xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision 3b8bcfcd96e64d199392550928be7c7665571bcb)
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_ethdev_driver.h>
8 #include <rte_kvargs.h>
9 #include <rte_log.h>
10 #include <rte_malloc.h>
11 #include <rte_bus_vdev.h>
12 
13 /* Unluckily, container_of is defined by both DPDK and MUSDK,
14  * we'll declare only one version.
15  *
16  * Note that it is not used in this PMD anyway.
17  */
18 #ifdef container_of
19 #undef container_of
20 #endif
21 
22 #include <fcntl.h>
23 #include <linux/ethtool.h>
24 #include <linux/sockios.h>
25 #include <net/if.h>
26 #include <net/if_arp.h>
27 #include <sys/ioctl.h>
28 #include <sys/socket.h>
29 #include <sys/stat.h>
30 #include <sys/types.h>
31 
32 #include "mrvl_ethdev.h"
33 #include "mrvl_qos.h"
34 
35 /* bitmask with reserved hifs */
36 #define MRVL_MUSDK_HIFS_RESERVED 0x0F
37 /* bitmask with reserved bpools */
38 #define MRVL_MUSDK_BPOOLS_RESERVED 0x07
39 /* bitmask with reserved kernel RSS tables */
40 #define MRVL_MUSDK_RSS_RESERVED 0x01
41 /* maximum number of available hifs */
42 #define MRVL_MUSDK_HIFS_MAX 9
43 
44 /* prefetch shift */
45 #define MRVL_MUSDK_PREFETCH_SHIFT 2
46 
47 /* TCAM has 25 entries reserved for uc/mc filter entries */
48 #define MRVL_MAC_ADDRS_MAX 25
49 #define MRVL_MATCH_LEN 16
50 #define MRVL_PKT_EFFEC_OFFS (MRVL_PKT_OFFS + MV_MH_SIZE)
51 /* Maximum allowable packet size */
52 #define MRVL_PKT_SIZE_MAX (10240 - MV_MH_SIZE)
53 
54 #define MRVL_IFACE_NAME_ARG "iface"
55 #define MRVL_CFG_ARG "cfg"
56 
57 #define MRVL_BURST_SIZE 64
58 
59 #define MRVL_ARP_LENGTH 28
60 
61 #define MRVL_COOKIE_ADDR_INVALID ~0ULL
62 
63 #define MRVL_COOKIE_HIGH_ADDR_SHIFT	(sizeof(pp2_cookie_t) * 8)
64 #define MRVL_COOKIE_HIGH_ADDR_MASK	(~0ULL << MRVL_COOKIE_HIGH_ADDR_SHIFT)
65 
66 /* Memory size (in bytes) for MUSDK dma buffers */
67 #define MRVL_MUSDK_DMA_MEMSIZE 41943040
68 
69 /** Port Rx offload capabilities */
70 #define MRVL_RX_OFFLOADS (DEV_RX_OFFLOAD_VLAN_FILTER | \
71 			  DEV_RX_OFFLOAD_JUMBO_FRAME | \
72 			  DEV_RX_OFFLOAD_CRC_STRIP | \
73 			  DEV_RX_OFFLOAD_CHECKSUM)
74 
75 /** Port Tx offloads capabilities */
76 #define MRVL_TX_OFFLOADS (DEV_TX_OFFLOAD_IPV4_CKSUM | \
77 			  DEV_TX_OFFLOAD_UDP_CKSUM | \
78 			  DEV_TX_OFFLOAD_TCP_CKSUM)
79 
80 static const char * const valid_args[] = {
81 	MRVL_IFACE_NAME_ARG,
82 	MRVL_CFG_ARG,
83 	NULL
84 };
85 
86 static int used_hifs = MRVL_MUSDK_HIFS_RESERVED;
87 static struct pp2_hif *hifs[RTE_MAX_LCORE];
88 static int used_bpools[PP2_NUM_PKT_PROC] = {
89 	MRVL_MUSDK_BPOOLS_RESERVED,
90 	MRVL_MUSDK_BPOOLS_RESERVED
91 };
92 
93 struct pp2_bpool *mrvl_port_to_bpool_lookup[RTE_MAX_ETHPORTS];
94 int mrvl_port_bpool_size[PP2_NUM_PKT_PROC][PP2_BPOOL_NUM_POOLS][RTE_MAX_LCORE];
95 uint64_t cookie_addr_high = MRVL_COOKIE_ADDR_INVALID;
96 
97 int mrvl_logtype;
98 
99 struct mrvl_ifnames {
100 	const char *names[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
101 	int idx;
102 };
103 
104 /*
105  * To use buffer harvesting based on loopback port shadow queue structure
106  * was introduced for buffers information bookkeeping.
107  *
108  * Before sending the packet, related buffer information (pp2_buff_inf) is
109  * stored in shadow queue. After packet is transmitted no longer used
110  * packet buffer is released back to it's original hardware pool,
111  * on condition it originated from interface.
112  * In case it  was generated by application itself i.e: mbuf->port field is
113  * 0xff then its released to software mempool.
114  */
115 struct mrvl_shadow_txq {
116 	int head;           /* write index - used when sending buffers */
117 	int tail;           /* read index - used when releasing buffers */
118 	u16 size;           /* queue occupied size */
119 	u16 num_to_release; /* number of buffers sent, that can be released */
120 	struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
121 };
122 
123 struct mrvl_rxq {
124 	struct mrvl_priv *priv;
125 	struct rte_mempool *mp;
126 	int queue_id;
127 	int port_id;
128 	int cksum_enabled;
129 	uint64_t bytes_recv;
130 	uint64_t drop_mac;
131 };
132 
133 struct mrvl_txq {
134 	struct mrvl_priv *priv;
135 	int queue_id;
136 	int port_id;
137 	uint64_t bytes_sent;
138 	struct mrvl_shadow_txq shadow_txqs[RTE_MAX_LCORE];
139 	int tx_deferred_start;
140 };
141 
142 static int mrvl_lcore_first;
143 static int mrvl_lcore_last;
144 static int mrvl_dev_num;
145 
146 static int mrvl_fill_bpool(struct mrvl_rxq *rxq, int num);
147 static inline void mrvl_free_sent_buffers(struct pp2_ppio *ppio,
148 			struct pp2_hif *hif, unsigned int core_id,
149 			struct mrvl_shadow_txq *sq, int qid, int force);
150 
151 #define MRVL_XSTATS_TBL_ENTRY(name) { \
152 	#name, offsetof(struct pp2_ppio_statistics, name),	\
153 	sizeof(((struct pp2_ppio_statistics *)0)->name)		\
154 }
155 
156 /* Table with xstats data */
157 static struct {
158 	const char *name;
159 	unsigned int offset;
160 	unsigned int size;
161 } mrvl_xstats_tbl[] = {
162 	MRVL_XSTATS_TBL_ENTRY(rx_bytes),
163 	MRVL_XSTATS_TBL_ENTRY(rx_packets),
164 	MRVL_XSTATS_TBL_ENTRY(rx_unicast_packets),
165 	MRVL_XSTATS_TBL_ENTRY(rx_errors),
166 	MRVL_XSTATS_TBL_ENTRY(rx_fullq_dropped),
167 	MRVL_XSTATS_TBL_ENTRY(rx_bm_dropped),
168 	MRVL_XSTATS_TBL_ENTRY(rx_early_dropped),
169 	MRVL_XSTATS_TBL_ENTRY(rx_fifo_dropped),
170 	MRVL_XSTATS_TBL_ENTRY(rx_cls_dropped),
171 	MRVL_XSTATS_TBL_ENTRY(tx_bytes),
172 	MRVL_XSTATS_TBL_ENTRY(tx_packets),
173 	MRVL_XSTATS_TBL_ENTRY(tx_unicast_packets),
174 	MRVL_XSTATS_TBL_ENTRY(tx_errors)
175 };
176 
177 static inline int
178 mrvl_get_bpool_size(int pp2_id, int pool_id)
179 {
180 	int i;
181 	int size = 0;
182 
183 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
184 		size += mrvl_port_bpool_size[pp2_id][pool_id][i];
185 
186 	return size;
187 }
188 
189 static inline int
190 mrvl_reserve_bit(int *bitmap, int max)
191 {
192 	int n = sizeof(*bitmap) * 8 - __builtin_clz(*bitmap);
193 
194 	if (n >= max)
195 		return -1;
196 
197 	*bitmap |= 1 << n;
198 
199 	return n;
200 }
201 
202 static int
203 mrvl_init_hif(int core_id)
204 {
205 	struct pp2_hif_params params;
206 	char match[MRVL_MATCH_LEN];
207 	int ret;
208 
209 	ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
210 	if (ret < 0) {
211 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
212 		return ret;
213 	}
214 
215 	snprintf(match, sizeof(match), "hif-%d", ret);
216 	memset(&params, 0, sizeof(params));
217 	params.match = match;
218 	params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
219 	ret = pp2_hif_init(&params, &hifs[core_id]);
220 	if (ret) {
221 		MRVL_LOG(ERR, "Failed to initialize hif %d", core_id);
222 		return ret;
223 	}
224 
225 	return 0;
226 }
227 
228 static inline struct pp2_hif*
229 mrvl_get_hif(struct mrvl_priv *priv, int core_id)
230 {
231 	int ret;
232 
233 	if (likely(hifs[core_id] != NULL))
234 		return hifs[core_id];
235 
236 	rte_spinlock_lock(&priv->lock);
237 
238 	ret = mrvl_init_hif(core_id);
239 	if (ret < 0) {
240 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
241 		goto out;
242 	}
243 
244 	if (core_id < mrvl_lcore_first)
245 		mrvl_lcore_first = core_id;
246 
247 	if (core_id > mrvl_lcore_last)
248 		mrvl_lcore_last = core_id;
249 out:
250 	rte_spinlock_unlock(&priv->lock);
251 
252 	return hifs[core_id];
253 }
254 
255 /**
256  * Configure rss based on dpdk rss configuration.
257  *
258  * @param priv
259  *   Pointer to private structure.
260  * @param rss_conf
261  *   Pointer to RSS configuration.
262  *
263  * @return
264  *   0 on success, negative error value otherwise.
265  */
266 static int
267 mrvl_configure_rss(struct mrvl_priv *priv, struct rte_eth_rss_conf *rss_conf)
268 {
269 	if (rss_conf->rss_key)
270 		MRVL_LOG(WARNING, "Changing hash key is not supported");
271 
272 	if (rss_conf->rss_hf == 0) {
273 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
274 	} else if (rss_conf->rss_hf & ETH_RSS_IPV4) {
275 		priv->ppio_params.inqs_params.hash_type =
276 			PP2_PPIO_HASH_T_2_TUPLE;
277 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP) {
278 		priv->ppio_params.inqs_params.hash_type =
279 			PP2_PPIO_HASH_T_5_TUPLE;
280 		priv->rss_hf_tcp = 1;
281 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) {
282 		priv->ppio_params.inqs_params.hash_type =
283 			PP2_PPIO_HASH_T_5_TUPLE;
284 		priv->rss_hf_tcp = 0;
285 	} else {
286 		return -EINVAL;
287 	}
288 
289 	return 0;
290 }
291 
292 /**
293  * Ethernet device configuration.
