xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision 08aa6271c86a561b66c6dd91f9a54fa2f12bc859)
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 struct mrvl_ifnames {
98 	const char *names[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
99 	int idx;
100 };
101 
102 /*
103  * To use buffer harvesting based on loopback port shadow queue structure
104  * was introduced for buffers information bookkeeping.
105  *
106  * Before sending the packet, related buffer information (pp2_buff_inf) is
107  * stored in shadow queue. After packet is transmitted no longer used
108  * packet buffer is released back to it's original hardware pool,
109  * on condition it originated from interface.
110  * In case it  was generated by application itself i.e: mbuf->port field is
111  * 0xff then its released to software mempool.
112  */
113 struct mrvl_shadow_txq {
114 	int head;           /* write index - used when sending buffers */
115 	int tail;           /* read index - used when releasing buffers */
116 	u16 size;           /* queue occupied size */
117 	u16 num_to_release; /* number of buffers sent, that can be released */
118 	struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
119 };
120 
121 struct mrvl_rxq {
122 	struct mrvl_priv *priv;
123 	struct rte_mempool *mp;
124 	int queue_id;
125 	int port_id;
126 	int cksum_enabled;
127 	uint64_t bytes_recv;
128 	uint64_t drop_mac;
129 };
130 
131 struct mrvl_txq {
132 	struct mrvl_priv *priv;
133 	int queue_id;
134 	int port_id;
135 	uint64_t bytes_sent;
136 	struct mrvl_shadow_txq shadow_txqs[RTE_MAX_LCORE];
137 	int tx_deferred_start;
138 };
139 
140 static int mrvl_lcore_first;
141 static int mrvl_lcore_last;
142 static int mrvl_dev_num;
143 
144 static int mrvl_fill_bpool(struct mrvl_rxq *rxq, int num);
145 static inline void mrvl_free_sent_buffers(struct pp2_ppio *ppio,
146 			struct pp2_hif *hif, unsigned int core_id,
147 			struct mrvl_shadow_txq *sq, int qid, int force);
148 
149 #define MRVL_XSTATS_TBL_ENTRY(name) { \
150 	#name, offsetof(struct pp2_ppio_statistics, name),	\
151 	sizeof(((struct pp2_ppio_statistics *)0)->name)		\
152 }
153 
154 /* Table with xstats data */
155 static struct {
156 	const char *name;
157 	unsigned int offset;
158 	unsigned int size;
159 } mrvl_xstats_tbl[] = {
160 	MRVL_XSTATS_TBL_ENTRY(rx_bytes),
161 	MRVL_XSTATS_TBL_ENTRY(rx_packets),
162 	MRVL_XSTATS_TBL_ENTRY(rx_unicast_packets),
163 	MRVL_XSTATS_TBL_ENTRY(rx_errors),
164 	MRVL_XSTATS_TBL_ENTRY(rx_fullq_dropped),
165 	MRVL_XSTATS_TBL_ENTRY(rx_bm_dropped),
166 	MRVL_XSTATS_TBL_ENTRY(rx_early_dropped),
167 	MRVL_XSTATS_TBL_ENTRY(rx_fifo_dropped),
168 	MRVL_XSTATS_TBL_ENTRY(rx_cls_dropped),
169 	MRVL_XSTATS_TBL_ENTRY(tx_bytes),
170 	MRVL_XSTATS_TBL_ENTRY(tx_packets),
171 	MRVL_XSTATS_TBL_ENTRY(tx_unicast_packets),
172 	MRVL_XSTATS_TBL_ENTRY(tx_errors)
173 };
174 
175 static inline int
176 mrvl_get_bpool_size(int pp2_id, int pool_id)
177 {
178 	int i;
179 	int size = 0;
180 
181 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
182 		size += mrvl_port_bpool_size[pp2_id][pool_id][i];
183 
184 	return size;
185 }
186 
187 static inline int
188 mrvl_reserve_bit(int *bitmap, int max)
189 {
190 	int n = sizeof(*bitmap) * 8 - __builtin_clz(*bitmap);
191 
192 	if (n >= max)
193 		return -1;
194 
195 	*bitmap |= 1 << n;
196 
197 	return n;
198 }
199 
200 static int
201 mrvl_init_hif(int core_id)
202 {
203 	struct pp2_hif_params params;
204 	char match[MRVL_MATCH_LEN];
205 	int ret;
206 
207 	ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
208 	if (ret < 0) {
209 		RTE_LOG(ERR, PMD, "Failed to allocate hif %d\n", core_id);
210 		return ret;
211 	}
212 
213 	snprintf(match, sizeof(match), "hif-%d", ret);
214 	memset(&params, 0, sizeof(params));
215 	params.match = match;
216 	params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
217 	ret = pp2_hif_init(&params, &hifs[core_id]);
218 	if (ret) {
219 		RTE_LOG(ERR, PMD, "Failed to initialize hif %d\n", core_id);
220 		return ret;
221 	}
222 
223 	return 0;
224 }
225 
226 static inline struct pp2_hif*
227 mrvl_get_hif(struct mrvl_priv *priv, int core_id)
228 {
229 	int ret;
230 
231 	if (likely(hifs[core_id] != NULL))
232 		return hifs[core_id];
233 
234 	rte_spinlock_lock(&priv->lock);
235 
236 	ret = mrvl_init_hif(core_id);
237 	if (ret < 0) {
238 		RTE_LOG(ERR, PMD, "Failed to allocate hif %d\n", core_id);
239 		goto out;
240 	}
241 
242 	if (core_id < mrvl_lcore_first)
243 		mrvl_lcore_first = core_id;
244 
245 	if (core_id > mrvl_lcore_last)
246 		mrvl_lcore_last = core_id;
247 out:
248 	rte_spinlock_unlock(&priv->lock);
249 
250 	return hifs[core_id];
251 }
252 
253 /**
254  * Configure rss based on dpdk rss configuration.
255  *
256  * @param priv
257  *   Pointer to private structure.
258  * @param rss_conf
259  *   Pointer to RSS configuration.
260  *
261  * @return
262  *   0 on success, negative error value otherwise.
263  */
264 static int
265 mrvl_configure_rss(struct mrvl_priv *priv, struct rte_eth_rss_conf *rss_conf)
266 {
267 	if (rss_conf->rss_key)
268 		RTE_LOG(WARNING, PMD, "Changing hash key is not supported\n");
269 
270 	if (rss_conf->rss_hf == 0) {
271 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
272 	} else if (rss_conf->rss_hf & ETH_RSS_IPV4) {
273 		priv->ppio_params.inqs_params.hash_type =
274 			PP2_PPIO_HASH_T_2_TUPLE;
275 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP) {
276 		priv->ppio_params.inqs_params.hash_type =
277 			PP2_PPIO_HASH_T_5_TUPLE;
278 		priv->rss_hf_tcp = 1;
279 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) {
280 		priv->ppio_params.inqs_params.hash_type =
281 			PP2_PPIO_HASH_T_5_TUPLE;
282 		priv->rss_hf_tcp = 0;
283 	} else {
284 		return -EINVAL;
285 	}
286 
287 	return 0;
288 }
289 
290 /**
291  * Ethernet device configuration.
292  *
293  * Prepare the driver for a given number of TX and RX queues and
294  * configure RSS.
295  *
296  * @param dev
297  *   Pointer to Ethernet device structure.
298  *
299  * @return
300  *   0 on success, negative error value otherwise.
