xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision 5997b0a827619b085852a00cb78af31ba4edac7c)
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 <rte_mvep_common.h>
33 #include "mrvl_ethdev.h"
34 #include "mrvl_qos.h"
35 
36 /* bitmask with reserved hifs */
37 #define MRVL_MUSDK_HIFS_RESERVED 0x0F
38 /* bitmask with reserved bpools */
39 #define MRVL_MUSDK_BPOOLS_RESERVED 0x07
40 /* bitmask with reserved kernel RSS tables */
41 #define MRVL_MUSDK_RSS_RESERVED 0x01
42 /* maximum number of available hifs */
43 #define MRVL_MUSDK_HIFS_MAX 9
44 
45 /* prefetch shift */
46 #define MRVL_MUSDK_PREFETCH_SHIFT 2
47 
48 /* TCAM has 25 entries reserved for uc/mc filter entries */
49 #define MRVL_MAC_ADDRS_MAX 25
50 #define MRVL_MATCH_LEN 16
51 #define MRVL_PKT_EFFEC_OFFS (MRVL_PKT_OFFS + MV_MH_SIZE)
52 /* Maximum allowable packet size */
53 #define MRVL_PKT_SIZE_MAX (10240 - MV_MH_SIZE)
54 
55 #define MRVL_IFACE_NAME_ARG "iface"
56 #define MRVL_CFG_ARG "cfg"
57 
58 #define MRVL_BURST_SIZE 64
59 
60 #define MRVL_ARP_LENGTH 28
61 
62 #define MRVL_COOKIE_ADDR_INVALID ~0ULL
63 
64 #define MRVL_COOKIE_HIGH_ADDR_SHIFT	(sizeof(pp2_cookie_t) * 8)
65 #define MRVL_COOKIE_HIGH_ADDR_MASK	(~0ULL << MRVL_COOKIE_HIGH_ADDR_SHIFT)
66 
67 /** Port Rx offload capabilities */
68 #define MRVL_RX_OFFLOADS (DEV_RX_OFFLOAD_VLAN_FILTER | \
69 			  DEV_RX_OFFLOAD_JUMBO_FRAME | \
70 			  DEV_RX_OFFLOAD_CHECKSUM)
71 
72 /** Port Tx offloads capabilities */
73 #define MRVL_TX_OFFLOADS (DEV_TX_OFFLOAD_IPV4_CKSUM | \
74 			  DEV_TX_OFFLOAD_UDP_CKSUM | \
75 			  DEV_TX_OFFLOAD_TCP_CKSUM)
76 
77 static const char * const valid_args[] = {
78 	MRVL_IFACE_NAME_ARG,
79 	MRVL_CFG_ARG,
80 	NULL
81 };
82 
83 static int used_hifs = MRVL_MUSDK_HIFS_RESERVED;
84 static struct pp2_hif *hifs[RTE_MAX_LCORE];
85 static int used_bpools[PP2_NUM_PKT_PROC] = {
86 	[0 ... PP2_NUM_PKT_PROC - 1] = MRVL_MUSDK_BPOOLS_RESERVED
87 };
88 
89 static struct pp2_bpool *mrvl_port_to_bpool_lookup[RTE_MAX_ETHPORTS];
90 static int mrvl_port_bpool_size[PP2_NUM_PKT_PROC][PP2_BPOOL_NUM_POOLS][RTE_MAX_LCORE];
91 static uint64_t cookie_addr_high = MRVL_COOKIE_ADDR_INVALID;
92 
93 int mrvl_logtype;
94 
95 struct mrvl_ifnames {
96 	const char *names[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
97 	int idx;
98 };
99 
100 /*
101  * To use buffer harvesting based on loopback port shadow queue structure
102  * was introduced for buffers information bookkeeping.
103  *
104  * Before sending the packet, related buffer information (pp2_buff_inf) is
105  * stored in shadow queue. After packet is transmitted no longer used
106  * packet buffer is released back to it's original hardware pool,
107  * on condition it originated from interface.
108  * In case it  was generated by application itself i.e: mbuf->port field is
109  * 0xff then its released to software mempool.
110  */
111 struct mrvl_shadow_txq {
112 	int head;           /* write index - used when sending buffers */
113 	int tail;           /* read index - used when releasing buffers */
114 	u16 size;           /* queue occupied size */
115 	u16 num_to_release; /* number of buffers sent, that can be released */
116 	struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
117 };
118 
119 struct mrvl_rxq {
120 	struct mrvl_priv *priv;
121 	struct rte_mempool *mp;
122 	int queue_id;
123 	int port_id;
124 	int cksum_enabled;
125 	uint64_t bytes_recv;
126 	uint64_t drop_mac;
127 };
128 
129 struct mrvl_txq {
130 	struct mrvl_priv *priv;
131 	int queue_id;
132 	int port_id;
133 	uint64_t bytes_sent;
134 	struct mrvl_shadow_txq shadow_txqs[RTE_MAX_LCORE];
135 	int tx_deferred_start;
136 };
137 
138 static int mrvl_lcore_first;
139 static int mrvl_lcore_last;
140 static int mrvl_dev_num;
141 
142 static int mrvl_fill_bpool(struct mrvl_rxq *rxq, int num);
143 static inline void mrvl_free_sent_buffers(struct pp2_ppio *ppio,
144 			struct pp2_hif *hif, unsigned int core_id,
145 			struct mrvl_shadow_txq *sq, int qid, int force);
146 
147 #define MRVL_XSTATS_TBL_ENTRY(name) { \
148 	#name, offsetof(struct pp2_ppio_statistics, name),	\
149 	sizeof(((struct pp2_ppio_statistics *)0)->name)		\
150 }
151 
152 /* Table with xstats data */
153 static struct {
154 	const char *name;
155 	unsigned int offset;
156 	unsigned int size;
157 } mrvl_xstats_tbl[] = {
158 	MRVL_XSTATS_TBL_ENTRY(rx_bytes),
159 	MRVL_XSTATS_TBL_ENTRY(rx_packets),
160 	MRVL_XSTATS_TBL_ENTRY(rx_unicast_packets),
161 	MRVL_XSTATS_TBL_ENTRY(rx_errors),
162 	MRVL_XSTATS_TBL_ENTRY(rx_fullq_dropped),
163 	MRVL_XSTATS_TBL_ENTRY(rx_bm_dropped),
164 	MRVL_XSTATS_TBL_ENTRY(rx_early_dropped),
165 	MRVL_XSTATS_TBL_ENTRY(rx_fifo_dropped),
166 	MRVL_XSTATS_TBL_ENTRY(rx_cls_dropped),
167 	MRVL_XSTATS_TBL_ENTRY(tx_bytes),
168 	MRVL_XSTATS_TBL_ENTRY(tx_packets),
169 	MRVL_XSTATS_TBL_ENTRY(tx_unicast_packets),
170 	MRVL_XSTATS_TBL_ENTRY(tx_errors)
171 };
172 
173 static inline int
174 mrvl_get_bpool_size(int pp2_id, int pool_id)
175 {
176 	int i;
177 	int size = 0;
178 
179 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
180 		size += mrvl_port_bpool_size[pp2_id][pool_id][i];
181 
182 	return size;
183 }
184 
185 static inline int
186 mrvl_reserve_bit(int *bitmap, int max)
187 {
188 	int n = sizeof(*bitmap) * 8 - __builtin_clz(*bitmap);
189 
190 	if (n >= max)
191 		return -1;
192 
193 	*bitmap |= 1 << n;
194 
195 	return n;
196 }
197 
198 static int
199 mrvl_init_hif(int core_id)
200 {
201 	struct pp2_hif_params params;
202 	char match[MRVL_MATCH_LEN];
203 	int ret;
204 
205 	ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
206 	if (ret < 0) {
207 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
208 		return ret;
209 	}
210 
211 	snprintf(match, sizeof(match), "hif-%d", ret);
212 	memset(&params, 0, sizeof(params));
213 	params.match = match;
214 	params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
215 	ret = pp2_hif_init(&params, &hifs[core_id]);
216 	if (ret) {
217 		MRVL_LOG(ERR, "Failed to initialize hif %d", core_id);
218 		return ret;
219 	}
220 
221 	return 0;
222 }
223 
224 static inline struct pp2_hif*
225 mrvl_get_hif(struct mrvl_priv *priv, int core_id)
226 {
227 	int ret;
228 
229 	if (likely(hifs[core_id] != NULL))
230 		return hifs[core_id];
231 
232 	rte_spinlock_lock(&priv->lock);
233 
234 	ret = mrvl_init_hif(core_id);
235 	if (ret < 0) {
236 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
237 		goto out;
238 	}
239 
240 	if (core_id < mrvl_lcore_first)
241 		mrvl_lcore_first = core_id;
242 
243 	if (core_id > mrvl_lcore_last)
244 		mrvl_lcore_last = core_id;
245 out:
246 	rte_spinlock_unlock(&priv->lock);
247 
248 	return hifs[core_id];
249 }
250 
251 /**
252  * Configure rss based on dpdk rss configuration.
253  *
254  * @param priv
255  *   Pointer to private structure.
256  * @param rss_conf
257  *   Pointer to RSS configuration.
258  *
259  * @return
260  *   0 on success, negative error value otherwise.
261  */
262 static int
263 mrvl_configure_rss(struct mrvl_priv *priv, struct rte_eth_rss_conf *rss_conf)
264 {
265 	if (rss_conf->rss_key)
266 		MRVL_LOG(WARNING, "Changing hash key is not supported");
267 
268 	if (rss_conf->rss_hf == 0) {
269 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
270 	} else if (rss_conf->rss_hf & ETH_RSS_IPV4) {
271 		priv->ppio_params.inqs_params.hash_type =
272 			PP2_PPIO_HASH_T_2_TUPLE;
273 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP) {
274 		priv->ppio_params.inqs_params.hash_type =
275 			PP2_PPIO_HASH_T_5_TUPLE;
276 		priv->rss_hf_tcp = 1;
277 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) {
278 		priv->ppio_params.inqs_params.hash_type =
279 			PP2_PPIO_HASH_T_5_TUPLE;
280 		priv->rss_hf_tcp = 0;
281 	} else {
282 		return -EINVAL;
283 	}
284 
285 	return 0;
286 }
287 
288 /**
289  * Ethernet device configuration.
290  *
291  * Prepare the driver for a given number of TX and RX queues and
292  * configure RSS.
293  *
294  * @param dev
295  *   Pointer to Ethernet device structure.
296  *
297  * @return
298  *   0 on success, negative error value otherwise.