294  *
295  * Prepare the driver for a given number of TX and RX queues and
296  * configure RSS.
297  *
298  * @param dev
299  *   Pointer to Ethernet device structure.
300  *
301  * @return
302  *   0 on success, negative error value otherwise.
303  */
304 static int
305 mrvl_dev_configure(struct rte_eth_dev *dev)
306 {
307 	struct mrvl_priv *priv = dev->data->dev_private;
308 	int ret;
309 
310 	if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_NONE &&
311 	    dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
312 		MRVL_LOG(INFO, "Unsupported rx multi queue mode %d",
313 			dev->data->dev_conf.rxmode.mq_mode);
314 		return -EINVAL;
315 	}
316 
317 	if (!(dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_CRC_STRIP)) {
318 		MRVL_LOG(INFO,
319 			"L2 CRC stripping is always enabled in hw");
320 		dev->data->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_CRC_STRIP;
321 	}
322 
323 	if (dev->data->dev_conf.rxmode.split_hdr_size) {
324 		MRVL_LOG(INFO, "Split headers not supported");
325 		return -EINVAL;
326 	}
327 
328 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
329 		dev->data->mtu = dev->data->dev_conf.rxmode.max_rx_pkt_len -
330 				 ETHER_HDR_LEN - ETHER_CRC_LEN;
331 
332 	ret = mrvl_configure_rxqs(priv, dev->data->port_id,
333 				  dev->data->nb_rx_queues);
334 	if (ret < 0)
335 		return ret;
336 
337 	ret = mrvl_configure_txqs(priv, dev->data->port_id,
338 				  dev->data->nb_tx_queues);
339 	if (ret < 0)
340 		return ret;
341 
342 	priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
343 	priv->ppio_params.maintain_stats = 1;
344 	priv->nb_rx_queues = dev->data->nb_rx_queues;
345 
346 	if (dev->data->nb_rx_queues == 1 &&
347 	    dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
348 		MRVL_LOG(WARNING, "Disabling hash for 1 rx queue");
349 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
350 
351 		return 0;
352 	}
353 
354 	return mrvl_configure_rss(priv,
355 				  &dev->data->dev_conf.rx_adv_conf.rss_conf);
356 }
357 
358 /**
359  * DPDK callback to change the MTU.
360  *
361  * Setting the MTU affects hardware MRU (packets larger than the MRU
362  * will be dropped).
363  *
364  * @param dev
365  *   Pointer to Ethernet device structure.
366  * @param mtu
367  *   New MTU.
368  *
369  * @return
370  *   0 on success, negative error value otherwise.
371  */
372 static int
373 mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
374 {
375 	struct mrvl_priv *priv = dev->data->dev_private;
376 	/* extra MV_MH_SIZE bytes are required for Marvell tag */
377 	uint16_t mru = mtu + MV_MH_SIZE + ETHER_HDR_LEN + ETHER_CRC_LEN;
378 	int ret;
379 
380 	if (mtu < ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX)
381 		return -EINVAL;
382 
383 	if (!priv->ppio)
384 		return 0;
385 
386 	ret = pp2_ppio_set_mru(priv->ppio, mru);
387 	if (ret)
388 		return ret;
389 
390 	return pp2_ppio_set_mtu(priv->ppio, mtu);
391 }
392 
393 /**
394  * DPDK callback to bring the link up.
395  *
396  * @param dev
397  *   Pointer to Ethernet device structure.
398  *
399  * @return
400  *   0 on success, negative error value otherwise.
401  */
402 static int
403 mrvl_dev_set_link_up(struct rte_eth_dev *dev)
404 {
405 	struct mrvl_priv *priv = dev->data->dev_private;
406 	int ret;
407 
408 	if (!priv->ppio)
409 		return -EPERM;
410 
411 	ret = pp2_ppio_enable(priv->ppio);
412 	if (ret)
413 		return ret;
414 
415 	/*
416 	 * mtu/mru can be updated if pp2_ppio_enable() was called at least once
417 	 * as pp2_ppio_enable() changes port->t_mode from default 0 to
418 	 * PP2_TRAFFIC_INGRESS_EGRESS.
419 	 *
420 	 * Set mtu to default DPDK value here.
421 	 */
422 	ret = mrvl_mtu_set(dev, dev->data->mtu);
423 	if (ret)
424 		pp2_ppio_disable(priv->ppio);
425 
426 	return ret;
427 }
428 
429 /**
430  * DPDK callback to bring the link down.
431  *
432  * @param dev
433  *   Pointer to Ethernet device structure.
434  *
435  * @return
436  *   0 on success, negative error value otherwise.
437  */
438 static int
439 mrvl_dev_set_link_down(struct rte_eth_dev *dev)
440 {
441 	struct mrvl_priv *priv = dev->data->dev_private;
442 
443 	if (!priv->ppio)
444 		return -EPERM;
445 
446 	return pp2_ppio_disable(priv->ppio);
447 }
448 
449 /**
450  * DPDK callback to start tx queue.
451  *
452  * @param dev
453  *   Pointer to Ethernet device structure.
454  * @param queue_id
455  *   Transmit queue index.
456  *
457  * @return
458  *   0 on success, negative error value otherwise.
459  */
460 static int
461 mrvl_tx_queue_start(struct rte_eth_dev *dev, uint16_t queue_id)
462 {
463 	struct mrvl_priv *priv = dev->data->dev_private;
464 	int ret;
465 
466 	if (!priv)
467 		return -EPERM;
468 
469 	/* passing 1 enables given tx queue */
470 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 1);
471 	if (ret) {
472 		MRVL_LOG(ERR, "Failed to start txq %d", queue_id);
473 		return ret;
474 	}
475 
476 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
477 
478 	return 0;
479 }
480 
481 /**
482  * DPDK callback to stop tx queue.
483  *
484  * @param dev
485  *   Pointer to Ethernet device structure.
486  * @param queue_id
487  *   Transmit queue index.
488  *
489  * @return
490  *   0 on success, negative error value otherwise.
491  */
492 static int
493 mrvl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t queue_id)
494 {
495 	struct mrvl_priv *priv = dev->data->dev_private;
496 	int ret;
497 
498 	if (!priv->ppio)
499 		return -EPERM;
500 
501 	/* passing 0 disables given tx queue */
502 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 0);
503 	if (ret) {
504 		MRVL_LOG(ERR, "Failed to stop txq %d", queue_id);
505 		return ret;
506 	}
507 
508 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
509 
510 	return 0;
511 }
512 
513 /**
514  * DPDK callback to start the device.
515  *
516  * @param dev
517  *   Pointer to Ethernet device structure.
518  *
519  * @return
520  *   0 on success, negative errno value on failure.
521  */
522 static int
523 mrvl_dev_start(struct rte_eth_dev *dev)
524 {
525 	struct mrvl_priv *priv = dev->data->dev_private;
526 	char match[MRVL_MATCH_LEN];
527 	int ret = 0, i, def_init_size;
528 
529 	snprintf(match, sizeof(match), "ppio-%d:%d",
530 		 priv->pp_id, priv->ppio_id);
531 	priv->ppio_params.match = match;
532 
533 	/*
534 	 * Calculate the minimum bpool size for refill feature as follows:
535 	 * 2 default burst sizes multiply by number of rx queues.
536 	 * If the bpool size will be below this value, new buffers will
537 	 * be added to the pool.
538 	 */
539 	priv->bpool_min_size = priv->nb_rx_queues * MRVL_BURST_SIZE * 2;
540 
541 	/* In case initial bpool size configured in queues setup is
542 	 * smaller than minimum size add more buffers
543 	 */
544 	def_init_size = priv->bpool_min_size + MRVL_BURST_SIZE * 2;
545 	if (priv->bpool_init_size < def_init_size) {
546 		int buffs_to_add = def_init_size - priv->bpool_init_size;
547 
548 		priv->bpool_init_size += buffs_to_add;
549 		ret = mrvl_fill_bpool(dev->data->rx_queues[0], buffs_to_add);
550 		if (ret)
551 			MRVL_LOG(ERR, "Failed to add buffers to bpool");
552 	}
553 
554 	/*
555 	 * Calculate the maximum bpool size for refill feature as follows:
556 	 * maximum number of descriptors in rx queue multiply by number
557 	 * of rx queues plus minimum bpool size.
558 	 * In case the bpool size will exceed this value, superfluous buffers
559 	 * will be removed
560 	 */
561 	priv->bpool_max_size = (priv->nb_rx_queues * MRVL_PP2_RXD_MAX) +
562 				priv->bpool_min_size;
563 
564 	ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
565 	if (ret) {
566 		MRVL_LOG(ERR, "Failed to init ppio");
567 		return ret;
568 	}
569 
570 	/*
571 	 * In case there are some some stale uc/mc mac addresses flush them
572 	 * here. It cannot be done during mrvl_dev_close() as port information
573 	 * is already gone at that point (due to pp2_ppio_deinit() in
574 	 * mrvl_dev_stop()).
575 	 */
576 	if (!priv->uc_mc_flushed) {
577 		ret = pp2_ppio_flush_mac_addrs(priv->ppio, 1, 1);
578 		if (ret) {
579 			MRVL_LOG(ERR,
580 				"Failed to flush uc/mc filter list");
581 			goto out;
582 		}
583 		priv->uc_mc_flushed = 1;
584 	}
585 
586 	if (!priv->vlan_flushed) {
587 		ret = pp2_ppio_flush_vlan(priv->ppio);
588 		if (ret) {
589 			MRVL_LOG(ERR, "Failed to flush vlan list");
590 			/*
591 			 * TODO
592 			 * once pp2_ppio_flush_vlan() is supported jump to out
593 			 * goto out;
594 			 */
595 		}
596 		priv->vlan_flushed = 1;
597 	}
598 
599 	/* For default QoS config, don't start classifier. */
600 	if (mrvl_qos_cfg) {
601 		ret = mrvl_start_qos_mapping(priv);
602 		if (ret) {
603 			MRVL_LOG(ERR, "Failed to setup QoS mapping");
604 			goto out;
605 		}
606 	}
607 
608 	ret = mrvl_dev_set_link_up(dev);
609 	if (ret) {
610 		MRVL_LOG(ERR, "Failed to set link up");
611 		goto out;
612 	}
613 
614 	/* start tx queues */
615 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
616 		struct mrvl_txq *txq = dev->data->tx_queues[i];
617 
618 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
619 
620 		if (!txq->tx_deferred_start)
621 			continue;
622 
623 		/*
624 		 * All txqs are started by default. Stop them
625 		 * so that tx_deferred_start works as expected.
