xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision d2c37f739c67af628bf98fffcbf289f4b49b0332)
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 0x0F
37 /* maximum number of available hifs */
38 #define MRVL_MUSDK_HIFS_MAX 9
39 
40 /* prefetch shift */
41 #define MRVL_MUSDK_PREFETCH_SHIFT 2
42 
43 /* TCAM has 25 entries reserved for uc/mc filter entries */
44 #define MRVL_MAC_ADDRS_MAX 25
45 #define MRVL_MATCH_LEN 16
46 #define MRVL_PKT_EFFEC_OFFS (MRVL_PKT_OFFS + MV_MH_SIZE)
47 /* Maximum allowable packet size */
48 #define MRVL_PKT_SIZE_MAX (10240 - MV_MH_SIZE)
49 
50 #define MRVL_IFACE_NAME_ARG "iface"
51 #define MRVL_CFG_ARG "cfg"
52 
53 #define MRVL_BURST_SIZE 64
54 
55 #define MRVL_ARP_LENGTH 28
56 
57 #define MRVL_COOKIE_ADDR_INVALID ~0ULL
58 #define MRVL_COOKIE_HIGH_ADDR_MASK 0xffffff0000000000
59 
60 /** Port Rx offload capabilities */
61 #define MRVL_RX_OFFLOADS (DEV_RX_OFFLOAD_VLAN_FILTER | \
62 			  DEV_RX_OFFLOAD_JUMBO_FRAME | \
63 			  DEV_RX_OFFLOAD_CHECKSUM)
64 
65 /** Port Tx offloads capabilities */
66 #define MRVL_TX_OFFLOADS (DEV_TX_OFFLOAD_IPV4_CKSUM | \
67 			  DEV_TX_OFFLOAD_UDP_CKSUM | \
68 			  DEV_TX_OFFLOAD_TCP_CKSUM | \
69 			  DEV_TX_OFFLOAD_MULTI_SEGS)
70 
71 static const char * const valid_args[] = {
72 	MRVL_IFACE_NAME_ARG,
73 	MRVL_CFG_ARG,
74 	NULL
75 };
76 
77 static int used_hifs = MRVL_MUSDK_HIFS_RESERVED;
78 static struct pp2_hif *hifs[RTE_MAX_LCORE];
79 static int used_bpools[PP2_NUM_PKT_PROC] = {
80 	[0 ... PP2_NUM_PKT_PROC - 1] = MRVL_MUSDK_BPOOLS_RESERVED
81 };
82 
83 static struct pp2_bpool *mrvl_port_to_bpool_lookup[RTE_MAX_ETHPORTS];
84 static int mrvl_port_bpool_size[PP2_NUM_PKT_PROC][PP2_BPOOL_NUM_POOLS][RTE_MAX_LCORE];
85 static uint64_t cookie_addr_high = MRVL_COOKIE_ADDR_INVALID;
86 
87 struct mrvl_ifnames {
88 	const char *names[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
89 	int idx;
90 };
91 
92 /*
93  * To use buffer harvesting based on loopback port shadow queue structure
94  * was introduced for buffers information bookkeeping.
95  *
96  * Before sending the packet, related buffer information (pp2_buff_inf) is
97  * stored in shadow queue. After packet is transmitted no longer used
98  * packet buffer is released back to it's original hardware pool,
99  * on condition it originated from interface.
100  * In case it  was generated by application itself i.e: mbuf->port field is
101  * 0xff then its released to software mempool.
102  */
103 struct mrvl_shadow_txq {
104 	int head;           /* write index - used when sending buffers */
105 	int tail;           /* read index - used when releasing buffers */
106 	u16 size;           /* queue occupied size */
107 	u16 num_to_release; /* number of descriptors sent, that can be
108 			     * released
109 			     */
110 	struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
111 };
112 
113 struct mrvl_rxq {
114 	struct mrvl_priv *priv;
115 	struct rte_mempool *mp;
116 	int queue_id;
117 	int port_id;
118 	int cksum_enabled;
119 	uint64_t bytes_recv;
120 	uint64_t drop_mac;
121 };
122 
123 struct mrvl_txq {
124 	struct mrvl_priv *priv;
125 	int queue_id;
126 	int port_id;
127 	uint64_t bytes_sent;
128 	struct mrvl_shadow_txq shadow_txqs[RTE_MAX_LCORE];
129 	int tx_deferred_start;
130 };
131 
132 static int mrvl_lcore_first;
133 static int mrvl_lcore_last;
134 static int mrvl_dev_num;
135 
136 static int mrvl_fill_bpool(struct mrvl_rxq *rxq, int num);
137 static inline void mrvl_free_sent_buffers(struct pp2_ppio *ppio,
138 			struct pp2_hif *hif, unsigned int core_id,
139 			struct mrvl_shadow_txq *sq, int qid, int force);
140 
141 static uint16_t mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
142 				  uint16_t nb_pkts);
143 static uint16_t mrvl_tx_sg_pkt_burst(void *txq,	struct rte_mbuf **tx_pkts,
144 				     uint16_t nb_pkts);
145 static int rte_pmd_mrvl_remove(struct rte_vdev_device *vdev);
146 static void mrvl_deinit_pp2(void);
147 static void mrvl_deinit_hifs(void);
148 
149 
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->isolated)
999 		return -ENOTSUP;
1000 
1001 	if (!priv->ppio)
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->isolated)
1029 		return -ENOTSUP;
1030 
1031 	if (!priv->ppio)
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->isolated)
1059 		return -ENOTSUP;
1060 
1061 	if (!priv->ppio)
1062 		return 0;
1063 
1064 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
1065 	if (ret) {
1066 		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
1067 		return -EAGAIN;
1068 	}
1069 
1070 	return 0;
1071 }
1072 
1073 /**
1074  * DPDK callback to disable allmulticast mode.
1075  *
1076  * @param dev
1077  *   Pointer to Ethernet device structure.
1078  *
1079  * @return
1080  *   0 on success, negative error value otherwise.
1081  */
1082 static int
1083 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
1084 {
1085 	struct mrvl_priv *priv = dev->data->dev_private;
1086 	int ret;
1087 
1088 	if (priv->isolated)
1089 		return -ENOTSUP;
1090 
1091 	if (!priv->ppio)
1092 		return 0;
1093 
1094 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
1095 	if (ret) {
1096 		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
1097 		return -EAGAIN;
1098 	}
1099 
1100 	return 0;
1101 }
1102 
1103 /**
1104  * DPDK callback to remove a MAC address.
1105  *
1106  * @param dev
1107  *   Pointer to Ethernet device structure.
1108  * @param index
1109  *   MAC address index.
1110  */
1111 static void
1112 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1113 {
1114 	struct mrvl_priv *priv = dev->data->dev_private;
1115 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1116 	int ret;
1117 
1118 	if (priv->isolated)
1119 		return;
1120 
1121 	if (!priv->ppio)
1122 		return;
1123 
1124 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
1125 				       dev->data->mac_addrs[index].addr_bytes);
1126 	if (ret) {
1127 		rte_ether_format_addr(buf, sizeof(buf),
1128 				  &dev->data->mac_addrs[index]);
1129 		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
1130 	}
1131 }
1132 
1133 /**
1134  * DPDK callback to add a MAC address.
1135  *
1136  * @param dev
1137  *   Pointer to Ethernet device structure.
1138  * @param mac_addr
1139  *   MAC address to register.
1140  * @param index
1141  *   MAC address index.
1142  * @param vmdq
1143  *   VMDq pool index to associate address with (unused).
1144  *
1145  * @return
1146  *   0 on success, negative error value otherwise.
1147  */
1148 static int
1149 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1150 		  uint32_t index, uint32_t vmdq __rte_unused)
1151 {
1152 	struct mrvl_priv *priv = dev->data->dev_private;
1153 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1154 	int ret;
1155 
1156 	if (priv->isolated)
1157 		return -ENOTSUP;
1158 
1159 	if (!priv->ppio)
1160 		return 0;
1161 
1162 	if (index == 0)
1163 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1164 		return -1;
1165 
1166 	/*
1167 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1168 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1169 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1170 	 * 21 respectively.