299  */
300 static int
301 mrvl_dev_configure(struct rte_eth_dev *dev)
302 {
303 	struct mrvl_priv *priv = dev->data->dev_private;
304 	int ret;
305 
306 	if (priv->ppio) {
307 		MRVL_LOG(INFO, "Device reconfiguration is not supported");
308 		return -EINVAL;
309 	}
310 
311 	if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_NONE &&
312 	    dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
313 		MRVL_LOG(INFO, "Unsupported rx multi queue mode %d",
314 			dev->data->dev_conf.rxmode.mq_mode);
315 		return -EINVAL;
316 	}
317 
318 	if (dev->data->dev_conf.rxmode.split_hdr_size) {
319 		MRVL_LOG(INFO, "Split headers not supported");
320 		return -EINVAL;
321 	}
322 
323 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
324 		dev->data->mtu = dev->data->dev_conf.rxmode.max_rx_pkt_len -
325 				 ETHER_HDR_LEN - ETHER_CRC_LEN;
326 
327 	ret = mrvl_configure_rxqs(priv, dev->data->port_id,
328 				  dev->data->nb_rx_queues);
329 	if (ret < 0)
330 		return ret;
331 
332 	ret = mrvl_configure_txqs(priv, dev->data->port_id,
333 				  dev->data->nb_tx_queues);
334 	if (ret < 0)
335 		return ret;
336 
337 	priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
338 	priv->ppio_params.maintain_stats = 1;
339 	priv->nb_rx_queues = dev->data->nb_rx_queues;
340 
341 	if (dev->data->nb_rx_queues == 1 &&
342 	    dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
343 		MRVL_LOG(WARNING, "Disabling hash for 1 rx queue");
344 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
345 
346 		return 0;
347 	}
348 
349 	return mrvl_configure_rss(priv,
350 				  &dev->data->dev_conf.rx_adv_conf.rss_conf);
351 }
352 
353 /**
354  * DPDK callback to change the MTU.
355  *
356  * Setting the MTU affects hardware MRU (packets larger than the MRU
357  * will be dropped).
358  *
359  * @param dev
360  *   Pointer to Ethernet device structure.
361  * @param mtu
362  *   New MTU.
363  *
364  * @return
365  *   0 on success, negative error value otherwise.
366  */
367 static int
368 mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
369 {
370 	struct mrvl_priv *priv = dev->data->dev_private;
371 	/* extra MV_MH_SIZE bytes are required for Marvell tag */
372 	uint16_t mru = mtu + MV_MH_SIZE + ETHER_HDR_LEN + ETHER_CRC_LEN;
373 	int ret;
374 
375 	if (mtu < ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX)
376 		return -EINVAL;
377 
378 	if (!priv->ppio)
379 		return 0;
380 
381 	ret = pp2_ppio_set_mru(priv->ppio, mru);
382 	if (ret)
383 		return ret;
384 
385 	return pp2_ppio_set_mtu(priv->ppio, mtu);
386 }
387 
388 /**
389  * DPDK callback to bring the link up.
390  *
391  * @param dev
392  *   Pointer to Ethernet device structure.
393  *
394  * @return
395  *   0 on success, negative error value otherwise.
396  */
397 static int
398 mrvl_dev_set_link_up(struct rte_eth_dev *dev)
399 {
400 	struct mrvl_priv *priv = dev->data->dev_private;
401 	int ret;
402 
403 	if (!priv->ppio)
404 		return -EPERM;
405 
406 	ret = pp2_ppio_enable(priv->ppio);
407 	if (ret)
408 		return ret;
409 
410 	/*
411 	 * mtu/mru can be updated if pp2_ppio_enable() was called at least once
412 	 * as pp2_ppio_enable() changes port->t_mode from default 0 to
413 	 * PP2_TRAFFIC_INGRESS_EGRESS.
414 	 *
415 	 * Set mtu to default DPDK value here.
416 	 */
417 	ret = mrvl_mtu_set(dev, dev->data->mtu);
418 	if (ret)
419 		pp2_ppio_disable(priv->ppio);
420 
421 	return ret;
422 }
423 
424 /**
425  * DPDK callback to bring the link down.
426  *
427  * @param dev
428  *   Pointer to Ethernet device structure.
429  *
430  * @return
431  *   0 on success, negative error value otherwise.
432  */
433 static int
434 mrvl_dev_set_link_down(struct rte_eth_dev *dev)
435 {
436 	struct mrvl_priv *priv = dev->data->dev_private;
437 
438 	if (!priv->ppio)
439 		return -EPERM;
440 
441 	return pp2_ppio_disable(priv->ppio);
442 }
443 
444 /**
445  * DPDK callback to start tx queue.
446  *
447  * @param dev
448  *   Pointer to Ethernet device structure.
449  * @param queue_id
450  *   Transmit queue index.
451  *
452  * @return
453  *   0 on success, negative error value otherwise.
454  */
455 static int
456 mrvl_tx_queue_start(struct rte_eth_dev *dev, uint16_t queue_id)
457 {
458 	struct mrvl_priv *priv = dev->data->dev_private;
459 	int ret;
460 
461 	if (!priv)
462 		return -EPERM;
463 
464 	/* passing 1 enables given tx queue */
465 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 1);
466 	if (ret) {
467 		MRVL_LOG(ERR, "Failed to start txq %d", queue_id);
468 		return ret;
469 	}
470 
471 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
472 
473 	return 0;
474 }
475 
476 /**
477  * DPDK callback to stop tx queue.
478  *
479  * @param dev
480  *   Pointer to Ethernet device structure.
481  * @param queue_id
482  *   Transmit queue index.
483  *
484  * @return
485  *   0 on success, negative error value otherwise.
486  */
487 static int
488 mrvl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t queue_id)
489 {
490 	struct mrvl_priv *priv = dev->data->dev_private;
491 	int ret;
492 
493 	if (!priv->ppio)
494 		return -EPERM;
495 
496 	/* passing 0 disables given tx queue */
497 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 0);
498 	if (ret) {
499 		MRVL_LOG(ERR, "Failed to stop txq %d", queue_id);
500 		return ret;
501 	}
502 
503 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
504 
505 	return 0;
506 }
507 
508 /**
509  * DPDK callback to start the device.
510  *
511  * @param dev
512  *   Pointer to Ethernet device structure.
513  *
514  * @return
515  *   0 on success, negative errno value on failure.
516  */
517 static int
518 mrvl_dev_start(struct rte_eth_dev *dev)
519 {
520 	struct mrvl_priv *priv = dev->data->dev_private;
521 	char match[MRVL_MATCH_LEN];
522 	int ret = 0, i, def_init_size;
523 
524 	if (priv->ppio)
525 		return mrvl_dev_set_link_up(dev);
526 
527 	snprintf(match, sizeof(match), "ppio-%d:%d",
528 		 priv->pp_id, priv->ppio_id);
529 	priv->ppio_params.match = match;
530 
531 	/*
532 	 * Calculate the minimum bpool size for refill feature as follows:
533 	 * 2 default burst sizes multiply by number of rx queues.
534 	 * If the bpool size will be below this value, new buffers will
535 	 * be added to the pool.
536 	 */
537 	priv->bpool_min_size = priv->nb_rx_queues * MRVL_BURST_SIZE * 2;
538 
539 	/* In case initial bpool size configured in queues setup is
540 	 * smaller than minimum size add more buffers
541 	 */
542 	def_init_size = priv->bpool_min_size + MRVL_BURST_SIZE * 2;
543 	if (priv->bpool_init_size < def_init_size) {
544 		int buffs_to_add = def_init_size - priv->bpool_init_size;
545 
546 		priv->bpool_init_size += buffs_to_add;
547 		ret = mrvl_fill_bpool(dev->data->rx_queues[0], buffs_to_add);
548 		if (ret)
549 			MRVL_LOG(ERR, "Failed to add buffers to bpool");
550 	}
551 
552 	/*
553 	 * Calculate the maximum bpool size for refill feature as follows:
554 	 * maximum number of descriptors in rx queue multiply by number
555 	 * of rx queues plus minimum bpool size.
556 	 * In case the bpool size will exceed this value, superfluous buffers
557 	 * will be removed
558 	 */
559 	priv->bpool_max_size = (priv->nb_rx_queues * MRVL_PP2_RXD_MAX) +
560 				priv->bpool_min_size;
561 
562 	ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
563 	if (ret) {
564 		MRVL_LOG(ERR, "Failed to init ppio");
565 		return ret;
566 	}
567 
568 	/*
569 	 * In case there are some some stale uc/mc mac addresses flush them
570 	 * here. It cannot be done during mrvl_dev_close() as port information
571 	 * is already gone at that point (due to pp2_ppio_deinit() in
572 	 * mrvl_dev_stop()).
573 	 */
574 	if (!priv->uc_mc_flushed) {
575 		ret = pp2_ppio_flush_mac_addrs(priv->ppio, 1, 1);
576 		if (ret) {
577 			MRVL_LOG(ERR,
578 				"Failed to flush uc/mc filter list");
579 			goto out;
580 		}
581 		priv->uc_mc_flushed = 1;
582 	}
583 
584 	if (!priv->vlan_flushed) {
585 		ret = pp2_ppio_flush_vlan(priv->ppio);
586 		if (ret) {
587 			MRVL_LOG(ERR, "Failed to flush vlan list");
588 			/*
589 			 * TODO
590 			 * once pp2_ppio_flush_vlan() is supported jump to out
591 			 * goto out;
592 			 */
593 		}
594 		priv->vlan_flushed = 1;
595 	}
596 
597 	/* For default QoS config, don't start classifier. */
598 	if (mrvl_qos_cfg) {
599 		ret = mrvl_start_qos_mapping(priv);
600 		if (ret) {
601 			MRVL_LOG(ERR, "Failed to setup QoS mapping");
602 			goto out;
603 		}
604 	}
605 
606 	ret = mrvl_dev_set_link_up(dev);
607 	if (ret) {
608 		MRVL_LOG(ERR, "Failed to set link up");
609 		goto out;
610 	}
611 
612 	/* start tx queues */
613 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
614 		struct mrvl_txq *txq = dev->data->tx_queues[i];
615 
616 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
617 
618 		if (!txq->tx_deferred_start)
619 			continue;
620 
621 		/*
622 		 * All txqs are started by default. Stop them
623 		 * so that tx_deferred_start works as expected.
624 		 */
625 		ret = mrvl_tx_queue_stop(dev, i);
626 		if (ret)
627 			goto out;
628 	}
629 
630 	return 0;
631 out:
632 	MRVL_LOG(ERR, "Failed to start device");
633 	pp2_ppio_deinit(priv->ppio);
634 	return ret;
635 }
636 
637 /**
638  * Flush receive queues.
639  *
640  * @param dev
641  *   Pointer to Ethernet device structure.
