			goto out;
685 		}
686 	}
687 
688 	ret = mrvl_dev_set_link_up(dev);
689 	if (ret) {
690 		MRVL_LOG(ERR, "Failed to set link up");
691 		goto out;
692 	}
693 
694 	/* start tx queues */
695 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
696 		struct mrvl_txq *txq = dev->data->tx_queues[i];
697 
698 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
699 
700 		if (!txq->tx_deferred_start)
701 			continue;
702 
703 		/*
704 		 * All txqs are started by default. Stop them
705 		 * so that tx_deferred_start works as expected.
706 		 */
707 		ret = mrvl_tx_queue_stop(dev, i);
708 		if (ret)
709 			goto out;
710 	}
711 
712 	mrvl_flow_init(dev);
713 	mrvl_mtr_init(dev);
714 	mrvl_set_tx_function(dev);
715 
716 	return 0;
717 out:
718 	MRVL_LOG(ERR, "Failed to start device");
719 	pp2_ppio_deinit(priv->ppio);
720 	return ret;
721 }
722 
723 /**
724  * Flush receive queues.
725  *
726  * @param dev
727  *   Pointer to Ethernet device structure.
728  */
729 static void
730 mrvl_flush_rx_queues(struct rte_eth_dev *dev)
731 {
732 	int i;
733 
734 	MRVL_LOG(INFO, "Flushing rx queues");
735 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
736 		int ret, num;
737 
738 		do {
739 			struct mrvl_rxq *q = dev->data->rx_queues[i];
740 			struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];
741 
742 			num = MRVL_PP2_RXD_MAX;
743 			ret = pp2_ppio_recv(q->priv->ppio,
744 					    q->priv->rxq_map[q->queue_id].tc,
745 					    q->priv->rxq_map[q->queue_id].inq,
746 					    descs, (uint16_t *)&num);
747 		} while (ret == 0 && num);
748 	}
749 }
750 
751 /**
752  * Flush transmit shadow queues.
753  *
754  * @param dev
755  *   Pointer to Ethernet device structure.
756  */
757 static void
758 mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
759 {
760 	int i, j;
761 	struct mrvl_txq *txq;
762 
763 	MRVL_LOG(INFO, "Flushing tx shadow queues");
764 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
765 		txq = (struct mrvl_txq *)dev->data->tx_queues[i];
766 
767 		for (j = 0; j < RTE_MAX_LCORE; j++) {
768 			struct mrvl_shadow_txq *sq;
769 
770 			if (!hifs[j])
771 				continue;
772 
773 			sq = &txq->shadow_txqs[j];
774 			mrvl_free_sent_buffers(txq->priv->ppio,
775 				hifs[j], j, sq, txq->queue_id, 1);
776 			while (sq->tail != sq->head) {
777 				uint64_t addr = cookie_addr_high |
778 					sq->ent[sq->tail].buff.cookie;
779 				rte_pktmbuf_free(
780 					(struct rte_mbuf *)addr);
781 				sq->tail = (sq->tail + 1) &
782 					    MRVL_PP2_TX_SHADOWQ_MASK;
783 			}
784 			memset(sq, 0, sizeof(*sq));
785 		}
786 	}
787 }
788 
789 /**
790  * Flush hardware bpool (buffer-pool).
791  *
792  * @param dev
793  *   Pointer to Ethernet device structure.
794  */
795 static void
796 mrvl_flush_bpool(struct rte_eth_dev *dev)
797 {
798 	struct mrvl_priv *priv = dev->data->dev_private;
799 	struct pp2_hif *hif;
800 	uint32_t num;
801 	int ret;
802 	unsigned int core_id = rte_lcore_id();
803 
804 	if (core_id == LCORE_ID_ANY)
805 		core_id = rte_get_main_lcore();
806 
807 	hif = mrvl_get_hif(priv, core_id);
808 
809 	ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
810 	if (ret) {
811 		MRVL_LOG(ERR, "Failed to get bpool buffers number");
812 		return;
813 	}
814 
815 	while (num--) {
816 		struct pp2_buff_inf inf;
817 		uint64_t addr;
818 
819 		ret = pp2_bpool_get_buff(hif, priv->bpool, &inf);
820 		if (ret)
821 			break;
822 
823 		addr = cookie_addr_high | inf.cookie;
824 		rte_pktmbuf_free((struct rte_mbuf *)addr);
825 	}
826 }
827 
828 /**
829  * DPDK callback to stop the device.
830  *
831  * @param dev
832  *   Pointer to Ethernet device structure.
833  */
834 static int
835 mrvl_dev_stop(struct rte_eth_dev *dev)
836 {
837 	return mrvl_dev_set_link_down(dev);
838 }
839 
840 /**
841  * DPDK callback to close the device.
842  *
843  * @param dev
844  *   Pointer to Ethernet device structure.
845  */
846 static int
847 mrvl_dev_close(struct rte_eth_dev *dev)
848 {
849 	struct mrvl_priv *priv = dev->data->dev_private;
850 	size_t i;
851 
852 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
853 		return 0;
854 
855 	mrvl_flush_rx_queues(dev);
856 	mrvl_flush_tx_shadow_queues(dev);
857 	mrvl_flow_deinit(dev);
858 	mrvl_mtr_deinit(dev);
859 
860 	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
861 		struct pp2_ppio_tc_params *tc_params =
862 			&priv->ppio_params.inqs_params.tcs_params[i];
863 
864 		if (tc_params->inqs_params) {
865 			rte_free(tc_params->inqs_params);
866 			tc_params->inqs_params = NULL;
867 		}
868 	}
869 
870 	if (priv->cls_tbl) {
871 		pp2_cls_tbl_deinit(priv->cls_tbl);
872 		priv->cls_tbl = NULL;
873 	}
874 
875 	if (priv->qos_tbl) {
876 		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
877 		priv->qos_tbl = NULL;
878 	}
879 
880 	mrvl_flush_bpool(dev);
881 	mrvl_tm_deinit(dev);
882 
883 	if (priv->ppio) {
884 		pp2_ppio_deinit(priv->ppio);
885 		priv->ppio = NULL;
886 	}
887 
888 	/* policer must be released after ppio deinitialization */
889 	if (priv->default_policer) {
890 		pp2_cls_plcr_deinit(priv->default_policer);
891 		priv->default_policer = NULL;
892 	}
893 
894 
895 	if (priv->bpool) {
896 		pp2_bpool_deinit(priv->bpool);
897 		used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
898 		priv->bpool = NULL;
899 	}
900 
901 	mrvl_dev_num--;
902 
903 	if (mrvl_dev_num == 0) {
904 		MRVL_LOG(INFO, "Perform MUSDK deinit");
905 		mrvl_deinit_hifs();
906 		mrvl_deinit_pp2();
907 		rte_mvep_deinit(MVEP_MOD_T_PP2);
908 	}
909 
910 	return 0;
911 }
912 
913 /**
914  * DPDK callback to retrieve physical link information.
915  *
916  * @param dev
917  *   Pointer to Ethernet device structure.
918  * @param wait_to_complete
919  *   Wait for request completion (ignored).
920  *
921  * @return
922  *   0 on success, negative error value otherwise.
923  */
924 static int
925 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
926 {
927 	/*
928 	 * TODO
929 	 * once MUSDK provides necessary API use it here
930 	 */
931 	struct mrvl_priv *priv = dev->data->dev_private;
932 	struct ethtool_cmd edata;
933 	struct ifreq req;
934 	int ret, fd, link_up;
935 
936 	if (!priv->ppio)
937 		return -EPERM;
938 
939 	edata.cmd = ETHTOOL_GSET;
940 
941 	strcpy(req.ifr_name, dev->data->name);
942 	req.ifr_data = (void *)&edata;
943 
944 	fd = socket(AF_INET, SOCK_DGRAM, 0);
945 	if (fd == -1)
946 		return -EFAULT;
947 
948 	ret = ioctl(fd, SIOCETHTOOL, &req);
949 	if (ret == -1) {
950 		close(fd);
951 		return -EFAULT;
952 	}
953 
954 	close(fd);
955 
956 	switch (ethtool_cmd_speed(&edata)) {
957 	case SPEED_10:
958 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
959 		break;
960 	case SPEED_100:
961 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
962 		break;
963 	case SPEED_1000:
964 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
965 		break;
966 	case SPEED_10000:
967 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
968 		break;
969 	default:
970 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
971 	}
972 
973 	dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
974 							 ETH_LINK_HALF_DUPLEX;
975 	dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
976 							   ETH_LINK_FIXED;
977 	pp2_ppio_get_link_state(priv->ppio, &link_up);
978 	dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
979 
980 	return 0;
981 }
982 
983 /**
984  * DPDK callback to enable promiscuous mode.
985  *
986  * @param dev
987  *   Pointer to Ethernet device structure.