1171 	 *
1172 	 * If more than uc_filter_max uc addresses were added to filter list
1173 	 * then NIC will switch to promiscuous mode automatically.
1174 	 *
1175 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1176 	 * were added to filter list then NIC will switch to all-multicast mode
1177 	 * automatically.
1178 	 */
1179 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1180 	if (ret) {
1181 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1182 		MRVL_LOG(ERR, "Failed to add mac %s", buf);
1183 		return -1;
1184 	}
1185 
1186 	return 0;
1187 }
1188 
1189 /**
1190  * DPDK callback to set the primary MAC address.
1191  *
1192  * @param dev
1193  *   Pointer to Ethernet device structure.
1194  * @param mac_addr
1195  *   MAC address to register.
1196  *
1197  * @return
1198  *   0 on success, negative error value otherwise.
1199  */
1200 static int
1201 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1202 {
1203 	struct mrvl_priv *priv = dev->data->dev_private;
1204 	int ret;
1205 
1206 	if (priv->isolated)
1207 		return -ENOTSUP;
1208 
1209 	if (!priv->ppio)
1210 		return 0;
1211 
1212 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1213 	if (ret) {
1214 		char buf[RTE_ETHER_ADDR_FMT_SIZE];
1215 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1216 		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1217 	}
1218 
1219 	return ret;
1220 }
1221 
1222 /**
1223  * DPDK callback to get device statistics.
1224  *
1225  * @param dev
1226  *   Pointer to Ethernet device structure.
1227  * @param stats
1228  *   Stats structure output buffer.
1229  *
1230  * @return
1231  *   0 on success, negative error value otherwise.
1232  */
1233 static int
1234 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1235 {
1236 	struct mrvl_priv *priv = dev->data->dev_private;
1237 	struct pp2_ppio_statistics ppio_stats;
1238 	uint64_t drop_mac = 0;
1239 	unsigned int i, idx, ret;
1240 
1241 	if (!priv->ppio)
1242 		return -EPERM;
1243 
1244 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1245 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1246 		struct pp2_ppio_inq_statistics rx_stats;
1247 
1248 		if (!rxq)
1249 			continue;
1250 
1251 		idx = rxq->queue_id;
1252 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1253 			MRVL_LOG(ERR,
1254 				"rx queue %d stats out of range (0 - %d)",
1255 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1256 			continue;
1257 		}
1258 
1259 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1260 						  priv->rxq_map[idx].tc,
1261 						  priv->rxq_map[idx].inq,
1262 						  &rx_stats, 0);
1263 		if (unlikely(ret)) {
1264 			MRVL_LOG(ERR,
1265 				"Failed to update rx queue %d stats", idx);
1266 			break;
1267 		}
1268 
1269 		stats->q_ibytes[idx] = rxq->bytes_recv;
1270 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1271 		stats->q_errors[idx] = rx_stats.drop_early +
1272 				       rx_stats.drop_fullq +
1273 				       rx_stats.drop_bm +
1274 				       rxq->drop_mac;
1275 		stats->ibytes += rxq->bytes_recv;
1276 		drop_mac += rxq->drop_mac;
1277 	}
1278 
1279 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1280 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1281 		struct pp2_ppio_outq_statistics tx_stats;
1282 
1283 		if (!txq)
1284 			continue;
1285 
1286 		idx = txq->queue_id;
1287 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1288 			MRVL_LOG(ERR,
1289 				"tx queue %d stats out of range (0 - %d)",
1290 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1291 		}
1292 
1293 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1294 						   &tx_stats, 0);
1295 		if (unlikely(ret)) {
1296 			MRVL_LOG(ERR,
1297 				"Failed to update tx queue %d stats", idx);
1298 			break;
1299 		}
1300 
1301 		stats->q_opackets[idx] = tx_stats.deq_desc;
1302 		stats->q_obytes[idx] = txq->bytes_sent;
1303 		stats->obytes += txq->bytes_sent;
1304 	}
1305 
1306 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1307 	if (unlikely(ret)) {
1308 		MRVL_LOG(ERR, "Failed to update port statistics");
1309 		return ret;
1310 	}
1311 
1312 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1313 	stats->opackets += ppio_stats.tx_packets;
1314 	stats->imissed += ppio_stats.rx_fullq_dropped +
1315 			  ppio_stats.rx_bm_dropped +
1316 			  ppio_stats.rx_early_dropped +
1317 			  ppio_stats.rx_fifo_dropped +
1318 			  ppio_stats.rx_cls_dropped;
1319 	stats->ierrors = drop_mac;
1320 
1321 	return 0;
1322 }
1323 
1324 /**
1325  * DPDK callback to clear device statistics.
1326  *
1327  * @param dev
1328  *   Pointer to Ethernet device structure.
1329  *
1330  * @return
1331  *   0 on success, negative error value otherwise.
1332  */
1333 static int
1334 mrvl_stats_reset(struct rte_eth_dev *dev)
1335 {
1336 	struct mrvl_priv *priv = dev->data->dev_private;
1337 	int i;
1338 
1339 	if (!priv->ppio)
1340 		return 0;
1341 
1342 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1343 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1344 
1345 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1346 					    priv->rxq_map[i].inq, NULL, 1);
1347 		rxq->bytes_recv = 0;
1348 		rxq->drop_mac = 0;
1349 	}
1350 
1351 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1352 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1353 
1354 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1355 		txq->bytes_sent = 0;
1356 	}
1357 
1358 	return pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1359 }
1360 
1361 /**
1362  * DPDK callback to get extended statistics.
1363  *
1364  * @param dev
1365  *   Pointer to Ethernet device structure.
1366  * @param stats
1367  *   Pointer to xstats table.
1368  * @param n
1369  *   Number of entries in xstats table.
1370  * @return
1371  *   Negative value on error, number of read xstats otherwise.
1372  */
1373 static int
1374 mrvl_xstats_get(struct rte_eth_dev *dev,
1375 		struct rte_eth_xstat *stats, unsigned int n)
1376 {
1377 	struct mrvl_priv *priv = dev->data->dev_private;
1378 	struct pp2_ppio_statistics ppio_stats;
1379 	unsigned int i;
1380 
1381 	if (!stats)
1382 		return 0;
1383 
1384 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1385 	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1386 		uint64_t val;
1387 
1388 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1389 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1390 					    mrvl_xstats_tbl[i].offset);
1391 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1392 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1393 					    mrvl_xstats_tbl[i].offset);
1394 		else
1395 			return -EINVAL;
1396 
1397 		stats[i].id = i;
1398 		stats[i].value = val;
1399 	}
1400 
1401 	return n;
1402 }
1403 
1404 /**
1405  * DPDK callback to reset extended statistics.
1406  *
1407  * @param dev
1408  *   Pointer to Ethernet device structure.
1409  *
1410  * @return
1411  *   0 on success, negative error value otherwise.
1412  */
1413 static int
1414 mrvl_xstats_reset(struct rte_eth_dev *dev)
1415 {
1416 	return mrvl_stats_reset(dev);
1417 }
1418 
1419 /**
1420  * DPDK callback to get extended statistics names.
1421  *
1422  * @param dev (unused)
1423  *   Pointer to Ethernet device structure.
1424  * @param xstats_names
1425  *   Pointer to xstats names table.
1426  * @param size
1427  *   Size of the xstats names table.
1428  * @return
1429  *   Number of read names.
1430  */
1431 static int
1432 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1433 		      struct rte_eth_xstat_name *xstats_names,
1434 		      unsigned int size)
1435 {
1436 	unsigned int i;
1437 
1438 	if (!xstats_names)
1439 		return RTE_DIM(mrvl_xstats_tbl);
1440 
1441 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1442 		strlcpy(xstats_names[i].name, mrvl_xstats_tbl[i].name,
1443 			RTE_ETH_XSTATS_NAME_SIZE);
1444 
1445 	return size;
1446 }
1447 
1448 /**
1449  * DPDK callback to get information about the device.
1450  *
1451  * @param dev
1452  *   Pointer to Ethernet device structure (unused).
1453  * @param info
1454  *   Info structure output buffer.