642  */
643 static void
644 mrvl_flush_rx_queues(struct rte_eth_dev *dev)
645 {
646 	int i;
647 
648 	MRVL_LOG(INFO, "Flushing rx queues");
649 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
650 		int ret, num;
651 
652 		do {
653 			struct mrvl_rxq *q = dev->data->rx_queues[i];
654 			struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];
655 
656 			num = MRVL_PP2_RXD_MAX;
657 			ret = pp2_ppio_recv(q->priv->ppio,
658 					    q->priv->rxq_map[q->queue_id].tc,
659 					    q->priv->rxq_map[q->queue_id].inq,
660 					    descs, (uint16_t *)&num);
661 		} while (ret == 0 && num);
662 	}
663 }
664 
665 /**
666  * Flush transmit shadow queues.
667  *
668  * @param dev
669  *   Pointer to Ethernet device structure.
670  */
671 static void
672 mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
673 {
674 	int i, j;
675 	struct mrvl_txq *txq;
676 
677 	MRVL_LOG(INFO, "Flushing tx shadow queues");
678 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
679 		txq = (struct mrvl_txq *)dev->data->tx_queues[i];
680 
681 		for (j = 0; j < RTE_MAX_LCORE; j++) {
682 			struct mrvl_shadow_txq *sq;
683 
684 			if (!hifs[j])
685 				continue;
686 
687 			sq = &txq->shadow_txqs[j];
688 			mrvl_free_sent_buffers(txq->priv->ppio,
689 				hifs[j], j, sq, txq->queue_id, 1);
690 			while (sq->tail != sq->head) {
691 				uint64_t addr = cookie_addr_high |
692 					sq->ent[sq->tail].buff.cookie;
693 				rte_pktmbuf_free(
694 					(struct rte_mbuf *)addr);
695 				sq->tail = (sq->tail + 1) &
696 					    MRVL_PP2_TX_SHADOWQ_MASK;
697 			}
698 			memset(sq, 0, sizeof(*sq));
699 		}
700 	}
701 }
702 
703 /**
704  * Flush hardware bpool (buffer-pool).
705  *
706  * @param dev
707  *   Pointer to Ethernet device structure.
708  */
709 static void
710 mrvl_flush_bpool(struct rte_eth_dev *dev)
711 {
712 	struct mrvl_priv *priv = dev->data->dev_private;
713 	struct pp2_hif *hif;
714 	uint32_t num;
715 	int ret;
716 	unsigned int core_id = rte_lcore_id();
717 
718 	if (core_id == LCORE_ID_ANY)
719 		core_id = 0;
720 
721 	hif = mrvl_get_hif(priv, core_id);
722 
723 	ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
724 	if (ret) {
725 		MRVL_LOG(ERR, "Failed to get bpool buffers number");
726 		return;
727 	}
728 
729 	while (num--) {
730 		struct pp2_buff_inf inf;
731 		uint64_t addr;
732 
733 		ret = pp2_bpool_get_buff(hif, priv->bpool, &inf);
734 		if (ret)
735 			break;
736 
737 		addr = cookie_addr_high | inf.cookie;
738 		rte_pktmbuf_free((struct rte_mbuf *)addr);
739 	}
740 }
741 
742 /**
743  * DPDK callback to stop the device.
744  *
745  * @param dev
746  *   Pointer to Ethernet device structure.
747  */
748 static void
749 mrvl_dev_stop(struct rte_eth_dev *dev)
750 {
751 	mrvl_dev_set_link_down(dev);
752 }
753 
754 /**
755  * DPDK callback to close the device.
756  *
757  * @param dev
758  *   Pointer to Ethernet device structure.
759  */
760 static void
761 mrvl_dev_close(struct rte_eth_dev *dev)
762 {
763 	struct mrvl_priv *priv = dev->data->dev_private;
764 	size_t i;
765 
766 	mrvl_flush_rx_queues(dev);
767 	mrvl_flush_tx_shadow_queues(dev);
768 
769 	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
770 		struct pp2_ppio_tc_params *tc_params =
771 			&priv->ppio_params.inqs_params.tcs_params[i];
772 
773 		if (tc_params->inqs_params) {
774 			rte_free(tc_params->inqs_params);
775 			tc_params->inqs_params = NULL;
776 		}
777 	}
778 
779 	if (priv->cls_tbl) {
780 		pp2_cls_tbl_deinit(priv->cls_tbl);
781 		priv->cls_tbl = NULL;
782 	}
783 
784 	if (priv->qos_tbl) {
785 		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
786 		priv->qos_tbl = NULL;
787 	}
788 
789 	mrvl_flush_bpool(dev);
790 
791 	if (priv->ppio) {
792 		pp2_ppio_deinit(priv->ppio);
793 		priv->ppio = NULL;
794 	}
795 
796 	/* policer must be released after ppio deinitialization */
797 	if (priv->policer) {
798 		pp2_cls_plcr_deinit(priv->policer);
799 		priv->policer = NULL;
800 	}
801 }
802 
803 /**
804  * DPDK callback to retrieve physical link information.
805  *
806  * @param dev
807  *   Pointer to Ethernet device structure.
808  * @param wait_to_complete
809  *   Wait for request completion (ignored).
810  *
811  * @return
812  *   0 on success, negative error value otherwise.
813  */
814 static int
815 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
816 {
817 	/*
818 	 * TODO
819 	 * once MUSDK provides necessary API use it here
820 	 */
821 	struct mrvl_priv *priv = dev->data->dev_private;
822 	struct ethtool_cmd edata;
823 	struct ifreq req;
824 	int ret, fd, link_up;
825 
826 	if (!priv->ppio)
827 		return -EPERM;
828 
829 	edata.cmd = ETHTOOL_GSET;
830 
831 	strcpy(req.ifr_name, dev->data->name);
832 	req.ifr_data = (void *)&edata;
833 
834 	fd = socket(AF_INET, SOCK_DGRAM, 0);
835 	if (fd == -1)
836 		return -EFAULT;
837 
838 	ret = ioctl(fd, SIOCETHTOOL, &req);
839 	if (ret == -1) {
840 		close(fd);
841 		return -EFAULT;
842 	}
843 
844 	close(fd);
845 
846 	switch (ethtool_cmd_speed(&edata)) {
847 	case SPEED_10:
848 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
849 		break;
850 	case SPEED_100:
851 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
852 		break;
853 	case SPEED_1000:
854 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
855 		break;
856 	case SPEED_10000:
857 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
858 		break;
859 	default:
860 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
861 	}
862 
863 	dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
864 							 ETH_LINK_HALF_DUPLEX;
865 	dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
866 							   ETH_LINK_FIXED;
867 	pp2_ppio_get_link_state(priv->ppio, &link_up);
868 	dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
869 
870 	return 0;
871 }
872 
873 /**
874  * DPDK callback to enable promiscuous mode.
875  *
876  * @param dev
877  *   Pointer to Ethernet device structure.
878  */
879 static void
880 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
881 {
882 	struct mrvl_priv *priv = dev->data->dev_private;
883 	int ret;
884 
885 	if (!priv->ppio)
886 		return;
887 
888 	if (priv->isolated)
889 		return;
890 
891 	ret = pp2_ppio_set_promisc(priv->ppio, 1);
892 	if (ret)
893 		MRVL_LOG(ERR, "Failed to enable promiscuous mode");
894 }
895 
896 /**
897  * DPDK callback to enable allmulti mode.
898  *
899  * @param dev
900  *   Pointer to Ethernet device structure.
901  */
902 static void
903 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
904 {
905 	struct mrvl_priv *priv = dev->data->dev_private;
906 	int ret;
907 
908 	if (!priv->ppio)
909 		return;
910 
911 	if (priv->isolated)
912 		return;
913 
914 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
915 	if (ret)
916 		MRVL_LOG(ERR, "Failed enable all-multicast mode");
917 }
918 
919 /**
920  * DPDK callback to disable promiscuous mode.
921  *
922  * @param dev
923  *   Pointer to Ethernet device structure.
924  */
925 static void
926 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
927 {
928 	struct mrvl_priv *priv = dev->data->dev_private;
929 	int ret;
930 
931 	if (!priv->ppio)
932 		return;
933 
934 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
935 	if (ret)
936 		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
937 }
938 
939 /**
940  * DPDK callback to disable allmulticast mode.
941  *
942  * @param dev
943  *   Pointer to Ethernet device structure.
944  */
945 static void
946 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
947 {
948 	struct mrvl_priv *priv = dev->data->dev_private;
949 	int ret;
950 
951 	if (!priv->ppio)
952 		return;
953 
954 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
955 	if (ret)
956 		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
957 }
958 
959 /**
960  * DPDK callback to remove a MAC address.
961  *
962  * @param dev
963  *   Pointer to Ethernet device structure.
964  * @param index
965  *   MAC address index.
966  */
967 static void
968 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
969 {
970 	struct mrvl_priv *priv = dev->data->dev_private;
971 	char buf[ETHER_ADDR_FMT_SIZE];
972 	int ret;
973 
974 	if (!priv->ppio)
975 		return;
976 
977 	if (priv->isolated)
978 		return;
979 
980 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
981 				       dev->data->mac_addrs[index].addr_bytes);
982 	if (ret) {
983 		ether_format_addr(buf, sizeof(buf),
984 				  &dev->data->mac_addrs[index]);
985 		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
986 	}
987 }
988 
989 /**
990  * DPDK callback to add a MAC address.
991  *
992  * @param dev
993  *   Pointer to Ethernet device structure.
994  * @param mac_addr
995  *   MAC address to register.
996  * @param index
997  *   MAC address index.
998  * @param vmdq
999  *   VMDq pool index to associate address with (unused).
1000  *
1001  * @return
1002  *   0 on success, negative error value otherwise.
1003  */
1004 static int
1005 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1006 		  uint32_t index, uint32_t vmdq __rte_unused)
1007 {
1008 	struct mrvl_priv *priv = dev->data->dev_private;
1009 	char buf[ETHER_ADDR_FMT_SIZE];
1010 	int ret;
1011 
1012 	if (priv->isolated)
1013 		return -ENOTSUP;
1014 
1015 	if (index == 0)
1016 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1017 		return -1;
1018 
1019 	if (!priv->ppio)
1020 		return 0;
1021 
1022 	/*
1023 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1024 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1025 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1026 	 * 21 respectively.
1027 	 *
1028 	 * If more than uc_filter_max uc addresses were added to filter list
1029 	 * then NIC will switch to promiscuous mode automatically.
1030 	 *
1031 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1032 	 * were added to filter list then NIC will switch to all-multicast mode
1033 	 * automatically.
1034 	 */
1035 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1036 	if (ret) {
1037 		ether_format_addr(buf, sizeof(buf), mac_addr);
1038 		MRVL_LOG(ERR, "Failed to add mac %s", buf);
1039 		return -1;
1040 	}
1041 
1042 	return 0;
1043 }
1044 
1045 /**
1046  * DPDK callback to set the primary MAC address.