988  *
989  * @return
990  *   0 on success, negative error value otherwise.
991  */
992 static int
993 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
994 {
995 	struct mrvl_priv *priv = dev->data->dev_private;
996 	int ret;
997 
998 	if (!priv->ppio)
999 		return 0;
1000 
1001 	if (priv->isolated)
1002 		return 0;
1003 
1004 	ret = pp2_ppio_set_promisc(priv->ppio, 1);
1005 	if (ret) {
1006 		MRVL_LOG(ERR, "Failed to enable promiscuous mode");
1007 		return -EAGAIN;
1008 	}
1009 
1010 	return 0;
1011 }
1012 
1013 /**
1014  * DPDK callback to enable allmulti mode.
1015  *
1016  * @param dev
1017  *   Pointer to Ethernet device structure.
1018  *
1019  * @return
1020  *   0 on success, negative error value otherwise.
1021  */
1022 static int
1023 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
1024 {
1025 	struct mrvl_priv *priv = dev->data->dev_private;
1026 	int ret;
1027 
1028 	if (!priv->ppio)
1029 		return 0;
1030 
1031 	if (priv->isolated)
1032 		return 0;
1033 
1034 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
1035 	if (ret) {
1036 		MRVL_LOG(ERR, "Failed enable all-multicast mode");
1037 		return -EAGAIN;
1038 	}
1039 
1040 	return 0;
1041 }
1042 
1043 /**
1044  * DPDK callback to disable promiscuous mode.
1045  *
1046  * @param dev
1047  *   Pointer to Ethernet device structure.
1048  *
1049  * @return
1050  *   0 on success, negative error value otherwise.
1051  */
1052 static int
1053 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
1054 {
1055 	struct mrvl_priv *priv = dev->data->dev_private;
1056 	int ret;
1057 
1058 	if (!priv->ppio)
1059 		return 0;
1060 
1061 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
1062 	if (ret) {
1063 		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
1064 		return -EAGAIN;
1065 	}
1066 
1067 	return 0;
1068 }
1069 
1070 /**
1071  * DPDK callback to disable allmulticast mode.
1072  *
1073  * @param dev
1074  *   Pointer to Ethernet device structure.
1075  *
1076  * @return
1077  *   0 on success, negative error value otherwise.
1078  */
1079 static int
1080 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
1081 {
1082 	struct mrvl_priv *priv = dev->data->dev_private;
1083 	int ret;
1084 
1085 	if (!priv->ppio)
1086 		return 0;
1087 
1088 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
1089 	if (ret) {
1090 		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
1091 		return -EAGAIN;
1092 	}
1093 
1094 	return 0;
1095 }
1096 
1097 /**
1098  * DPDK callback to remove a MAC address.
1099  *
1100  * @param dev
1101  *   Pointer to Ethernet device structure.
1102  * @param index
1103  *   MAC address index.
1104  */
1105 static void
1106 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1107 {
1108 	struct mrvl_priv *priv = dev->data->dev_private;
1109 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1110 	int ret;
1111 
1112 	if (!priv->ppio)
1113 		return;
1114 
1115 	if (priv->isolated)
1116 		return;
1117 
1118 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
1119 				       dev->data->mac_addrs[index].addr_bytes);
1120 	if (ret) {
1121 		rte_ether_format_addr(buf, sizeof(buf),
1122 				  &dev->data->mac_addrs[index]);
1123 		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
1124 	}
1125 }
1126 
1127 /**
1128  * DPDK callback to add a MAC address.
1129  *
1130  * @param dev
1131  *   Pointer to Ethernet device structure.
1132  * @param mac_addr
1133  *   MAC address to register.
1134  * @param index
1135  *   MAC address index.
1136  * @param vmdq
1137  *   VMDq pool index to associate address with (unused).
1138  *
1139  * @return
1140  *   0 on success, negative error value otherwise.
1141  */
1142 static int
1143 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1144 		  uint32_t index, uint32_t vmdq __rte_unused)
1145 {
1146 	struct mrvl_priv *priv = dev->data->dev_private;
1147 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1148 	int ret;
1149 
1150 	if (priv->isolated)
1151 		return -ENOTSUP;
1152 
1153 	if (index == 0)
1154 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1155 		return -1;
1156 
1157 	if (!priv->ppio)
1158 		return 0;
1159 
1160 	/*
1161 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1162 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1163 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1164 	 * 21 respectively.
1165 	 *
1166 	 * If more than uc_filter_max uc addresses were added to filter list
1167 	 * then NIC will switch to promiscuous mode automatically.
1168 	 *
1169 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1170 	 * were added to filter list then NIC will switch to all-multicast mode
1171 	 * automatically.
1172 	 */
1173 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1174 	if (ret) {
1175 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1176 		MRVL_LOG(ERR, "Failed to add mac %s", buf);
1177 		return -1;
1178 	}
1179 
1180 	return 0;
1181 }
1182 
1183 /**
1184  * DPDK callback to set the primary MAC address.
1185  *
1186  * @param dev
1187  *   Pointer to Ethernet device structure.
1188  * @param mac_addr
1189  *   MAC address to register.
1190  *
1191  * @return
1192  *   0 on success, negative error value otherwise.
1193  */
1194 static int
1195 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1196 {
1197 	struct mrvl_priv *priv = dev->data->dev_private;
1198 	int ret;
1199 
1200 	if (!priv->ppio)
1201 		return 0;
1202 
1203 	if (priv->isolated)
1204 		return -ENOTSUP;
1205 
1206 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1207 	if (ret) {
1208 		char buf[RTE_ETHER_ADDR_FMT_SIZE];
1209 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1210 		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1211 	}
1212 
1213 	return ret;
1214 }
1215 
1216 /**
1217  * DPDK callback to get device statistics.
1218  *
1219  * @param dev
1220  *   Pointer to Ethernet device structure.
1221  * @param stats
1222  *   Stats structure output buffer.
1223  *
1224  * @return
1225  *   0 on success, negative error value otherwise.
1226  */
1227 static int
1228 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1229 {
1230 	struct mrvl_priv *priv = dev->data->dev_private;
1231 	struct pp2_ppio_statistics ppio_stats;
1232 	uint64_t drop_mac = 0;
1233 	unsigned int i, idx, ret;
1234 
1235 	if (!priv->ppio)
1236 		return -EPERM;
1237 
1238 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1239 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1240 		struct pp2_ppio_inq_statistics rx_stats;
1241 
1242 		if (!rxq)
1243 			continue;
1244 
1245 		idx = rxq->queue_id;
1246 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1247 			MRVL_LOG(ERR,
1248 				"rx queue %d stats out of range (0 - %d)",
1249 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1250 			continue;
1251 		}
1252 
1253 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1254 						  priv->rxq_map[idx].tc,
1255 						  priv->rxq_map[idx].inq,
1256 						  &rx_stats, 0);
1257 		if (unlikely(ret)) {
1258 			MRVL_LOG(ERR,
1259 				"Failed to update rx queue %d stats", idx);
1260 			break;
1261 		}
1262 
1263 		stats->q_ibytes[idx] = rxq->bytes_recv;
1264 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1265 		stats->q_errors[idx] = rx_stats.drop_early +
1266 				       rx_stats.drop_fullq +
1267 				       rx_stats.drop_bm +
1268 				       rxq->drop_mac;
1269 		stats->ibytes += rxq->bytes_recv;
1270 		drop_mac += rxq->drop_mac;
1271 	}
1272 
1273 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1274 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1275 		struct pp2_ppio_outq_statistics tx_stats;
1276 
1277 		if (!txq)
1278 			continue;
1279 
1280 		idx = txq->queue_id;
1281 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1282 			MRVL_LOG(ERR,
1283 				"tx queue %d stats out of range (0 - %d)",
1284 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1285 		}
1286 
1287 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1288 						   &tx_stats, 0);
1289 		if (unlikely(ret)) {
1290 			MRVL_LOG(ERR,
1291 				"Failed to update tx queue %d stats", idx);
1292 			break;
1293 		}
1294 
1295 		stats->q_opackets[idx] = tx_stats.deq_desc;
1296 		stats->q_obytes[idx] = txq->bytes_sent;
1297 		stats->obytes += txq->bytes_sent;
1298 	}
1299 
1300 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1301 	if (unlikely(ret)) {
1302 		MRVL_LOG(ERR, "Failed to update port statistics");
1303 		return ret;
1304 	}
1305 
1306 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1307 	stats->opackets += ppio_stats.tx_packets;
1308 	stats->imissed += ppio_stats.rx_fullq_dropped +
1309 			  ppio_stats.rx_bm_dropped +
1310 			  ppio_stats.rx_early_dropped +
1311 			  ppio_stats.rx_fifo_dropped +
1312 			  ppio_stats.rx_cls_dropped;
1313 	stats->ierrors = drop_mac;
1314 
1315 	return 0;
1316 }
1317 
1318 /**
1319  * DPDK callback to clear device statistics.