301  */
302 static int
303 mrvl_dev_configure(struct rte_eth_dev *dev)
304 {
305 	struct mrvl_priv *priv = dev->data->dev_private;
306 	int ret;
307 
308 	if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_NONE &&
309 	    dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
310 		RTE_LOG(INFO, PMD, "Unsupported rx multi queue mode %d\n",
311 			dev->data->dev_conf.rxmode.mq_mode);
312 		return -EINVAL;
313 	}
314 
315 	if (!(dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_CRC_STRIP)) {
316 		RTE_LOG(INFO, PMD,
317 			"L2 CRC stripping is always enabled in hw\n");
318 		dev->data->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_CRC_STRIP;
319 	}
320 
321 	if (dev->data->dev_conf.rxmode.split_hdr_size) {
322 		RTE_LOG(INFO, PMD, "Split headers not supported\n");
323 		return -EINVAL;
324 	}
325 
326 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
327 		dev->data->mtu = dev->data->dev_conf.rxmode.max_rx_pkt_len -
328 				 ETHER_HDR_LEN - ETHER_CRC_LEN;
329 
330 	ret = mrvl_configure_rxqs(priv, dev->data->port_id,
331 				  dev->data->nb_rx_queues);
332 	if (ret < 0)
333 		return ret;
334 
335 	ret = mrvl_configure_txqs(priv, dev->data->port_id,
336 				  dev->data->nb_tx_queues);
337 	if (ret < 0)
338 		return ret;
339 
340 	priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
341 	priv->ppio_params.maintain_stats = 1;
342 	priv->nb_rx_queues = dev->data->nb_rx_queues;
343 
344 	if (dev->data->nb_rx_queues == 1 &&
345 	    dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
346 		RTE_LOG(WARNING, PMD, "Disabling hash for 1 rx queue\n");
347 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
348 
349 		return 0;
350 	}
351 
352 	return mrvl_configure_rss(priv,
353 				  &dev->data->dev_conf.rx_adv_conf.rss_conf);
354 }
355 
356 /**
357  * DPDK callback to change the MTU.
358  *
359  * Setting the MTU affects hardware MRU (packets larger than the MRU
360  * will be dropped).
361  *
362  * @param dev
363  *   Pointer to Ethernet device structure.
364  * @param mtu
365  *   New MTU.
366  *
367  * @return
368  *   0 on success, negative error value otherwise.
369  */
370 static int
371 mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
372 {
373 	struct mrvl_priv *priv = dev->data->dev_private;
374 	/* extra MV_MH_SIZE bytes are required for Marvell tag */
375 	uint16_t mru = mtu + MV_MH_SIZE + ETHER_HDR_LEN + ETHER_CRC_LEN;
376 	int ret;
377 
378 	if (mtu < ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX)
379 		return -EINVAL;
380 
381 	if (!priv->ppio)
382 		return 0;
383 
384 	ret = pp2_ppio_set_mru(priv->ppio, mru);
385 	if (ret)
386 		return ret;
387 
388 	return pp2_ppio_set_mtu(priv->ppio, mtu);
389 }
390 
391 /**
392  * DPDK callback to bring the link up.
393  *
394  * @param dev
395  *   Pointer to Ethernet device structure.
396  *
397  * @return
398  *   0 on success, negative error value otherwise.
399  */
400 static int
401 mrvl_dev_set_link_up(struct rte_eth_dev *dev)
402 {
403 	struct mrvl_priv *priv = dev->data->dev_private;
404 	int ret;
405 
406 	if (!priv->ppio)
407 		return -EPERM;
408 
409 	ret = pp2_ppio_enable(priv->ppio);
410 	if (ret)
411 		return ret;
412 
413 	/*
414 	 * mtu/mru can be updated if pp2_ppio_enable() was called at least once
415 	 * as pp2_ppio_enable() changes port->t_mode from default 0 to
416 	 * PP2_TRAFFIC_INGRESS_EGRESS.
417 	 *
418 	 * Set mtu to default DPDK value here.
419 	 */
420 	ret = mrvl_mtu_set(dev, dev->data->mtu);
421 	if (ret)
422 		pp2_ppio_disable(priv->ppio);
423 
424 	return ret;
425 }
426 
427 /**
428  * DPDK callback to bring the link down.
429  *
430  * @param dev
431  *   Pointer to Ethernet device structure.
432  *
433  * @return
434  *   0 on success, negative error value otherwise.
435  */
436 static int
437 mrvl_dev_set_link_down(struct rte_eth_dev *dev)
438 {
439 	struct mrvl_priv *priv = dev->data->dev_private;
440 
441 	if (!priv->ppio)
442 		return -EPERM;
443 
444 	return pp2_ppio_disable(priv->ppio);
445 }
446 
447 /**
448  * DPDK callback to start tx queue.
449  *
450  * @param dev
451  *   Pointer to Ethernet device structure.
452  * @param queue_id
453  *   Transmit queue index.
454  *
455  * @return
456  *   0 on success, negative error value otherwise.
457  */
458 static int
459 mrvl_tx_queue_start(struct rte_eth_dev *dev, uint16_t queue_id)
460 {
461 	struct mrvl_priv *priv = dev->data->dev_private;
462 	int ret;
463 
464 	if (!priv)
465 		return -EPERM;
466 
467 	/* passing 1 enables given tx queue */
468 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 1);
469 	if (ret) {
470 		RTE_LOG(ERR, PMD, "Failed to start txq %d\n", queue_id);
471 		return ret;
472 	}
473 
474 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
475 
476 	return 0;
477 }
478 
479 /**
480  * DPDK callback to stop tx queue.
481  *
482  * @param dev
483  *   Pointer to Ethernet device structure.
484  * @param queue_id
485  *   Transmit queue index.
486  *
487  * @return
488  *   0 on success, negative error value otherwise.
489  */
490 static int
491 mrvl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t queue_id)
492 {
493 	struct mrvl_priv *priv = dev->data->dev_private;
494 	int ret;
495 
496 	if (!priv->ppio)
497 		return -EPERM;
498 
499 	/* passing 0 disables given tx queue */
500 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 0);
501 	if (ret) {
502 		RTE_LOG(ERR, PMD, "Failed to stop txq %d\n", queue_id);
503 		return ret;
504 	}
505 
506 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
507 
508 	return 0;
509 }
510 
511 /**
512  * DPDK callback to start the device.
513  *
514  * @param dev
515  *   Pointer to Ethernet device structure.
516  *
517  * @return
518  *   0 on success, negative errno value on failure.
519  */
520 static int
521 mrvl_dev_start(struct rte_eth_dev *dev)
522 {
523 	struct mrvl_priv *priv = dev->data->dev_private;
524 	char match[MRVL_MATCH_LEN];
525 	int ret = 0, i, def_init_size;
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 			RTE_LOG(ERR, PMD, "Failed to add buffers to bpool\n");
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 		RTE_LOG(ERR, PMD, "Failed to init ppio\n");
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 			RTE_LOG(ERR, PMD,
578 				"Failed to flush uc/mc filter list\n");
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 			RTE_LOG(ERR, PMD, "Failed to flush vlan list\n");
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 			RTE_LOG(ERR, PMD, "Failed to setup QoS mapping\n");
602 			goto out;
603 		}
604 	}
605 
606 	ret = mrvl_dev_set_link_up(dev);
607 	if (ret) {
608 		RTE_LOG(ERR, PMD, "Failed to set link up\n");
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 	RTE_LOG(ERR, PMD, "Failed to start device\n");
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 	RTE_LOG(INFO, PMD, "Flushing rx queues\n");
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 	RTE_LOG(INFO, PMD, "Flushing tx shadow queues\n");
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 		RTE_LOG(ERR, PMD, "Failed to get bpool buffers number\n");
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 	struct mrvl_priv *priv = dev->data->dev_private;
752 
753 	mrvl_dev_set_link_down(dev);
754 	mrvl_flush_rx_queues(dev);
755 	mrvl_flush_tx_shadow_queues(dev);
756 	if (priv->cls_tbl) {
757 		pp2_cls_tbl_deinit(priv->cls_tbl);
758 		priv->cls_tbl = NULL;
759 	}
760 	if (priv->qos_tbl) {
761 		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
762 		priv->qos_tbl = NULL;
763 	}
764 	if (priv->ppio)
765 		pp2_ppio_deinit(priv->ppio);
766 	priv->ppio = NULL;
767 
768 	/* policer must be released after ppio deinitialization */
769 	if (priv->policer) {
770 		pp2_cls_plcr_deinit(priv->policer);
771 		priv->policer = NULL;
772 	}
773 }
774 
775 /**
776  * DPDK callback to close the device.