626 		 */
627 		ret = mrvl_tx_queue_stop(dev, i);
628 		if (ret)
629 			goto out;
630 	}
631 
632 	return 0;
633 out:
634 	MRVL_LOG(ERR, "Failed to start device");
635 	pp2_ppio_deinit(priv->ppio);
636 	return ret;
637 }
638 
639 /**
640  * Flush receive queues.
641  *
642  * @param dev
643  *   Pointer to Ethernet device structure.
644  */
645 static void
646 mrvl_flush_rx_queues(struct rte_eth_dev *dev)
647 {
648 	int i;
649 
650 	MRVL_LOG(INFO, "Flushing rx queues");
651 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
652 		int ret, num;
653 
654 		do {
655 			struct mrvl_rxq *q = dev->data->rx_queues[i];
656 			struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];
657 
658 			num = MRVL_PP2_RXD_MAX;
659 			ret = pp2_ppio_recv(q->priv->ppio,
660 					    q->priv->rxq_map[q->queue_id].tc,
661 					    q->priv->rxq_map[q->queue_id].inq,
662 					    descs, (uint16_t *)&num);
663 		} while (ret == 0 && num);
664 	}
665 }
666 
667 /**
668  * Flush transmit shadow queues.
669  *
670  * @param dev
671  *   Pointer to Ethernet device structure.
672  */
673 static void
674 mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
675 {
676 	int i, j;
677 	struct mrvl_txq *txq;
678 
679 	MRVL_LOG(INFO, "Flushing tx shadow queues");
680 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
681 		txq = (struct mrvl_txq *)dev->data->tx_queues[i];
682 
683 		for (j = 0; j < RTE_MAX_LCORE; j++) {
684 			struct mrvl_shadow_txq *sq;
685 
686 			if (!hifs[j])
687 				continue;
688 
689 			sq = &txq->shadow_txqs[j];
690 			mrvl_free_sent_buffers(txq->priv->ppio,
691 				hifs[j], j, sq, txq->queue_id, 1);
692 			while (sq->tail != sq->head) {
693 				uint64_t addr = cookie_addr_high |
694 					sq->ent[sq->tail].buff.cookie;
695 				rte_pktmbuf_free(
696 					(struct rte_mbuf *)addr);
697 				sq->tail = (sq->tail + 1) &
698 					    MRVL_PP2_TX_SHADOWQ_MASK;
699 			}
700 			memset(sq, 0, sizeof(*sq));
701 		}
702 	}
703 }
704 
705 /**
706  * Flush hardware bpool (buffer-pool).
707  *
708  * @param dev
709  *   Pointer to Ethernet device structure.
710  */
711 static void
712 mrvl_flush_bpool(struct rte_eth_dev *dev)
713 {
714 	struct mrvl_priv *priv = dev->data->dev_private;
715 	struct pp2_hif *hif;
716 	uint32_t num;
717 	int ret;
718 	unsigned int core_id = rte_lcore_id();
719 
720 	if (core_id == LCORE_ID_ANY)
721 		core_id = 0;
722 
723 	hif = mrvl_get_hif(priv, core_id);
724 
725 	ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
726 	if (ret) {
727 		MRVL_LOG(ERR, "Failed to get bpool buffers number");
728 		return;
729 	}
730 
731 	while (num--) {
732 		struct pp2_buff_inf inf;
733 		uint64_t addr;
734 
735 		ret = pp2_bpool_get_buff(hif, priv->bpool, &inf);
736 		if (ret)
737 			break;
738 
739 		addr = cookie_addr_high | inf.cookie;
740 		rte_pktmbuf_free((struct rte_mbuf *)addr);
741 	}
742 }
743 
744 /**
745  * DPDK callback to stop the device.
746  *
747  * @param dev
748  *   Pointer to Ethernet device structure.
749  */
750 static void
751 mrvl_dev_stop(struct rte_eth_dev *dev)
752 {
753 	struct mrvl_priv *priv = dev->data->dev_private;
754 
755 	mrvl_dev_set_link_down(dev);
756 	mrvl_flush_rx_queues(dev);
757 	mrvl_flush_tx_shadow_queues(dev);
758 	if (priv->cls_tbl) {
759 		pp2_cls_tbl_deinit(priv->cls_tbl);
760 		priv->cls_tbl = NULL;
761 	}
762 	if (priv->qos_tbl) {
763 		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
764 		priv->qos_tbl = NULL;
765 	}
766 	if (priv->ppio)
767 		pp2_ppio_deinit(priv->ppio);
768 	priv->ppio = NULL;
769 
770 	/* policer must be released after ppio deinitialization */
771 	if (priv->policer) {
772 		pp2_cls_plcr_deinit(priv->policer);
773 		priv->policer = NULL;
774 	}
775 }
776 
777 /**
778  * DPDK callback to close the device.
779  *
780  * @param dev
781  *   Pointer to Ethernet device structure.
782  */
783 static void
784 mrvl_dev_close(struct rte_eth_dev *dev)
785 {
786 	struct mrvl_priv *priv = dev->data->dev_private;
787 	size_t i;
788 
789 	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
790 		struct pp2_ppio_tc_params *tc_params =
791 			&priv->ppio_params.inqs_params.tcs_params[i];
792 
793 		if (tc_params->inqs_params) {
794 			rte_free(tc_params->inqs_params);
795 			tc_params->inqs_params = NULL;
796 		}
797 	}
798 
799 	mrvl_flush_bpool(dev);
800 }
801 
802 /**
803  * DPDK callback to retrieve physical link information.
804  *
805  * @param dev
806  *   Pointer to Ethernet device structure.
807  * @param wait_to_complete
808  *   Wait for request completion (ignored).
809  *
810  * @return
811  *   0 on success, negative error value otherwise.
812  */
813 static int
814 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
815 {
816 	/*
817 	 * TODO
818 	 * once MUSDK provides necessary API use it here
819 	 */
820 	struct mrvl_priv *priv = dev->data->dev_private;
821 	struct ethtool_cmd edata;
822 	struct ifreq req;
823 	int ret, fd, link_up;
824 
825 	if (!priv->ppio)
826 		return -EPERM;
827 
828 	edata.cmd = ETHTOOL_GSET;
829 
830 	strcpy(req.ifr_name, dev->data->name);
831 	req.ifr_data = (void *)&edata;
832 
833 	fd = socket(AF_INET, SOCK_DGRAM, 0);
834 	if (fd == -1)
835 		return -EFAULT;
836 
837 	ret = ioctl(fd, SIOCETHTOOL, &req);
838 	if (ret == -1) {
839 		close(fd);
840 		return -EFAULT;
841 	}
842 
843 	close(fd);
844 
845 	switch (ethtool_cmd_speed(&edata)) {
846 	case SPEED_10:
847 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
848 		break;
849 	case SPEED_100:
850 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
851 		break;
852 	case SPEED_1000:
853 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
854 		break;
855 	case SPEED_10000:
856 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
857 		break;
858 	default:
859 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
860 	}
861 
862 	dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
863 							 ETH_LINK_HALF_DUPLEX;
864 	dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
865 							   ETH_LINK_FIXED;
866 	pp2_ppio_get_link_state(priv->ppio, &link_up);
867 	dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
868 
869 	return 0;
870 }
871 
872 /**
873  * DPDK callback to enable promiscuous mode.
874  *
875  * @param dev
876  *   Pointer to Ethernet device structure.
877  */
878 static void
879 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
880 {
881 	struct mrvl_priv *priv = dev->data->dev_private;
882 	int ret;
883 
884 	if (!priv->ppio)
885 		return;
886 
887 	if (priv->isolated)
888 		return;
889 
890 	ret = pp2_ppio_set_promisc(priv->ppio, 1);
891 	if (ret)
892 		MRVL_LOG(ERR, "Failed to enable promiscuous mode");
893 }
894 
895 /**
896  * DPDK callback to enable allmulti mode.
897  *
898  * @param dev
899  *   Pointer to Ethernet device structure.
900  */
901 static void
902 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
903 {
904 	struct mrvl_priv *priv = dev->data->dev_private;
905 	int ret;
906 
907 	if (!priv->ppio)
908 		return;
909 
910 	if (priv->isolated)
911 		return;
912 
913 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
914 	if (ret)
915 		MRVL_LOG(ERR, "Failed enable all-multicast mode");
916 }
917 
918 /**
919  * DPDK callback to disable promiscuous mode.
920  *
921  * @param dev
922  *   Pointer to Ethernet device structure.
923  */
924 static void
925 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
926 {
927 	struct mrvl_priv *priv = dev->data->dev_private;
928 	int ret;
929 
930 	if (!priv->ppio)
931 		return;
932 
933 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
934 	if (ret)
935 		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
936 }
937 
938 /**
939  * DPDK callback to disable allmulticast mode.
940  *
941  * @param dev
942  *   Pointer to Ethernet device structure.
943  */
944 static void
945 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
946 {
947 	struct mrvl_priv *priv = dev->data->dev_private;
948 	int ret;
949 
950 	if (!priv->ppio)
951 		return;
952 
953 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
954 	if (ret)
955 		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
956 }
957 
958 /**
959  * DPDK callback to remove a MAC address.
960  *
961  * @param dev
962  *   Pointer to Ethernet device structure.
963  * @param index
964  *   MAC address index.
965  */
966 static void
967 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
968 {
969 	struct mrvl_priv *priv = dev->data->dev_private;
970 	char buf[ETHER_ADDR_FMT_SIZE];
971 	int ret;
972 
973 	if (!priv->ppio)
974 		return;
975 
976 	if (priv->isolated)
977 		return;
978 
979 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
980 				       dev->data->mac_addrs[index].addr_bytes);
981 	if (ret) {
982 		ether_format_addr(buf, sizeof(buf),
983 				  &dev->data->mac_addrs[index]);
984 		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
985 	}
986 }
987 
988 /**
989  * DPDK callback to add a MAC address.
990  *
991  * @param dev
992  *   Pointer to Ethernet device structure.
993  * @param mac_addr
994  *   MAC address to register.
995  * @param index
996  *   MAC address index.
997  * @param vmdq
998  *   VMDq pool index to associate address with (unused).
999  *
1000  * @return
1001  *   0 on success, negative error value otherwise.
1002  */
1003 static int
1004 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1005 		  uint32_t index, uint32_t vmdq __rte_unused)
1006 {
1007 	struct mrvl_priv *priv = dev->data->dev_private;
1008 	char buf[ETHER_ADDR_FMT_SIZE];
1009 	int ret;
1010 
1011 	if (priv->isolated)
1012 		return -ENOTSUP;
1013 
1014 	if (index == 0)
1015 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1016 		return -1;
1017 
1018 	if (!priv->ppio)
1019 		return 0;
1020 
1021 	/*
1022 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1023 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1024 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1025 	 * 21 respectively.
1026 	 *
1027 	 * If more than uc_filter_max uc addresses were added to filter list
1028 	 * then NIC will switch to promiscuous mode automatically.