1455  */
1456 static int
1457 mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
1458 		   struct rte_eth_dev_info *info)
1459 {
1460 	info->speed_capa = ETH_LINK_SPEED_10M |
1461 			   ETH_LINK_SPEED_100M |
1462 			   ETH_LINK_SPEED_1G |
1463 			   ETH_LINK_SPEED_10G;
1464 
1465 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1466 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1467 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1468 
1469 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1470 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1471 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1472 
1473 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1474 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1475 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1476 
1477 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1478 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1479 
1480 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1481 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1482 
1483 	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1484 				       ETH_RSS_NONFRAG_IPV4_TCP |
1485 				       ETH_RSS_NONFRAG_IPV4_UDP;
1486 
1487 	/* By default packets are dropped if no descriptors are available */
1488 	info->default_rxconf.rx_drop_en = 1;
1489 
1490 	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1491 
1492 	return 0;
1493 }
1494 
1495 /**
1496  * Return supported packet types.
1497  *
1498  * @param dev
1499  *   Pointer to Ethernet device structure (unused).
1500  *
1501  * @return
1502  *   Const pointer to the table with supported packet types.
1503  */
1504 static const uint32_t *
1505 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1506 {
1507 	static const uint32_t ptypes[] = {
1508 		RTE_PTYPE_L2_ETHER,
1509 		RTE_PTYPE_L2_ETHER_VLAN,
1510 		RTE_PTYPE_L2_ETHER_QINQ,
1511 		RTE_PTYPE_L3_IPV4,
1512 		RTE_PTYPE_L3_IPV4_EXT,
1513 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1514 		RTE_PTYPE_L3_IPV6,
1515 		RTE_PTYPE_L3_IPV6_EXT,
1516 		RTE_PTYPE_L2_ETHER_ARP,
1517 		RTE_PTYPE_L4_TCP,
1518 		RTE_PTYPE_L4_UDP
1519 	};
1520 
1521 	return ptypes;
1522 }
1523 
1524 /**
1525  * DPDK callback to get information about specific receive queue.
1526  *
1527  * @param dev
1528  *   Pointer to Ethernet device structure.
1529  * @param rx_queue_id
1530  *   Receive queue index.
1531  * @param qinfo
1532  *   Receive queue information structure.
1533  */
1534 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1535 			      struct rte_eth_rxq_info *qinfo)
1536 {
1537 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1538 	struct mrvl_priv *priv = dev->data->dev_private;
1539 	int inq = priv->rxq_map[rx_queue_id].inq;
1540 	int tc = priv->rxq_map[rx_queue_id].tc;
1541 	struct pp2_ppio_tc_params *tc_params =
1542 		&priv->ppio_params.inqs_params.tcs_params[tc];
1543 
1544 	qinfo->mp = q->mp;
1545 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1546 }
1547 
1548 /**
1549  * DPDK callback to get information about specific transmit queue.
1550  *
1551  * @param dev
1552  *   Pointer to Ethernet device structure.
1553  * @param tx_queue_id
1554  *   Transmit queue index.
1555  * @param qinfo
1556  *   Transmit queue information structure.
1557  */
1558 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1559 			      struct rte_eth_txq_info *qinfo)
1560 {
1561 	struct mrvl_priv *priv = dev->data->dev_private;
1562 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1563 
1564 	qinfo->nb_desc =
1565 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1566 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1567 }
1568 
1569 /**
1570  * DPDK callback to Configure a VLAN filter.
1571  *
1572  * @param dev
1573  *   Pointer to Ethernet device structure.
1574  * @param vlan_id
1575  *   VLAN ID to filter.
1576  * @param on
1577  *   Toggle filter.
1578  *
1579  * @return
1580  *   0 on success, negative error value otherwise.
1581  */
1582 static int
1583 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1584 {
1585 	struct mrvl_priv *priv = dev->data->dev_private;
1586 
1587 	if (priv->isolated)
1588 		return -ENOTSUP;
1589 
1590 	if (!priv->ppio)
1591 		return 0;
1592 
1593 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1594 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1595 }
1596 
1597 /**
1598  * Release buffers to hardware bpool (buffer-pool)
1599  *
1600  * @param rxq
1601  *   Receive queue pointer.
1602  * @param num
1603  *   Number of buffers to release to bpool.
1604  *
1605  * @return
1606  *   0 on success, negative error value otherwise.
1607  */
1608 static int
1609 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1610 {
1611 	struct buff_release_entry entries[num];
1612 	struct rte_mbuf *mbufs[num];
1613 	int i, ret;
1614 	unsigned int core_id;
1615 	struct pp2_hif *hif;
1616 	struct pp2_bpool *bpool;
1617 
1618 	core_id = rte_lcore_id();
1619 	if (core_id == LCORE_ID_ANY)
1620 		core_id = rte_get_main_lcore();
1621 
1622 	hif = mrvl_get_hif(rxq->priv, core_id);
1623 	if (!hif)
1624 		return -1;
1625 
1626 	bpool = rxq->priv->bpool;
1627 
1628 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1629 	if (ret)
1630 		return ret;
1631 
1632 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1633 		cookie_addr_high =
1634 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1635 
1636 	for (i = 0; i < num; i++) {
1637 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1638 			!= cookie_addr_high) {
1639 			MRVL_LOG(ERR,
1640 				"mbuf virtual addr high is out of range "
1641 				"0x%x instead of 0x%x\n",
1642 				(uint32_t)((uint64_t)mbufs[i] >> 32),
1643 				(uint32_t)(cookie_addr_high >> 32));
1644 			goto out;
1645 		}
1646 
1647 		entries[i].buff.addr =
1648 			rte_mbuf_data_iova_default(mbufs[i]);
1649 		entries[i].buff.cookie = (uintptr_t)mbufs[i];
1650 		entries[i].bpool = bpool;
1651 	}
1652 
1653 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1654 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1655 
1656 	if (i != num)
1657 		goto out;
1658 
1659 	return 0;
1660 out:
1661 	for (; i < num; i++)
1662 		rte_pktmbuf_free(mbufs[i]);
1663 
1664 	return -1;
1665 }
1666 
1667 /**
1668  * DPDK callback to configure the receive queue.
1669  *
1670  * @param dev
1671  *   Pointer to Ethernet device structure.
1672  * @param idx
1673  *   RX queue index.
1674  * @param desc
1675  *   Number of descriptors to configure in queue.
1676  * @param socket
1677  *   NUMA socket on which memory must be allocated.
1678  * @param conf
1679  *   Thresholds parameters.
1680  * @param mp
1681  *   Memory pool for buffer allocations.
1682  *
1683  * @return
1684  *   0 on success, negative error value otherwise.
1685  */
1686 static int
1687 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1688 		    unsigned int socket,
1689 		    const struct rte_eth_rxconf *conf,
1690 		    struct rte_mempool *mp)
1691 {
1692 	struct mrvl_priv *priv = dev->data->dev_private;
1693 	struct mrvl_rxq *rxq;
1694 	uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
1695 	uint32_t max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1696 	int ret, tc, inq;
1697 	uint64_t offloads;
1698 
1699 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
1700 
1701 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1702 		/*
1703 		 * Unknown TC mapping, mapping will not have a correct queue.
1704 		 */
1705 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
1706 			idx, priv->ppio_id);
1707 		return -EFAULT;
1708 	}
1709 
1710 	frame_size = buf_size - RTE_PKTMBUF_HEADROOM -
1711 		     MRVL_PKT_EFFEC_OFFS + RTE_ETHER_CRC_LEN;
1712 	if (frame_size < max_rx_pkt_len) {
1713 		MRVL_LOG(WARNING,
1714 			"Mbuf size must be increased to %u bytes to hold up "
1715 			"to %u bytes of data.",
1716 			buf_size + max_rx_pkt_len - frame_size,
1717 			max_rx_pkt_len);
1718 		dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
1719 		MRVL_LOG(INFO, "Setting max rx pkt len to %u",
1720 			dev->data->dev_conf.rxmode.max_rx_pkt_len);
1721 	}
1722 
1723 	if (dev->data->rx_queues[idx]) {
1724 		rte_free(dev->data->rx_queues[idx]);
1725 		dev->data->rx_queues[idx] = NULL;
1726 	}
1727 
1728 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
1729 	if (!rxq)
1730 		return -ENOMEM;
1731 
1732 	rxq->priv = priv;
1733 	rxq->mp = mp;
1734 	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
1735 	rxq->queue_id = idx;
1736 	rxq->port_id = dev->data->port_id;
1737 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
1738 
1739 	tc = priv->rxq_map[rxq->queue_id].tc,
1740 	inq = priv->rxq_map[rxq->queue_id].inq;
1741 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
1742 		desc;
1743 
1744 	ret = mrvl_fill_bpool(rxq, desc);
1745 	if (ret) {
1746 		rte_free(rxq);
1747 		return ret;
1748 	}
1749 
1750 	priv->bpool_init_size += desc;
1751 
1752 	dev->data->rx_queues[idx] = rxq;
1753 
1754 	return 0;
1755 }
1756 
1757 /**
1758  * DPDK callback to release the receive queue.