777  *
778  * @param dev
779  *   Pointer to Ethernet device structure.
780  */
781 static void
782 mrvl_dev_close(struct rte_eth_dev *dev)
783 {
784 	struct mrvl_priv *priv = dev->data->dev_private;
785 	size_t i;
786 
787 	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
788 		struct pp2_ppio_tc_params *tc_params =
789 			&priv->ppio_params.inqs_params.tcs_params[i];
790 
791 		if (tc_params->inqs_params) {
792 			rte_free(tc_params->inqs_params);
793 			tc_params->inqs_params = NULL;
794 		}
795 	}
796 
797 	mrvl_flush_bpool(dev);
798 }
799 
800 /**
801  * DPDK callback to retrieve physical link information.
802  *
803  * @param dev
804  *   Pointer to Ethernet device structure.
805  * @param wait_to_complete
806  *   Wait for request completion (ignored).
807  *
808  * @return
809  *   0 on success, negative error value otherwise.
810  */
811 static int
812 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
813 {
814 	/*
815 	 * TODO
816 	 * once MUSDK provides necessary API use it here
817 	 */
818 	struct mrvl_priv *priv = dev->data->dev_private;
819 	struct ethtool_cmd edata;
820 	struct ifreq req;
821 	int ret, fd, link_up;
822 
823 	if (!priv->ppio)
824 		return -EPERM;
825 
826 	edata.cmd = ETHTOOL_GSET;
827 
828 	strcpy(req.ifr_name, dev->data->name);
829 	req.ifr_data = (void *)&edata;
830 
831 	fd = socket(AF_INET, SOCK_DGRAM, 0);
832 	if (fd == -1)
833 		return -EFAULT;
834 
835 	ret = ioctl(fd, SIOCETHTOOL, &req);
836 	if (ret == -1) {
837 		close(fd);
838 		return -EFAULT;
839 	}
840 
841 	close(fd);
842 
843 	switch (ethtool_cmd_speed(&edata)) {
844 	case SPEED_10:
845 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
846 		break;
847 	case SPEED_100:
848 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
849 		break;
850 	case SPEED_1000:
851 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
852 		break;
853 	case SPEED_10000:
854 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
855 		break;
856 	default:
857 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
858 	}
859 
860 	dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
861 							 ETH_LINK_HALF_DUPLEX;
862 	dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
863 							   ETH_LINK_FIXED;
864 	pp2_ppio_get_link_state(priv->ppio, &link_up);
865 	dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
866 
867 	return 0;
868 }
869 
870 /**
871  * DPDK callback to enable promiscuous mode.
872  *
873  * @param dev
874  *   Pointer to Ethernet device structure.
875  */
876 static void
877 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
878 {
879 	struct mrvl_priv *priv = dev->data->dev_private;
880 	int ret;
881 
882 	if (!priv->ppio)
883 		return;
884 
885 	if (priv->isolated)
886 		return;
887 
888 	ret = pp2_ppio_set_promisc(priv->ppio, 1);
889 	if (ret)
890 		RTE_LOG(ERR, PMD, "Failed to enable promiscuous mode\n");
891 }
892 
893 /**
894  * DPDK callback to enable allmulti mode.
895  *
896  * @param dev
897  *   Pointer to Ethernet device structure.
898  */
899 static void
900 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
901 {
902 	struct mrvl_priv *priv = dev->data->dev_private;
903 	int ret;
904 
905 	if (!priv->ppio)
906 		return;
907 
908 	if (priv->isolated)
909 		return;
910 
911 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
912 	if (ret)
913 		RTE_LOG(ERR, PMD, "Failed enable all-multicast mode\n");
914 }
915 
916 /**
917  * DPDK callback to disable promiscuous mode.
918  *
919  * @param dev
920  *   Pointer to Ethernet device structure.
921  */
922 static void
923 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
924 {
925 	struct mrvl_priv *priv = dev->data->dev_private;
926 	int ret;
927 
928 	if (!priv->ppio)
929 		return;
930 
931 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
932 	if (ret)
933 		RTE_LOG(ERR, PMD, "Failed to disable promiscuous mode\n");
934 }
935 
936 /**
937  * DPDK callback to disable allmulticast mode.
938  *
939  * @param dev
940  *   Pointer to Ethernet device structure.
941  */
942 static void
943 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
944 {
945 	struct mrvl_priv *priv = dev->data->dev_private;
946 	int ret;
947 
948 	if (!priv->ppio)
949 		return;
950 
951 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
952 	if (ret)
953 		RTE_LOG(ERR, PMD, "Failed to disable all-multicast mode\n");
954 }
955 
956 /**
957  * DPDK callback to remove a MAC address.
958  *
959  * @param dev
960  *   Pointer to Ethernet device structure.
961  * @param index
962  *   MAC address index.
963  */
964 static void
965 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
966 {
967 	struct mrvl_priv *priv = dev->data->dev_private;
968 	char buf[ETHER_ADDR_FMT_SIZE];
969 	int ret;
970 
971 	if (!priv->ppio)
972 		return;
973 
974 	if (priv->isolated)
975 		return;
976 
977 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
978 				       dev->data->mac_addrs[index].addr_bytes);
979 	if (ret) {
980 		ether_format_addr(buf, sizeof(buf),
981 				  &dev->data->mac_addrs[index]);
982 		RTE_LOG(ERR, PMD, "Failed to remove mac %s\n", buf);
983 	}
984 }
985 
986 /**
987  * DPDK callback to add a MAC address.
988  *
989  * @param dev
990  *   Pointer to Ethernet device structure.
991  * @param mac_addr
992  *   MAC address to register.
993  * @param index
994  *   MAC address index.
995  * @param vmdq
996  *   VMDq pool index to associate address with (unused).
997  *
998  * @return
999  *   0 on success, negative error value otherwise.
1000  */
1001 static int
1002 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1003 		  uint32_t index, uint32_t vmdq __rte_unused)
1004 {
1005 	struct mrvl_priv *priv = dev->data->dev_private;
1006 	char buf[ETHER_ADDR_FMT_SIZE];
1007 	int ret;
1008 
1009 	if (priv->isolated)
1010 		return -ENOTSUP;
1011 
1012 	if (index == 0)
1013 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1014 		return -1;
1015 
1016 	if (!priv->ppio)
1017 		return 0;
1018 
1019 	/*
1020 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1021 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1022 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1023 	 * 21 respectively.
1024 	 *
1025 	 * If more than uc_filter_max uc addresses were added to filter list
1026 	 * then NIC will switch to promiscuous mode automatically.