1029 	 *
1030 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1031 	 * were added to filter list then NIC will switch to all-multicast mode
1032 	 * automatically.
1033 	 */
1034 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1035 	if (ret) {
1036 		ether_format_addr(buf, sizeof(buf), mac_addr);
1037 		MRVL_LOG(ERR, "Failed to add mac %s", buf);
1038 		return -1;
1039 	}
1040 
1041 	return 0;
1042 }
1043 
1044 /**
1045  * DPDK callback to set the primary MAC address.
1046  *
1047  * @param dev
1048  *   Pointer to Ethernet device structure.
1049  * @param mac_addr
1050  *   MAC address to register.
1051  *
1052  * @return
1053  *   0 on success, negative error value otherwise.
1054  */
1055 static int
1056 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1057 {
1058 	struct mrvl_priv *priv = dev->data->dev_private;
1059 	int ret;
1060 
1061 	if (!priv->ppio)
1062 		return 0;
1063 
1064 	if (priv->isolated)
1065 		return -ENOTSUP;
1066 
1067 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1068 	if (ret) {
1069 		char buf[ETHER_ADDR_FMT_SIZE];
1070 		ether_format_addr(buf, sizeof(buf), mac_addr);
1071 		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1072 	}
1073 
1074 	return ret;
1075 }
1076 
1077 /**
1078  * DPDK callback to get device statistics.
1079  *
1080  * @param dev
1081  *   Pointer to Ethernet device structure.
1082  * @param stats
1083  *   Stats structure output buffer.
1084  *
1085  * @return
1086  *   0 on success, negative error value otherwise.
1087  */
1088 static int
1089 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1090 {
1091 	struct mrvl_priv *priv = dev->data->dev_private;
1092 	struct pp2_ppio_statistics ppio_stats;
1093 	uint64_t drop_mac = 0;
1094 	unsigned int i, idx, ret;
1095 
1096 	if (!priv->ppio)
1097 		return -EPERM;
1098 
1099 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1100 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1101 		struct pp2_ppio_inq_statistics rx_stats;
1102 
1103 		if (!rxq)
1104 			continue;
1105 
1106 		idx = rxq->queue_id;
1107 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1108 			MRVL_LOG(ERR,
1109 				"rx queue %d stats out of range (0 - %d)",
1110 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1111 			continue;
1112 		}
1113 
1114 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1115 						  priv->rxq_map[idx].tc,
1116 						  priv->rxq_map[idx].inq,
1117 						  &rx_stats, 0);
1118 		if (unlikely(ret)) {
1119 			MRVL_LOG(ERR,
1120 				"Failed to update rx queue %d stats", idx);
1121 			break;
1122 		}
1123 
1124 		stats->q_ibytes[idx] = rxq->bytes_recv;
1125 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1126 		stats->q_errors[idx] = rx_stats.drop_early +
1127 				       rx_stats.drop_fullq +
1128 				       rx_stats.drop_bm +
1129 				       rxq->drop_mac;
1130 		stats->ibytes += rxq->bytes_recv;
1131 		drop_mac += rxq->drop_mac;
1132 	}
1133 
1134 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1135 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1136 		struct pp2_ppio_outq_statistics tx_stats;
1137 
1138 		if (!txq)
1139 			continue;
1140 
1141 		idx = txq->queue_id;
1142 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1143 			MRVL_LOG(ERR,
1144 				"tx queue %d stats out of range (0 - %d)",
1145 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1146 		}
1147 
1148 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1149 						   &tx_stats, 0);
1150 		if (unlikely(ret)) {
1151 			MRVL_LOG(ERR,
1152 				"Failed to update tx queue %d stats", idx);
1153 			break;
1154 		}
1155 
1156 		stats->q_opackets[idx] = tx_stats.deq_desc;
1157 		stats->q_obytes[idx] = txq->bytes_sent;
1158 		stats->obytes += txq->bytes_sent;
1159 	}
1160 
1161 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1162 	if (unlikely(ret)) {
1163 		MRVL_LOG(ERR, "Failed to update port statistics");
1164 		return ret;
1165 	}
1166 
1167 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1168 	stats->opackets += ppio_stats.tx_packets;
1169 	stats->imissed += ppio_stats.rx_fullq_dropped +
1170 			  ppio_stats.rx_bm_dropped +
1171 			  ppio_stats.rx_early_dropped +
1172 			  ppio_stats.rx_fifo_dropped +
1173 			  ppio_stats.rx_cls_dropped;
1174 	stats->ierrors = drop_mac;
1175 
1176 	return 0;
1177 }
1178 
1179 /**
1180  * DPDK callback to clear device statistics.
1181  *
1182  * @param dev
1183  *   Pointer to Ethernet device structure.
1184  */
1185 static void
1186 mrvl_stats_reset(struct rte_eth_dev *dev)
1187 {
1188 	struct mrvl_priv *priv = dev->data->dev_private;
1189 	int i;
1190 
1191 	if (!priv->ppio)
1192 		return;
1193 
1194 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1195 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1196 
1197 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1198 					    priv->rxq_map[i].inq, NULL, 1);
1199 		rxq->bytes_recv = 0;
1200 		rxq->drop_mac = 0;
1201 	}
1202 
1203 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1204 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1205 
1206 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1207 		txq->bytes_sent = 0;
1208 	}
1209 
1210 	pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1211 }
1212 
1213 /**
1214  * DPDK callback to get extended statistics.
1215  *
1216  * @param dev
1217  *   Pointer to Ethernet device structure.
1218  * @param stats
1219  *   Pointer to xstats table.
1220  * @param n
1221  *   Number of entries in xstats table.
1222  * @return
1223  *   Negative value on error, number of read xstats otherwise.
1224  */
1225 static int
1226 mrvl_xstats_get(struct rte_eth_dev *dev,
1227 		struct rte_eth_xstat *stats, unsigned int n)
1228 {
1229 	struct mrvl_priv *priv = dev->data->dev_private;
1230 	struct pp2_ppio_statistics ppio_stats;
1231 	unsigned int i;
1232 
1233 	if (!stats)
1234 		return 0;
1235 
1236 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1237 	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1238 		uint64_t val;
1239 
1240 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1241 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1242 					    mrvl_xstats_tbl[i].offset);
1243 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1244 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1245 					    mrvl_xstats_tbl[i].offset);
1246 		else
1247 			return -EINVAL;
1248 
1249 		stats[i].id = i;
1250 		stats[i].value = val;
1251 	}
1252 
1253 	return n;
1254 }
1255 
1256 /**
1257  * DPDK callback to reset extended statistics.
1258  *
1259  * @param dev
1260  *   Pointer to Ethernet device structure.
1261  */
1262 static void
1263 mrvl_xstats_reset(struct rte_eth_dev *dev)
1264 {
1265 	mrvl_stats_reset(dev);
1266 }
1267 
1268 /**
1269  * DPDK callback to get extended statistics names.
1270  *
1271  * @param dev (unused)
1272  *   Pointer to Ethernet device structure.
1273  * @param xstats_names
1274  *   Pointer to xstats names table.
1275  * @param size
1276  *   Size of the xstats names table.
1277  * @return
1278  *   Number of read names.
1279  */
1280 static int
1281 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1282 		      struct rte_eth_xstat_name *xstats_names,
1283 		      unsigned int size)
1284 {
1285 	unsigned int i;
1286 
1287 	if (!xstats_names)
1288 		return RTE_DIM(mrvl_xstats_tbl);
1289 
1290 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1291 		snprintf(xstats_names[i].name, RTE_ETH_XSTATS_NAME_SIZE, "%s",
1292 			 mrvl_xstats_tbl[i].name);
1293 
1294 	return size;
1295 }
1296 
1297 /**
1298  * DPDK callback to get information about the device.
1299  *
1300  * @param dev
1301  *   Pointer to Ethernet device structure (unused).
1302  * @param info
1303  *   Info structure output buffer.
1304  */
1305 static void
1306 mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
1307 		   struct rte_eth_dev_info *info)
1308 {
1309 	info->speed_capa = ETH_LINK_SPEED_10M |
1310 			   ETH_LINK_SPEED_100M |
1311 			   ETH_LINK_SPEED_1G |
1312 			   ETH_LINK_SPEED_10G;
1313 
1314 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1315 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1316 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1317 
1318 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1319 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1320 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1321 
1322 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1323 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1324 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1325 
1326 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1327 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1328 
1329 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1330 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1331 
1332 	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1333 				       ETH_RSS_NONFRAG_IPV4_TCP |
1334 				       ETH_RSS_NONFRAG_IPV4_UDP;
1335 
1336 	/* By default packets are dropped if no descriptors are available */
1337 	info->default_rxconf.rx_drop_en = 1;
1338 	info->default_rxconf.offloads = DEV_RX_OFFLOAD_CRC_STRIP;
1339 
1340 	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1341 }
1342 
1343 /**
1344  * Return supported packet types.
1345  *
1346  * @param dev
1347  *   Pointer to Ethernet device structure (unused).
1348  *
1349  * @return
1350  *   Const pointer to the table with supported packet types.
1351  */
1352 static const uint32_t *
1353 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1354 {
1355 	static const uint32_t ptypes[] = {
1356 		RTE_PTYPE_L2_ETHER,
1357 		RTE_PTYPE_L3_IPV4,
1358 		RTE_PTYPE_L3_IPV4_EXT,
1359 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1360 		RTE_PTYPE_L3_IPV6,
1361 		RTE_PTYPE_L3_IPV6_EXT,
1362 		RTE_PTYPE_L2_ETHER_ARP,
1363 		RTE_PTYPE_L4_TCP,
1364 		RTE_PTYPE_L4_UDP
1365 	};
1366 
1367 	return ptypes;
1368 }
1369 
1370 /**
1371  * DPDK callback to get information about specific receive queue.
1372  *
1373  * @param dev
1374  *   Pointer to Ethernet device structure.
1375  * @param rx_queue_id
1376  *   Receive queue index.
1377  * @param qinfo
1378  *   Receive queue information structure.
1379  */
1380 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1381 			      struct rte_eth_rxq_info *qinfo)
1382 {
1383 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1384 	struct mrvl_priv *priv = dev->data->dev_private;
1385 	int inq = priv->rxq_map[rx_queue_id].inq;
1386 	int tc = priv->rxq_map[rx_queue_id].tc;
1387 	struct pp2_ppio_tc_params *tc_params =
1388 		&priv->ppio_params.inqs_params.tcs_params[tc];
1389 
1390 	qinfo->mp = q->mp;
1391 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1392 }
1393 
1394 /**
1395  * DPDK callback to get information about specific transmit queue.
1396  *
1397  * @param dev
1398  *   Pointer to Ethernet device structure.
1399  * @param tx_queue_id
1400  *   Transmit queue index.
1401  * @param qinfo
1402  *   Transmit queue information structure.