1759  *
1760  * @param rxq
1761  *   Generic receive queue pointer.
1762  */
1763 static void
1764 mrvl_rx_queue_release(void *rxq)
1765 {
1766 	struct mrvl_rxq *q = rxq;
1767 	struct pp2_ppio_tc_params *tc_params;
1768 	int i, num, tc, inq;
1769 	struct pp2_hif *hif;
1770 	unsigned int core_id = rte_lcore_id();
1771 
1772 	if (core_id == LCORE_ID_ANY)
1773 		core_id = rte_get_main_lcore();
1774 
1775 	if (!q)
1776 		return;
1777 
1778 	hif = mrvl_get_hif(q->priv, core_id);
1779 
1780 	if (!hif)
1781 		return;
1782 
1783 	tc = q->priv->rxq_map[q->queue_id].tc;
1784 	inq = q->priv->rxq_map[q->queue_id].inq;
1785 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
1786 	num = tc_params->inqs_params[inq].size;
1787 	for (i = 0; i < num; i++) {
1788 		struct pp2_buff_inf inf;
1789 		uint64_t addr;
1790 
1791 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
1792 		addr = cookie_addr_high | inf.cookie;
1793 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1794 	}
1795 
1796 	rte_free(q);
1797 }
1798 
1799 /**
1800  * DPDK callback to configure the transmit queue.
1801  *
1802  * @param dev
1803  *   Pointer to Ethernet device structure.
1804  * @param idx
1805  *   Transmit queue index.
1806  * @param desc
1807  *   Number of descriptors to configure in the queue.
1808  * @param socket
1809  *   NUMA socket on which memory must be allocated.
1810  * @param conf
1811  *   Tx queue configuration parameters.
1812  *
1813  * @return
1814  *   0 on success, negative error value otherwise.
1815  */
1816 static int
1817 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1818 		    unsigned int socket,
1819 		    const struct rte_eth_txconf *conf)
1820 {
1821 	struct mrvl_priv *priv = dev->data->dev_private;
1822 	struct mrvl_txq *txq;
1823 
1824 	if (dev->data->tx_queues[idx]) {
1825 		rte_free(dev->data->tx_queues[idx]);
1826 		dev->data->tx_queues[idx] = NULL;
1827 	}
1828 
1829 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
1830 	if (!txq)
1831 		return -ENOMEM;
1832 
1833 	txq->priv = priv;
1834 	txq->queue_id = idx;
1835 	txq->port_id = dev->data->port_id;
1836 	txq->tx_deferred_start = conf->tx_deferred_start;
1837 	dev->data->tx_queues[idx] = txq;
1838 
1839 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
1840 
1841 	return 0;
1842 }
1843 
1844 /**
1845  * DPDK callback to release the transmit queue.
1846  *
1847  * @param txq
1848  *   Generic transmit queue pointer.
1849  */
1850 static void
1851 mrvl_tx_queue_release(void *txq)
1852 {
1853 	struct mrvl_txq *q = txq;
1854 
1855 	if (!q)
1856 		return;
1857 
1858 	rte_free(q);
1859 }
1860 
1861 /**
1862  * DPDK callback to get flow control configuration.
1863  *
1864  * @param dev
1865  *  Pointer to Ethernet device structure.
1866  * @param fc_conf
1867  *  Pointer to the flow control configuration.
1868  *
1869  * @return
1870  *  0 on success, negative error value otherwise.
1871  */
1872 static int
1873 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1874 {
1875 	struct mrvl_priv *priv = dev->data->dev_private;
1876 	int ret, en;
1877 
1878 	if (!priv)
1879 		return -EPERM;
1880 
1881 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
1882 	if (ret) {
1883 		MRVL_LOG(ERR, "Failed to read rx pause state");
1884 		return ret;
1885 	}
1886 
1887 	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
1888 
1889 	return 0;
1890 }
1891 
1892 /**
1893  * DPDK callback to set flow control configuration.
1894  *
1895  * @param dev
1896  *  Pointer to Ethernet device structure.
1897  * @param fc_conf
1898  *  Pointer to the flow control configuration.
1899  *
1900  * @return
1901  *  0 on success, negative error value otherwise.
1902  */
1903 static int
1904 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1905 {
1906 	struct mrvl_priv *priv = dev->data->dev_private;
1907 
1908 	if (!priv)
1909 		return -EPERM;
1910 
1911 	if (fc_conf->high_water ||
1912 	    fc_conf->low_water ||
1913 	    fc_conf->pause_time ||
1914 	    fc_conf->mac_ctrl_frame_fwd ||
1915 	    fc_conf->autoneg) {
1916 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
1917 
1918 		return -EINVAL;
1919 	}
1920 
1921 	if (fc_conf->mode == RTE_FC_NONE ||
1922 	    fc_conf->mode == RTE_FC_RX_PAUSE) {
1923 		int ret, en;
1924 
1925 		en = fc_conf->mode == RTE_FC_NONE ? 0 : 1;
1926 		ret = pp2_ppio_set_rx_pause(priv->ppio, en);
1927 		if (ret)
1928 			MRVL_LOG(ERR,
1929 				"Failed to change flowctrl on RX side");
1930 
1931 		return ret;
1932 	}
1933 
1934 	return 0;
1935 }
1936 
1937 /**
1938  * Update RSS hash configuration
1939  *
1940  * @param dev
1941  *   Pointer to Ethernet device structure.
1942  * @param rss_conf
1943  *   Pointer to RSS configuration.
1944  *
1945  * @return
1946  *   0 on success, negative error value otherwise.
1947  */
1948 static int
1949 mrvl_rss_hash_update(struct rte_eth_dev *dev,
1950 		     struct rte_eth_rss_conf *rss_conf)
1951 {
1952 	struct mrvl_priv *priv = dev->data->dev_private;
1953 
1954 	if (priv->isolated)
1955 		return -ENOTSUP;
1956 
1957 	return mrvl_configure_rss(priv, rss_conf);
1958 }
1959 
1960 /**
1961  * DPDK callback to get RSS hash configuration.
1962  *
1963  * @param dev
1964  *   Pointer to Ethernet device structure.
1965  * @rss_conf
1966  *   Pointer to RSS configuration.
1967  *
1968  * @return
1969  *   Always 0.
1970  */
1971 static int
1972 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
1973 		       struct rte_eth_rss_conf *rss_conf)
1974 {
1975 	struct mrvl_priv *priv = dev->data->dev_private;
1976 	enum pp2_ppio_hash_type hash_type =
1977 		priv->ppio_params.inqs_params.hash_type;
1978 
1979 	rss_conf->rss_key = NULL;
1980 
1981 	if (hash_type == PP2_PPIO_HASH_T_NONE)
1982 		rss_conf->rss_hf = 0;
1983 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
1984 		rss_conf->rss_hf = ETH_RSS_IPV4;
1985 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
1986 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
1987 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
1988 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
1989 
1990 	return 0;
1991 }
1992 
1993 /**
1994  * DPDK callback to get rte_flow callbacks.
1995  *
1996  * @param dev
1997  *   Pointer to the device structure.
1998  * @param filer_type
1999  *   Flow filter type.
2000  * @param filter_op
2001  *   Flow filter operation.
2002  * @param arg
2003  *   Pointer to pass the flow ops.
2004  *
2005  * @return
2006  *   0 on success, negative error value otherwise.