1027 	 *
1028 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1029 	 * were added to filter list then NIC will switch to all-multicast mode
1030 	 * automatically.
1031 	 */
1032 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1033 	if (ret) {
1034 		ether_format_addr(buf, sizeof(buf), mac_addr);
1035 		RTE_LOG(ERR, PMD, "Failed to add mac %s\n", buf);
1036 		return -1;
1037 	}
1038 
1039 	return 0;
1040 }
1041 
1042 /**
1043  * DPDK callback to set the primary MAC address.
1044  *
1045  * @param dev
1046  *   Pointer to Ethernet device structure.
1047  * @param mac_addr
1048  *   MAC address to register.
1049  *
1050  * @return
1051  *   0 on success, negative error value otherwise.
1052  */
1053 static int
1054 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1055 {
1056 	struct mrvl_priv *priv = dev->data->dev_private;
1057 	int ret;
1058 
1059 	if (!priv->ppio)
1060 		return 0;
1061 
1062 	if (priv->isolated)
1063 		return -ENOTSUP;
1064 
1065 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1066 	if (ret) {
1067 		char buf[ETHER_ADDR_FMT_SIZE];
1068 		ether_format_addr(buf, sizeof(buf), mac_addr);
1069 		RTE_LOG(ERR, PMD, "Failed to set mac to %s\n", buf);
1070 	}
1071 
1072 	return ret;
1073 }
1074 
1075 /**
1076  * DPDK callback to get device statistics.
1077  *
1078  * @param dev
1079  *   Pointer to Ethernet device structure.
1080  * @param stats
1081  *   Stats structure output buffer.
1082  *
1083  * @return
1084  *   0 on success, negative error value otherwise.
1085  */
1086 static int
1087 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1088 {
1089 	struct mrvl_priv *priv = dev->data->dev_private;
1090 	struct pp2_ppio_statistics ppio_stats;
1091 	uint64_t drop_mac = 0;
1092 	unsigned int i, idx, ret;
1093 
1094 	if (!priv->ppio)
1095 		return -EPERM;
1096 
1097 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1098 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1099 		struct pp2_ppio_inq_statistics rx_stats;
1100 
1101 		if (!rxq)
1102 			continue;
1103 
1104 		idx = rxq->queue_id;
1105 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1106 			RTE_LOG(ERR, PMD,
1107 				"rx queue %d stats out of range (0 - %d)\n",
1108 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1109 			continue;
1110 		}
1111 
1112 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1113 						  priv->rxq_map[idx].tc,
1114 						  priv->rxq_map[idx].inq,
1115 						  &rx_stats, 0);
1116 		if (unlikely(ret)) {
1117 			RTE_LOG(ERR, PMD,
1118 				"Failed to update rx queue %d stats\n", idx);
1119 			break;
1120 		}
1121 
1122 		stats->q_ibytes[idx] = rxq->bytes_recv;
1123 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1124 		stats->q_errors[idx] = rx_stats.drop_early +
1125 				       rx_stats.drop_fullq +
1126 				       rx_stats.drop_bm +
1127 				       rxq->drop_mac;
1128 		stats->ibytes += rxq->bytes_recv;
1129 		drop_mac += rxq->drop_mac;
1130 	}
1131 
1132 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1133 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1134 		struct pp2_ppio_outq_statistics tx_stats;
1135 
1136 		if (!txq)
1137 			continue;
1138 
1139 		idx = txq->queue_id;
1140 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1141 			RTE_LOG(ERR, PMD,
1142 				"tx queue %d stats out of range (0 - %d)\n",
1143 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1144 		}
1145 
1146 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1147 						   &tx_stats, 0);
1148 		if (unlikely(ret)) {
1149 			RTE_LOG(ERR, PMD,
1150 				"Failed to update tx queue %d stats\n", idx);
1151 			break;
1152 		}
1153 
1154 		stats->q_opackets[idx] = tx_stats.deq_desc;
1155 		stats->q_obytes[idx] = txq->bytes_sent;
1156 		stats->obytes += txq->bytes_sent;
1157 	}
1158 
1159 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1160 	if (unlikely(ret)) {
1161 		RTE_LOG(ERR, PMD, "Failed to update port statistics\n");
1162 		return ret;
1163 	}
1164 
1165 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1166 	stats->opackets += ppio_stats.tx_packets;
1167 	stats->imissed += ppio_stats.rx_fullq_dropped +
1168 			  ppio_stats.rx_bm_dropped +
1169 			  ppio_stats.rx_early_dropped +
1170 			  ppio_stats.rx_fifo_dropped +
1171 			  ppio_stats.rx_cls_dropped;
1172 	stats->ierrors = drop_mac;
1173 
1174 	return 0;
1175 }
1176 
1177 /**
1178  * DPDK callback to clear device statistics.
1179  *
1180  * @param dev
1181  *   Pointer to Ethernet device structure.
1182  */
1183 static void
1184 mrvl_stats_reset(struct rte_eth_dev *dev)
1185 {
1186 	struct mrvl_priv *priv = dev->data->dev_private;
1187 	int i;
1188 
1189 	if (!priv->ppio)
1190 		return;
1191 
1192 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1193 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1194 
1195 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1196 					    priv->rxq_map[i].inq, NULL, 1);
1197 		rxq->bytes_recv = 0;
1198 		rxq->drop_mac = 0;
1199 	}
1200 
1201 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1202 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1203 
1204 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1205 		txq->bytes_sent = 0;
1206 	}
1207 
1208 	pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1209 }
1210 
1211 /**
1212  * DPDK callback to get extended statistics.
1213  *
1214  * @param dev
1215  *   Pointer to Ethernet device structure.
1216  * @param stats
1217  *   Pointer to xstats table.
1218  * @param n
1219  *   Number of entries in xstats table.
1220  * @return
1221  *   Negative value on error, number of read xstats otherwise.
1222  */
1223 static int
1224 mrvl_xstats_get(struct rte_eth_dev *dev,
1225 		struct rte_eth_xstat *stats, unsigned int n)
1226 {
1227 	struct mrvl_priv *priv = dev->data->dev_private;
1228 	struct pp2_ppio_statistics ppio_stats;
1229 	unsigned int i;
1230 
1231 	if (!stats)
1232 		return 0;
1233 
1234 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1235 	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1236 		uint64_t val;
1237 
1238 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1239 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1240 					    mrvl_xstats_tbl[i].offset);
1241 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1242 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1243 					    mrvl_xstats_tbl[i].offset);
1244 		else
1245 			return -EINVAL;
1246 
1247 		stats[i].id = i;
1248 		stats[i].value = val;
1249 	}
1250 
1251 	return n;
1252 }
1253 
1254 /**
1255  * DPDK callback to reset extended statistics.
1256  *
1257  * @param dev
1258  *   Pointer to Ethernet device structure.
1259  */
1260 static void
1261 mrvl_xstats_reset(struct rte_eth_dev *dev)
1262 {
1263 	mrvl_stats_reset(dev);
1264 }
1265 
1266 /**
1267  * DPDK callback to get extended statistics names.
1268  *
1269  * @param dev (unused)
1270  *   Pointer to Ethernet device structure.
1271  * @param xstats_names
1272  *   Pointer to xstats names table.
1273  * @param size
1274  *   Size of the xstats names table.
1275  * @return
1276  *   Number of read names.