1403  */
1404 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1405 			      struct rte_eth_txq_info *qinfo)
1406 {
1407 	struct mrvl_priv *priv = dev->data->dev_private;
1408 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1409 
1410 	qinfo->nb_desc =
1411 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1412 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1413 }
1414 
1415 /**
1416  * DPDK callback to Configure a VLAN filter.
1417  *
1418  * @param dev
1419  *   Pointer to Ethernet device structure.
1420  * @param vlan_id
1421  *   VLAN ID to filter.
1422  * @param on
1423  *   Toggle filter.
1424  *
1425  * @return
1426  *   0 on success, negative error value otherwise.
1427  */
1428 static int
1429 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1430 {
1431 	struct mrvl_priv *priv = dev->data->dev_private;
1432 
1433 	if (!priv->ppio)
1434 		return -EPERM;
1435 
1436 	if (priv->isolated)
1437 		return -ENOTSUP;
1438 
1439 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1440 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1441 }
1442 
1443 /**
1444  * Release buffers to hardware bpool (buffer-pool)
1445  *
1446  * @param rxq
1447  *   Receive queue pointer.
1448  * @param num
1449  *   Number of buffers to release to bpool.
1450  *
1451  * @return
1452  *   0 on success, negative error value otherwise.
1453  */
1454 static int
1455 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1456 {
1457 	struct buff_release_entry entries[MRVL_PP2_RXD_MAX];
1458 	struct rte_mbuf *mbufs[MRVL_PP2_RXD_MAX];
1459 	int i, ret;
1460 	unsigned int core_id;
1461 	struct pp2_hif *hif;
1462 	struct pp2_bpool *bpool;
1463 
1464 	core_id = rte_lcore_id();
1465 	if (core_id == LCORE_ID_ANY)
1466 		core_id = 0;
1467 
1468 	hif = mrvl_get_hif(rxq->priv, core_id);
1469 	if (!hif)
1470 		return -1;
1471 
1472 	bpool = rxq->priv->bpool;
1473 
1474 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1475 	if (ret)
1476 		return ret;
1477 
1478 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1479 		cookie_addr_high =
1480 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1481 
1482 	for (i = 0; i < num; i++) {
1483 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1484 			!= cookie_addr_high) {
1485 			MRVL_LOG(ERR,
1486 				"mbuf virtual addr high 0x%lx out of range",
1487 				(uint64_t)mbufs[i] >> 32);
1488 			goto out;
1489 		}
1490 
1491 		entries[i].buff.addr =
1492 			rte_mbuf_data_iova_default(mbufs[i]);
1493 		entries[i].buff.cookie = (pp2_cookie_t)(uint64_t)mbufs[i];
1494 		entries[i].bpool = bpool;
1495 	}
1496 
1497 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1498 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1499 
1500 	if (i != num)
1501 		goto out;
1502 
1503 	return 0;
1504 out:
1505 	for (; i < num; i++)
1506 		rte_pktmbuf_free(mbufs[i]);
1507 
1508 	return -1;
1509 }
1510 
1511 /**
1512  * DPDK callback to configure the receive queue.
1513  *
1514  * @param dev
1515  *   Pointer to Ethernet device structure.
1516  * @param idx
1517  *   RX queue index.
1518  * @param desc
1519  *   Number of descriptors to configure in queue.
1520  * @param socket
1521  *   NUMA socket on which memory must be allocated.
1522  * @param conf
1523  *   Thresholds parameters.
1524  * @param mp
1525  *   Memory pool for buffer allocations.
1526  *
1527  * @return
1528  *   0 on success, negative error value otherwise.
1529  */
1530 static int
1531 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1532 		    unsigned int socket,
1533 		    const struct rte_eth_rxconf *conf,
1534 		    struct rte_mempool *mp)
1535 {
1536 	struct mrvl_priv *priv = dev->data->dev_private;
1537 	struct mrvl_rxq *rxq;
1538 	uint32_t min_size,
1539 		 max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1540 	int ret, tc, inq;
1541 	uint64_t offloads;
1542 
1543 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
1544 
1545 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1546 		/*
1547 		 * Unknown TC mapping, mapping will not have a correct queue.
1548 		 */
1549 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
1550 			idx, priv->ppio_id);
1551 		return -EFAULT;
1552 	}
1553 
1554 	min_size = rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM -
1555 		   MRVL_PKT_EFFEC_OFFS;
1556 	if (min_size < max_rx_pkt_len) {
1557 		MRVL_LOG(ERR,
1558 			"Mbuf size must be increased to %u bytes to hold up to %u bytes of data.",
1559 			max_rx_pkt_len + RTE_PKTMBUF_HEADROOM +
1560 			MRVL_PKT_EFFEC_OFFS,
1561 			max_rx_pkt_len);
1562 		return -EINVAL;
1563 	}
1564 
1565 	if (dev->data->rx_queues[idx]) {
1566 		rte_free(dev->data->rx_queues[idx]);
1567 		dev->data->rx_queues[idx] = NULL;
1568 	}
1569 
1570 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
1571 	if (!rxq)
1572 		return -ENOMEM;
1573 
1574 	rxq->priv = priv;
1575 	rxq->mp = mp;
1576 	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
1577 	rxq->queue_id = idx;
1578 	rxq->port_id = dev->data->port_id;
1579 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
1580 
1581 	tc = priv->rxq_map[rxq->queue_id].tc,
1582 	inq = priv->rxq_map[rxq->queue_id].inq;
1583 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
1584 		desc;
1585 
1586 	ret = mrvl_fill_bpool(rxq, desc);
1587 	if (ret) {
1588 		rte_free(rxq);
1589 		return ret;
1590 	}
1591 
1592 	priv->bpool_init_size += desc;
1593 
1594 	dev->data->rx_queues[idx] = rxq;
1595 
1596 	return 0;
1597 }
1598 
1599 /**
1600  * DPDK callback to release the receive queue.
1601  *
1602  * @param rxq
1603  *   Generic receive queue pointer.
1604  */
1605 static void
1606 mrvl_rx_queue_release(void *rxq)
1607 {
1608 	struct mrvl_rxq *q = rxq;
1609 	struct pp2_ppio_tc_params *tc_params;
1610 	int i, num, tc, inq;
1611 	struct pp2_hif *hif;
1612 	unsigned int core_id = rte_lcore_id();
1613 
1614 	if (core_id == LCORE_ID_ANY)
1615 		core_id = 0;
1616 
1617 	if (!q)
1618 		return;
1619 
1620 	hif = mrvl_get_hif(q->priv, core_id);
1621 
1622 	if (!hif)
1623 		return;
1624 
1625 	tc = q->priv->rxq_map[q->queue_id].tc;
1626 	inq = q->priv->rxq_map[q->queue_id].inq;
1627 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
1628 	num = tc_params->inqs_params[inq].size;
1629 	for (i = 0; i < num; i++) {
1630 		struct pp2_buff_inf inf;
1631 		uint64_t addr;
1632 
1633 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
1634 		addr = cookie_addr_high | inf.cookie;
1635 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1636 	}
1637 
1638 	rte_free(q);
1639 }
1640 
1641 /**
1642  * DPDK callback to configure the transmit queue.
1643  *
1644  * @param dev
1645  *   Pointer to Ethernet device structure.
1646  * @param idx
1647  *   Transmit queue index.
1648  * @param desc
1649  *   Number of descriptors to configure in the queue.
1650  * @param socket
1651  *   NUMA socket on which memory must be allocated.
1652  * @param conf
1653  *   Tx queue configuration parameters.
1654  *
1655  * @return
1656  *   0 on success, negative error value otherwise.
1657  */
1658 static int
1659 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1660 		    unsigned int socket,
1661 		    const struct rte_eth_txconf *conf)
1662 {
1663 	struct mrvl_priv *priv = dev->data->dev_private;
1664 	struct mrvl_txq *txq;
1665 
1666 	if (dev->data->tx_queues[idx]) {
1667 		rte_free(dev->data->tx_queues[idx]);
1668 		dev->data->tx_queues[idx] = NULL;
1669 	}
1670 
1671 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
1672 	if (!txq)
1673 		return -ENOMEM;
1674 
1675 	txq->priv = priv;
1676 	txq->queue_id = idx;
1677 	txq->port_id = dev->data->port_id;
1678 	txq->tx_deferred_start = conf->tx_deferred_start;
1679 	dev->data->tx_queues[idx] = txq;
1680 
1681 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
1682 
1683 	return 0;
1684 }
1685 
1686 /**
1687  * DPDK callback to release the transmit queue.
1688  *
1689  * @param txq
1690  *   Generic transmit queue pointer.
1691  */
1692 static void
1693 mrvl_tx_queue_release(void *txq)
1694 {
1695 	struct mrvl_txq *q = txq;
1696 
1697 	if (!q)
1698 		return;
1699 
1700 	rte_free(q);
1701 }
1702 
1703 /**
1704  * DPDK callback to get flow control configuration.
1705  *
1706  * @param dev
1707  *  Pointer to Ethernet device structure.
1708  * @param fc_conf
1709  *  Pointer to the flow control configuration.
1710  *
1711  * @return
1712  *  0 on success, negative error value otherwise.
1713  */
1714 static int
1715 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1716 {
1717 	struct mrvl_priv *priv = dev->data->dev_private;
1718 	int ret, en;
1719 
1720 	if (!priv)
1721 		return -EPERM;
1722 
1723 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
1724 	if (ret) {
1725 		MRVL_LOG(ERR, "Failed to read rx pause state");
1726 		return ret;
1727 	}
1728 
1729 	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
1730 
1731 	return 0;
1732 }
1733 
1734 /**
1735  * DPDK callback to set flow control configuration.
1736  *
1737  * @param dev
1738  *  Pointer to Ethernet device structure.
1739  * @param fc_conf
1740  *  Pointer to the flow control configuration.
1741  *
1742  * @return
1743  *  0 on success, negative error value otherwise.
1744  */
1745 static int
1746 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1747 {
1748 	struct mrvl_priv *priv = dev->data->dev_private;
1749 
1750 	if (!priv)
1751 		return -EPERM;
1752 
1753 	if (fc_conf->high_water ||
1754 	    fc_conf->low_water ||
1755 	    fc_conf->pause_time ||
1756 	    fc_conf->mac_ctrl_frame_fwd ||
1757 	    fc_conf->autoneg) {
1758 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
1759 
1760 		return -EINVAL;
1761 	}
1762 
1763 	if (fc_conf->mode == RTE_FC_NONE ||
1764 	    fc_conf->mode == RTE_FC_RX_PAUSE) {
1765 		int ret, en;
1766 
1767 		en = fc_conf->mode == RTE_FC_NONE ? 0 : 1;
1768 		ret = pp2_ppio_set_rx_pause(priv->ppio, en);
1769 		if (ret)
1770 			MRVL_LOG(ERR,
1771 				"Failed to change flowctrl on RX side");
1772 
1773 		return ret;
1774 	}
1775 
1776 	return 0;
1777 }
1778 
1779 /**
1780  * Update RSS hash configuration
1781  *
1782  * @param dev
1783  *   Pointer to Ethernet device structure.