2007  */
2008 static int
2009 mrvl_eth_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
2010 		     enum rte_filter_type filter_type,
2011 		     enum rte_filter_op filter_op, void *arg)
2012 {
2013 	switch (filter_type) {
2014 	case RTE_ETH_FILTER_GENERIC:
2015 		if (filter_op != RTE_ETH_FILTER_GET)
2016 			return -EINVAL;
2017 		*(const void **)arg = &mrvl_flow_ops;
2018 		return 0;
2019 	default:
2020 		MRVL_LOG(WARNING, "Filter type (%d) not supported",
2021 				filter_type);
2022 		return -EINVAL;
2023 	}
2024 }
2025 
2026 /**
2027  * DPDK callback to get rte_mtr callbacks.
2028  *
2029  * @param dev
2030  *   Pointer to the device structure.
2031  * @param ops
2032  *   Pointer to pass the mtr ops.
2033  *
2034  * @return
2035  *   Always 0.
2036  */
2037 static int
2038 mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2039 {
2040 	*(const void **)ops = &mrvl_mtr_ops;
2041 
2042 	return 0;
2043 }
2044 
2045 /**
2046  * DPDK callback to get rte_tm callbacks.
2047  *
2048  * @param dev
2049  *   Pointer to the device structure.
2050  * @param ops
2051  *   Pointer to pass the tm ops.
2052  *
2053  * @return
2054  *   Always 0.
2055  */
2056 static int
2057 mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2058 {
2059 	*(const void **)ops = &mrvl_tm_ops;
2060 
2061 	return 0;
2062 }
2063 
2064 static const struct eth_dev_ops mrvl_ops = {
2065 	.dev_configure = mrvl_dev_configure,
2066 	.dev_start = mrvl_dev_start,
2067 	.dev_stop = mrvl_dev_stop,
2068 	.dev_set_link_up = mrvl_dev_set_link_up,
2069 	.dev_set_link_down = mrvl_dev_set_link_down,
2070 	.dev_close = mrvl_dev_close,
2071 	.link_update = mrvl_link_update,
2072 	.promiscuous_enable = mrvl_promiscuous_enable,
2073 	.allmulticast_enable = mrvl_allmulticast_enable,
2074 	.promiscuous_disable = mrvl_promiscuous_disable,
2075 	.allmulticast_disable = mrvl_allmulticast_disable,
2076 	.mac_addr_remove = mrvl_mac_addr_remove,
2077 	.mac_addr_add = mrvl_mac_addr_add,
2078 	.mac_addr_set = mrvl_mac_addr_set,
2079 	.mtu_set = mrvl_mtu_set,
2080 	.stats_get = mrvl_stats_get,
2081 	.stats_reset = mrvl_stats_reset,
2082 	.xstats_get = mrvl_xstats_get,
2083 	.xstats_reset = mrvl_xstats_reset,
2084 	.xstats_get_names = mrvl_xstats_get_names,
2085 	.dev_infos_get = mrvl_dev_infos_get,
2086 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
2087 	.rxq_info_get = mrvl_rxq_info_get,
2088 	.txq_info_get = mrvl_txq_info_get,
2089 	.vlan_filter_set = mrvl_vlan_filter_set,
2090 	.tx_queue_start = mrvl_tx_queue_start,
2091 	.tx_queue_stop = mrvl_tx_queue_stop,
2092 	.rx_queue_setup = mrvl_rx_queue_setup,
2093 	.rx_queue_release = mrvl_rx_queue_release,
2094 	.tx_queue_setup = mrvl_tx_queue_setup,
2095 	.tx_queue_release = mrvl_tx_queue_release,
2096 	.flow_ctrl_get = mrvl_flow_ctrl_get,
2097 	.flow_ctrl_set = mrvl_flow_ctrl_set,
2098 	.rss_hash_update = mrvl_rss_hash_update,
2099 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
2100 	.filter_ctrl = mrvl_eth_filter_ctrl,
2101 	.mtr_ops_get = mrvl_mtr_ops_get,
2102 	.tm_ops_get = mrvl_tm_ops_get,
2103 };
2104 
2105 /**
2106  * Return packet type information and l3/l4 offsets.
2107  *
2108  * @param desc
2109  *   Pointer to the received packet descriptor.
2110  * @param l3_offset
2111  *   l3 packet offset.
2112  * @param l4_offset
2113  *   l4 packet offset.
2114  *
2115  * @return
2116  *   Packet type information.
2117  */
2118 static inline uint64_t
2119 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
2120 				    uint8_t *l3_offset, uint8_t *l4_offset)
2121 {
2122 	enum pp2_inq_l3_type l3_type;
2123 	enum pp2_inq_l4_type l4_type;
2124 	enum pp2_inq_vlan_tag vlan_tag;
2125 	uint64_t packet_type;
2126 
2127 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
2128 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
2129 	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
2130 
2131 	packet_type = RTE_PTYPE_L2_ETHER;
2132 
2133 	switch (vlan_tag) {
2134 	case PP2_INQ_VLAN_TAG_SINGLE:
2135 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
2136 		break;
2137 	case PP2_INQ_VLAN_TAG_DOUBLE:
2138 	case PP2_INQ_VLAN_TAG_TRIPLE:
2139 		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
2140 		break;
2141 	default:
2142 		break;
2143 	}
2144 
2145 	switch (l3_type) {
2146 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
2147 		packet_type |= RTE_PTYPE_L3_IPV4;
2148 		break;
2149 	case PP2_INQ_L3_TYPE_IPV4_OK:
2150 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
2151 		break;
2152 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
2153 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
2154 		break;
2155 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
2156 		packet_type |= RTE_PTYPE_L3_IPV6;
2157 		break;
2158 	case PP2_INQ_L3_TYPE_IPV6_EXT:
2159 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
2160 		break;
2161 	case PP2_INQ_L3_TYPE_ARP:
2162 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
2163 		/*
2164 		 * In case of ARP l4_offset is set to wrong value.
2165 		 * Set it to proper one so that later on mbuf->l3_len can be
2166 		 * calculated subtracting l4_offset and l3_offset.
2167 		 */
2168 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
2169 		break;
2170 	default:
2171 		break;
2172 	}
2173 
2174 	switch (l4_type) {
2175 	case PP2_INQ_L4_TYPE_TCP:
2176 		packet_type |= RTE_PTYPE_L4_TCP;
2177 		break;
2178 	case PP2_INQ_L4_TYPE_UDP:
2179 		packet_type |= RTE_PTYPE_L4_UDP;
2180 		break;
2181 	default:
2182 		break;
2183 	}
2184 
2185 	return packet_type;
2186 }
2187 
2188 /**
2189  * Get offload information from the received packet descriptor.
2190  *
2191  * @param desc
2192  *   Pointer to the received packet descriptor.
2193  *
2194  * @return
2195  *   Mbuf offload flags.
2196  */
2197 static inline uint64_t
2198 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc)
2199 {
2200 	uint64_t flags;
2201 	enum pp2_inq_desc_status status;
2202 
2203 	status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2204 	if (unlikely(status != PP2_DESC_ERR_OK))
2205 		flags = PKT_RX_IP_CKSUM_BAD;
2206 	else
2207 		flags = PKT_RX_IP_CKSUM_GOOD;
2208 
2209 	status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2210 	if (unlikely(status != PP2_DESC_ERR_OK))
2211 		flags |= PKT_RX_L4_CKSUM_BAD;
2212 	else
2213 		flags |= PKT_RX_L4_CKSUM_GOOD;
2214 
2215 	return flags;
2216 }
2217 
2218 /**
2219  * DPDK callback for receive.
2220  *
2221  * @param rxq
2222  *   Generic pointer to the receive queue.
2223  * @param rx_pkts
2224  *   Array to store received packets.
2225  * @param nb_pkts
2226  *   Maximum number of packets in array.
2227  *
2228  * @return
2229  *   Number of packets successfully received.