1320  *
1321  * @param dev
1322  *   Pointer to Ethernet device structure.
1323  *
1324  * @return
1325  *   0 on success, negative error value otherwise.
1326  */
1327 static int
1328 mrvl_stats_reset(struct rte_eth_dev *dev)
1329 {
1330 	struct mrvl_priv *priv = dev->data->dev_private;
1331 	int i;
1332 
1333 	if (!priv->ppio)
1334 		return 0;
1335 
1336 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1337 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1338 
1339 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1340 					    priv->rxq_map[i].inq, NULL, 1);
1341 		rxq->bytes_recv = 0;
1342 		rxq->drop_mac = 0;
1343 	}
1344 
1345 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1346 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1347 
1348 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1349 		txq->bytes_sent = 0;
1350 	}
1351 
1352 	return pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1353 }
1354 
1355 /**
1356  * DPDK callback to get extended statistics.
1357  *
1358  * @param dev
1359  *   Pointer to Ethernet device structure.
1360  * @param stats
1361  *   Pointer to xstats table.
1362  * @param n
1363  *   Number of entries in xstats table.
1364  * @return
1365  *   Negative value on error, number of read xstats otherwise.
1366  */
1367 static int
1368 mrvl_xstats_get(struct rte_eth_dev *dev,
1369 		struct rte_eth_xstat *stats, unsigned int n)
1370 {
1371 	struct mrvl_priv *priv = dev->data->dev_private;
1372 	struct pp2_ppio_statistics ppio_stats;
1373 	unsigned int i;
1374 
1375 	if (!stats)
1376 		return 0;
1377 
1378 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1379 	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1380 		uint64_t val;
1381 
1382 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1383 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1384 					    mrvl_xstats_tbl[i].offset);
1385 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1386 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1387 					    mrvl_xstats_tbl[i].offset);
1388 		else
1389 			return -EINVAL;
1390 
1391 		stats[i].id = i;
1392 		stats[i].value = val;
1393 	}
1394 
1395 	return n;
1396 }
1397 
1398 /**
1399  * DPDK callback to reset extended statistics.
1400  *
1401  * @param dev
1402  *   Pointer to Ethernet device structure.
1403  *
1404  * @return
1405  *   0 on success, negative error value otherwise.
1406  */
1407 static int
1408 mrvl_xstats_reset(struct rte_eth_dev *dev)
1409 {
1410 	return mrvl_stats_reset(dev);
1411 }
1412 
1413 /**
1414  * DPDK callback to get extended statistics names.
1415  *
1416  * @param dev (unused)
1417  *   Pointer to Ethernet device structure.
1418  * @param xstats_names
1419  *   Pointer to xstats names table.
1420  * @param size
1421  *   Size of the xstats names table.
1422  * @return
1423  *   Number of read names.
1424  */
1425 static int
1426 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1427 		      struct rte_eth_xstat_name *xstats_names,
1428 		      unsigned int size)
1429 {
1430 	unsigned int i;
1431 
1432 	if (!xstats_names)
1433 		return RTE_DIM(mrvl_xstats_tbl);
1434 
1435 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1436 		strlcpy(xstats_names[i].name, mrvl_xstats_tbl[i].name,
1437 			RTE_ETH_XSTATS_NAME_SIZE);
1438 
1439 	return size;
1440 }
1441 
1442 /**
1443  * DPDK callback to get information about the device.
1444  *
1445  * @param dev
1446  *   Pointer to Ethernet device structure (unused).
1447  * @param info
1448  *   Info structure output buffer.
1449  */
1450 static int
1451 mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
1452 		   struct rte_eth_dev_info *info)
1453 {
1454 	info->speed_capa = ETH_LINK_SPEED_10M |
1455 			   ETH_LINK_SPEED_100M |
1456 			   ETH_LINK_SPEED_1G |
1457 			   ETH_LINK_SPEED_10G;
1458 
1459 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1460 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1461 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1462 
1463 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1464 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1465 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1466 
1467 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1468 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1469 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1470 
1471 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1472 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1473 
1474 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1475 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1476 
1477 	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1478 				       ETH_RSS_NONFRAG_IPV4_TCP |
1479 				       ETH_RSS_NONFRAG_IPV4_UDP;
1480 
1481 	/* By default packets are dropped if no descriptors are available */
1482 	info->default_rxconf.rx_drop_en = 1;
1483 
1484 	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1485 
1486 	return 0;
1487 }
1488 
1489 /**
1490  * Return supported packet types.
1491  *
1492  * @param dev
1493  *   Pointer to Ethernet device structure (unused).
1494  *
1495  * @return
1496  *   Const pointer to the table with supported packet types.
1497  */
1498 static const uint32_t *
1499 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1500 {
1501 	static const uint32_t ptypes[] = {
1502 		RTE_PTYPE_L2_ETHER,
1503 		RTE_PTYPE_L2_ETHER_VLAN,
1504 		RTE_PTYPE_L2_ETHER_QINQ,
1505 		RTE_PTYPE_L3_IPV4,
1506 		RTE_PTYPE_L3_IPV4_EXT,
1507 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1508 		RTE_PTYPE_L3_IPV6,
1509 		RTE_PTYPE_L3_IPV6_EXT,
1510 		RTE_PTYPE_L2_ETHER_ARP,
1511 		RTE_PTYPE_L4_TCP,
1512 		RTE_PTYPE_L4_UDP
1513 	};
1514 
1515 	return ptypes;
1516 }
1517 
1518 /**
1519  * DPDK callback to get information about specific receive queue.
1520  *
1521  * @param dev
1522  *   Pointer to Ethernet device structure.
1523  * @param rx_queue_id
1524  *   Receive queue index.
1525  * @param qinfo
1526  *   Receive queue information structure.
1527  */
1528 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1529 			      struct rte_eth_rxq_info *qinfo)
1530 {
1531 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1532 	struct mrvl_priv *priv = dev->data->dev_private;
1533 	int inq = priv->rxq_map[rx_queue_id].inq;
1534 	int tc = priv->rxq_map[rx_queue_id].tc;
1535 	struct pp2_ppio_tc_params *tc_params =
1536 		&priv->ppio_params.inqs_params.tcs_params[tc];
1537 
1538 	qinfo->mp = q->mp;
1539 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1540 }
1541 
1542 /**
1543  * DPDK callback to get information about specific transmit queue.
1544  *
1545  * @param dev
1546  *   Pointer to Ethernet device structure.
1547  * @param tx_queue_id
1548  *   Transmit queue index.
1549  * @param qinfo
1550  *   Transmit queue information structure.
1551  */
1552 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1553 			      struct rte_eth_txq_info *qinfo)
1554 {
1555 	struct mrvl_priv *priv = dev->data->dev_private;
1556 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1557 
1558 	qinfo->nb_desc =
1559 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1560 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1561 }
1562 
1563 /**
1564  * DPDK callback to Configure a VLAN filter.
1565  *
1566  * @param dev
1567  *   Pointer to Ethernet device structure.
1568  * @param vlan_id
1569  *   VLAN ID to filter.
1570  * @param on
1571  *   Toggle filter.
1572  *
1573  * @return
1574  *   0 on success, negative error value otherwise.
1575  */
1576 static int
1577 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1578 {
1579 	struct mrvl_priv *priv = dev->data->dev_private;
1580 
1581 	if (!priv->ppio)
1582 		return -EPERM;
1583 
1584 	if (priv->isolated)
1585 		return -ENOTSUP;
1586 
1587 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1588 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1589 }
1590 
1591 /**
1592  * Release buffers to hardware bpool (buffer-pool)
1593  *
1594  * @param rxq
1595  *   Receive queue pointer.
1596  * @param num
1597  *   Number of buffers to release to bpool.
1598  *
1599  * @return
1600  *   0 on success, negative error value otherwise.