1277  */
1278 static int
1279 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1280 		      struct rte_eth_xstat_name *xstats_names,
1281 		      unsigned int size)
1282 {
1283 	unsigned int i;
1284 
1285 	if (!xstats_names)
1286 		return RTE_DIM(mrvl_xstats_tbl);
1287 
1288 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1289 		snprintf(xstats_names[i].name, RTE_ETH_XSTATS_NAME_SIZE, "%s",
1290 			 mrvl_xstats_tbl[i].name);
1291 
1292 	return size;
1293 }
1294 
1295 /**
1296  * DPDK callback to get information about the device.
1297  *
1298  * @param dev
1299  *   Pointer to Ethernet device structure (unused).
1300  * @param info
1301  *   Info structure output buffer.
1302  */
1303 static void
1304 mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
1305 		   struct rte_eth_dev_info *info)
1306 {
1307 	info->speed_capa = ETH_LINK_SPEED_10M |
1308 			   ETH_LINK_SPEED_100M |
1309 			   ETH_LINK_SPEED_1G |
1310 			   ETH_LINK_SPEED_10G;
1311 
1312 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1313 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1314 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1315 
1316 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1317 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1318 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1319 
1320 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1321 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1322 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1323 
1324 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1325 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1326 
1327 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1328 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1329 
1330 	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1331 				       ETH_RSS_NONFRAG_IPV4_TCP |
1332 				       ETH_RSS_NONFRAG_IPV4_UDP;
1333 
1334 	/* By default packets are dropped if no descriptors are available */
1335 	info->default_rxconf.rx_drop_en = 1;
1336 	info->default_rxconf.offloads = DEV_RX_OFFLOAD_CRC_STRIP;
1337 
1338 	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1339 }
1340 
1341 /**
1342  * Return supported packet types.
1343  *
1344  * @param dev
1345  *   Pointer to Ethernet device structure (unused).
1346  *
1347  * @return
1348  *   Const pointer to the table with supported packet types.
1349  */
1350 static const uint32_t *
1351 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1352 {
1353 	static const uint32_t ptypes[] = {
1354 		RTE_PTYPE_L2_ETHER,
1355 		RTE_PTYPE_L3_IPV4,
1356 		RTE_PTYPE_L3_IPV4_EXT,
1357 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1358 		RTE_PTYPE_L3_IPV6,
1359 		RTE_PTYPE_L3_IPV6_EXT,
1360 		RTE_PTYPE_L2_ETHER_ARP,
1361 		RTE_PTYPE_L4_TCP,
1362 		RTE_PTYPE_L4_UDP
1363 	};
1364 
1365 	return ptypes;
1366 }
1367 
1368 /**
1369  * DPDK callback to get information about specific receive queue.
1370  *
1371  * @param dev
1372  *   Pointer to Ethernet device structure.
1373  * @param rx_queue_id
1374  *   Receive queue index.
1375  * @param qinfo
1376  *   Receive queue information structure.
1377  */
1378 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1379 			      struct rte_eth_rxq_info *qinfo)
1380 {
1381 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1382 	struct mrvl_priv *priv = dev->data->dev_private;
1383 	int inq = priv->rxq_map[rx_queue_id].inq;
1384 	int tc = priv->rxq_map[rx_queue_id].tc;
1385 	struct pp2_ppio_tc_params *tc_params =
1386 		&priv->ppio_params.inqs_params.tcs_params[tc];
1387 
1388 	qinfo->mp = q->mp;
1389 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1390 }
1391 
1392 /**
1393  * DPDK callback to get information about specific transmit queue.
1394  *
1395  * @param dev
1396  *   Pointer to Ethernet device structure.
1397  * @param tx_queue_id
1398  *   Transmit queue index.
1399  * @param qinfo
1400  *   Transmit queue information structure.
1401  */
1402 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1403 			      struct rte_eth_txq_info *qinfo)
1404 {
1405 	struct mrvl_priv *priv = dev->data->dev_private;
1406 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1407 
1408 	qinfo->nb_desc =
1409 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1410 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1411 }
1412 
1413 /**
1414  * DPDK callback to Configure a VLAN filter.
1415  *
1416  * @param dev
1417  *   Pointer to Ethernet device structure.
1418  * @param vlan_id
1419  *   VLAN ID to filter.
1420  * @param on
1421  *   Toggle filter.
1422  *
1423  * @return
1424  *   0 on success, negative error value otherwise.
1425  */
1426 static int
1427 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1428 {
1429 	struct mrvl_priv *priv = dev->data->dev_private;
1430 
1431 	if (!priv->ppio)
1432 		return -EPERM;
1433 
1434 	if (priv->isolated)
1435 		return -ENOTSUP;
1436 
1437 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1438 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1439 }
1440 
1441 /**
1442  * Release buffers to hardware bpool (buffer-pool)
1443  *
1444  * @param rxq
1445  *   Receive queue pointer.
1446  * @param num
1447  *   Number of buffers to release to bpool.
1448  *
1449  * @return
1450  *   0 on success, negative error value otherwise.
1451  */
1452 static int
1453 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1454 {
1455 	struct buff_release_entry entries[MRVL_PP2_RXD_MAX];
1456 	struct rte_mbuf *mbufs[MRVL_PP2_RXD_MAX];
1457 	int i, ret;
1458 	unsigned int core_id;
1459 	struct pp2_hif *hif;
1460 	struct pp2_bpool *bpool;
1461 
1462 	core_id = rte_lcore_id();
1463 	if (core_id == LCORE_ID_ANY)
1464 		core_id = 0;
1465 
1466 	hif = mrvl_get_hif(rxq->priv, core_id);
1467 	if (!hif)
1468 		return -1;
1469 
1470 	bpool = rxq->priv->bpool;
1471 
1472 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1473 	if (ret)
1474 		return ret;
1475 
1476 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1477 		cookie_addr_high =
1478 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1479 
1480 	for (i = 0; i < num; i++) {
1481 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1482 			!= cookie_addr_high) {
1483 			RTE_LOG(ERR, PMD,
1484 				"mbuf virtual addr high 0x%lx out of range\n",
1485 				(uint64_t)mbufs[i] >> 32);
1486 			goto out;
1487 		}
1488 
1489 		entries[i].buff.addr =
1490 			rte_mbuf_data_iova_default(mbufs[i]);
1491 		entries[i].buff.cookie = (pp2_cookie_t)(uint64_t)mbufs[i];
1492 		entries[i].bpool = bpool;
1493 	}
1494 
1495 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1496 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1497 
1498 	if (i != num)
1499 		goto out;
1500 
1501 	return 0;
1502 out:
1503 	for (; i < num; i++)
1504 		rte_pktmbuf_free(mbufs[i]);
1505 
1506 	return -1;
1507 }
1508 
1509 /**
1510  * DPDK callback to configure the receive queue.
1511  *
1512  * @param dev
1513  *   Pointer to Ethernet device structure.
1514  * @param idx
1515  *   RX queue index.
1516  * @param desc
1517  *   Number of descriptors to configure in queue.
1518  * @param socket
1519  *   NUMA socket on which memory must be allocated.
1520  * @param conf
1521  *   Thresholds parameters.
1522  * @param mp
1523  *   Memory pool for buffer allocations.
1524  *
1525  * @return
1526  *   0 on success, negative error value otherwise.
1527  */
1528 static int
1529 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1530 		    unsigned int socket,
1531 		    const struct rte_eth_rxconf *conf,
1532 		    struct rte_mempool *mp)
1533 {
1534 	struct mrvl_priv *priv = dev->data->dev_private;
1535 	struct mrvl_rxq *rxq;
1536 	uint32_t min_size,
1537 		 max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1538 	int ret, tc, inq;
1539 	uint64_t offloads;
1540 
1541 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
1542 
1543 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1544 		/*
1545 		 * Unknown TC mapping, mapping will not have a correct queue.