1047  *
1048  * @param dev
1049  *   Pointer to Ethernet device structure.
1050  * @param mac_addr
1051  *   MAC address to register.
1052  *
1053  * @return
1054  *   0 on success, negative error value otherwise.
1055  */
1056 static int
1057 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1058 {
1059 	struct mrvl_priv *priv = dev->data->dev_private;
1060 	int ret;
1061 
1062 	if (!priv->ppio)
1063 		return 0;
1064 
1065 	if (priv->isolated)
1066 		return -ENOTSUP;
1067 
1068 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1069 	if (ret) {
1070 		char buf[ETHER_ADDR_FMT_SIZE];
1071 		ether_format_addr(buf, sizeof(buf), mac_addr);
1072 		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1073 	}
1074 
1075 	return ret;
1076 }
1077 
1078 /**
1079  * DPDK callback to get device statistics.
1080  *
1081  * @param dev
1082  *   Pointer to Ethernet device structure.
1083  * @param stats
1084  *   Stats structure output buffer.
1085  *
1086  * @return
1087  *   0 on success, negative error value otherwise.
1088  */
1089 static int
1090 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1091 {
1092 	struct mrvl_priv *priv = dev->data->dev_private;
1093 	struct pp2_ppio_statistics ppio_stats;
1094 	uint64_t drop_mac = 0;
1095 	unsigned int i, idx, ret;
1096 
1097 	if (!priv->ppio)
1098 		return -EPERM;
1099 
1100 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1101 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1102 		struct pp2_ppio_inq_statistics rx_stats;
1103 
1104 		if (!rxq)
1105 			continue;
1106 
1107 		idx = rxq->queue_id;
1108 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1109 			MRVL_LOG(ERR,
1110 				"rx queue %d stats out of range (0 - %d)",
1111 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1112 			continue;
1113 		}
1114 
1115 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1116 						  priv->rxq_map[idx].tc,
1117 						  priv->rxq_map[idx].inq,
1118 						  &rx_stats, 0);
1119 		if (unlikely(ret)) {
1120 			MRVL_LOG(ERR,
1121 				"Failed to update rx queue %d stats", idx);
1122 			break;
1123 		}
1124 
1125 		stats->q_ibytes[idx] = rxq->bytes_recv;
1126 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1127 		stats->q_errors[idx] = rx_stats.drop_early +
1128 				       rx_stats.drop_fullq +
1129 				       rx_stats.drop_bm +
1130 				       rxq->drop_mac;
1131 		stats->ibytes += rxq->bytes_recv;
1132 		drop_mac += rxq->drop_mac;
1133 	}
1134 
1135 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1136 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1137 		struct pp2_ppio_outq_statistics tx_stats;
1138 
1139 		if (!txq)
1140 			continue;
1141 
1142 		idx = txq->queue_id;
1143 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1144 			MRVL_LOG(ERR,
1145 				"tx queue %d stats out of range (0 - %d)",
1146 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1147 		}
1148 
1149 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1150 						   &tx_stats, 0);
1151 		if (unlikely(ret)) {
1152 			MRVL_LOG(ERR,
1153 				"Failed to update tx queue %d stats", idx);
1154 			break;
1155 		}
1156 
1157 		stats->q_opackets[idx] = tx_stats.deq_desc;
1158 		stats->q_obytes[idx] = txq->bytes_sent;
1159 		stats->obytes += txq->bytes_sent;
1160 	}
1161 
1162 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1163 	if (unlikely(ret)) {
1164 		MRVL_LOG(ERR, "Failed to update port statistics");
1165 		return ret;
1166 	}
1167 
1168 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1169 	stats->opackets += ppio_stats.tx_packets;
1170 	stats->imissed += ppio_stats.rx_fullq_dropped +
1171 			  ppio_stats.rx_bm_dropped +
1172 			  ppio_stats.rx_early_dropped +
1173 			  ppio_stats.rx_fifo_dropped +
1174 			  ppio_stats.rx_cls_dropped;
1175 	stats->ierrors = drop_mac;
1176 
1177 	return 0;
1178 }
1179 
1180 /**
1181  * DPDK callback to clear device statistics.
1182  *
1183  * @param dev
1184  *   Pointer to Ethernet device structure.
1185  */
1186 static void
1187 mrvl_stats_reset(struct rte_eth_dev *dev)
1188 {
1189 	struct mrvl_priv *priv = dev->data->dev_private;
1190 	int i;
1191 
1192 	if (!priv->ppio)
1193 		return;
1194 
1195 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1196 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1197 
1198 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1199 					    priv->rxq_map[i].inq, NULL, 1);
1200 		rxq->bytes_recv = 0;
1201 		rxq->drop_mac = 0;
1202 	}
1203 
1204 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1205 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1206 
1207 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1208 		txq->bytes_sent = 0;
1209 	}
1210 
1211 	pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1212 }
1213 
1214 /**
1215  * DPDK callback to get extended statistics.
1216  *
1217  * @param dev
1218  *   Pointer to Ethernet device structure.
1219  * @param stats
1220  *   Pointer to xstats table.
1221  * @param n
1222  *   Number of entries in xstats table.
1223  * @return
1224  *   Negative value on error, number of read xstats otherwise.
1225  */
1226 static int
1227 mrvl_xstats_get(struct rte_eth_dev *dev,
1228 		struct rte_eth_xstat *stats, unsigned int n)
1229 {
1230 	struct mrvl_priv *priv = dev->data->dev_private;
1231 	struct pp2_ppio_statistics ppio_stats;
1232 	unsigned int i;
1233 
1234 	if (!stats)
1235 		return 0;
1236 
1237 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1238 	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1239 		uint64_t val;
1240 
1241 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1242 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1243 					    mrvl_xstats_tbl[i].offset);
1244 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1245 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1246 					    mrvl_xstats_tbl[i].offset);
1247 		else
1248 			return -EINVAL;
1249 
1250 		stats[i].id = i;
1251 		stats[i].value = val;
1252 	}
1253 
1254 	return n;
1255 }
1256 
1257 /**
1258  * DPDK callback to reset extended statistics.
1259  *
1260  * @param dev
1261  *   Pointer to Ethernet device structure.
1262  */
1263 static void
1264 mrvl_xstats_reset(struct rte_eth_dev *dev)
1265 {
1266 	mrvl_stats_reset(dev);
1267 }
1268 
1269 /**
1270  * DPDK callback to get extended statistics names.
1271  *
1272  * @param dev (unused)
1273  *   Pointer to Ethernet device structure.
1274  * @param xstats_names
1275  *   Pointer to xstats names table.
1276  * @param size
1277  *   Size of the xstats names table.
1278  * @return
1279  *   Number of read names.
1280  */
1281 static int
1282 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1283 		      struct rte_eth_xstat_name *xstats_names,
1284 		      unsigned int size)
1285 {
1286 	unsigned int i;
1287 
1288 	if (!xstats_names)
1289 		return RTE_DIM(mrvl_xstats_tbl);
1290 
1291 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1292 		snprintf(xstats_names[i].name, RTE_ETH_XSTATS_NAME_SIZE, "%s",
1293 			 mrvl_xstats_tbl[i].name);
1294 
1295 	return size;
1296 }
1297 
1298 /**
1299  * DPDK callback to get information about the device.
1300  *
1301  * @param dev
1302  *   Pointer to Ethernet device structure (unused).
1303  * @param info
1304  *   Info structure output buffer.
1305  */
1306 static void
1307 mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
1308 		   struct rte_eth_dev_info *info)
1309 {
1310 	info->speed_capa = ETH_LINK_SPEED_10M |
1311 			   ETH_LINK_SPEED_100M |
1312 			   ETH_LINK_SPEED_1G |
1313 			   ETH_LINK_SPEED_10G;
1314 
1315 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1316 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1317 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1318 
1319 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1320 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1321 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1322 
1323 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1324 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1325 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1326 
1327 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1328 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1329 
1330 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1331 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1332 
1333 	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1334 				       ETH_RSS_NONFRAG_IPV4_TCP |
1335 				       ETH_RSS_NONFRAG_IPV4_UDP;
1336 
1337 	/* By default packets are dropped if no descriptors are available */
1338 	info->default_rxconf.rx_drop_en = 1;
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_L2_ETHER_VLAN,
1358 		RTE_PTYPE_L2_ETHER_QINQ,
1359 		RTE_PTYPE_L3_IPV4,
1360 		RTE_PTYPE_L3_IPV4_EXT,
1361 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1362 		RTE_PTYPE_L3_IPV6,
1363 		RTE_PTYPE_L3_IPV6_EXT,
1364 		RTE_PTYPE_L2_ETHER_ARP,
1365 		RTE_PTYPE_L4_TCP,
1366 		RTE_PTYPE_L4_UDP
1367 	};
1368 
1369 	return ptypes;
1370 }
1371 
1372 /**
1373  * DPDK callback to get information about specific receive queue.
1374  *
1375  * @param dev
1376  *   Pointer to Ethernet device structure.
1377  * @param rx_queue_id
1378  *   Receive queue index.
1379  * @param qinfo
1380  *   Receive queue information structure.
1381  */
1382 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1383 			      struct rte_eth_rxq_info *qinfo)
1384 {
1385 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1386 	struct mrvl_priv *priv = dev->data->dev_private;
1387 	int inq = priv->rxq_map[rx_queue_id].inq;
1388 	int tc = priv->rxq_map[rx_queue_id].tc;
1389 	struct pp2_ppio_tc_params *tc_params =
1390 		&priv->ppio_params.inqs_params.tcs_params[tc];
1391 
1392 	qinfo->mp = q->mp;
1393 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1394 }
1395 
1396 /**
1397  * DPDK callback to get information about specific transmit queue.
1398  *
1399  * @param dev
1400  *   Pointer to Ethernet device structure.
1401  * @param tx_queue_id
1402  *   Transmit queue index.
1403  * @param qinfo
1404  *   Transmit queue information structure.
1405  */
1406 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1407 			      struct rte_eth_txq_info *qinfo)
1408 {
1409 	struct mrvl_priv *priv = dev->data->dev_private;
1410 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1411 
1412 	qinfo->nb_desc =
1413 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1414 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1415 }
1416 
1417 /**
1418  * DPDK callback to Configure a VLAN filter.
1419  *
1420  * @param dev
1421  *   Pointer to Ethernet device structure.
1422  * @param vlan_id
1423  *   VLAN ID to filter.
1424  * @param on
1425  *   Toggle filter.
1426  *
1427  * @return
1428  *   0 on success, negative error value otherwise.