2230  */
2231 static uint16_t
2232 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2233 {
2234 	struct mrvl_rxq *q = rxq;
2235 	struct pp2_ppio_desc descs[nb_pkts];
2236 	struct pp2_bpool *bpool;
2237 	int i, ret, rx_done = 0;
2238 	int num;
2239 	struct pp2_hif *hif;
2240 	unsigned int core_id = rte_lcore_id();
2241 
2242 	hif = mrvl_get_hif(q->priv, core_id);
2243 
2244 	if (unlikely(!q->priv->ppio || !hif))
2245 		return 0;
2246 
2247 	bpool = q->priv->bpool;
2248 
2249 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2250 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2251 	if (unlikely(ret < 0))
2252 		return 0;
2253 
2254 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2255 
2256 	for (i = 0; i < nb_pkts; i++) {
2257 		struct rte_mbuf *mbuf;
2258 		uint8_t l3_offset, l4_offset;
2259 		enum pp2_inq_desc_status status;
2260 		uint64_t addr;
2261 
2262 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2263 			struct pp2_ppio_desc *pref_desc;
2264 			u64 pref_addr;
2265 
2266 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2267 			pref_addr = cookie_addr_high |
2268 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2269 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2270 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2271 		}
2272 
2273 		addr = cookie_addr_high |
2274 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2275 		mbuf = (struct rte_mbuf *)addr;
2276 		rte_pktmbuf_reset(mbuf);
2277 
2278 		/* drop packet in case of mac, overrun or resource error */
2279 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2280 		if (unlikely(status != PP2_DESC_ERR_OK)) {
2281 			struct pp2_buff_inf binf = {
2282 				.addr = rte_mbuf_data_iova_default(mbuf),
2283 				.cookie = (uint64_t)mbuf,
2284 			};
2285 
2286 			pp2_bpool_put_buff(hif, bpool, &binf);
2287 			mrvl_port_bpool_size
2288 				[bpool->pp2_id][bpool->id][core_id]++;
2289 			q->drop_mac++;
2290 			continue;
2291 		}
2292 
2293 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2294 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2295 		mbuf->data_len = mbuf->pkt_len;
2296 		mbuf->port = q->port_id;
2297 		mbuf->packet_type =
2298 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2299 							    &l3_offset,
2300 							    &l4_offset);
2301 		mbuf->l2_len = l3_offset;
2302 		mbuf->l3_len = l4_offset - l3_offset;
2303 
2304 		if (likely(q->cksum_enabled))
2305 			mbuf->ol_flags = mrvl_desc_to_ol_flags(&descs[i]);
2306 
2307 		rx_pkts[rx_done++] = mbuf;
2308 		q->bytes_recv += mbuf->pkt_len;
2309 	}
2310 
2311 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2312 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2313 
2314 		if (unlikely(num <= q->priv->bpool_min_size ||
2315 			     (!rx_done && num < q->priv->bpool_init_size))) {
2316 			mrvl_fill_bpool(q, MRVL_BURST_SIZE);
2317 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2318 			int i;
2319 			int pkt_to_remove = num - q->priv->bpool_init_size;
2320 			struct rte_mbuf *mbuf;
2321 			struct pp2_buff_inf buff;
2322 
2323 			for (i = 0; i < pkt_to_remove; i++) {
2324 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2325 				if (ret)
2326 					break;
2327 				mbuf = (struct rte_mbuf *)
2328 					(cookie_addr_high | buff.cookie);
2329 				rte_pktmbuf_free(mbuf);
2330 			}
2331 			mrvl_port_bpool_size
2332 				[bpool->pp2_id][bpool->id][core_id] -= i;
2333 		}
2334 		rte_spinlock_unlock(&q->priv->lock);
2335 	}
2336 
2337 	return rx_done;
2338 }
2339 
2340 /**
2341  * Prepare offload information.
2342  *
2343  * @param ol_flags
2344  *   Offload flags.
2345  * @param packet_type
2346  *   Packet type bitfield.
2347  * @param l3_type
2348  *   Pointer to the pp2_ouq_l3_type structure.
2349  * @param l4_type
2350  *   Pointer to the pp2_outq_l4_type structure.
2351  * @param gen_l3_cksum
2352  *   Will be set to 1 in case l3 checksum is computed.
2353  * @param l4_cksum
2354  *   Will be set to 1 in case l4 checksum is computed.
2355  *
2356  * @return
2357  *   0 on success, negative error value otherwise.
2358  */
2359 static inline int
2360 mrvl_prepare_proto_info(uint64_t ol_flags, uint32_t packet_type,
2361 			enum pp2_outq_l3_type *l3_type,
2362 			enum pp2_outq_l4_type *l4_type,
2363 			int *gen_l3_cksum,
2364 			int *gen_l4_cksum)
2365 {
2366 	/*
2367 	 * Based on ol_flags prepare information
2368 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2369 	 * for offloading.
2370 	 */
2371 	if (ol_flags & PKT_TX_IPV4) {
2372 		*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2373 		*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2374 	} else if (ol_flags & PKT_TX_IPV6) {
2375 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2376 		/* no checksum for ipv6 header */
2377 		*gen_l3_cksum = 0;
2378 	} else {
2379 		/* if something different then stop processing */
2380 		return -1;
2381 	}
2382 
2383 	ol_flags &= PKT_TX_L4_MASK;
2384 	if ((packet_type & RTE_PTYPE_L4_TCP) &&
2385 	    ol_flags == PKT_TX_TCP_CKSUM) {
2386 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2387 		*gen_l4_cksum = 1;
2388 	} else if ((packet_type & RTE_PTYPE_L4_UDP) &&
2389 		   ol_flags == PKT_TX_UDP_CKSUM) {
2390 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2391 		*gen_l4_cksum = 1;
2392 	} else {
2393 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2394 		/* no checksum for other type */
2395 		*gen_l4_cksum = 0;
2396 	}
2397 
2398 	return 0;
2399 }
2400 
2401 /**
2402  * Release already sent buffers to bpool (buffer-pool).
2403  *
2404  * @param ppio
2405  *   Pointer to the port structure.
2406  * @param hif
2407  *   Pointer to the MUSDK hardware interface.
2408  * @param sq
2409  *   Pointer to the shadow queue.
2410  * @param qid
2411  *   Queue id number.
2412  * @param force
2413  *   Force releasing packets.
2414  */
2415 static inline void
2416 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2417 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2418 		       int qid, int force)
2419 {
2420 	struct buff_release_entry *entry;
2421 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2422 	int i;
2423 
2424 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2425 
2426 	sq->num_to_release += nb_done;
2427 
2428 	if (likely(!force &&
2429 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2430 		return;
2431 
2432 	nb_done = sq->num_to_release;
2433 	sq->num_to_release = 0;
2434 
2435 	for (i = 0; i < nb_done; i++) {
2436 		entry = &sq->ent[sq->tail + num];
2437 		if (unlikely(!entry->buff.addr)) {
2438 			MRVL_LOG(ERR,
2439 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2440 				sq->tail, (u64)entry->buff.cookie,
2441 				(u64)entry->buff.addr);
2442 			skip_bufs = 1;
2443 			goto skip;
2444 		}
2445 
2446 		if (unlikely(!entry->bpool)) {
2447 			struct rte_mbuf *mbuf;
2448 
2449 			mbuf = (struct rte_mbuf *)entry->buff.cookie;
2450 			rte_pktmbuf_free(mbuf);
2451 			skip_bufs = 1;
2452 			goto skip;
2453 		}
2454 
2455 		mrvl_port_bpool_size
2456 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2457 		num++;
2458 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2459 			goto skip;
2460 		continue;
2461 skip:
2462 		if (likely(num))
2463 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2464 		num += skip_bufs;
2465 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2466 		sq->size -= num;
2467 		num = 0;
2468 		skip_bufs = 0;
2469 	}
2470 
2471 	if (likely(num)) {
2472 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2473 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2474 		sq->size -= num;
2475 	}
2476 }
2477 
2478 /**
2479  * DPDK callback for transmit.
2480  *
2481  * @param txq
2482  *   Generic pointer transmit queue.
2483  * @param tx_pkts
2484  *   Packets to transmit.
2485  * @param nb_pkts
2486  *   Number of packets in array.
2487  *
2488  * @return
2489  *   Number of packets successfully transmitted.