1546 		 */
1547 		RTE_LOG(ERR, PMD, "Unknown TC mapping for queue %hu eth%hhu\n",
1548 			idx, priv->ppio_id);
1549 		return -EFAULT;
1550 	}
1551 
1552 	min_size = rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM -
1553 		   MRVL_PKT_EFFEC_OFFS;
1554 	if (min_size < max_rx_pkt_len) {
1555 		RTE_LOG(ERR, PMD,
1556 			"Mbuf size must be increased to %u bytes to hold up to %u bytes of data.\n",
1557 			max_rx_pkt_len + RTE_PKTMBUF_HEADROOM +
1558 			MRVL_PKT_EFFEC_OFFS,
1559 			max_rx_pkt_len);
1560 		return -EINVAL;
1561 	}
1562 
1563 	if (dev->data->rx_queues[idx]) {
1564 		rte_free(dev->data->rx_queues[idx]);
1565 		dev->data->rx_queues[idx] = NULL;
1566 	}
1567 
1568 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
1569 	if (!rxq)
1570 		return -ENOMEM;
1571 
1572 	rxq->priv = priv;
1573 	rxq->mp = mp;
1574 	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
1575 	rxq->queue_id = idx;
1576 	rxq->port_id = dev->data->port_id;
1577 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
1578 
1579 	tc = priv->rxq_map[rxq->queue_id].tc,
1580 	inq = priv->rxq_map[rxq->queue_id].inq;
1581 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
1582 		desc;
1583 
1584 	ret = mrvl_fill_bpool(rxq, desc);
1585 	if (ret) {
1586 		rte_free(rxq);
1587 		return ret;
1588 	}
1589 
1590 	priv->bpool_init_size += desc;
1591 
1592 	dev->data->rx_queues[idx] = rxq;
1593 
1594 	return 0;
1595 }
1596 
1597 /**
1598  * DPDK callback to release the receive queue.
1599  *
1600  * @param rxq
1601  *   Generic receive queue pointer.
1602  */
1603 static void
1604 mrvl_rx_queue_release(void *rxq)
1605 {
1606 	struct mrvl_rxq *q = rxq;
1607 	struct pp2_ppio_tc_params *tc_params;
1608 	int i, num, tc, inq;
1609 	struct pp2_hif *hif;
1610 	unsigned int core_id = rte_lcore_id();
1611 
1612 	if (core_id == LCORE_ID_ANY)
1613 		core_id = 0;
1614 
1615 	hif = mrvl_get_hif(q->priv, core_id);
1616 
1617 	if (!q || !hif)
1618 		return;
1619 
1620 	tc = q->priv->rxq_map[q->queue_id].tc;
1621 	inq = q->priv->rxq_map[q->queue_id].inq;
1622 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
1623 	num = tc_params->inqs_params[inq].size;
1624 	for (i = 0; i < num; i++) {
1625 		struct pp2_buff_inf inf;
1626 		uint64_t addr;
1627 
1628 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
1629 		addr = cookie_addr_high | inf.cookie;
1630 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1631 	}
1632 
1633 	rte_free(q);
1634 }
1635 
1636 /**
1637  * DPDK callback to configure the transmit queue.
1638  *
1639  * @param dev
1640  *   Pointer to Ethernet device structure.
1641  * @param idx
1642  *   Transmit queue index.
1643  * @param desc
1644  *   Number of descriptors to configure in the queue.
1645  * @param socket
1646  *   NUMA socket on which memory must be allocated.
1647  * @param conf
1648  *   Tx queue configuration parameters.
1649  *
1650  * @return
1651  *   0 on success, negative error value otherwise.
1652  */
1653 static int
1654 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1655 		    unsigned int socket,
1656 		    const struct rte_eth_txconf *conf)
1657 {
1658 	struct mrvl_priv *priv = dev->data->dev_private;
1659 	struct mrvl_txq *txq;
1660 
1661 	if (dev->data->tx_queues[idx]) {
1662 		rte_free(dev->data->tx_queues[idx]);
1663 		dev->data->tx_queues[idx] = NULL;
1664 	}
1665 
1666 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
1667 	if (!txq)
1668 		return -ENOMEM;
1669 
1670 	txq->priv = priv;
1671 	txq->queue_id = idx;
1672 	txq->port_id = dev->data->port_id;
1673 	txq->tx_deferred_start = conf->tx_deferred_start;
1674 	dev->data->tx_queues[idx] = txq;
1675 
1676 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
1677 
1678 	return 0;
1679 }
1680 
1681 /**
1682  * DPDK callback to release the transmit queue.
1683  *
1684  * @param txq
1685  *   Generic transmit queue pointer.
1686  */
1687 static void
1688 mrvl_tx_queue_release(void *txq)
1689 {
1690 	struct mrvl_txq *q = txq;
1691 
1692 	if (!q)
1693 		return;
1694 
1695 	rte_free(q);
1696 }
1697 
1698 /**
1699  * DPDK callback to get flow control configuration.
1700  *
1701  * @param dev
1702  *  Pointer to Ethernet device structure.
1703  * @param fc_conf
1704  *  Pointer to the flow control configuration.
1705  *
1706  * @return
1707  *  0 on success, negative error value otherwise.
1708  */
1709 static int
1710 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1711 {
1712 	struct mrvl_priv *priv = dev->data->dev_private;
1713 	int ret, en;
1714 
1715 	if (!priv)
1716 		return -EPERM;
1717 
1718 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
1719 	if (ret) {
1720 		RTE_LOG(ERR, PMD, "Failed to read rx pause state\n");
1721 		return ret;
1722 	}
1723 
1724 	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
1725 
1726 	return 0;
1727 }
1728 
1729 /**
1730  * DPDK callback to set flow control configuration.
1731  *
1732  * @param dev
1733  *  Pointer to Ethernet device structure.
1734  * @param fc_conf
1735  *  Pointer to the flow control configuration.
1736  *
1737  * @return
1738  *  0 on success, negative error value otherwise.
1739  */
1740 static int
1741 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1742 {
1743 	struct mrvl_priv *priv = dev->data->dev_private;
1744 
1745 	if (!priv)
1746 		return -EPERM;
1747 
1748 	if (fc_conf->high_water ||
1749 	    fc_conf->low_water ||
1750 	    fc_conf->pause_time ||
1751 	    fc_conf->mac_ctrl_frame_fwd ||
1752 	    fc_conf->autoneg) {
1753 		RTE_LOG(ERR, PMD, "Flowctrl parameter is not supported\n");
1754 
1755 		return -EINVAL;
1756 	}
1757 
1758 	if (fc_conf->mode == RTE_FC_NONE ||
1759 	    fc_conf->mode == RTE_FC_RX_PAUSE) {
1760 		int ret, en;
1761 
1762 		en = fc_conf->mode == RTE_FC_NONE ? 0 : 1;
1763 		ret = pp2_ppio_set_rx_pause(priv->ppio, en);
1764 		if (ret)
1765 			RTE_LOG(ERR, PMD,
1766 				"Failed to change flowctrl on RX side\n");
1767 
1768 		return ret;
1769 	}
1770 
1771 	return 0;
1772 }
1773 
1774 /**
1775  * Update RSS hash configuration
1776  *
1777  * @param dev
1778  *   Pointer to Ethernet device structure.
1779  * @param rss_conf
1780  *   Pointer to RSS configuration.
1781  *
1782  * @return
1783  *   0 on success, negative error value otherwise.