1429  */
1430 static int
1431 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1432 {
1433 	struct mrvl_priv *priv = dev->data->dev_private;
1434 
1435 	if (!priv->ppio)
1436 		return -EPERM;
1437 
1438 	if (priv->isolated)
1439 		return -ENOTSUP;
1440 
1441 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1442 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1443 }
1444 
1445 /**
1446  * Release buffers to hardware bpool (buffer-pool)
1447  *
1448  * @param rxq
1449  *   Receive queue pointer.
1450  * @param num
1451  *   Number of buffers to release to bpool.
1452  *
1453  * @return
1454  *   0 on success, negative error value otherwise.
1455  */
1456 static int
1457 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1458 {
1459 	struct buff_release_entry entries[MRVL_PP2_RXD_MAX];
1460 	struct rte_mbuf *mbufs[MRVL_PP2_RXD_MAX];
1461 	int i, ret;
1462 	unsigned int core_id;
1463 	struct pp2_hif *hif;
1464 	struct pp2_bpool *bpool;
1465 
1466 	core_id = rte_lcore_id();
1467 	if (core_id == LCORE_ID_ANY)
1468 		core_id = 0;
1469 
1470 	hif = mrvl_get_hif(rxq->priv, core_id);
1471 	if (!hif)
1472 		return -1;
1473 
1474 	bpool = rxq->priv->bpool;
1475 
1476 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1477 	if (ret)
1478 		return ret;
1479 
1480 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1481 		cookie_addr_high =
1482 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1483 
1484 	for (i = 0; i < num; i++) {
1485 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1486 			!= cookie_addr_high) {
1487 			MRVL_LOG(ERR,
1488 				"mbuf virtual addr high 0x%lx out of range",
1489 				(uint64_t)mbufs[i] >> 32);
1490 			goto out;
1491 		}
1492 
1493 		entries[i].buff.addr =
1494 			rte_mbuf_data_iova_default(mbufs[i]);
1495 		entries[i].buff.cookie = (pp2_cookie_t)(uint64_t)mbufs[i];
1496 		entries[i].bpool = bpool;
1497 	}
1498 
1499 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1500 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1501 
1502 	if (i != num)
1503 		goto out;
1504 
1505 	return 0;
1506 out:
1507 	for (; i < num; i++)
1508 		rte_pktmbuf_free(mbufs[i]);
1509 
1510 	return -1;
1511 }
1512 
1513 /**
1514  * DPDK callback to configure the receive queue.
1515  *
1516  * @param dev
1517  *   Pointer to Ethernet device structure.
1518  * @param idx
1519  *   RX queue index.
1520  * @param desc
1521  *   Number of descriptors to configure in queue.
1522  * @param socket
1523  *   NUMA socket on which memory must be allocated.
1524  * @param conf
1525  *   Thresholds parameters.
1526  * @param mp
1527  *   Memory pool for buffer allocations.
1528  *
1529  * @return
1530  *   0 on success, negative error value otherwise.
1531  */
1532 static int
1533 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1534 		    unsigned int socket,
1535 		    const struct rte_eth_rxconf *conf,
1536 		    struct rte_mempool *mp)
1537 {
1538 	struct mrvl_priv *priv = dev->data->dev_private;
1539 	struct mrvl_rxq *rxq;
1540 	uint32_t min_size,
1541 		 max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1542 	int ret, tc, inq;
1543 	uint64_t offloads;
1544 
1545 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
1546 
1547 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1548 		/*
1549 		 * Unknown TC mapping, mapping will not have a correct queue.
1550 		 */
1551 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
1552 			idx, priv->ppio_id);
1553 		return -EFAULT;
1554 	}
1555 
1556 	min_size = rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM -
1557 		   MRVL_PKT_EFFEC_OFFS;
1558 	if (min_size < max_rx_pkt_len) {
1559 		MRVL_LOG(ERR,
1560 			"Mbuf size must be increased to %u bytes to hold up to %u bytes of data.",
1561 			max_rx_pkt_len + RTE_PKTMBUF_HEADROOM +
1562 			MRVL_PKT_EFFEC_OFFS,
1563 			max_rx_pkt_len);
1564 		return -EINVAL;
1565 	}
1566 
1567 	if (dev->data->rx_queues[idx]) {
1568 		rte_free(dev->data->rx_queues[idx]);
1569 		dev->data->rx_queues[idx] = NULL;
1570 	}
1571 
1572 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
1573 	if (!rxq)
1574 		return -ENOMEM;
1575 
1576 	rxq->priv = priv;
1577 	rxq->mp = mp;
1578 	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
1579 	rxq->queue_id = idx;
1580 	rxq->port_id = dev->data->port_id;
1581 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
1582 
1583 	tc = priv->rxq_map[rxq->queue_id].tc,
1584 	inq = priv->rxq_map[rxq->queue_id].inq;
1585 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
1586 		desc;
1587 
1588 	ret = mrvl_fill_bpool(rxq, desc);
1589 	if (ret) {
1590 		rte_free(rxq);
1591 		return ret;
1592 	}
1593 
1594 	priv->bpool_init_size += desc;
1595 
1596 	dev->data->rx_queues[idx] = rxq;
1597 
1598 	return 0;
1599 }
1600 
1601 /**
1602  * DPDK callback to release the receive queue.
1603  *
1604  * @param rxq
1605  *   Generic receive queue pointer.
1606  */
1607 static void
1608 mrvl_rx_queue_release(void *rxq)
1609 {
1610 	struct mrvl_rxq *q = rxq;
1611 	struct pp2_ppio_tc_params *tc_params;
1612 	int i, num, tc, inq;
1613 	struct pp2_hif *hif;
1614 	unsigned int core_id = rte_lcore_id();
1615 
1616 	if (core_id == LCORE_ID_ANY)
1617 		core_id = 0;
1618 
1619 	if (!q)
1620 		return;
1621 
1622 	hif = mrvl_get_hif(q->priv, core_id);
1623 
1624 	if (!hif)
1625 		return;
1626 
1627 	tc = q->priv->rxq_map[q->queue_id].tc;
1628 	inq = q->priv->rxq_map[q->queue_id].inq;
1629 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
1630 	num = tc_params->inqs_params[inq].size;
1631 	for (i = 0; i < num; i++) {
1632 		struct pp2_buff_inf inf;
1633 		uint64_t addr;
1634 
1635 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
1636 		addr = cookie_addr_high | inf.cookie;
1637 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1638 	}
1639 
1640 	rte_free(q);
1641 }
1642 
1643 /**
1644  * DPDK callback to configure the transmit queue.
1645  *
1646  * @param dev
1647  *   Pointer to Ethernet device structure.
1648  * @param idx
1649  *   Transmit queue index.
1650  * @param desc
1651  *   Number of descriptors to configure in the queue.
1652  * @param socket
1653  *   NUMA socket on which memory must be allocated.
1654  * @param conf
1655  *   Tx queue configuration parameters.
1656  *
1657  * @return
1658  *   0 on success, negative error value otherwise.
1659  */
1660 static int
1661 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1662 		    unsigned int socket,
1663 		    const struct rte_eth_txconf *conf)
1664 {
1665 	struct mrvl_priv *priv = dev->data->dev_private;
1666 	struct mrvl_txq *txq;
1667 
1668 	if (dev->data->tx_queues[idx]) {
1669 		rte_free(dev->data->tx_queues[idx]);
1670 		dev->data->tx_queues[idx] = NULL;
1671 	}
1672 
1673 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
1674 	if (!txq)
1675 		return -ENOMEM;
1676 
1677 	txq->priv = priv;
1678 	txq->queue_id = idx;
1679 	txq->port_id = dev->data->port_id;
1680 	txq->tx_deferred_start = conf->tx_deferred_start;
1681 	dev->data->tx_queues[idx] = txq;
1682 
1683 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
1684 
1685 	return 0;
1686 }
1687 
1688 /**
1689  * DPDK callback to release the transmit queue.
1690  *
1691  * @param txq
1692  *   Generic transmit queue pointer.
1693  */
1694 static void
1695 mrvl_tx_queue_release(void *txq)
1696 {
1697 	struct mrvl_txq *q = txq;
1698 
1699 	if (!q)
1700 		return;
1701 
1702 	rte_free(q);
1703 }
1704 
1705 /**
1706  * DPDK callback to get flow control configuration.
1707  *
1708  * @param dev
1709  *  Pointer to Ethernet device structure.
1710  * @param fc_conf
1711  *  Pointer to the flow control configuration.
1712  *
1713  * @return
1714  *  0 on success, negative error value otherwise.
1715  */
1716 static int
1717 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1718 {
1719 	struct mrvl_priv *priv = dev->data->dev_private;
1720 	int ret, en;
1721 
1722 	if (!priv)
1723 		return -EPERM;
1724 
1725 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
1726 	if (ret) {
1727 		MRVL_LOG(ERR, "Failed to read rx pause state");
1728 		return ret;
1729 	}
1730 
1731 	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
1732 
1733 	return 0;
1734 }
1735 
1736 /**
1737  * DPDK callback to set flow control configuration.
1738  *
1739  * @param dev
1740  *  Pointer to Ethernet device structure.
1741  * @param fc_conf
1742  *  Pointer to the flow control configuration.
1743  *
1744  * @return
1745  *  0 on success, negative error value otherwise.
1746  */
1747 static int
1748 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1749 {
1750 	struct mrvl_priv *priv = dev->data->dev_private;
1751 
1752 	if (!priv)
1753 		return -EPERM;
1754 
1755 	if (fc_conf->high_water ||
1756 	    fc_conf->low_water ||
1757 	    fc_conf->pause_time ||
1758 	    fc_conf->mac_ctrl_frame_fwd ||
1759 	    fc_conf->autoneg) {
1760 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
1761 
1762 		return -EINVAL;
1763 	}
1764 
1765 	if (fc_conf->mode == RTE_FC_NONE ||
1766 	    fc_conf->mode == RTE_FC_RX_PAUSE) {
1767 		int ret, en;
1768 
1769 		en = fc_conf->mode == RTE_FC_NONE ? 0 : 1;
1770 		ret = pp2_ppio_set_rx_pause(priv->ppio, en);
1771 		if (ret)
1772 			MRVL_LOG(ERR,
1773 				"Failed to change flowctrl on RX side");
1774 
1775 		return ret;
1776 	}
1777 
1778 	return 0;
1779 }
1780 
1781 /**
1782  * Update RSS hash configuration
1783  *
1784  * @param dev
1785  *   Pointer to Ethernet device structure.
1786  * @param rss_conf
1787  *   Pointer to RSS configuration.
1788  *
1789  * @return
1790  *   0 on success, negative error value otherwise.