1601  */
1602 static int
1603 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1604 {
1605 	struct buff_release_entry entries[num];
1606 	struct rte_mbuf *mbufs[num];
1607 	int i, ret;
1608 	unsigned int core_id;
1609 	struct pp2_hif *hif;
1610 	struct pp2_bpool *bpool;
1611 
1612 	core_id = rte_lcore_id();
1613 	if (core_id == LCORE_ID_ANY)
1614 		core_id = rte_get_main_lcore();
1615 
1616 	hif = mrvl_get_hif(rxq->priv, core_id);
1617 	if (!hif)
1618 		return -1;
1619 
1620 	bpool = rxq->priv->bpool;
1621 
1622 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1623 	if (ret)
1624 		return ret;
1625 
1626 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1627 		cookie_addr_high =
1628 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1629 
1630 	for (i = 0; i < num; i++) {
1631 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1632 			!= cookie_addr_high) {
1633 			MRVL_LOG(ERR,
1634 				"mbuf virtual addr high 0x%lx out of range",
1635 				(uint64_t)mbufs[i] >> 32);
1636 			goto out;
1637 		}
1638 
1639 		entries[i].buff.addr =
1640 			rte_mbuf_data_iova_default(mbufs[i]);
1641 		entries[i].buff.cookie = (uint64_t)mbufs[i];
1642 		entries[i].bpool = bpool;
1643 	}
1644 
1645 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1646 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1647 
1648 	if (i != num)
1649 		goto out;
1650 
1651 	return 0;
1652 out:
1653 	for (; i < num; i++)
1654 		rte_pktmbuf_free(mbufs[i]);
1655 
1656 	return -1;
1657 }
1658 
1659 /**
1660  * DPDK callback to configure the receive queue.
1661  *
1662  * @param dev
1663  *   Pointer to Ethernet device structure.
1664  * @param idx
1665  *   RX queue index.
1666  * @param desc
1667  *   Number of descriptors to configure in queue.
1668  * @param socket
1669  *   NUMA socket on which memory must be allocated.
1670  * @param conf
1671  *   Thresholds parameters.
1672  * @param mp
1673  *   Memory pool for buffer allocations.
1674  *
1675  * @return
1676  *   0 on success, negative error value otherwise.
1677  */
1678 static int
1679 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1680 		    unsigned int socket,
1681 		    const struct rte_eth_rxconf *conf,
1682 		    struct rte_mempool *mp)
1683 {
1684 	struct mrvl_priv *priv = dev->data->dev_private;
1685 	struct mrvl_rxq *rxq;
1686 	uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
1687 	uint32_t max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1688 	int ret, tc, inq;
1689 	uint64_t offloads;
1690 
1691 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
1692 
1693 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1694 		/*
1695 		 * Unknown TC mapping, mapping will not have a correct queue.
1696 		 */
1697 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
1698 			idx, priv->ppio_id);
1699 		return -EFAULT;
1700 	}
1701 
1702 	frame_size = buf_size - RTE_PKTMBUF_HEADROOM -
1703 		     MRVL_PKT_EFFEC_OFFS + RTE_ETHER_CRC_LEN;
1704 	if (frame_size < max_rx_pkt_len) {
1705 		MRVL_LOG(WARNING,
1706 			"Mbuf size must be increased to %u bytes to hold up "
1707 			"to %u bytes of data.",
1708 			buf_size + max_rx_pkt_len - frame_size,
1709 			max_rx_pkt_len);
1710 		dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
1711 		MRVL_LOG(INFO, "Setting max rx pkt len to %u",
1712 			dev->data->dev_conf.rxmode.max_rx_pkt_len);
1713 	}
1714 
1715 	if (dev->data->rx_queues[idx]) {
1716 		rte_free(dev->data->rx_queues[idx]);
1717 		dev->data->rx_queues[idx] = NULL;
1718 	}
1719 
1720 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
1721 	if (!rxq)
1722 		return -ENOMEM;
1723 
1724 	rxq->priv = priv;
1725 	rxq->mp = mp;
1726 	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
1727 	rxq->queue_id = idx;
1728 	rxq->port_id = dev->data->port_id;
1729 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
1730 
1731 	tc = priv->rxq_map[rxq->queue_id].tc,
1732 	inq = priv->rxq_map[rxq->queue_id].inq;
1733 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
1734 		desc;
1735 
1736 	ret = mrvl_fill_bpool(rxq, desc);
1737 	if (ret) {
1738 		rte_free(rxq);
1739 		return ret;
1740 	}
1741 
1742 	priv->bpool_init_size += desc;
1743 
1744 	dev->data->rx_queues[idx] = rxq;
1745 
1746 	return 0;
1747 }
1748 
1749 /**
1750  * DPDK callback to release the receive queue.
1751  *
1752  * @param rxq
1753  *   Generic receive queue pointer.
1754  */
1755 static void
1756 mrvl_rx_queue_release(void *rxq)
1757 {
1758 	struct mrvl_rxq *q = rxq;
1759 	struct pp2_ppio_tc_params *tc_params;
1760 	int i, num, tc, inq;
1761 	struct pp2_hif *hif;
1762 	unsigned int core_id = rte_lcore_id();
1763 
1764 	if (core_id == LCORE_ID_ANY)
1765 		core_id = rte_get_main_lcore();
1766 
1767 	if (!q)
1768 		return;
1769 
1770 	hif = mrvl_get_hif(q->priv, core_id);
1771 
1772 	if (!hif)
1773 		return;
1774 
1775 	tc = q->priv->rxq_map[q->queue_id].tc;
1776 	inq = q->priv->rxq_map[q->queue_id].inq;
1777 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
1778 	num = tc_params->inqs_params[inq].size;
1779 	for (i = 0; i < num; i++) {
1780 		struct pp2_buff_inf inf;
1781 		uint64_t addr;
1782 
1783 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
1784 		addr = cookie_addr_high | inf.cookie;
1785 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1786 	}
1787 
1788 	rte_free(q);
1789 }
1790 
1791 /**
1792  * DPDK callback to configure the transmit queue.
1793  *
1794  * @param dev
1795  *   Pointer to Ethernet device structure.
1796  * @param idx
1797  *   Transmit queue index.
1798  * @param desc
1799  *   Number of descriptors to configure in the queue.
1800  * @param socket
1801  *   NUMA socket on which memory must be allocated.
1802  * @param conf
1803  *   Tx queue configuration parameters.
1804  *
1805  * @return
1806  *   0 on success, negative error value otherwise.
1807  */
1808 static int
1809 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1810 		    unsigned int socket,
1811 		    const struct rte_eth_txconf *conf)
1812 {
1813 	struct mrvl_priv *priv = dev->data->dev_private;
1814 	struct mrvl_txq *txq;
1815 
1816 	if (dev->data->tx_queues[idx]) {
1817 		rte_free(dev->data->tx_queues[idx]);
1818 		dev->data->tx_queues[idx] = NULL;
1819 	}
1820 
1821 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
1822 	if (!txq)
1823 		return -ENOMEM;
1824 
1825 	txq->priv = priv;
1826 	txq->queue_id = idx;
1827 	txq->port_id = dev->data->port_id;
1828 	txq->tx_deferred_start = conf->tx_deferred_start;
1829 	dev->data->tx_queues[idx] = txq;
1830 
1831 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
1832 
1833 	return 0;
1834 }
1835 
1836 /**
1837  * DPDK callback to release the transmit queue.
1838  *
1839  * @param txq
1840  *   Generic transmit queue pointer.
1841  */
1842 static void
1843 mrvl_tx_queue_release(void *txq)
1844 {
1845 	struct mrvl_txq *q = txq;
1846 
1847 	if (!q)
1848 		return;
1849 
1850 	rte_free(q);
1851 }
1852 
1853 /**
1854  * DPDK callback to get flow control configuration.
1855  *
1856  * @param dev
1857  *  Pointer to Ethernet device structure.
1858  * @param fc_conf
1859  *  Pointer to the flow control configuration.
1860  *
1861  * @return
1862  *  0 on success, negative error value otherwise.
1863  */
1864 static int
1865 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1866 {
1867 	struct mrvl_priv *priv = dev->data->dev_private;
1868 	int ret, en;
1869 
1870 	if (!priv)
1871 		return -EPERM;
1872 
1873 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
1874 	if (ret) {
1875 		MRVL_LOG(ERR, "Failed to read rx pause state");
1876 		return ret;
1877 	}
1878 
1879 	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
1880 
1881 	return 0;
1882 }
1883 
1884 /**
1885  * DPDK callback to set flow control configuration.
1886  *
1887  * @param dev
1888  *  Pointer to Ethernet device structure.
1889  * @param fc_conf
1890  *  Pointer to the flow control configuration.
1891  *
1892  * @return
1893  *  0 on success, negative error value otherwise.
1894  */
1895 static int
1896 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1897 {
1898 	struct mrvl_priv *priv = dev->data->dev_private;
1899 
1900 	if (!priv)
1901 		return -EPERM;
1902 
1903 	if (fc_conf->high_water ||
1904 	    fc_conf->low_water ||
1905 	    fc_conf->pause_time ||
1906 	    fc_conf->mac_ctrl_frame_fwd ||
1907 	    fc_conf->autoneg) {
1908 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
1909 
1910 		return -EINVAL;
1911 	}
1912 
1913 	if (fc_conf->mode == RTE_FC_NONE ||
1914 	    fc_conf->mode == RTE_FC_RX_PAUSE) {
1915 		int ret, en;
1916 
1917 		en = fc_conf->mode == RTE_FC_NONE ? 0 : 1;
1918 		ret = pp2_ppio_set_rx_pause(priv->ppio, en);
1919 		if (ret)
1920 			MRVL_LOG(ERR,
1921 				"Failed to change flowctrl on RX side");
1922 
1923 		return ret;
1924 	}
1925 
1926 	return 0;
1927 }
1928 
1929 /**
1930  * Update RSS hash configuration
1931  *
1932  * @param dev
1933  *   Pointer to Ethernet device structure.