1784  * @param rss_conf
1785  *   Pointer to RSS configuration.
1786  *
1787  * @return
1788  *   0 on success, negative error value otherwise.
1789  */
1790 static int
1791 mrvl_rss_hash_update(struct rte_eth_dev *dev,
1792 		     struct rte_eth_rss_conf *rss_conf)
1793 {
1794 	struct mrvl_priv *priv = dev->data->dev_private;
1795 
1796 	if (priv->isolated)
1797 		return -ENOTSUP;
1798 
1799 	return mrvl_configure_rss(priv, rss_conf);
1800 }
1801 
1802 /**
1803  * DPDK callback to get RSS hash configuration.
1804  *
1805  * @param dev
1806  *   Pointer to Ethernet device structure.
1807  * @rss_conf
1808  *   Pointer to RSS configuration.
1809  *
1810  * @return
1811  *   Always 0.
1812  */
1813 static int
1814 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
1815 		       struct rte_eth_rss_conf *rss_conf)
1816 {
1817 	struct mrvl_priv *priv = dev->data->dev_private;
1818 	enum pp2_ppio_hash_type hash_type =
1819 		priv->ppio_params.inqs_params.hash_type;
1820 
1821 	rss_conf->rss_key = NULL;
1822 
1823 	if (hash_type == PP2_PPIO_HASH_T_NONE)
1824 		rss_conf->rss_hf = 0;
1825 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
1826 		rss_conf->rss_hf = ETH_RSS_IPV4;
1827 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
1828 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
1829 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
1830 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
1831 
1832 	return 0;
1833 }
1834 
1835 /**
1836  * DPDK callback to get rte_flow callbacks.
1837  *
1838  * @param dev
1839  *   Pointer to the device structure.
1840  * @param filer_type
1841  *   Flow filter type.
1842  * @param filter_op
1843  *   Flow filter operation.
1844  * @param arg
1845  *   Pointer to pass the flow ops.
1846  *
1847  * @return
1848  *   0 on success, negative error value otherwise.
1849  */
1850 static int
1851 mrvl_eth_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
1852 		     enum rte_filter_type filter_type,
1853 		     enum rte_filter_op filter_op, void *arg)
1854 {
1855 	switch (filter_type) {
1856 	case RTE_ETH_FILTER_GENERIC:
1857 		if (filter_op != RTE_ETH_FILTER_GET)
1858 			return -EINVAL;
1859 		*(const void **)arg = &mrvl_flow_ops;
1860 		return 0;
1861 	default:
1862 		MRVL_LOG(WARNING, "Filter type (%d) not supported",
1863 				filter_type);
1864 		return -EINVAL;
1865 	}
1866 }
1867 
1868 static const struct eth_dev_ops mrvl_ops = {
1869 	.dev_configure = mrvl_dev_configure,
1870 	.dev_start = mrvl_dev_start,
1871 	.dev_stop = mrvl_dev_stop,
1872 	.dev_set_link_up = mrvl_dev_set_link_up,
1873 	.dev_set_link_down = mrvl_dev_set_link_down,
1874 	.dev_close = mrvl_dev_close,
1875 	.link_update = mrvl_link_update,
1876 	.promiscuous_enable = mrvl_promiscuous_enable,
1877 	.allmulticast_enable = mrvl_allmulticast_enable,
1878 	.promiscuous_disable = mrvl_promiscuous_disable,
1879 	.allmulticast_disable = mrvl_allmulticast_disable,
1880 	.mac_addr_remove = mrvl_mac_addr_remove,
1881 	.mac_addr_add = mrvl_mac_addr_add,
1882 	.mac_addr_set = mrvl_mac_addr_set,
1883 	.mtu_set = mrvl_mtu_set,
1884 	.stats_get = mrvl_stats_get,
1885 	.stats_reset = mrvl_stats_reset,
1886 	.xstats_get = mrvl_xstats_get,
1887 	.xstats_reset = mrvl_xstats_reset,
1888 	.xstats_get_names = mrvl_xstats_get_names,
1889 	.dev_infos_get = mrvl_dev_infos_get,
1890 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
1891 	.rxq_info_get = mrvl_rxq_info_get,
1892 	.txq_info_get = mrvl_txq_info_get,
1893 	.vlan_filter_set = mrvl_vlan_filter_set,
1894 	.tx_queue_start = mrvl_tx_queue_start,
1895 	.tx_queue_stop = mrvl_tx_queue_stop,
1896 	.rx_queue_setup = mrvl_rx_queue_setup,
1897 	.rx_queue_release = mrvl_rx_queue_release,
1898 	.tx_queue_setup = mrvl_tx_queue_setup,
1899 	.tx_queue_release = mrvl_tx_queue_release,
1900 	.flow_ctrl_get = mrvl_flow_ctrl_get,
1901 	.flow_ctrl_set = mrvl_flow_ctrl_set,
1902 	.rss_hash_update = mrvl_rss_hash_update,
1903 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
1904 	.filter_ctrl = mrvl_eth_filter_ctrl,
1905 };
1906 
1907 /**
1908  * Return packet type information and l3/l4 offsets.
1909  *
1910  * @param desc
1911  *   Pointer to the received packet descriptor.
1912  * @param l3_offset
1913  *   l3 packet offset.
1914  * @param l4_offset
1915  *   l4 packet offset.
1916  *
1917  * @return
1918  *   Packet type information.
1919  */
1920 static inline uint64_t
1921 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
1922 				    uint8_t *l3_offset, uint8_t *l4_offset)
1923 {
1924 	enum pp2_inq_l3_type l3_type;
1925 	enum pp2_inq_l4_type l4_type;
1926 	uint64_t packet_type;
1927 
1928 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
1929 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
1930 
1931 	packet_type = RTE_PTYPE_L2_ETHER;
1932 
1933 	switch (l3_type) {
1934 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
1935 		packet_type |= RTE_PTYPE_L3_IPV4;
1936 		break;
1937 	case PP2_INQ_L3_TYPE_IPV4_OK:
1938 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
1939 		break;
1940 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
1941 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
1942 		break;
1943 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
1944 		packet_type |= RTE_PTYPE_L3_IPV6;
1945 		break;
1946 	case PP2_INQ_L3_TYPE_IPV6_EXT:
1947 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
1948 		break;
1949 	case PP2_INQ_L3_TYPE_ARP:
1950 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
1951 		/*
1952 		 * In case of ARP l4_offset is set to wrong value.
1953 		 * Set it to proper one so that later on mbuf->l3_len can be
1954 		 * calculated subtracting l4_offset and l3_offset.
1955 		 */
1956 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
1957 		break;
1958 	default:
1959 		MRVL_LOG(DEBUG, "Failed to recognise l3 packet type");
1960 		break;
1961 	}
1962 
1963 	switch (l4_type) {
1964 	case PP2_INQ_L4_TYPE_TCP:
1965 		packet_type |= RTE_PTYPE_L4_TCP;
1966 		break;
1967 	case PP2_INQ_L4_TYPE_UDP:
1968 		packet_type |= RTE_PTYPE_L4_UDP;
1969 		break;
1970 	default:
1971 		MRVL_LOG(DEBUG, "Failed to recognise l4 packet type");
1972 		break;
1973 	}
1974 
1975 	return packet_type;
1976 }
1977 
1978 /**
1979  * Get offload information from the received packet descriptor.
1980  *
1981  * @param desc
1982  *   Pointer to the received packet descriptor.
1983  *
1984  * @return
1985  *   Mbuf offload flags.
1986  */
1987 static inline uint64_t
1988 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc)
1989 {
1990 	uint64_t flags;
1991 	enum pp2_inq_desc_status status;
1992 
1993 	status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
1994 	if (unlikely(status != PP2_DESC_ERR_OK))
1995 		flags = PKT_RX_IP_CKSUM_BAD;
1996 	else
1997 		flags = PKT_RX_IP_CKSUM_GOOD;
1998 
1999 	status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2000 	if (unlikely(status != PP2_DESC_ERR_OK))
2001 		flags |= PKT_RX_L4_CKSUM_BAD;
2002 	else
2003 		flags |= PKT_RX_L4_CKSUM_GOOD;
2004 
2005 	return flags;
2006 }
2007 
2008 /**
2009  * DPDK callback for receive.
2010  *
2011  * @param rxq
2012  *   Generic pointer to the receive queue.
2013  * @param rx_pkts
2014  *   Array to store received packets.
2015  * @param nb_pkts
2016  *   Maximum number of packets in array.
2017  *
2018  * @return
2019  *   Number of packets successfully received.
2020  */
2021 static uint16_t
2022 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2023 {
2024 	struct mrvl_rxq *q = rxq;
2025 	struct pp2_ppio_desc descs[nb_pkts];
2026 	struct pp2_bpool *bpool;
2027 	int i, ret, rx_done = 0;
2028 	int num;
2029 	struct pp2_hif *hif;
2030 	unsigned int core_id = rte_lcore_id();
2031 
2032 	hif = mrvl_get_hif(q->priv, core_id);
2033 
2034 	if (unlikely(!q->priv->ppio || !hif))
2035 		return 0;
2036 
2037 	bpool = q->priv->bpool;
2038 
2039 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2040 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2041 	if (unlikely(ret < 0)) {
2042 		MRVL_LOG(ERR, "Failed to receive packets");
2043 		return 0;
2044 	}
2045 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2046 
2047 	for (i = 0; i < nb_pkts; i++) {
2048 		struct rte_mbuf *mbuf;
2049 		uint8_t l3_offset, l4_offset;
2050 		enum pp2_inq_desc_status status;
2051 		uint64_t addr;
2052 
2053 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2054 			struct pp2_ppio_desc *pref_desc;
2055 			u64 pref_addr;
2056 
2057 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2058 			pref_addr = cookie_addr_high |
2059 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2060 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2061 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2062 		}
2063 
2064 		addr = cookie_addr_high |
2065 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2066 		mbuf = (struct rte_mbuf *)addr;
2067 		rte_pktmbuf_reset(mbuf);
2068 
2069 		/* drop packet in case of mac, overrun or resource error */
2070 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2071 		if (unlikely(status != PP2_DESC_ERR_OK)) {
2072 			struct pp2_buff_inf binf = {
2073 				.addr = rte_mbuf_data_iova_default(mbuf),
2074 				.cookie = (pp2_cookie_t)(uint64_t)mbuf,
2075 			};
2076 
2077 			pp2_bpool_put_buff(hif, bpool, &binf);
2078 			mrvl_port_bpool_size
2079 				[bpool->pp2_id][bpool->id][core_id]++;
2080 			q->drop_mac++;
2081 			continue;
2082 		}
2083 
2084 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2085 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2086 		mbuf->data_len = mbuf->pkt_len;
2087 		mbuf->port = q->port_id;
2088 		mbuf->packet_type =
2089 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2090 							    &l3_offset,
2091 							    &l4_offset);
2092 		mbuf->l2_len = l3_offset;
2093 		mbuf->l3_len = l4_offset - l3_offset;
2094 
2095 		if (likely(q->cksum_enabled))
2096 			mbuf->ol_flags = mrvl_desc_to_ol_flags(&descs[i]);
2097 
2098 		rx_pkts[rx_done++] = mbuf;
2099 		q->bytes_recv += mbuf->pkt_len;
2100 	}
2101 
2102 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2103 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2104 
2105 		if (unlikely(num <= q->priv->bpool_min_size ||
2106 			     (!rx_done && num < q->priv->bpool_init_size))) {
2107 			ret = mrvl_fill_bpool(q, MRVL_BURST_SIZE);
2108 			if (ret)
2109 				MRVL_LOG(ERR, "Failed to fill bpool");
2110 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2111 			int i;
2112 			int pkt_to_remove = num - q->priv->bpool_init_size;
2113 			struct rte_mbuf *mbuf;
2114 			struct pp2_buff_inf buff;
2115 
2116 			MRVL_LOG(DEBUG,
2117 				"port-%d:%d: bpool %d oversize - remove %d buffers (pool size: %d -> %d)",
2118 				bpool->pp2_id, q->priv->ppio->port_id,
2119 				bpool->id, pkt_to_remove, num,
2120 				q->priv->bpool_init_size);
2121 
2122 			for (i = 0; i < pkt_to_remove; i++) {
2123 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2124 				if (ret)
2125 					break;
2126 				mbuf = (struct rte_mbuf *)
2127 					(cookie_addr_high | buff.cookie);
2128 				rte_pktmbuf_free(mbuf);
2129 			}
2130 			mrvl_port_bpool_size
2131 				[bpool->pp2_id][bpool->id][core_id] -= i;
2132 		}
2133 		rte_spinlock_unlock(&q->priv->lock);
2134 	}
2135 
2136 	return rx_done;
2137 }
2138 
2139 /**
2140  * Prepare offload information.