1784  */
1785 static int
1786 mrvl_rss_hash_update(struct rte_eth_dev *dev,
1787 		     struct rte_eth_rss_conf *rss_conf)
1788 {
1789 	struct mrvl_priv *priv = dev->data->dev_private;
1790 
1791 	if (priv->isolated)
1792 		return -ENOTSUP;
1793 
1794 	return mrvl_configure_rss(priv, rss_conf);
1795 }
1796 
1797 /**
1798  * DPDK callback to get RSS hash configuration.
1799  *
1800  * @param dev
1801  *   Pointer to Ethernet device structure.
1802  * @rss_conf
1803  *   Pointer to RSS configuration.
1804  *
1805  * @return
1806  *   Always 0.
1807  */
1808 static int
1809 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
1810 		       struct rte_eth_rss_conf *rss_conf)
1811 {
1812 	struct mrvl_priv *priv = dev->data->dev_private;
1813 	enum pp2_ppio_hash_type hash_type =
1814 		priv->ppio_params.inqs_params.hash_type;
1815 
1816 	rss_conf->rss_key = NULL;
1817 
1818 	if (hash_type == PP2_PPIO_HASH_T_NONE)
1819 		rss_conf->rss_hf = 0;
1820 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
1821 		rss_conf->rss_hf = ETH_RSS_IPV4;
1822 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
1823 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
1824 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
1825 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
1826 
1827 	return 0;
1828 }
1829 
1830 /**
1831  * DPDK callback to get rte_flow callbacks.
1832  *
1833  * @param dev
1834  *   Pointer to the device structure.
1835  * @param filer_type
1836  *   Flow filter type.
1837  * @param filter_op
1838  *   Flow filter operation.
1839  * @param arg
1840  *   Pointer to pass the flow ops.
1841  *
1842  * @return
1843  *   0 on success, negative error value otherwise.
1844  */
1845 static int
1846 mrvl_eth_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
1847 		     enum rte_filter_type filter_type,
1848 		     enum rte_filter_op filter_op, void *arg)
1849 {
1850 	switch (filter_type) {
1851 	case RTE_ETH_FILTER_GENERIC:
1852 		if (filter_op != RTE_ETH_FILTER_GET)
1853 			return -EINVAL;
1854 		*(const void **)arg = &mrvl_flow_ops;
1855 		return 0;
1856 	default:
1857 		RTE_LOG(WARNING, PMD, "Filter type (%d) not supported",
1858 				filter_type);
1859 		return -EINVAL;
1860 	}
1861 }
1862 
1863 static const struct eth_dev_ops mrvl_ops = {
1864 	.dev_configure = mrvl_dev_configure,
1865 	.dev_start = mrvl_dev_start,
1866 	.dev_stop = mrvl_dev_stop,
1867 	.dev_set_link_up = mrvl_dev_set_link_up,
1868 	.dev_set_link_down = mrvl_dev_set_link_down,
1869 	.dev_close = mrvl_dev_close,
1870 	.link_update = mrvl_link_update,
1871 	.promiscuous_enable = mrvl_promiscuous_enable,
1872 	.allmulticast_enable = mrvl_allmulticast_enable,
1873 	.promiscuous_disable = mrvl_promiscuous_disable,
1874 	.allmulticast_disable = mrvl_allmulticast_disable,
1875 	.mac_addr_remove = mrvl_mac_addr_remove,
1876 	.mac_addr_add = mrvl_mac_addr_add,
1877 	.mac_addr_set = mrvl_mac_addr_set,
1878 	.mtu_set = mrvl_mtu_set,
1879 	.stats_get = mrvl_stats_get,
1880 	.stats_reset = mrvl_stats_reset,
1881 	.xstats_get = mrvl_xstats_get,
1882 	.xstats_reset = mrvl_xstats_reset,
1883 	.xstats_get_names = mrvl_xstats_get_names,
1884 	.dev_infos_get = mrvl_dev_infos_get,
1885 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
1886 	.rxq_info_get = mrvl_rxq_info_get,
1887 	.txq_info_get = mrvl_txq_info_get,
1888 	.vlan_filter_set = mrvl_vlan_filter_set,
1889 	.tx_queue_start = mrvl_tx_queue_start,
1890 	.tx_queue_stop = mrvl_tx_queue_stop,
1891 	.rx_queue_setup = mrvl_rx_queue_setup,
1892 	.rx_queue_release = mrvl_rx_queue_release,
1893 	.tx_queue_setup = mrvl_tx_queue_setup,
1894 	.tx_queue_release = mrvl_tx_queue_release,
1895 	.flow_ctrl_get = mrvl_flow_ctrl_get,
1896 	.flow_ctrl_set = mrvl_flow_ctrl_set,
1897 	.rss_hash_update = mrvl_rss_hash_update,
1898 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
1899 	.filter_ctrl = mrvl_eth_filter_ctrl,
1900 };
1901 
1902 /**
1903  * Return packet type information and l3/l4 offsets.
1904  *
1905  * @param desc
1906  *   Pointer to the received packet descriptor.
1907  * @param l3_offset
1908  *   l3 packet offset.
1909  * @param l4_offset
1910  *   l4 packet offset.
1911  *
1912  * @return
1913  *   Packet type information.
1914  */
1915 static inline uint64_t
1916 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
1917 				    uint8_t *l3_offset, uint8_t *l4_offset)
1918 {
1919 	enum pp2_inq_l3_type l3_type;
1920 	enum pp2_inq_l4_type l4_type;
1921 	uint64_t packet_type;
1922 
1923 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
1924 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
1925 
1926 	packet_type = RTE_PTYPE_L2_ETHER;
1927 
1928 	switch (l3_type) {
1929 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
1930 		packet_type |= RTE_PTYPE_L3_IPV4;
1931 		break;
1932 	case PP2_INQ_L3_TYPE_IPV4_OK:
1933 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
1934 		break;
1935 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
1936 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
1937 		break;
1938 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
1939 		packet_type |= RTE_PTYPE_L3_IPV6;
1940 		break;
1941 	case PP2_INQ_L3_TYPE_IPV6_EXT:
1942 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
1943 		break;
1944 	case PP2_INQ_L3_TYPE_ARP:
1945 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
1946 		/*
1947 		 * In case of ARP l4_offset is set to wrong value.
1948 		 * Set it to proper one so that later on mbuf->l3_len can be
1949 		 * calculated subtracting l4_offset and l3_offset.
1950 		 */
1951 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
1952 		break;
1953 	default:
1954 		RTE_LOG(DEBUG, PMD, "Failed to recognise l3 packet type\n");
1955 		break;
1956 	}
1957 
1958 	switch (l4_type) {
1959 	case PP2_INQ_L4_TYPE_TCP:
1960 		packet_type |= RTE_PTYPE_L4_TCP;
1961 		break;
1962 	case PP2_INQ_L4_TYPE_UDP:
1963 		packet_type |= RTE_PTYPE_L4_UDP;
1964 		break;
1965 	default:
1966 		RTE_LOG(DEBUG, PMD, "Failed to recognise l4 packet type\n");
1967 		break;
1968 	}
1969 
1970 	return packet_type;
1971 }
1972 
1973 /**
1974  * Get offload information from the received packet descriptor.
1975  *
1976  * @param desc
1977  *   Pointer to the received packet descriptor.
1978  *
1979  * @return
1980  *   Mbuf offload flags.