2490  */
2491 static uint16_t
2492 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2493 {
2494 	struct mrvl_txq *q = txq;
2495 	struct mrvl_shadow_txq *sq;
2496 	struct pp2_hif *hif;
2497 	struct pp2_ppio_desc descs[nb_pkts];
2498 	unsigned int core_id = rte_lcore_id();
2499 	int i, ret, bytes_sent = 0;
2500 	uint16_t num, sq_free_size;
2501 	uint64_t addr;
2502 
2503 	hif = mrvl_get_hif(q->priv, core_id);
2504 	sq = &q->shadow_txqs[core_id];
2505 
2506 	if (unlikely(!q->priv->ppio || !hif))
2507 		return 0;
2508 
2509 	if (sq->size)
2510 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2511 				       sq, q->queue_id, 0);
2512 
2513 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2514 	if (unlikely(nb_pkts > sq_free_size))
2515 		nb_pkts = sq_free_size;
2516 
2517 	for (i = 0; i < nb_pkts; i++) {
2518 		struct rte_mbuf *mbuf = tx_pkts[i];
2519 		int gen_l3_cksum, gen_l4_cksum;
2520 		enum pp2_outq_l3_type l3_type;
2521 		enum pp2_outq_l4_type l4_type;
2522 
2523 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2524 			struct rte_mbuf *pref_pkt_hdr;
2525 
2526 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2527 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2528 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2529 		}
2530 
2531 		mrvl_fill_shadowq(sq, mbuf);
2532 		mrvl_fill_desc(&descs[i], mbuf);
2533 
2534 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2535 		/*
2536 		 * in case unsupported ol_flags were passed
2537 		 * do not update descriptor offload information
2538 		 */
2539 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2540 					      &l3_type, &l4_type, &gen_l3_cksum,
2541 					      &gen_l4_cksum);
2542 		if (unlikely(ret))
2543 			continue;
2544 
2545 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2546 						  mbuf->l2_len,
2547 						  mbuf->l2_len + mbuf->l3_len,
2548 						  gen_l3_cksum, gen_l4_cksum);
2549 	}
2550 
2551 	num = nb_pkts;
2552 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2553 	/* number of packets that were not sent */
2554 	if (unlikely(num > nb_pkts)) {
2555 		for (i = nb_pkts; i < num; i++) {
2556 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2557 				MRVL_PP2_TX_SHADOWQ_MASK;
2558 			addr = sq->ent[sq->head].buff.cookie;
2559 			bytes_sent -=
2560 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2561 		}
2562 		sq->size -= num - nb_pkts;
2563 	}
2564 
2565 	q->bytes_sent += bytes_sent;
2566 
2567 	return nb_pkts;
2568 }
2569 
2570 /** DPDK callback for S/G transmit.
2571  *
2572  * @param txq
2573  *   Generic pointer transmit queue.
2574  * @param tx_pkts
2575  *   Packets to transmit.
2576  * @param nb_pkts
2577  *   Number of packets in array.
2578  *
2579  * @return
2580  *   Number of packets successfully transmitted.
2581  */
2582 static uint16_t
2583 mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
2584 		     uint16_t nb_pkts)
2585 {
2586 	struct mrvl_txq *q = txq;
2587 	struct mrvl_shadow_txq *sq;
2588 	struct pp2_hif *hif;
2589 	struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
2590 	struct pp2_ppio_sg_pkts pkts;
2591 	uint8_t frags[nb_pkts];
2592 	unsigned int core_id = rte_lcore_id();
2593 	int i, j, ret, bytes_sent = 0;
2594 	int tail, tail_first;
2595 	uint16_t num, sq_free_size;
2596 	uint16_t nb_segs, total_descs = 0;
2597 	uint64_t addr;
2598 
2599 	hif = mrvl_get_hif(q->priv, core_id);
2600 	sq = &q->shadow_txqs[core_id];
2601 	pkts.frags = frags;
2602 	pkts.num = 0;
2603 
2604 	if (unlikely(!q->priv->ppio || !hif))
2605 		return 0;
2606 
2607 	if (sq->size)
2608 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2609 				       sq, q->queue_id, 0);
2610 
2611 	/* Save shadow queue free size */
2612 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2613 
2614 	tail = 0;
2615 	for (i = 0; i < nb_pkts; i++) {
2616 		struct rte_mbuf *mbuf = tx_pkts[i];
2617 		struct rte_mbuf *seg = NULL;
2618 		int gen_l3_cksum, gen_l4_cksum;
2619 		enum pp2_outq_l3_type l3_type;
2620 		enum pp2_outq_l4_type l4_type;
2621 
2622 		nb_segs = mbuf->nb_segs;
2623 		tail_first = tail;
2624 		total_descs += nb_segs;
2625 
2626 		/*
2627 		 * Check if total_descs does not exceed
2628 		 * shadow queue free size
2629 		 */
2630 		if (unlikely(total_descs > sq_free_size)) {
2631 			total_descs -= nb_segs;
2632 			break;
2633 		}
2634 
2635 		/* Check if nb_segs does not exceed the max nb of desc per
2636 		 * fragmented packet
2637 		 */
2638 		if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
2639 			total_descs -= nb_segs;
2640 			RTE_LOG(ERR, PMD,
2641 				"Too many segments. Packet won't be sent.\n");
2642 			break;
2643 		}
2644 
2645 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2646 			struct rte_mbuf *pref_pkt_hdr;
2647 
2648 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2649 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2650 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2651 		}
2652 
2653 		pkts.frags[pkts.num] = nb_segs;
2654 		pkts.num++;
2655 
2656 		seg = mbuf;
2657 		for (j = 0; j < nb_segs - 1; j++) {
2658 			/* For the subsequent segments, set shadow queue
2659 			 * buffer to NULL
2660 			 */
2661 			mrvl_fill_shadowq(sq, NULL);
2662 			mrvl_fill_desc(&descs[tail], seg);
2663 
2664 			tail++;
2665 			seg = seg->next;
2666 		}
2667 		/* Put first mbuf info in last shadow queue entry */
2668 		mrvl_fill_shadowq(sq, mbuf);
2669 		/* Update descriptor with last segment */
2670 		mrvl_fill_desc(&descs[tail++], seg);
2671 
2672 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2673 		/* In case unsupported ol_flags were passed
2674 		 * do not update descriptor offload information
2675 		 */
2676 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2677 					      &l3_type, &l4_type, &gen_l3_cksum,
2678 					      &gen_l4_cksum);
2679 		if (unlikely(ret))
2680 			continue;
2681 
2682 		pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
2683 						  l4_type, mbuf->l2_len,
2684 						  mbuf->l2_len + mbuf->l3_len,
2685 						  gen_l3_cksum, gen_l4_cksum);
2686 	}
2687 
2688 	num = total_descs;
2689 	pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
2690 			 &total_descs, &pkts);
2691 	/* number of packets that were not sent */
2692 	if (unlikely(num > total_descs)) {
2693 		for (i = total_descs; i < num; i++) {
2694 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2695 				MRVL_PP2_TX_SHADOWQ_MASK;
2696 
2697 			addr = sq->ent[sq->head].buff.cookie;
2698 			if (addr)
2699 				bytes_sent -=
2700 					rte_pktmbuf_pkt_len((struct rte_mbuf *)
2701 						(cookie_addr_high | addr));
2702 		}
2703 		sq->size -= num - total_descs;
2704 		nb_pkts = pkts.num;
2705 	}
2706 
2707 	q->bytes_sent += bytes_sent;
2708 
2709 	return nb_pkts;
2710 }
2711 
2712 /**
2713  * Initialize packet processor.
2714  *
2715  * @return
2716  *   0 on success, negative error value otherwise.
2717  */
2718 static int
2719 mrvl_init_pp2(void)
2720 {
2721 	struct pp2_init_params init_params;
2722 
2723 	memset(&init_params, 0, sizeof(init_params));
2724 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
2725 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
2726 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
2727 
2728 	return pp2_init(&init_params);
2729 }
2730 
2731 /**
2732  * Deinitialize packet processor.
2733  *
2734  * @return
2735  *   0 on success, negative error value otherwise.
2736  */
2737 static void
2738 mrvl_deinit_pp2(void)
2739 {
2740 	pp2_deinit();
2741 }
2742 
2743 /**
2744  * Create private device structure.