2141  *
2142  * @param ol_flags
2143  *   Offload flags.
2144  * @param packet_type
2145  *   Packet type bitfield.
2146  * @param l3_type
2147  *   Pointer to the pp2_ouq_l3_type structure.
2148  * @param l4_type
2149  *   Pointer to the pp2_outq_l4_type structure.
2150  * @param gen_l3_cksum
2151  *   Will be set to 1 in case l3 checksum is computed.
2152  * @param l4_cksum
2153  *   Will be set to 1 in case l4 checksum is computed.
2154  *
2155  * @return
2156  *   0 on success, negative error value otherwise.
2157  */
2158 static inline int
2159 mrvl_prepare_proto_info(uint64_t ol_flags, uint32_t packet_type,
2160 			enum pp2_outq_l3_type *l3_type,
2161 			enum pp2_outq_l4_type *l4_type,
2162 			int *gen_l3_cksum,
2163 			int *gen_l4_cksum)
2164 {
2165 	/*
2166 	 * Based on ol_flags prepare information
2167 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2168 	 * for offloading.
2169 	 */
2170 	if (ol_flags & PKT_TX_IPV4) {
2171 		*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2172 		*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2173 	} else if (ol_flags & PKT_TX_IPV6) {
2174 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2175 		/* no checksum for ipv6 header */
2176 		*gen_l3_cksum = 0;
2177 	} else {
2178 		/* if something different then stop processing */
2179 		return -1;
2180 	}
2181 
2182 	ol_flags &= PKT_TX_L4_MASK;
2183 	if ((packet_type & RTE_PTYPE_L4_TCP) &&
2184 	    ol_flags == PKT_TX_TCP_CKSUM) {
2185 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2186 		*gen_l4_cksum = 1;
2187 	} else if ((packet_type & RTE_PTYPE_L4_UDP) &&
2188 		   ol_flags == PKT_TX_UDP_CKSUM) {
2189 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2190 		*gen_l4_cksum = 1;
2191 	} else {
2192 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2193 		/* no checksum for other type */
2194 		*gen_l4_cksum = 0;
2195 	}
2196 
2197 	return 0;
2198 }
2199 
2200 /**
2201  * Release already sent buffers to bpool (buffer-pool).
2202  *
2203  * @param ppio
2204  *   Pointer to the port structure.
2205  * @param hif
2206  *   Pointer to the MUSDK hardware interface.
2207  * @param sq
2208  *   Pointer to the shadow queue.
2209  * @param qid
2210  *   Queue id number.
2211  * @param force
2212  *   Force releasing packets.
2213  */
2214 static inline void
2215 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2216 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2217 		       int qid, int force)
2218 {
2219 	struct buff_release_entry *entry;
2220 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2221 	int i;
2222 
2223 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2224 
2225 	sq->num_to_release += nb_done;
2226 
2227 	if (likely(!force &&
2228 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2229 		return;
2230 
2231 	nb_done = sq->num_to_release;
2232 	sq->num_to_release = 0;
2233 
2234 	for (i = 0; i < nb_done; i++) {
2235 		entry = &sq->ent[sq->tail + num];
2236 		if (unlikely(!entry->buff.addr)) {
2237 			MRVL_LOG(ERR,
2238 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2239 				sq->tail, (u64)entry->buff.cookie,
2240 				(u64)entry->buff.addr);
2241 			skip_bufs = 1;
2242 			goto skip;
2243 		}
2244 
2245 		if (unlikely(!entry->bpool)) {
2246 			struct rte_mbuf *mbuf;
2247 
2248 			mbuf = (struct rte_mbuf *)
2249 			       (cookie_addr_high | entry->buff.cookie);
2250 			rte_pktmbuf_free(mbuf);
2251 			skip_bufs = 1;
2252 			goto skip;
2253 		}
2254 
2255 		mrvl_port_bpool_size
2256 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2257 		num++;
2258 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2259 			goto skip;
2260 		continue;
2261 skip:
2262 		if (likely(num))
2263 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2264 		num += skip_bufs;
2265 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2266 		sq->size -= num;
2267 		num = 0;
2268 		skip_bufs = 0;
2269 	}
2270 
2271 	if (likely(num)) {
2272 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2273 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2274 		sq->size -= num;
2275 	}
2276 }
2277 
2278 /**
2279  * DPDK callback for transmit.
2280  *
2281  * @param txq
2282  *   Generic pointer transmit queue.
2283  * @param tx_pkts
2284  *   Packets to transmit.
2285  * @param nb_pkts
2286  *   Number of packets in array.
2287  *
2288  * @return
2289  *   Number of packets successfully transmitted.
2290  */
2291 static uint16_t
2292 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2293 {
2294 	struct mrvl_txq *q = txq;
2295 	struct mrvl_shadow_txq *sq;
2296 	struct pp2_hif *hif;
2297 	struct pp2_ppio_desc descs[nb_pkts];
2298 	unsigned int core_id = rte_lcore_id();
2299 	int i, ret, bytes_sent = 0;
2300 	uint16_t num, sq_free_size;
2301 	uint64_t addr;
2302 
2303 	hif = mrvl_get_hif(q->priv, core_id);
2304 	sq = &q->shadow_txqs[core_id];
2305 
2306 	if (unlikely(!q->priv->ppio || !hif))
2307 		return 0;
2308 
2309 	if (sq->size)
2310 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2311 				       sq, q->queue_id, 0);
2312 
2313 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2314 	if (unlikely(nb_pkts > sq_free_size)) {
2315 		MRVL_LOG(DEBUG,
2316 			"No room in shadow queue for %d packets! %d packets will be sent.",
2317 			nb_pkts, sq_free_size);
2318 		nb_pkts = sq_free_size;
2319 	}
2320 
2321 	for (i = 0; i < nb_pkts; i++) {
2322 		struct rte_mbuf *mbuf = tx_pkts[i];
2323 		int gen_l3_cksum, gen_l4_cksum;
2324 		enum pp2_outq_l3_type l3_type;
2325 		enum pp2_outq_l4_type l4_type;
2326 
2327 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2328 			struct rte_mbuf *pref_pkt_hdr;
2329 
2330 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2331 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2332 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2333 		}
2334 
2335 		sq->ent[sq->head].buff.cookie = (pp2_cookie_t)(uint64_t)mbuf;
2336 		sq->ent[sq->head].buff.addr =
2337 			rte_mbuf_data_iova_default(mbuf);
2338 		sq->ent[sq->head].bpool =
2339 			(unlikely(mbuf->port >= RTE_MAX_ETHPORTS ||
2340 			 mbuf->refcnt > 1)) ? NULL :
2341 			 mrvl_port_to_bpool_lookup[mbuf->port];
2342 		sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
2343 		sq->size++;
2344 
2345 		pp2_ppio_outq_desc_reset(&descs[i]);
2346 		pp2_ppio_outq_desc_set_phys_addr(&descs[i],
2347 						 rte_pktmbuf_iova(mbuf));
2348 		pp2_ppio_outq_desc_set_pkt_offset(&descs[i], 0);
2349 		pp2_ppio_outq_desc_set_pkt_len(&descs[i],
2350 					       rte_pktmbuf_pkt_len(mbuf));
2351 
2352 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2353 		/*
2354 		 * in case unsupported ol_flags were passed
2355 		 * do not update descriptor offload information
2356 		 */
2357 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2358 					      &l3_type, &l4_type, &gen_l3_cksum,
2359 					      &gen_l4_cksum);
2360 		if (unlikely(ret))
2361 			continue;
2362 
2363 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2364 						  mbuf->l2_len,
2365 						  mbuf->l2_len + mbuf->l3_len,
2366 						  gen_l3_cksum, gen_l4_cksum);
2367 	}
2368 
2369 	num = nb_pkts;
2370 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2371 	/* number of packets that were not sent */
2372 	if (unlikely(num > nb_pkts)) {
2373 		for (i = nb_pkts; i < num; i++) {
2374 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2375 				MRVL_PP2_TX_SHADOWQ_MASK;
2376 			addr = cookie_addr_high | sq->ent[sq->head].buff.cookie;
2377 			bytes_sent -=
2378 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2379 		}
2380 		sq->size -= num - nb_pkts;
2381 	}
2382 
2383 	q->bytes_sent += bytes_sent;
2384 
2385 	return nb_pkts;
2386 }
2387 
2388 /**
2389  * Initialize packet processor.
2390  *
2391  * @return
2392  *   0 on success, negative error value otherwise.
2393  */
2394 static int
2395 mrvl_init_pp2(void)
2396 {
2397 	struct pp2_init_params init_params;
2398 
2399 	memset(&init_params, 0, sizeof(init_params));
2400 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
2401 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
2402 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
2403 
2404 	return pp2_init(&init_params);
2405 }
2406 
2407 /**
2408  * Deinitialize packet processor.