1791  */
1792 static int
1793 mrvl_rss_hash_update(struct rte_eth_dev *dev,
1794 		     struct rte_eth_rss_conf *rss_conf)
1795 {
1796 	struct mrvl_priv *priv = dev->data->dev_private;
1797 
1798 	if (priv->isolated)
1799 		return -ENOTSUP;
1800 
1801 	return mrvl_configure_rss(priv, rss_conf);
1802 }
1803 
1804 /**
1805  * DPDK callback to get RSS hash configuration.
1806  *
1807  * @param dev
1808  *   Pointer to Ethernet device structure.
1809  * @rss_conf
1810  *   Pointer to RSS configuration.
1811  *
1812  * @return
1813  *   Always 0.
1814  */
1815 static int
1816 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
1817 		       struct rte_eth_rss_conf *rss_conf)
1818 {
1819 	struct mrvl_priv *priv = dev->data->dev_private;
1820 	enum pp2_ppio_hash_type hash_type =
1821 		priv->ppio_params.inqs_params.hash_type;
1822 
1823 	rss_conf->rss_key = NULL;
1824 
1825 	if (hash_type == PP2_PPIO_HASH_T_NONE)
1826 		rss_conf->rss_hf = 0;
1827 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
1828 		rss_conf->rss_hf = ETH_RSS_IPV4;
1829 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
1830 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
1831 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
1832 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
1833 
1834 	return 0;
1835 }
1836 
1837 /**
1838  * DPDK callback to get rte_flow callbacks.
1839  *
1840  * @param dev
1841  *   Pointer to the device structure.
1842  * @param filer_type
1843  *   Flow filter type.
1844  * @param filter_op
1845  *   Flow filter operation.
1846  * @param arg
1847  *   Pointer to pass the flow ops.
1848  *
1849  * @return
1850  *   0 on success, negative error value otherwise.
1851  */
1852 static int
1853 mrvl_eth_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
1854 		     enum rte_filter_type filter_type,
1855 		     enum rte_filter_op filter_op, void *arg)
1856 {
1857 	switch (filter_type) {
1858 	case RTE_ETH_FILTER_GENERIC:
1859 		if (filter_op != RTE_ETH_FILTER_GET)
1860 			return -EINVAL;
1861 		*(const void **)arg = &mrvl_flow_ops;
1862 		return 0;
1863 	default:
1864 		MRVL_LOG(WARNING, "Filter type (%d) not supported",
1865 				filter_type);
1866 		return -EINVAL;
1867 	}
1868 }
1869 
1870 static const struct eth_dev_ops mrvl_ops = {
1871 	.dev_configure = mrvl_dev_configure,
1872 	.dev_start = mrvl_dev_start,
1873 	.dev_stop = mrvl_dev_stop,
1874 	.dev_set_link_up = mrvl_dev_set_link_up,
1875 	.dev_set_link_down = mrvl_dev_set_link_down,
1876 	.dev_close = mrvl_dev_close,
1877 	.link_update = mrvl_link_update,
1878 	.promiscuous_enable = mrvl_promiscuous_enable,
1879 	.allmulticast_enable = mrvl_allmulticast_enable,
1880 	.promiscuous_disable = mrvl_promiscuous_disable,
1881 	.allmulticast_disable = mrvl_allmulticast_disable,
1882 	.mac_addr_remove = mrvl_mac_addr_remove,
1883 	.mac_addr_add = mrvl_mac_addr_add,
1884 	.mac_addr_set = mrvl_mac_addr_set,
1885 	.mtu_set = mrvl_mtu_set,
1886 	.stats_get = mrvl_stats_get,
1887 	.stats_reset = mrvl_stats_reset,
1888 	.xstats_get = mrvl_xstats_get,
1889 	.xstats_reset = mrvl_xstats_reset,
1890 	.xstats_get_names = mrvl_xstats_get_names,
1891 	.dev_infos_get = mrvl_dev_infos_get,
1892 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
1893 	.rxq_info_get = mrvl_rxq_info_get,
1894 	.txq_info_get = mrvl_txq_info_get,
1895 	.vlan_filter_set = mrvl_vlan_filter_set,
1896 	.tx_queue_start = mrvl_tx_queue_start,
1897 	.tx_queue_stop = mrvl_tx_queue_stop,
1898 	.rx_queue_setup = mrvl_rx_queue_setup,
1899 	.rx_queue_release = mrvl_rx_queue_release,
1900 	.tx_queue_setup = mrvl_tx_queue_setup,
1901 	.tx_queue_release = mrvl_tx_queue_release,
1902 	.flow_ctrl_get = mrvl_flow_ctrl_get,
1903 	.flow_ctrl_set = mrvl_flow_ctrl_set,
1904 	.rss_hash_update = mrvl_rss_hash_update,
1905 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
1906 	.filter_ctrl = mrvl_eth_filter_ctrl,
1907 };
1908 
1909 /**
1910  * Return packet type information and l3/l4 offsets.
1911  *
1912  * @param desc
1913  *   Pointer to the received packet descriptor.
1914  * @param l3_offset
1915  *   l3 packet offset.
1916  * @param l4_offset
1917  *   l4 packet offset.
1918  *
1919  * @return
1920  *   Packet type information.
1921  */
1922 static inline uint64_t
1923 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
1924 				    uint8_t *l3_offset, uint8_t *l4_offset)
1925 {
1926 	enum pp2_inq_l3_type l3_type;
1927 	enum pp2_inq_l4_type l4_type;
1928 	enum pp2_inq_vlan_tag vlan_tag;
1929 	uint64_t packet_type;
1930 
1931 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
1932 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
1933 	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
1934 
1935 	packet_type = RTE_PTYPE_L2_ETHER;
1936 
1937 	switch (vlan_tag) {
1938 	case PP2_INQ_VLAN_TAG_SINGLE:
1939 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
1940 		break;
1941 	case PP2_INQ_VLAN_TAG_DOUBLE:
1942 	case PP2_INQ_VLAN_TAG_TRIPLE:
1943 		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
1944 		break;
1945 	default:
1946 		break;
1947 	}
1948 
1949 	switch (l3_type) {
1950 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
1951 		packet_type |= RTE_PTYPE_L3_IPV4;
1952 		break;
1953 	case PP2_INQ_L3_TYPE_IPV4_OK:
1954 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
1955 		break;
1956 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
1957 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
1958 		break;
1959 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
1960 		packet_type |= RTE_PTYPE_L3_IPV6;
1961 		break;
1962 	case PP2_INQ_L3_TYPE_IPV6_EXT:
1963 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
1964 		break;
1965 	case PP2_INQ_L3_TYPE_ARP:
1966 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
1967 		/*
1968 		 * In case of ARP l4_offset is set to wrong value.
1969 		 * Set it to proper one so that later on mbuf->l3_len can be
1970 		 * calculated subtracting l4_offset and l3_offset.
1971 		 */
1972 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
1973 		break;
1974 	default:
1975 		MRVL_LOG(DEBUG, "Failed to recognise l3 packet type");
1976 		break;
1977 	}
1978 
1979 	switch (l4_type) {
1980 	case PP2_INQ_L4_TYPE_TCP:
1981 		packet_type |= RTE_PTYPE_L4_TCP;
1982 		break;
1983 	case PP2_INQ_L4_TYPE_UDP:
1984 		packet_type |= RTE_PTYPE_L4_UDP;
1985 		break;
1986 	default:
1987 		MRVL_LOG(DEBUG, "Failed to recognise l4 packet type");
1988 		break;
1989 	}
1990 
1991 	return packet_type;
1992 }
1993 
1994 /**
1995  * Get offload information from the received packet descriptor.
1996  *
1997  * @param desc
1998  *   Pointer to the received packet descriptor.
1999  *
2000  * @return
2001  *   Mbuf offload flags.
2002  */
2003 static inline uint64_t
2004 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc)
2005 {
2006 	uint64_t flags;
2007 	enum pp2_inq_desc_status status;
2008 
2009 	status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2010 	if (unlikely(status != PP2_DESC_ERR_OK))
2011 		flags = PKT_RX_IP_CKSUM_BAD;
2012 	else
2013 		flags = PKT_RX_IP_CKSUM_GOOD;
2014 
2015 	status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2016 	if (unlikely(status != PP2_DESC_ERR_OK))
2017 		flags |= PKT_RX_L4_CKSUM_BAD;
2018 	else
2019 		flags |= PKT_RX_L4_CKSUM_GOOD;
2020 
2021 	return flags;
2022 }
2023 
2024 /**
2025  * DPDK callback for receive.
2026  *
2027  * @param rxq
2028  *   Generic pointer to the receive queue.
2029  * @param rx_pkts
2030  *   Array to store received packets.
2031  * @param nb_pkts
2032  *   Maximum number of packets in array.
2033  *
2034  * @return
2035  *   Number of packets successfully received.