1934  * @param rss_conf
1935  *   Pointer to RSS configuration.
1936  *
1937  * @return
1938  *   0 on success, negative error value otherwise.
1939  */
1940 static int
1941 mrvl_rss_hash_update(struct rte_eth_dev *dev,
1942 		     struct rte_eth_rss_conf *rss_conf)
1943 {
1944 	struct mrvl_priv *priv = dev->data->dev_private;
1945 
1946 	if (priv->isolated)
1947 		return -ENOTSUP;
1948 
1949 	return mrvl_configure_rss(priv, rss_conf);
1950 }
1951 
1952 /**
1953  * DPDK callback to get RSS hash configuration.
1954  *
1955  * @param dev
1956  *   Pointer to Ethernet device structure.
1957  * @rss_conf
1958  *   Pointer to RSS configuration.
1959  *
1960  * @return
1961  *   Always 0.
1962  */
1963 static int
1964 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
1965 		       struct rte_eth_rss_conf *rss_conf)
1966 {
1967 	struct mrvl_priv *priv = dev->data->dev_private;
1968 	enum pp2_ppio_hash_type hash_type =
1969 		priv->ppio_params.inqs_params.hash_type;
1970 
1971 	rss_conf->rss_key = NULL;
1972 
1973 	if (hash_type == PP2_PPIO_HASH_T_NONE)
1974 		rss_conf->rss_hf = 0;
1975 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
1976 		rss_conf->rss_hf = ETH_RSS_IPV4;
1977 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
1978 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
1979 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
1980 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
1981 
1982 	return 0;
1983 }
1984 
1985 /**
1986  * DPDK callback to get rte_flow callbacks.
1987  *
1988  * @param dev
1989  *   Pointer to the device structure.
1990  * @param filer_type
1991  *   Flow filter type.
1992  * @param filter_op
1993  *   Flow filter operation.
1994  * @param arg
1995  *   Pointer to pass the flow ops.
1996  *
1997  * @return
1998  *   0 on success, negative error value otherwise.
1999  */
2000 static int
2001 mrvl_eth_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
2002 		     enum rte_filter_type filter_type,
2003 		     enum rte_filter_op filter_op, void *arg)
2004 {
2005 	switch (filter_type) {
2006 	case RTE_ETH_FILTER_GENERIC:
2007 		if (filter_op != RTE_ETH_FILTER_GET)
2008 			return -EINVAL;
2009 		*(const void **)arg = &mrvl_flow_ops;
2010 		return 0;
2011 	default:
2012 		MRVL_LOG(WARNING, "Filter type (%d) not supported",
2013 				filter_type);
2014 		return -EINVAL;
2015 	}
2016 }
2017 
2018 /**
2019  * DPDK callback to get rte_mtr callbacks.
2020  *
2021  * @param dev
2022  *   Pointer to the device structure.
2023  * @param ops
2024  *   Pointer to pass the mtr ops.
2025  *
2026  * @return
2027  *   Always 0.
2028  */
2029 static int
2030 mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2031 {
2032 	*(const void **)ops = &mrvl_mtr_ops;
2033 
2034 	return 0;
2035 }
2036 
2037 /**
2038  * DPDK callback to get rte_tm callbacks.
2039  *
2040  * @param dev
2041  *   Pointer to the device structure.
2042  * @param ops
2043  *   Pointer to pass the tm ops.
2044  *
2045  * @return
2046  *   Always 0.
2047  */
2048 static int
2049 mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2050 {
2051 	*(const void **)ops = &mrvl_tm_ops;
2052 
2053 	return 0;
2054 }
2055 
2056 static const struct eth_dev_ops mrvl_ops = {
2057 	.dev_configure = mrvl_dev_configure,
2058 	.dev_start = mrvl_dev_start,
2059 	.dev_stop = mrvl_dev_stop,
2060 	.dev_set_link_up = mrvl_dev_set_link_up,
2061 	.dev_set_link_down = mrvl_dev_set_link_down,
2062 	.dev_close = mrvl_dev_close,
2063 	.link_update = mrvl_link_update,
2064 	.promiscuous_enable = mrvl_promiscuous_enable,
2065 	.allmulticast_enable = mrvl_allmulticast_enable,
2066 	.promiscuous_disable = mrvl_promiscuous_disable,
2067 	.allmulticast_disable = mrvl_allmulticast_disable,
2068 	.mac_addr_remove = mrvl_mac_addr_remove,
2069 	.mac_addr_add = mrvl_mac_addr_add,
2070 	.mac_addr_set = mrvl_mac_addr_set,
2071 	.mtu_set = mrvl_mtu_set,
2072 	.stats_get = mrvl_stats_get,
2073 	.stats_reset = mrvl_stats_reset,
2074 	.xstats_get = mrvl_xstats_get,
2075 	.xstats_reset = mrvl_xstats_reset,
2076 	.xstats_get_names = mrvl_xstats_get_names,
2077 	.dev_infos_get = mrvl_dev_infos_get,
2078 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
2079 	.rxq_info_get = mrvl_rxq_info_get,
2080 	.txq_info_get = mrvl_txq_info_get,
2081 	.vlan_filter_set = mrvl_vlan_filter_set,
2082 	.tx_queue_start = mrvl_tx_queue_start,
2083 	.tx_queue_stop = mrvl_tx_queue_stop,
2084 	.rx_queue_setup = mrvl_rx_queue_setup,
2085 	.rx_queue_release = mrvl_rx_queue_release,
2086 	.tx_queue_setup = mrvl_tx_queue_setup,
2087 	.tx_queue_release = mrvl_tx_queue_release,
2088 	.flow_ctrl_get = mrvl_flow_ctrl_get,
2089 	.flow_ctrl_set = mrvl_flow_ctrl_set,
2090 	.rss_hash_update = mrvl_rss_hash_update,
2091 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
2092 	.filter_ctrl = mrvl_eth_filter_ctrl,
2093 	.mtr_ops_get = mrvl_mtr_ops_get,
2094 	.tm_ops_get = mrvl_tm_ops_get,
2095 };
2096 
2097 /**
2098  * Return packet type information and l3/l4 offsets.
2099  *
2100  * @param desc
2101  *   Pointer to the received packet descriptor.
2102  * @param l3_offset
2103  *   l3 packet offset.
2104  * @param l4_offset
2105  *   l4 packet offset.
2106  *
2107  * @return
2108  *   Packet type information.
2109  */
2110 static inline uint64_t
2111 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
2112 				    uint8_t *l3_offset, uint8_t *l4_offset)
2113 {
2114 	enum pp2_inq_l3_type l3_type;
2115 	enum pp2_inq_l4_type l4_type;
2116 	enum pp2_inq_vlan_tag vlan_tag;
2117 	uint64_t packet_type;
2118 
2119 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
2120 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
2121 	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
2122 
2123 	packet_type = RTE_PTYPE_L2_ETHER;
2124 
2125 	switch (vlan_tag) {
2126 	case PP2_INQ_VLAN_TAG_SINGLE:
2127 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
2128 		break;
2129 	case PP2_INQ_VLAN_TAG_DOUBLE:
2130 	case PP2_INQ_VLAN_TAG_TRIPLE:
2131 		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
2132 		break;
2133 	default:
2134 		break;
2135 	}
2136 
2137 	switch (l3_type) {
2138 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
2139 		packet_type |= RTE_PTYPE_L3_IPV4;
2140 		break;
2141 	case PP2_INQ_L3_TYPE_IPV4_OK:
2142 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
2143 		break;
2144 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
2145 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
2146 		break;
2147 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
2148 		packet_type |= RTE_PTYPE_L3_IPV6;
2149 		break;
2150 	case PP2_INQ_L3_TYPE_IPV6_EXT:
2151 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
2152 		break;
2153 	case PP2_INQ_L3_TYPE_ARP:
2154 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
2155 		/*
2156 		 * In case of ARP l4_offset is set to wrong value.
2157 		 * Set it to proper one so that later on mbuf->l3_len can be
2158 		 * calculated subtracting l4_offset and l3_offset.
2159 		 */
2160 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
2161 		break;
2162 	default:
2163 		break;
2164 	}
2165 
2166 	switch (l4_type) {
2167 	case PP2_INQ_L4_TYPE_TCP:
2168 		packet_type |= RTE_PTYPE_L4_TCP;
2169 		break;
2170 	case PP2_INQ_L4_TYPE_UDP:
2171 		packet_type |= RTE_PTYPE_L4_UDP;
2172 		break;
2173 	default:
2174 		break;
2175 	}
2176 
2177 	return packet_type;
2178 }
2179 
2180 /**
2181  * Get offload information from the received packet descriptor.