2409  *
2410  * @return
2411  *   0 on success, negative error value otherwise.
2412  */
2413 static void
2414 mrvl_deinit_pp2(void)
2415 {
2416 	pp2_deinit();
2417 }
2418 
2419 /**
2420  * Create private device structure.
2421  *
2422  * @param dev_name
2423  *   Pointer to the port name passed in the initialization parameters.
2424  *
2425  * @return
2426  *   Pointer to the newly allocated private device structure.
2427  */
2428 static struct mrvl_priv *
2429 mrvl_priv_create(const char *dev_name)
2430 {
2431 	struct pp2_bpool_params bpool_params;
2432 	char match[MRVL_MATCH_LEN];
2433 	struct mrvl_priv *priv;
2434 	int ret, bpool_bit;
2435 
2436 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
2437 	if (!priv)
2438 		return NULL;
2439 
2440 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
2441 				       &priv->pp_id, &priv->ppio_id);
2442 	if (ret)
2443 		goto out_free_priv;
2444 
2445 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
2446 				     PP2_BPOOL_NUM_POOLS);
2447 	if (bpool_bit < 0)
2448 		goto out_free_priv;
2449 	priv->bpool_bit = bpool_bit;
2450 
2451 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
2452 		 priv->bpool_bit);
2453 	memset(&bpool_params, 0, sizeof(bpool_params));
2454 	bpool_params.match = match;
2455 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
2456 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
2457 	if (ret)
2458 		goto out_clear_bpool_bit;
2459 
2460 	priv->ppio_params.type = PP2_PPIO_T_NIC;
2461 	rte_spinlock_init(&priv->lock);
2462 
2463 	return priv;
2464 out_clear_bpool_bit:
2465 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2466 out_free_priv:
2467 	rte_free(priv);
2468 	return NULL;
2469 }
2470 
2471 /**
2472  * Create device representing Ethernet port.
2473  *
2474  * @param name
2475  *   Pointer to the port's name.
2476  *
2477  * @return
2478  *   0 on success, negative error value otherwise.
2479  */
2480 static int
2481 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
2482 {
2483 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
2484 	struct rte_eth_dev *eth_dev;
2485 	struct mrvl_priv *priv;
2486 	struct ifreq req;
2487 
2488 	eth_dev = rte_eth_dev_allocate(name);
2489 	if (!eth_dev)
2490 		return -ENOMEM;
2491 
2492 	priv = mrvl_priv_create(name);
2493 	if (!priv) {
2494 		ret = -ENOMEM;
2495 		goto out_free_dev;
2496 	}
2497 
2498 	eth_dev->data->mac_addrs =
2499 		rte_zmalloc("mac_addrs",
2500 			    ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
2501 	if (!eth_dev->data->mac_addrs) {
2502 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
2503 		ret = -ENOMEM;
2504 		goto out_free_priv;
2505 	}
2506 
2507 	memset(&req, 0, sizeof(req));
2508 	strcpy(req.ifr_name, name);
2509 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
2510 	if (ret)
2511 		goto out_free_mac;
2512 
2513 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
2514 	       req.ifr_addr.sa_data, ETHER_ADDR_LEN);
2515 
2516 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
2517 	eth_dev->tx_pkt_burst = mrvl_tx_pkt_burst;
2518 	eth_dev->data->kdrv = RTE_KDRV_NONE;
2519 	eth_dev->data->dev_private = priv;
2520 	eth_dev->device = &vdev->device;
2521 	eth_dev->dev_ops = &mrvl_ops;
2522 
2523 	rte_eth_dev_probing_finish(eth_dev);
2524 	return 0;
2525 out_free_mac:
2526 	rte_free(eth_dev->data->mac_addrs);
2527 out_free_dev:
2528 	rte_eth_dev_release_port(eth_dev);
2529 out_free_priv:
2530 	rte_free(priv);
2531 
2532 	return ret;
2533 }
2534 
2535 /**
2536  * Cleanup previously created device representing Ethernet port.
2537  *
2538  * @param name
2539  *   Pointer to the port name.
2540  */
2541 static void
2542 mrvl_eth_dev_destroy(const char *name)
2543 {
2544 	struct rte_eth_dev *eth_dev;
2545 	struct mrvl_priv *priv;
2546 
2547 	eth_dev = rte_eth_dev_allocated(name);
2548 	if (!eth_dev)
2549 		return;
2550 
2551 	priv = eth_dev->data->dev_private;
2552 	pp2_bpool_deinit(priv->bpool);
2553 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2554 	rte_free(priv);
2555 	rte_free(eth_dev->data->mac_addrs);
2556 	rte_eth_dev_release_port(eth_dev);
2557 }
2558 
2559 /**
2560  * Callback used by rte_kvargs_process() during argument parsing.
2561  *
2562  * @param key
2563  *   Pointer to the parsed key (unused).
2564  * @param value
2565  *   Pointer to the parsed value.
2566  * @param extra_args
2567  *   Pointer to the extra arguments which contains address of the
2568  *   table of pointers to parsed interface names.
2569  *
2570  * @return
2571  *   Always 0.
2572  */
2573 static int
2574 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
2575 		 void *extra_args)
2576 {
2577 	struct mrvl_ifnames *ifnames = extra_args;
2578 
2579 	ifnames->names[ifnames->idx++] = value;
2580 
2581 	return 0;
2582 }
2583 
2584 /**
2585  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
2586  */
2587 static void
2588 mrvl_deinit_hifs(void)
2589 {
2590 	int i;
2591 
2592 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
2593 		if (hifs[i])
2594 			pp2_hif_deinit(hifs[i]);
2595 	}
2596 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
2597 	memset(hifs, 0, sizeof(hifs));
2598 }
2599 
2600 /**
2601  * DPDK callback to register the virtual device.
2602  *
2603  * @param vdev
2604  *   Pointer to the virtual device.
2605  *
2606  * @return
2607  *   0 on success, negative error value otherwise.
2608  */
2609 static int
2610 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
2611 {
2612 	struct rte_kvargs *kvlist;
2613 	struct mrvl_ifnames ifnames;
2614 	int ret = -EINVAL;
2615 	uint32_t i, ifnum, cfgnum;
2616 	const char *params;
2617 
2618 	params = rte_vdev_device_args(vdev);
2619 	if (!params)
2620 		return -EINVAL;
2621 
2622 	kvlist = rte_kvargs_parse(params, valid_args);
2623 	if (!kvlist)
2624 		return -EINVAL;
2625 
2626 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
2627 	if (ifnum > RTE_DIM(ifnames.names))
2628 		goto out_free_kvlist;
2629 
2630 	ifnames.idx = 0;
2631 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
2632 			   mrvl_get_ifnames, &ifnames);
2633 
2634 
2635 	/*
2636 	 * The below system initialization should be done only once,
2637 	 * on the first provided configuration file
2638 	 */
2639 	if (!mrvl_qos_cfg) {
2640 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
2641 		MRVL_LOG(INFO, "Parsing config file!");
2642 		if (cfgnum > 1) {
2643 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
2644 			goto out_free_kvlist;
2645 		} else if (cfgnum == 1) {
2646 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
2647 					   mrvl_get_qoscfg, &mrvl_qos_cfg);
2648 		}
2649 	}
2650 
2651 	if (mrvl_dev_num)
2652 		goto init_devices;
2653 
2654 	MRVL_LOG(INFO, "Perform MUSDK initializations");
2655 	/*
2656 	 * ret == -EEXIST is correct, it means DMA
2657 	 * has been already initialized (by another PMD).
2658 	 */
2659 	ret = mv_sys_dma_mem_init(MRVL_MUSDK_DMA_MEMSIZE);
2660 	if (ret < 0) {
2661 		if (ret != -EEXIST)
2662 			goto out_free_kvlist;
2663 		else
2664 			MRVL_LOG(INFO,
2665 				"DMA memory has been already initialized by a different driver.");
2666 	}
2667 
2668 	ret = mrvl_init_pp2();
2669 	if (ret) {
2670 		MRVL_LOG(ERR, "Failed to init PP!");
2671 		goto out_deinit_dma;
2672 	}
2673 
2674 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
2675 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
2676 
2677 	mrvl_lcore_first = RTE_MAX_LCORE;
2678 	mrvl_lcore_last = 0;
2679 
2680 init_devices:
2681 	for (i = 0; i < ifnum; i++) {
2682 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
2683 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
2684 		if (ret)
2685 			goto out_cleanup;
2686 	}
2687 	mrvl_dev_num += ifnum;
2688 
2689 	rte_kvargs_free(kvlist);
2690 
2691 	return 0;
2692 out_cleanup:
2693 	for (; i > 0; i--)
2694 		mrvl_eth_dev_destroy(ifnames.names[i]);
2695 
2696 	if (mrvl_dev_num == 0)
2697 		mrvl_deinit_pp2();
2698 out_deinit_dma:
2699 	if (mrvl_dev_num == 0)
2700 		mv_sys_dma_mem_destroy();
2701 out_free_kvlist:
2702 	rte_kvargs_free(kvlist);
2703 
2704 	return ret;
2705 }
2706 
2707 /**
2708  * DPDK callback to remove virtual device.
2709  *
2710  * @param vdev
2711  *   Pointer to the removed virtual device.
2712  *
2713  * @return
2714  *   0 on success, negative error value otherwise.
2715  */
2716 static int
2717 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
2718 {
2719 	int i;
2720 	const char *name;
2721 
2722 	name = rte_vdev_device_name(vdev);
2723 	if (!name)
2724 		return -EINVAL;
2725 
2726 	MRVL_LOG(INFO, "Removing %s", name);
2727 
2728 	RTE_ETH_FOREACH_DEV(i) { /* FIXME: removing all devices! */
2729 		char ifname[RTE_ETH_NAME_MAX_LEN];
2730 
2731 		rte_eth_dev_get_name_by_port(i, ifname);
2732 		mrvl_eth_dev_destroy(ifname);
2733 		mrvl_dev_num--;
2734 	}
2735 
2736 	if (mrvl_dev_num == 0) {
2737 		MRVL_LOG(INFO, "Perform MUSDK deinit");
2738 		mrvl_deinit_hifs();
2739 		mrvl_deinit_pp2();
2740 		mv_sys_dma_mem_destroy();
2741 	}
2742 
2743 	return 0;
2744 }
2745 
2746 static struct rte_vdev_driver pmd_mrvl_drv = {
2747 	.probe = rte_pmd_mrvl_probe,
2748 	.remove = rte_pmd_mrvl_remove,
2749 };
2750 
2751 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
2752 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
2753 
2754 RTE_INIT(mrvl_init_log);
2755 static void
2756 mrvl_init_log(void)
2757 {
2758 	mrvl_logtype = rte_log_register("pmd.net.mvpp2");
2759 	if (mrvl_logtype >= 0)
2760 		rte_log_set_level(mrvl_logtype, RTE_LOG_NOTICE);
2761 }
2762