2036  */
2037 static uint16_t
2038 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2039 {
2040 	struct mrvl_rxq *q = rxq;
2041 	struct pp2_ppio_desc descs[nb_pkts];
2042 	struct pp2_bpool *bpool;
2043 	int i, ret, rx_done = 0;
2044 	int num;
2045 	struct pp2_hif *hif;
2046 	unsigned int core_id = rte_lcore_id();
2047 
2048 	hif = mrvl_get_hif(q->priv, core_id);
2049 
2050 	if (unlikely(!q->priv->ppio || !hif))
2051 		return 0;
2052 
2053 	bpool = q->priv->bpool;
2054 
2055 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2056 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2057 	if (unlikely(ret < 0)) {
2058 		MRVL_LOG(ERR, "Failed to receive packets");
2059 		return 0;
2060 	}
2061 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2062 
2063 	for (i = 0; i < nb_pkts; i++) {
2064 		struct rte_mbuf *mbuf;
2065 		uint8_t l3_offset, l4_offset;
2066 		enum pp2_inq_desc_status status;
2067 		uint64_t addr;
2068 
2069 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2070 			struct pp2_ppio_desc *pref_desc;
2071 			u64 pref_addr;
2072 
2073 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2074 			pref_addr = cookie_addr_high |
2075 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2076 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2077 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2078 		}
2079 
2080 		addr = cookie_addr_high |
2081 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2082 		mbuf = (struct rte_mbuf *)addr;
2083 		rte_pktmbuf_reset(mbuf);
2084 
2085 		/* drop packet in case of mac, overrun or resource error */
2086 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2087 		if (unlikely(status != PP2_DESC_ERR_OK)) {
2088 			struct pp2_buff_inf binf = {
2089 				.addr = rte_mbuf_data_iova_default(mbuf),
2090 				.cookie = (pp2_cookie_t)(uint64_t)mbuf,
2091 			};
2092 
2093 			pp2_bpool_put_buff(hif, bpool, &binf);
2094 			mrvl_port_bpool_size
2095 				[bpool->pp2_id][bpool->id][core_id]++;
2096 			q->drop_mac++;
2097 			continue;
2098 		}
2099 
2100 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2101 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2102 		mbuf->data_len = mbuf->pkt_len;
2103 		mbuf->port = q->port_id;
2104 		mbuf->packet_type =
2105 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2106 							    &l3_offset,
2107 							    &l4_offset);
2108 		mbuf->l2_len = l3_offset;
2109 		mbuf->l3_len = l4_offset - l3_offset;
2110 
2111 		if (likely(q->cksum_enabled))
2112 			mbuf->ol_flags = mrvl_desc_to_ol_flags(&descs[i]);
2113 
2114 		rx_pkts[rx_done++] = mbuf;
2115 		q->bytes_recv += mbuf->pkt_len;
2116 	}
2117 
2118 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2119 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2120 
2121 		if (unlikely(num <= q->priv->bpool_min_size ||
2122 			     (!rx_done && num < q->priv->bpool_init_size))) {
2123 			ret = mrvl_fill_bpool(q, MRVL_BURST_SIZE);
2124 			if (ret)
2125 				MRVL_LOG(ERR, "Failed to fill bpool");
2126 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2127 			int i;
2128 			int pkt_to_remove = num - q->priv->bpool_init_size;
2129 			struct rte_mbuf *mbuf;
2130 			struct pp2_buff_inf buff;
2131 
2132 			MRVL_LOG(DEBUG,
2133 				"port-%d:%d: bpool %d oversize - remove %d buffers (pool size: %d -> %d)",
2134 				bpool->pp2_id, q->priv->ppio->port_id,
2135 				bpool->id, pkt_to_remove, num,
2136 				q->priv->bpool_init_size);
2137 
2138 			for (i = 0; i < pkt_to_remove; i++) {
2139 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2140 				if (ret)
2141 					break;
2142 				mbuf = (struct rte_mbuf *)
2143 					(cookie_addr_high | buff.cookie);
2144 				rte_pktmbuf_free(mbuf);
2145 			}
2146 			mrvl_port_bpool_size
2147 				[bpool->pp2_id][bpool->id][core_id] -= i;
2148 		}
2149 		rte_spinlock_unlock(&q->priv->lock);
2150 	}
2151 
2152 	return rx_done;
2153 }
2154 
2155 /**
2156  * Prepare offload information.
2157  *
2158  * @param ol_flags
2159  *   Offload flags.
2160  * @param packet_type
2161  *   Packet type bitfield.
2162  * @param l3_type
2163  *   Pointer to the pp2_ouq_l3_type structure.
2164  * @param l4_type
2165  *   Pointer to the pp2_outq_l4_type structure.
2166  * @param gen_l3_cksum
2167  *   Will be set to 1 in case l3 checksum is computed.
2168  * @param l4_cksum
2169  *   Will be set to 1 in case l4 checksum is computed.
2170  *
2171  * @return
2172  *   0 on success, negative error value otherwise.
2173  */
2174 static inline int
2175 mrvl_prepare_proto_info(uint64_t ol_flags, uint32_t packet_type,
2176 			enum pp2_outq_l3_type *l3_type,
2177 			enum pp2_outq_l4_type *l4_type,
2178 			int *gen_l3_cksum,
2179 			int *gen_l4_cksum)
2180 {
2181 	/*
2182 	 * Based on ol_flags prepare information
2183 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2184 	 * for offloading.
2185 	 */
2186 	if (ol_flags & PKT_TX_IPV4) {
2187 		*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2188 		*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2189 	} else if (ol_flags & PKT_TX_IPV6) {
2190 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2191 		/* no checksum for ipv6 header */
2192 		*gen_l3_cksum = 0;
2193 	} else {
2194 		/* if something different then stop processing */
2195 		return -1;
2196 	}
2197 
2198 	ol_flags &= PKT_TX_L4_MASK;
2199 	if ((packet_type & RTE_PTYPE_L4_TCP) &&
2200 	    ol_flags == PKT_TX_TCP_CKSUM) {
2201 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2202 		*gen_l4_cksum = 1;
2203 	} else if ((packet_type & RTE_PTYPE_L4_UDP) &&
2204 		   ol_flags == PKT_TX_UDP_CKSUM) {
2205 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2206 		*gen_l4_cksum = 1;
2207 	} else {
2208 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2209 		/* no checksum for other type */
2210 		*gen_l4_cksum = 0;
2211 	}
2212 
2213 	return 0;
2214 }
2215 
2216 /**
2217  * Release already sent buffers to bpool (buffer-pool).
2218  *
2219  * @param ppio
2220  *   Pointer to the port structure.
2221  * @param hif
2222  *   Pointer to the MUSDK hardware interface.
2223  * @param sq
2224  *   Pointer to the shadow queue.
2225  * @param qid
2226  *   Queue id number.
2227  * @param force
2228  *   Force releasing packets.
2229  */
2230 static inline void
2231 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2232 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2233 		       int qid, int force)
2234 {
2235 	struct buff_release_entry *entry;
2236 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2237 	int i;
2238 
2239 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2240 
2241 	sq->num_to_release += nb_done;
2242 
2243 	if (likely(!force &&
2244 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2245 		return;
2246 
2247 	nb_done = sq->num_to_release;
2248 	sq->num_to_release = 0;
2249 
2250 	for (i = 0; i < nb_done; i++) {
2251 		entry = &sq->ent[sq->tail + num];
2252 		if (unlikely(!entry->buff.addr)) {
2253 			MRVL_LOG(ERR,
2254 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2255 				sq->tail, (u64)entry->buff.cookie,
2256 				(u64)entry->buff.addr);
2257 			skip_bufs = 1;
2258 			goto skip;
2259 		}
2260 
2261 		if (unlikely(!entry->bpool)) {
2262 			struct rte_mbuf *mbuf;
2263 
2264 			mbuf = (struct rte_mbuf *)
2265 			       (cookie_addr_high | entry->buff.cookie);
2266 			rte_pktmbuf_free(mbuf);
2267 			skip_bufs = 1;
2268 			goto skip;
2269 		}
2270 
2271 		mrvl_port_bpool_size
2272 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2273 		num++;
2274 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2275 			goto skip;
2276 		continue;
2277 skip:
2278 		if (likely(num))
2279 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2280 		num += skip_bufs;
2281 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2282 		sq->size -= num;
2283 		num = 0;
2284 		skip_bufs = 0;
2285 	}
2286 
2287 	if (likely(num)) {
2288 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2289 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2290 		sq->size -= num;
2291 	}
2292 }
2293 
2294 /**
2295  * DPDK callback for transmit.
2296  *
2297  * @param txq
2298  *   Generic pointer transmit queue.
2299  * @param tx_pkts
2300  *   Packets to transmit.
2301  * @param nb_pkts
2302  *   Number of packets in array.
2303  *
2304  * @return
2305  *   Number of packets successfully transmitted.
2306  */
2307 static uint16_t
2308 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2309 {
2310 	struct mrvl_txq *q = txq;
2311 	struct mrvl_shadow_txq *sq;
2312 	struct pp2_hif *hif;
2313 	struct pp2_ppio_desc descs[nb_pkts];
2314 	unsigned int core_id = rte_lcore_id();
2315 	int i, ret, bytes_sent = 0;
2316 	uint16_t num, sq_free_size;
2317 	uint64_t addr;
2318 
2319 	hif = mrvl_get_hif(q->priv, core_id);
2320 	sq = &q->shadow_txqs[core_id];
2321 
2322 	if (unlikely(!q->priv->ppio || !hif))
2323 		return 0;
2324 
2325 	if (sq->size)
2326 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2327 				       sq, q->queue_id, 0);
2328 
2329 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2330 	if (unlikely(nb_pkts > sq_free_size)) {
2331 		MRVL_LOG(DEBUG,
2332 			"No room in shadow queue for %d packets! %d packets will be sent.",
2333 			nb_pkts, sq_free_size);
2334 		nb_pkts = sq_free_size;
2335 	}
2336 
2337 	for (i = 0; i < nb_pkts; i++) {
2338 		struct rte_mbuf *mbuf = tx_pkts[i];
2339 		int gen_l3_cksum, gen_l4_cksum;
2340 		enum pp2_outq_l3_type l3_type;
2341 		enum pp2_outq_l4_type l4_type;
2342 
2343 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2344 			struct rte_mbuf *pref_pkt_hdr;
2345 
2346 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2347 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2348 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2349 		}
2350 
2351 		sq->ent[sq->head].buff.cookie = (pp2_cookie_t)(uint64_t)mbuf;
2352 		sq->ent[sq->head].buff.addr =
2353 			rte_mbuf_data_iova_default(mbuf);
2354 		sq->ent[sq->head].bpool =
2355 			(unlikely(mbuf->port >= RTE_MAX_ETHPORTS ||
2356 			 mbuf->refcnt > 1)) ? NULL :
2357 			 mrvl_port_to_bpool_lookup[mbuf->port];
2358 		sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
2359 		sq->size++;
2360 
2361 		pp2_ppio_outq_desc_reset(&descs[i]);
2362 		pp2_ppio_outq_desc_set_phys_addr(&descs[i],
2363 						 rte_pktmbuf_iova(mbuf));
2364 		pp2_ppio_outq_desc_set_pkt_offset(&descs[i], 0);
2365 		pp2_ppio_outq_desc_set_pkt_len(&descs[i],
2366 					       rte_pktmbuf_pkt_len(mbuf));
2367 
2368 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2369 		/*
2370 		 * in case unsupported ol_flags were passed
2371 		 * do not update descriptor offload information
2372 		 */
2373 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2374 					      &l3_type, &l4_type, &gen_l3_cksum,
2375 					      &gen_l4_cksum);
2376 		if (unlikely(ret))
2377 			continue;
2378 
2379 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2380 						  mbuf->l2_len,
2381 						  mbuf->l2_len + mbuf->l3_len,
2382 						  gen_l3_cksum, gen_l4_cksum);
2383 	}
2384 
2385 	num = nb_pkts;
2386 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2387 	/* number of packets that were not sent */
2388 	if (unlikely(num > nb_pkts)) {
2389 		for (i = nb_pkts; i < num; i++) {
2390 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2391 				MRVL_PP2_TX_SHADOWQ_MASK;
2392 			addr = cookie_addr_high | sq->ent[sq->head].buff.cookie;
2393 			bytes_sent -=
2394 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2395 		}
2396 		sq->size -= num - nb_pkts;
2397 	}
2398 
2399 	q->bytes_sent += bytes_sent;
2400 
2401 	return nb_pkts;
2402 }
2403 
2404 /**
2405  * Initialize packet processor.
2406  *
2407  * @return
2408  *   0 on success, negative error value otherwise.