2745  *
2746  * @param dev_name
2747  *   Pointer to the port name passed in the initialization parameters.
2748  *
2749  * @return
2750  *   Pointer to the newly allocated private device structure.
2751  */
2752 static struct mrvl_priv *
2753 mrvl_priv_create(const char *dev_name)
2754 {
2755 	struct pp2_bpool_params bpool_params;
2756 	char match[MRVL_MATCH_LEN];
2757 	struct mrvl_priv *priv;
2758 	int ret, bpool_bit;
2759 
2760 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
2761 	if (!priv)
2762 		return NULL;
2763 
2764 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
2765 				       &priv->pp_id, &priv->ppio_id);
2766 	if (ret)
2767 		goto out_free_priv;
2768 
2769 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
2770 				     PP2_BPOOL_NUM_POOLS);
2771 	if (bpool_bit < 0)
2772 		goto out_free_priv;
2773 	priv->bpool_bit = bpool_bit;
2774 
2775 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
2776 		 priv->bpool_bit);
2777 	memset(&bpool_params, 0, sizeof(bpool_params));
2778 	bpool_params.match = match;
2779 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
2780 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
2781 	if (ret)
2782 		goto out_clear_bpool_bit;
2783 
2784 	priv->ppio_params.type = PP2_PPIO_T_NIC;
2785 	rte_spinlock_init(&priv->lock);
2786 
2787 	return priv;
2788 out_clear_bpool_bit:
2789 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2790 out_free_priv:
2791 	rte_free(priv);
2792 	return NULL;
2793 }
2794 
2795 /**
2796  * Create device representing Ethernet port.
2797  *
2798  * @param name
2799  *   Pointer to the port's name.
2800  *
2801  * @return
2802  *   0 on success, negative error value otherwise.
2803  */
2804 static int
2805 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
2806 {
2807 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
2808 	struct rte_eth_dev *eth_dev;
2809 	struct mrvl_priv *priv;
2810 	struct ifreq req;
2811 
2812 	eth_dev = rte_eth_dev_allocate(name);
2813 	if (!eth_dev)
2814 		return -ENOMEM;
2815 
2816 	priv = mrvl_priv_create(name);
2817 	if (!priv) {
2818 		ret = -ENOMEM;
2819 		goto out_free;
2820 	}
2821 	eth_dev->data->dev_private = priv;
2822 
2823 	eth_dev->data->mac_addrs =
2824 		rte_zmalloc("mac_addrs",
2825 			    RTE_ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
2826 	if (!eth_dev->data->mac_addrs) {
2827 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
2828 		ret = -ENOMEM;
2829 		goto out_free;
2830 	}
2831 
2832 	memset(&req, 0, sizeof(req));
2833 	strcpy(req.ifr_name, name);
2834 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
2835 	if (ret)
2836 		goto out_free;
2837 
2838 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
2839 	       req.ifr_addr.sa_data, RTE_ETHER_ADDR_LEN);
2840 
2841 	eth_dev->device = &vdev->device;
2842 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
2843 	mrvl_set_tx_function(eth_dev);
2844 	eth_dev->dev_ops = &mrvl_ops;
2845 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2846 
2847 	rte_eth_dev_probing_finish(eth_dev);
2848 	return 0;
2849 out_free:
2850 	rte_eth_dev_release_port(eth_dev);
2851 
2852 	return ret;
2853 }
2854 
2855 /**
2856  * Callback used by rte_kvargs_process() during argument parsing.
2857  *
2858  * @param key
2859  *   Pointer to the parsed key (unused).
2860  * @param value
2861  *   Pointer to the parsed value.
2862  * @param extra_args
2863  *   Pointer to the extra arguments which contains address of the
2864  *   table of pointers to parsed interface names.
2865  *
2866  * @return
2867  *   Always 0.
2868  */
2869 static int
2870 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
2871 		 void *extra_args)
2872 {
2873 	struct mrvl_ifnames *ifnames = extra_args;
2874 
2875 	ifnames->names[ifnames->idx++] = value;
2876 
2877 	return 0;
2878 }
2879 
2880 /**
2881  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
2882  */
2883 static void
2884 mrvl_deinit_hifs(void)
2885 {
2886 	int i;
2887 
2888 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
2889 		if (hifs[i])
2890 			pp2_hif_deinit(hifs[i]);
2891 	}
2892 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
2893 	memset(hifs, 0, sizeof(hifs));
2894 }
2895 
2896 /**
2897  * DPDK callback to register the virtual device.
2898  *
2899  * @param vdev
2900  *   Pointer to the virtual device.
2901  *
2902  * @return
2903  *   0 on success, negative error value otherwise.
2904  */
2905 static int
2906 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
2907 {
2908 	struct rte_kvargs *kvlist;
2909 	struct mrvl_ifnames ifnames;
2910 	int ret = -EINVAL;
2911 	uint32_t i, ifnum, cfgnum;
2912 	const char *params;
2913 
2914 	params = rte_vdev_device_args(vdev);
2915 	if (!params)
2916 		return -EINVAL;
2917 
2918 	kvlist = rte_kvargs_parse(params, valid_args);
2919 	if (!kvlist)
2920 		return -EINVAL;
2921 
2922 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
2923 	if (ifnum > RTE_DIM(ifnames.names))
2924 		goto out_free_kvlist;
2925 
2926 	ifnames.idx = 0;
2927 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
2928 			   mrvl_get_ifnames, &ifnames);
2929 
2930 
2931 	/*
2932 	 * The below system initialization should be done only once,
2933 	 * on the first provided configuration file
2934 	 */
2935 	if (!mrvl_qos_cfg) {
2936 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
2937 		MRVL_LOG(INFO, "Parsing config file!");
2938 		if (cfgnum > 1) {
2939 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
2940 			goto out_free_kvlist;
2941 		} else if (cfgnum == 1) {
2942 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
2943 					   mrvl_get_qoscfg, &mrvl_qos_cfg);
2944 		}
2945 	}
2946 
2947 	if (mrvl_dev_num)
2948 		goto init_devices;
2949 
2950 	MRVL_LOG(INFO, "Perform MUSDK initializations");
2951 
2952 	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
2953 	if (ret)
2954 		goto out_free_kvlist;
2955 
2956 	ret = mrvl_init_pp2();
2957 	if (ret) {
2958 		MRVL_LOG(ERR, "Failed to init PP!");
2959 		rte_mvep_deinit(MVEP_MOD_T_PP2);
2960 		goto out_free_kvlist;
2961 	}
2962 
2963 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
2964 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
2965 
2966 	mrvl_lcore_first = RTE_MAX_LCORE;
2967 	mrvl_lcore_last = 0;
2968 
2969 init_devices:
2970 	for (i = 0; i < ifnum; i++) {
2971 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
2972 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
2973 		if (ret)
2974 			goto out_cleanup;
2975 		mrvl_dev_num++;
2976 	}
2977 
2978 	rte_kvargs_free(kvlist);
2979 
2980 	return 0;
2981 out_cleanup:
2982 	rte_pmd_mrvl_remove(vdev);
2983 
2984 out_free_kvlist:
2985 	rte_kvargs_free(kvlist);
2986 
2987 	return ret;
2988 }
2989 
2990 /**
2991  * DPDK callback to remove virtual device.
2992  *
2993  * @param vdev
2994  *   Pointer to the removed virtual device.
2995  *
2996  * @return
2997  *   0 on success, negative error value otherwise.
2998  */
2999 static int
3000 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
3001 {
3002 	uint16_t port_id;
3003 	int ret = 0;
3004 
3005 	RTE_ETH_FOREACH_DEV(port_id) {
3006 		if (rte_eth_devices[port_id].device != &vdev->device)
3007 			continue;
3008 		ret |= rte_eth_dev_close(port_id);
3009 	}
3010 
3011 	return ret == 0 ? 0 : -EIO;
3012 }
3013 
3014 static struct rte_vdev_driver pmd_mrvl_drv = {
3015 	.probe = rte_pmd_mrvl_probe,
3016 	.remove = rte_pmd_mrvl_remove,
3017 };
3018 
3019 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
3020 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
3021 RTE_LOG_REGISTER(mrvl_logtype, pmd.net.mvpp2, NOTICE);
3022