2182  *
2183  * @param desc
2184  *   Pointer to the received packet descriptor.
2185  *
2186  * @return
2187  *   Mbuf offload flags.
2188  */
2189 static inline uint64_t
2190 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc)
2191 {
2192 	uint64_t flags;
2193 	enum pp2_inq_desc_status status;
2194 
2195 	status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2196 	if (unlikely(status != PP2_DESC_ERR_OK))
2197 		flags = PKT_RX_IP_CKSUM_BAD;
2198 	else
2199 		flags = PKT_RX_IP_CKSUM_GOOD;
2200 
2201 	status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2202 	if (unlikely(status != PP2_DESC_ERR_OK))
2203 		flags |= PKT_RX_L4_CKSUM_BAD;
2204 	else
2205 		flags |= PKT_RX_L4_CKSUM_GOOD;
2206 
2207 	return flags;
2208 }
2209 
2210 /**
2211  * DPDK callback for receive.
2212  *
2213  * @param rxq
2214  *   Generic pointer to the receive queue.
2215  * @param rx_pkts
2216  *   Array to store received packets.
2217  * @param nb_pkts
2218  *   Maximum number of packets in array.
2219  *
2220  * @return
2221  *   Number of packets successfully received.
2222  */
2223 static uint16_t
2224 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2225 {
2226 	struct mrvl_rxq *q = rxq;
2227 	struct pp2_ppio_desc descs[nb_pkts];
2228 	struct pp2_bpool *bpool;
2229 	int i, ret, rx_done = 0;
2230 	int num;
2231 	struct pp2_hif *hif;
2232 	unsigned int core_id = rte_lcore_id();
2233 
2234 	hif = mrvl_get_hif(q->priv, core_id);
2235 
2236 	if (unlikely(!q->priv->ppio || !hif))
2237 		return 0;
2238 
2239 	bpool = q->priv->bpool;
2240 
2241 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2242 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2243 	if (unlikely(ret < 0))
2244 		return 0;
2245 
2246 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2247 
2248 	for (i = 0; i < nb_pkts; i++) {
2249 		struct rte_mbuf *mbuf;
2250 		uint8_t l3_offset, l4_offset;
2251 		enum pp2_inq_desc_status status;
2252 		uint64_t addr;
2253 
2254 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2255 			struct pp2_ppio_desc *pref_desc;
2256 			u64 pref_addr;
2257 
2258 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2259 			pref_addr = cookie_addr_high |
2260 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2261 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2262 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2263 		}
2264 
2265 		addr = cookie_addr_high |
2266 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2267 		mbuf = (struct rte_mbuf *)addr;
2268 		rte_pktmbuf_reset(mbuf);
2269 
2270 		/* drop packet in case of mac, overrun or resource error */
2271 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2272 		if (unlikely(status != PP2_DESC_ERR_OK)) {
2273 			struct pp2_buff_inf binf = {
2274 				.addr = rte_mbuf_data_iova_default(mbuf),
2275 				.cookie = (uint64_t)mbuf,
2276 			};
2277 
2278 			pp2_bpool_put_buff(hif, bpool, &binf);
2279 			mrvl_port_bpool_size
2280 				[bpool->pp2_id][bpool->id][core_id]++;
2281 			q->drop_mac++;
2282 			continue;
2283 		}
2284 
2285 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2286 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2287 		mbuf->data_len = mbuf->pkt_len;
2288 		mbuf->port = q->port_id;
2289 		mbuf->packet_type =
2290 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2291 							    &l3_offset,
2292 							    &l4_offset);
2293 		mbuf->l2_len = l3_offset;
2294 		mbuf->l3_len = l4_offset - l3_offset;
2295 
2296 		if (likely(q->cksum_enabled))
2297 			mbuf->ol_flags = mrvl_desc_to_ol_flags(&descs[i]);
2298 
2299 		rx_pkts[rx_done++] = mbuf;
2300 		q->bytes_recv += mbuf->pkt_len;
2301 	}
2302 
2303 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2304 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2305 
2306 		if (unlikely(num <= q->priv->bpool_min_size ||
2307 			     (!rx_done && num < q->priv->bpool_init_size))) {
2308 			mrvl_fill_bpool(q, MRVL_BURST_SIZE);
2309 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2310 			int i;
2311 			int pkt_to_remove = num - q->priv->bpool_init_size;
2312 			struct rte_mbuf *mbuf;
2313 			struct pp2_buff_inf buff;
2314 
2315 			for (i = 0; i < pkt_to_remove; i++) {
2316 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2317 				if (ret)
2318 					break;
2319 				mbuf = (struct rte_mbuf *)
2320 					(cookie_addr_high | buff.cookie);
2321 				rte_pktmbuf_free(mbuf);
2322 			}
2323 			mrvl_port_bpool_size
2324 				[bpool->pp2_id][bpool->id][core_id] -= i;
2325 		}
2326 		rte_spinlock_unlock(&q->priv->lock);
2327 	}
2328 
2329 	return rx_done;
2330 }
2331 
2332 /**
2333  * Prepare offload information.
2334  *
2335  * @param ol_flags
2336  *   Offload flags.
2337  * @param packet_type
2338  *   Packet type bitfield.
2339  * @param l3_type
2340  *   Pointer to the pp2_ouq_l3_type structure.
2341  * @param l4_type
2342  *   Pointer to the pp2_outq_l4_type structure.
2343  * @param gen_l3_cksum
2344  *   Will be set to 1 in case l3 checksum is computed.
2345  * @param l4_cksum
2346  *   Will be set to 1 in case l4 checksum is computed.
2347  *
2348  * @return
2349  *   0 on success, negative error value otherwise.
2350  */
2351 static inline int
2352 mrvl_prepare_proto_info(uint64_t ol_flags, uint32_t packet_type,
2353 			enum pp2_outq_l3_type *l3_type,
2354 			enum pp2_outq_l4_type *l4_type,
2355 			int *gen_l3_cksum,
2356 			int *gen_l4_cksum)
2357 {
2358 	/*
2359 	 * Based on ol_flags prepare information
2360 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2361 	 * for offloading.
2362 	 */
2363 	if (ol_flags & PKT_TX_IPV4) {
2364 		*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2365 		*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2366 	} else if (ol_flags & PKT_TX_IPV6) {
2367 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2368 		/* no checksum for ipv6 header */
2369 		*gen_l3_cksum = 0;
2370 	} else {
2371 		/* if something different then stop processing */
2372 		return -1;
2373 	}
2374 
2375 	ol_flags &= PKT_TX_L4_MASK;
2376 	if ((packet_type & RTE_PTYPE_L4_TCP) &&
2377 	    ol_flags == PKT_TX_TCP_CKSUM) {
2378 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2379 		*gen_l4_cksum = 1;
2380 	} else if ((packet_type & RTE_PTYPE_L4_UDP) &&
2381 		   ol_flags == PKT_TX_UDP_CKSUM) {
2382 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2383 		*gen_l4_cksum = 1;
2384 	} else {
2385 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2386 		/* no checksum for other type */
2387 		*gen_l4_cksum = 0;
2388 	}
2389 
2390 	return 0;
2391 }
2392 
2393 /**
2394  * Release already sent buffers to bpool (buffer-pool).
2395  *
2396  * @param ppio
2397  *   Pointer to the port structure.
2398  * @param hif
2399  *   Pointer to the MUSDK hardware interface.
2400  * @param sq
2401  *   Pointer to the shadow queue.
2402  * @param qid
2403  *   Queue id number.
2404  * @param force
2405  *   Force releasing packets.
2406  */
2407 static inline void
2408 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2409 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2410 		       int qid, int force)
2411 {
2412 	struct buff_release_entry *entry;
2413 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2414 	int i;
2415 
2416 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2417 
2418 	sq->num_to_release += nb_done;
2419 
2420 	if (likely(!force &&
2421 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2422 		return;
2423 
2424 	nb_done = sq->num_to_release;
2425 	sq->num_to_release = 0;
2426 
2427 	for (i = 0; i < nb_done; i++) {
2428 		entry = &sq->ent[sq->tail + num];
2429 		if (unlikely(!entry->buff.addr)) {
2430 			MRVL_LOG(ERR,
2431 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2432 				sq->tail, (u64)entry->buff.cookie,
2433 				(u64)entry->buff.addr);
2434 			skip_bufs = 1;
2435 			goto skip;
2436 		}
2437 
2438 		if (unlikely(!entry->bpool)) {
2439 			struct rte_mbuf *mbuf;
2440 
2441 			mbuf = (struct rte_mbuf *)
2442 			       (cookie_addr_high | entry->buff.cookie);
2443 			rte_pktmbuf_free(mbuf);
2444 			skip_bufs = 1;
2445 			goto skip;
2446 		}
2447 
2448 		mrvl_port_bpool_size
2449 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2450 		num++;
2451 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2452 			goto skip;
2453 		continue;
2454 skip:
2455 		if (likely(num))
2456 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2457 		num += skip_bufs;
2458 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2459 		sq->size -= num;
2460 		num = 0;
2461 		skip_bufs = 0;
2462 	}
2463 
2464 	if (likely(num)) {
2465 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2466 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2467 		sq->size -= num;
2468 	}
2469 }
2470 
2471 /**
2472  * DPDK callback for transmit.