2409  */
2410 static int
2411 mrvl_init_pp2(void)
2412 {
2413 	struct pp2_init_params init_params;
2414 
2415 	memset(&init_params, 0, sizeof(init_params));
2416 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
2417 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
2418 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
2419 
2420 	return pp2_init(&init_params);
2421 }
2422 
2423 /**
2424  * Deinitialize packet processor.
2425  *
2426  * @return
2427  *   0 on success, negative error value otherwise.
2428  */
2429 static void
2430 mrvl_deinit_pp2(void)
2431 {
2432 	pp2_deinit();
2433 }
2434 
2435 /**
2436  * Create private device structure.
2437  *
2438  * @param dev_name
2439  *   Pointer to the port name passed in the initialization parameters.
2440  *
2441  * @return
2442  *   Pointer to the newly allocated private device structure.
2443  */
2444 static struct mrvl_priv *
2445 mrvl_priv_create(const char *dev_name)
2446 {
2447 	struct pp2_bpool_params bpool_params;
2448 	char match[MRVL_MATCH_LEN];
2449 	struct mrvl_priv *priv;
2450 	int ret, bpool_bit;
2451 
2452 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
2453 	if (!priv)
2454 		return NULL;
2455 
2456 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
2457 				       &priv->pp_id, &priv->ppio_id);
2458 	if (ret)
2459 		goto out_free_priv;
2460 
2461 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
2462 				     PP2_BPOOL_NUM_POOLS);
2463 	if (bpool_bit < 0)
2464 		goto out_free_priv;
2465 	priv->bpool_bit = bpool_bit;
2466 
2467 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
2468 		 priv->bpool_bit);
2469 	memset(&bpool_params, 0, sizeof(bpool_params));
2470 	bpool_params.match = match;
2471 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
2472 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
2473 	if (ret)
2474 		goto out_clear_bpool_bit;
2475 
2476 	priv->ppio_params.type = PP2_PPIO_T_NIC;
2477 	rte_spinlock_init(&priv->lock);
2478 
2479 	return priv;
2480 out_clear_bpool_bit:
2481 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2482 out_free_priv:
2483 	rte_free(priv);
2484 	return NULL;
2485 }
2486 
2487 /**
2488  * Create device representing Ethernet port.
2489  *
2490  * @param name
2491  *   Pointer to the port's name.
2492  *
2493  * @return
2494  *   0 on success, negative error value otherwise.
2495  */
2496 static int
2497 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
2498 {
2499 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
2500 	struct rte_eth_dev *eth_dev;
2501 	struct mrvl_priv *priv;
2502 	struct ifreq req;
2503 
2504 	eth_dev = rte_eth_dev_allocate(name);
2505 	if (!eth_dev)
2506 		return -ENOMEM;
2507 
2508 	priv = mrvl_priv_create(name);
2509 	if (!priv) {
2510 		ret = -ENOMEM;
2511 		goto out_free_dev;
2512 	}
2513 
2514 	eth_dev->data->mac_addrs =
2515 		rte_zmalloc("mac_addrs",
2516 			    ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
2517 	if (!eth_dev->data->mac_addrs) {
2518 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
2519 		ret = -ENOMEM;
2520 		goto out_free_priv;
2521 	}
2522 
2523 	memset(&req, 0, sizeof(req));
2524 	strcpy(req.ifr_name, name);
2525 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
2526 	if (ret)
2527 		goto out_free_mac;
2528 
2529 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
2530 	       req.ifr_addr.sa_data, ETHER_ADDR_LEN);
2531 
2532 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
2533 	eth_dev->tx_pkt_burst = mrvl_tx_pkt_burst;
2534 	eth_dev->data->kdrv = RTE_KDRV_NONE;
2535 	eth_dev->data->dev_private = priv;
2536 	eth_dev->device = &vdev->device;
2537 	eth_dev->dev_ops = &mrvl_ops;
2538 
2539 	rte_eth_dev_probing_finish(eth_dev);
2540 	return 0;
2541 out_free_mac:
2542 	rte_free(eth_dev->data->mac_addrs);
2543 out_free_dev:
2544 	rte_eth_dev_release_port(eth_dev);
2545 out_free_priv:
2546 	rte_free(priv);
2547 
2548 	return ret;
2549 }
2550 
2551 /**
2552  * Cleanup previously created device representing Ethernet port.
2553  *
2554  * @param name
2555  *   Pointer to the port name.
2556  */
2557 static void
2558 mrvl_eth_dev_destroy(const char *name)
2559 {
2560 	struct rte_eth_dev *eth_dev;
2561 	struct mrvl_priv *priv;
2562 
2563 	eth_dev = rte_eth_dev_allocated(name);
2564 	if (!eth_dev)
2565 		return;
2566 
2567 	priv = eth_dev->data->dev_private;
2568 	pp2_bpool_deinit(priv->bpool);
2569 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2570 	rte_free(priv);
2571 	rte_free(eth_dev->data->mac_addrs);
2572 	rte_eth_dev_release_port(eth_dev);
2573 }
2574 
2575 /**
2576  * Callback used by rte_kvargs_process() during argument parsing.
2577  *
2578  * @param key
2579  *   Pointer to the parsed key (unused).
2580  * @param value
2581  *   Pointer to the parsed value.
2582  * @param extra_args
2583  *   Pointer to the extra arguments which contains address of the
2584  *   table of pointers to parsed interface names.
2585  *
2586  * @return
2587  *   Always 0.
2588  */
2589 static int
2590 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
2591 		 void *extra_args)
2592 {
2593 	struct mrvl_ifnames *ifnames = extra_args;
2594 
2595 	ifnames->names[ifnames->idx++] = value;
2596 
2597 	return 0;
2598 }
2599 
2600 /**
2601  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
2602  */
2603 static void
2604 mrvl_deinit_hifs(void)
2605 {
2606 	int i;
2607 
2608 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
2609 		if (hifs[i])
2610 			pp2_hif_deinit(hifs[i]);
2611 	}
2612 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
2613 	memset(hifs, 0, sizeof(hifs));
2614 }
2615 
2616 /**
2617  * DPDK callback to register the virtual device.
2618  *
2619  * @param vdev
2620  *   Pointer to the virtual device.
2621  *
2622  * @return
2623  *   0 on success, negative error value otherwise.
2624  */
2625 static int
2626 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
2627 {
2628 	struct rte_kvargs *kvlist;
2629 	struct mrvl_ifnames ifnames;
2630 	int ret = -EINVAL;
2631 	uint32_t i, ifnum, cfgnum;
2632 	const char *params;
2633 
2634 	params = rte_vdev_device_args(vdev);
2635 	if (!params)
2636 		return -EINVAL;
2637 
2638 	kvlist = rte_kvargs_parse(params, valid_args);
2639 	if (!kvlist)
2640 		return -EINVAL;
2641 
2642 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
2643 	if (ifnum > RTE_DIM(ifnames.names))
2644 		goto out_free_kvlist;
2645 
2646 	ifnames.idx = 0;
2647 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
2648 			   mrvl_get_ifnames, &ifnames);
2649 
2650 
2651 	/*
2652 	 * The below system initialization should be done only once,
2653 	 * on the first provided configuration file
2654 	 */
2655 	if (!mrvl_qos_cfg) {
2656 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
2657 		MRVL_LOG(INFO, "Parsing config file!");
2658 		if (cfgnum > 1) {
2659 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
2660 			goto out_free_kvlist;
2661 		} else if (cfgnum == 1) {
2662 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
2663 					   mrvl_get_qoscfg, &mrvl_qos_cfg);
2664 		}
2665 	}
2666 
2667 	if (mrvl_dev_num)
2668 		goto init_devices;
2669 
2670 	MRVL_LOG(INFO, "Perform MUSDK initializations");
2671 
2672 	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
2673 	if (ret)
2674 		goto out_free_kvlist;
2675 
2676 	ret = mrvl_init_pp2();
2677 	if (ret) {
2678 		MRVL_LOG(ERR, "Failed to init PP!");
2679 		rte_mvep_deinit(MVEP_MOD_T_PP2);
2680 		goto out_free_kvlist;
2681 	}
2682 
2683 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
2684 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
2685 
2686 	mrvl_lcore_first = RTE_MAX_LCORE;
2687 	mrvl_lcore_last = 0;
2688 
2689 init_devices:
2690 	for (i = 0; i < ifnum; i++) {
2691 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
2692 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
2693 		if (ret)
2694 			goto out_cleanup;
2695 	}
2696 	mrvl_dev_num += ifnum;
2697 
2698 	rte_kvargs_free(kvlist);
2699 
2700 	return 0;
2701 out_cleanup:
2702 	for (; i > 0; i--)
2703 		mrvl_eth_dev_destroy(ifnames.names[i]);
2704 
2705 	if (mrvl_dev_num == 0) {
2706 		mrvl_deinit_pp2();
2707 		rte_mvep_deinit(MVEP_MOD_T_PP2);
2708 	}
2709 out_free_kvlist:
2710 	rte_kvargs_free(kvlist);
2711 
2712 	return ret;
2713 }
2714 
2715 /**
2716  * DPDK callback to remove virtual device.
2717  *
2718  * @param vdev
2719  *   Pointer to the removed virtual device.
2720  *
2721  * @return
2722  *   0 on success, negative error value otherwise.
2723  */
2724 static int
2725 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
2726 {
2727 	int i;
2728 	const char *name;
2729 
2730 	name = rte_vdev_device_name(vdev);
2731 	if (!name)
2732 		return -EINVAL;
2733 
2734 	MRVL_LOG(INFO, "Removing %s", name);
2735 
2736 	RTE_ETH_FOREACH_DEV(i) { /* FIXME: removing all devices! */
2737 		char ifname[RTE_ETH_NAME_MAX_LEN];
2738 
2739 		rte_eth_dev_get_name_by_port(i, ifname);
2740 		mrvl_eth_dev_destroy(ifname);
2741 		mrvl_dev_num--;
2742 	}
2743 
2744 	if (mrvl_dev_num == 0) {
2745 		MRVL_LOG(INFO, "Perform MUSDK deinit");
2746 		mrvl_deinit_hifs();
2747 		mrvl_deinit_pp2();
2748 		rte_mvep_deinit(MVEP_MOD_T_PP2);
2749 	}
2750 
2751 	return 0;
2752 }
2753 
2754 static struct rte_vdev_driver pmd_mrvl_drv = {
2755 	.probe = rte_pmd_mrvl_probe,
2756 	.remove = rte_pmd_mrvl_remove,
2757 };
2758 
2759 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
2760 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
2761 
2762 RTE_INIT(mrvl_init_log)
2763 {
2764 	mrvl_logtype = rte_log_register("pmd.net.mvpp2");
2765 	if (mrvl_logtype >= 0)
2766 		rte_log_set_level(mrvl_logtype, RTE_LOG_NOTICE);
2767 }
2768