2473  *
2474  * @param txq
2475  *   Generic pointer transmit queue.
2476  * @param tx_pkts
2477  *   Packets to transmit.
2478  * @param nb_pkts
2479  *   Number of packets in array.
2480  *
2481  * @return
2482  *   Number of packets successfully transmitted.
2483  */
2484 static uint16_t
2485 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2486 {
2487 	struct mrvl_txq *q = txq;
2488 	struct mrvl_shadow_txq *sq;
2489 	struct pp2_hif *hif;
2490 	struct pp2_ppio_desc descs[nb_pkts];
2491 	unsigned int core_id = rte_lcore_id();
2492 	int i, ret, bytes_sent = 0;
2493 	uint16_t num, sq_free_size;
2494 	uint64_t addr;
2495 
2496 	hif = mrvl_get_hif(q->priv, core_id);
2497 	sq = &q->shadow_txqs[core_id];
2498 
2499 	if (unlikely(!q->priv->ppio || !hif))
2500 		return 0;
2501 
2502 	if (sq->size)
2503 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2504 				       sq, q->queue_id, 0);
2505 
2506 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2507 	if (unlikely(nb_pkts > sq_free_size))
2508 		nb_pkts = sq_free_size;
2509 
2510 	for (i = 0; i < nb_pkts; i++) {
2511 		struct rte_mbuf *mbuf = tx_pkts[i];
2512 		int gen_l3_cksum, gen_l4_cksum;
2513 		enum pp2_outq_l3_type l3_type;
2514 		enum pp2_outq_l4_type l4_type;
2515 
2516 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2517 			struct rte_mbuf *pref_pkt_hdr;
2518 
2519 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2520 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2521 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2522 		}
2523 
2524 		mrvl_fill_shadowq(sq, mbuf);
2525 		mrvl_fill_desc(&descs[i], mbuf);
2526 
2527 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2528 		/*
2529 		 * in case unsupported ol_flags were passed
2530 		 * do not update descriptor offload information
2531 		 */
2532 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2533 					      &l3_type, &l4_type, &gen_l3_cksum,
2534 					      &gen_l4_cksum);
2535 		if (unlikely(ret))
2536 			continue;
2537 
2538 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2539 						  mbuf->l2_len,
2540 						  mbuf->l2_len + mbuf->l3_len,
2541 						  gen_l3_cksum, gen_l4_cksum);
2542 	}
2543 
2544 	num = nb_pkts;
2545 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2546 	/* number of packets that were not sent */
2547 	if (unlikely(num > nb_pkts)) {
2548 		for (i = nb_pkts; i < num; i++) {
2549 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2550 				MRVL_PP2_TX_SHADOWQ_MASK;
2551 			addr = cookie_addr_high | sq->ent[sq->head].buff.cookie;
2552 			bytes_sent -=
2553 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2554 		}
2555 		sq->size -= num - nb_pkts;
2556 	}
2557 
2558 	q->bytes_sent += bytes_sent;
2559 
2560 	return nb_pkts;
2561 }
2562 
2563 /** DPDK callback for S/G transmit.
2564  *
2565  * @param txq
2566  *   Generic pointer transmit queue.
2567  * @param tx_pkts
2568  *   Packets to transmit.
2569  * @param nb_pkts
2570  *   Number of packets in array.
2571  *
2572  * @return
2573  *   Number of packets successfully transmitted.
2574  */
2575 static uint16_t
2576 mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
2577 		     uint16_t nb_pkts)
2578 {
2579 	struct mrvl_txq *q = txq;
2580 	struct mrvl_shadow_txq *sq;
2581 	struct pp2_hif *hif;
2582 	struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
2583 	struct pp2_ppio_sg_pkts pkts;
2584 	uint8_t frags[nb_pkts];
2585 	unsigned int core_id = rte_lcore_id();
2586 	int i, j, ret, bytes_sent = 0;
2587 	int tail, tail_first;
2588 	uint16_t num, sq_free_size;
2589 	uint16_t nb_segs, total_descs = 0;
2590 	uint64_t addr;
2591 
2592 	hif = mrvl_get_hif(q->priv, core_id);
2593 	sq = &q->shadow_txqs[core_id];
2594 	pkts.frags = frags;
2595 	pkts.num = 0;
2596 
2597 	if (unlikely(!q->priv->ppio || !hif))
2598 		return 0;
2599 
2600 	if (sq->size)
2601 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2602 				       sq, q->queue_id, 0);
2603 
2604 	/* Save shadow queue free size */
2605 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2606 
2607 	tail = 0;
2608 	for (i = 0; i < nb_pkts; i++) {
2609 		struct rte_mbuf *mbuf = tx_pkts[i];
2610 		struct rte_mbuf *seg = NULL;
2611 		int gen_l3_cksum, gen_l4_cksum;
2612 		enum pp2_outq_l3_type l3_type;
2613 		enum pp2_outq_l4_type l4_type;
2614 
2615 		nb_segs = mbuf->nb_segs;
2616 		tail_first = tail;
2617 		total_descs += nb_segs;
2618 
2619 		/*
2620 		 * Check if total_descs does not exceed
2621 		 * shadow queue free size
2622 		 */
2623 		if (unlikely(total_descs > sq_free_size)) {
2624 			total_descs -= nb_segs;
2625 			break;
2626 		}
2627 
2628 		/* Check if nb_segs does not exceed the max nb of desc per
2629 		 * fragmented packet
2630 		 */
2631 		if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
2632 			total_descs -= nb_segs;
2633 			RTE_LOG(ERR, PMD,
2634 				"Too many segments. Packet won't be sent.\n");
2635 			break;
2636 		}
2637 
2638 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2639 			struct rte_mbuf *pref_pkt_hdr;
2640 
2641 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2642 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2643 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2644 		}
2645 
2646 		pkts.frags[pkts.num] = nb_segs;
2647 		pkts.num++;
2648 
2649 		seg = mbuf;
2650 		for (j = 0; j < nb_segs - 1; j++) {
2651 			/* For the subsequent segments, set shadow queue
2652 			 * buffer to NULL
2653 			 */
2654 			mrvl_fill_shadowq(sq, NULL);
2655 			mrvl_fill_desc(&descs[tail], seg);
2656 
2657 			tail++;
2658 			seg = seg->next;
2659 		}
2660 		/* Put first mbuf info in last shadow queue entry */
2661 		mrvl_fill_shadowq(sq, mbuf);
2662 		/* Update descriptor with last segment */
2663 		mrvl_fill_desc(&descs[tail++], seg);
2664 
2665 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2666 		/* In case unsupported ol_flags were passed
2667 		 * do not update descriptor offload information
2668 		 */
2669 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2670 					      &l3_type, &l4_type, &gen_l3_cksum,
2671 					      &gen_l4_cksum);
2672 		if (unlikely(ret))
2673 			continue;
2674 
2675 		pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
2676 						  l4_type, mbuf->l2_len,
2677 						  mbuf->l2_len + mbuf->l3_len,
2678 						  gen_l3_cksum, gen_l4_cksum);
2679 	}
2680 
2681 	num = total_descs;
2682 	pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
2683 			 &total_descs, &pkts);
2684 	/* number of packets that were not sent */
2685 	if (unlikely(num > total_descs)) {
2686 		for (i = total_descs; i < num; i++) {
2687 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2688 				MRVL_PP2_TX_SHADOWQ_MASK;
2689 
2690 			addr = sq->ent[sq->head].buff.cookie;
2691 			if (addr)
2692 				bytes_sent -=
2693 					rte_pktmbuf_pkt_len((struct rte_mbuf *)
2694 						(cookie_addr_high | addr));
2695 		}
2696 		sq->size -= num - total_descs;
2697 		nb_pkts = pkts.num;
2698 	}
2699 
2700 	q->bytes_sent += bytes_sent;
2701 
2702 	return nb_pkts;
2703 }
2704 
2705 /**
2706  * Initialize packet processor.
2707  *
2708  * @return
2709  *   0 on success, negative error value otherwise.
2710  */
2711 static int
2712 mrvl_init_pp2(void)
2713 {
2714 	struct pp2_init_params init_params;
2715 
2716 	memset(&init_params, 0, sizeof(init_params));
2717 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
2718 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
2719 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
2720 
2721 	return pp2_init(&init_params);
2722 }
2723 
2724 /**
2725  * Deinitialize packet processor.
2726  *
2727  * @return
2728  *   0 on success, negative error value otherwise.
2729  */
2730 static void
2731 mrvl_deinit_pp2(void)
2732 {
2733 	pp2_deinit();
2734 }
2735 
2736 /**
2737  * Create private device structure.
2738  *
2739  * @param dev_name
2740  *   Pointer to the port name passed in the initialization parameters.
2741  *
2742  * @return
2743  *   Pointer to the newly allocated private device structure.
2744  */
2745 static struct mrvl_priv *
2746 mrvl_priv_create(const char *dev_name)
2747 {
2748 	struct pp2_bpool_params bpool_params;
2749 	char match[MRVL_MATCH_LEN];
2750 	struct mrvl_priv *priv;
2751 	int ret, bpool_bit;
2752 
2753 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
2754 	if (!priv)
2755 		return NULL;
2756 
2757 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
2758 				       &priv->pp_id, &priv->ppio_id);
2759 	if (ret)
2760 		goto out_free_priv;
2761 
2762 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
2763 				     PP2_BPOOL_NUM_POOLS);
2764 	if (bpool_bit < 0)
2765 		goto out_free_priv;
2766 	priv->bpool_bit = bpool_bit;
2767 
2768 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
2769 		 priv->bpool_bit);
2770 	memset(&bpool_params, 0, sizeof(bpool_params));
2771 	bpool_params.match = match;
2772 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
2773 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
2774 	if (ret)
2775 		goto out_clear_bpool_bit;
2776 
2777 	priv->ppio_params.type = PP2_PPIO_T_NIC;
2778 	rte_spinlock_init(&priv->lock);
2779 
2780 	return priv;
2781 out_clear_bpool_bit:
2782 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2783 out_free_priv:
2784 	rte_free(priv);
2785 	return NULL;
2786 }
2787 
2788 /**
2789  * Create device representing Ethernet port.
2790  *
2791  * @param name
2792  *   Pointer to the port's name.
2793  *
2794  * @return
2795  *   0 on success, negative error value otherwise.
2796  */
2797 static int
2798 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
2799 {
2800 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
2801 	struct rte_eth_dev *eth_dev;
2802 	struct mrvl_priv *priv;
2803 	struct ifreq req;
2804 
2805 	eth_dev = rte_eth_dev_allocate(name);
2806 	if (!eth_dev)
2807 		return -ENOMEM;
2808 
2809 	priv = mrvl_priv_create(name);
2810 	if (!priv) {
2811 		ret = -ENOMEM;
2812 		goto out_free;
2813 	}
2814 	eth_dev->data->dev_private = priv;
2815 
2816 	eth_dev->data->mac_addrs =
2817 		rte_zmalloc("mac_addrs",
2818 			    RTE_ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
2819 	if (!eth_dev->data->mac_addrs) {
2820 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
2821 		ret = -ENOMEM;
2822 		goto out_free;
2823 	}
2824 
2825 	memset(&req, 0, sizeof(req));
2826 	strcpy(req.ifr_name, name);
2827 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
2828 	if (ret)
2829 		goto out_free;
2830 
2831 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
2832 	       req.ifr_addr.sa_data, RTE_ETHER_ADDR_LEN);
2833 
2834 	eth_dev->device = &vdev->device;
2835 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
2836 	mrvl_set_tx_function(eth_dev);
2837 	eth_dev->dev_ops = &mrvl_ops;
2838 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2839 
2840 	rte_eth_dev_probing_finish(eth_dev);
2841 	return 0;
2842 out_free:
2843 	rte_eth_dev_release_port(eth_dev);
2844 
2845 	return ret;
2846 }
2847 
2848 /**
2849  * Callback used by rte_kvargs_process() during argument parsing.
2850  *
2851  * @param key
2852  *   Pointer to the parsed key (unused).
2853  * @param value
2854  *   Pointer to the parsed value.
2855  * @param extra_args
2856  *   Pointer to the extra arguments which contains address of the
2857  *   table of pointers to parsed interface names.
2858  *
2859  * @return
2860  *   Always 0.
2861  */
2862 static int
2863 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
2864 		 void *extra_args)
2865 {
2866 	struct mrvl_ifnames *ifnames = extra_args;
2867 
2868 	ifnames->names[ifnames->idx++] = value;
2869 
2870 	return 0;
2871 }
2872 
2873 /**
2874  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
2875  */
2876 static void
2877 mrvl_deinit_hifs(void)
2878 {
2879 	int i;
2880 
2881 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
2882 		if (hifs[i])
2883 			pp2_hif_deinit(hifs[i]);
2884 	}
2885 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
2886 	memset(hifs, 0, sizeof(hifs));
2887 }
2888 
2889 /**
2890  * DPDK callback to register the virtual device.
2891  *
2892  * @param vdev
2893  *   Pointer to the virtual device.
2894  *
2895  * @return
2896  *   0 on success, negative error value otherwise.
2897  */
2898 static int
2899 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
2900 {
2901 	struct rte_kvargs *kvlist;
2902 	struct mrvl_ifnames ifnames;
2903 	int ret = -EINVAL;
2904 	uint32_t i, ifnum, cfgnum;
2905 	const char *params;
2906 
2907 	params = rte_vdev_device_args(vdev);
2908 	if (!params)
2909 		return -EINVAL;
2910 
2911 	kvlist = rte_kvargs_parse(params, valid_args);
2912 	if (!kvlist)
2913 		return -EINVAL;
2914 
2915 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
2916 	if (ifnum > RTE_DIM(ifnames.names))
2917 		goto out_free_kvlist;
2918 
2919 	ifnames.idx = 0;
2920 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
2921 			   mrvl_get_ifnames, &ifnames);
2922 
2923 
2924 	/*
2925 	 * The below system initialization should be done only once,
2926 	 * on the first provided configuration file
2927 	 */
2928 	if (!mrvl_qos_cfg) {
2929 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
2930 		MRVL_LOG(INFO, "Parsing config file!");
2931 		if (cfgnum > 1) {
2932 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
2933 			goto out_free_kvlist;
2934 		} else if (cfgnum == 1) {
2935 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
2936 					   mrvl_get_qoscfg, &mrvl_qos_cfg);
2937 		}
2938 	}
2939 
2940 	if (mrvl_dev_num)
2941 		goto init_devices;
2942 
2943 	MRVL_LOG(INFO, "Perform MUSDK initializations");
2944 
2945 	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
2946 	if (ret)
2947 		goto out_free_kvlist;
2948 
2949 	ret = mrvl_init_pp2();
2950 	if (ret) {
2951 		MRVL_LOG(ERR, "Failed to init PP!");
2952 		rte_mvep_deinit(MVEP_MOD_T_PP2);
2953 		goto out_free_kvlist;
2954 	}
2955 
2956 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
2957 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
2958 
2959 	mrvl_lcore_first = RTE_MAX_LCORE;
2960 	mrvl_lcore_last = 0;
2961 
2962 init_devices:
2963 	for (i = 0; i < ifnum; i++) {
2964 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
2965 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
2966 		if (ret)
2967 			goto out_cleanup;
2968 		mrvl_dev_num++;
2969 	}
2970 
2971 	rte_kvargs_free(kvlist);
2972 
2973 	return 0;
2974 out_cleanup:
2975 	rte_pmd_mrvl_remove(vdev);
2976 
2977 out_free_kvlist:
2978 	rte_kvargs_free(kvlist);
2979 
2980 	return ret;
2981 }
2982 
2983 /**
2984  * DPDK callback to remove virtual device.
2985  *
2986  * @param vdev
2987  *   Pointer to the removed virtual device.
2988  *
2989  * @return
2990  *   0 on success, negative error value otherwise.
2991  */
2992 static int
2993 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
2994 {
2995 	uint16_t port_id;
2996 	int ret = 0;
2997 
2998 	RTE_ETH_FOREACH_DEV(port_id) {
2999 		if (rte_eth_devices[port_id].device != &vdev->device)
3000 			continue;
3001 		ret |= rte_eth_dev_close(port_id);
3002 	}
3003 
3004 	return ret == 0 ? 0 : -EIO;
3005 }
3006 
3007 static struct rte_vdev_driver pmd_mrvl_drv = {
3008 	.probe = rte_pmd_mrvl_probe,
3009 	.remove = rte_pmd_mrvl_remove,
3010 };
3011 
3012 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
3013 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
3014 RTE_LOG_REGISTER(mrvl_logtype, pmd.net.mvpp2, NOTICE);
3015