1981  */
1982 static inline uint64_t
1983 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc)
1984 {
1985 	uint64_t flags;
1986 	enum pp2_inq_desc_status status;
1987 
1988 	status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
1989 	if (unlikely(status != PP2_DESC_ERR_OK))
1990 		flags = PKT_RX_IP_CKSUM_BAD;
1991 	else
1992 		flags = PKT_RX_IP_CKSUM_GOOD;
1993 
1994 	status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
1995 	if (unlikely(status != PP2_DESC_ERR_OK))
1996 		flags |= PKT_RX_L4_CKSUM_BAD;
1997 	else
1998 		flags |= PKT_RX_L4_CKSUM_GOOD;
1999 
2000 	return flags;
2001 }
2002 
2003 /**
2004  * DPDK callback for receive.
2005  *
2006  * @param rxq
2007  *   Generic pointer to the receive queue.
2008  * @param rx_pkts
2009  *   Array to store received packets.
2010  * @param nb_pkts
2011  *   Maximum number of packets in array.
2012  *
2013  * @return
2014  *   Number of packets successfully received.
2015  */
2016 static uint16_t
2017 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2018 {
2019 	struct mrvl_rxq *q = rxq;
2020 	struct pp2_ppio_desc descs[nb_pkts];
2021 	struct pp2_bpool *bpool;
2022 	int i, ret, rx_done = 0;
2023 	int num;
2024 	struct pp2_hif *hif;
2025 	unsigned int core_id = rte_lcore_id();
2026 
2027 	hif = mrvl_get_hif(q->priv, core_id);
2028 
2029 	if (unlikely(!q->priv->ppio || !hif))
2030 		return 0;
2031 
2032 	bpool = q->priv->bpool;
2033 
2034 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2035 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2036 	if (unlikely(ret < 0)) {
2037 		RTE_LOG(ERR, PMD, "Failed to receive packets\n");
2038 		return 0;
2039 	}
2040 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2041 
2042 	for (i = 0; i < nb_pkts; i++) {
2043 		struct rte_mbuf *mbuf;
2044 		uint8_t l3_offset, l4_offset;
2045 		enum pp2_inq_desc_status status;
2046 		uint64_t addr;
2047 
2048 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2049 			struct pp2_ppio_desc *pref_desc;
2050 			u64 pref_addr;
2051 
2052 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2053 			pref_addr = cookie_addr_high |
2054 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2055 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2056 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2057 		}
2058 
2059 		addr = cookie_addr_high |
2060 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2061 		mbuf = (struct rte_mbuf *)addr;
2062 		rte_pktmbuf_reset(mbuf);
2063 
2064 		/* drop packet in case of mac, overrun or resource error */
2065 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2066 		if (unlikely(status != PP2_DESC_ERR_OK)) {
2067 			struct pp2_buff_inf binf = {
2068 				.addr = rte_mbuf_data_iova_default(mbuf),
2069 				.cookie = (pp2_cookie_t)(uint64_t)mbuf,
2070 			};
2071 
2072 			pp2_bpool_put_buff(hif, bpool, &binf);
2073 			mrvl_port_bpool_size
2074 				[bpool->pp2_id][bpool->id][core_id]++;
2075 			q->drop_mac++;
2076 			continue;
2077 		}
2078 
2079 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2080 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2081 		mbuf->data_len = mbuf->pkt_len;
2082 		mbuf->port = q->port_id;
2083 		mbuf->packet_type =
2084 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2085 							    &l3_offset,
2086 							    &l4_offset);
2087 		mbuf->l2_len = l3_offset;
2088 		mbuf->l3_len = l4_offset - l3_offset;
2089 
2090 		if (likely(q->cksum_enabled))
2091 			mbuf->ol_flags = mrvl_desc_to_ol_flags(&descs[i]);
2092 
2093 		rx_pkts[rx_done++] = mbuf;
2094 		q->bytes_recv += mbuf->pkt_len;
2095 	}
2096 
2097 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2098 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2099 
2100 		if (unlikely(num <= q->priv->bpool_min_size ||
2101 			     (!rx_done && num < q->priv->bpool_init_size))) {
2102 			ret = mrvl_fill_bpool(q, MRVL_BURST_SIZE);
2103 			if (ret)
2104 				RTE_LOG(ERR, PMD, "Failed to fill bpool\n");
2105 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2106 			int i;
2107 			int pkt_to_remove = num - q->priv->bpool_init_size;
2108 			struct rte_mbuf *mbuf;
2109 			struct pp2_buff_inf buff;
2110 
2111 			RTE_LOG(DEBUG, PMD,
2112 				"\nport-%d:%d: bpool %d oversize - remove %d buffers (pool size: %d -> %d)\n",
2113 				bpool->pp2_id, q->priv->ppio->port_id,
2114 				bpool->id, pkt_to_remove, num,
2115 				q->priv->bpool_init_size);
2116 
2117 			for (i = 0; i < pkt_to_remove; i++) {
2118 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2119 				if (ret)
2120 					break;
2121 				mbuf = (struct rte_mbuf *)
2122 					(cookie_addr_high | buff.cookie);
2123 				rte_pktmbuf_free(mbuf);
2124 			}
2125 			mrvl_port_bpool_size
2126 				[bpool->pp2_id][bpool->id][core_id] -= i;
2127 		}
2128 		rte_spinlock_unlock(&q->priv->lock);
2129 	}
2130 
2131 	return rx_done;
2132 }
2133 
2134 /**
2135  * Prepare offload information.
2136  *
2137  * @param ol_flags
2138  *   Offload flags.
2139  * @param packet_type
2140  *   Packet type bitfield.
2141  * @param l3_type
2142  *   Pointer to the pp2_ouq_l3_type structure.
2143  * @param l4_type
2144  *   Pointer to the pp2_outq_l4_type structure.
2145  * @param gen_l3_cksum
2146  *   Will be set to 1 in case l3 checksum is computed.
2147  * @param l4_cksum
2148  *   Will be set to 1 in case l4 checksum is computed.
2149  *
2150  * @return
2151  *   0 on success, negative error value otherwise.
2152  */
2153 static inline int
2154 mrvl_prepare_proto_info(uint64_t ol_flags, uint32_t packet_type,
2155 			enum pp2_outq_l3_type *l3_type,
2156 			enum pp2_outq_l4_type *l4_type,
2157 			int *gen_l3_cksum,
2158 			int *gen_l4_cksum)
2159 {
2160 	/*
2161 	 * Based on ol_flags prepare information
2162 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2163 	 * for offloading.
2164 	 */
2165 	if (ol_flags & PKT_TX_IPV4) {
2166 		*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2167 		*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2168 	} else if (ol_flags & PKT_TX_IPV6) {
2169 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2170 		/* no checksum for ipv6 header */
2171 		*gen_l3_cksum = 0;
2172 	} else {
2173 		/* if something different then stop processing */
2174 		return -1;
2175 	}
2176 
2177 	ol_flags &= PKT_TX_L4_MASK;
2178 	if ((packet_type & RTE_PTYPE_L4_TCP) &&
2179 	    ol_flags == PKT_TX_TCP_CKSUM) {
2180 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2181 		*gen_l4_cksum = 1;
2182 	} else if ((packet_type & RTE_PTYPE_L4_UDP) &&
2183 		   ol_flags == PKT_TX_UDP_CKSUM) {
2184 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2185 		*gen_l4_cksum = 1;
2186 	} else {
2187 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2188 		/* no checksum for other type */
2189 		*gen_l4_cksum = 0;
2190 	}
2191 
2192 	return 0;
2193 }
2194 
2195 /**
2196  * Release already sent buffers to bpool (buffer-pool).
2197  *
2198  * @param ppio
2199  *   Pointer to the port structure.
2200  * @param hif
2201  *   Pointer to the MUSDK hardware interface.
2202  * @param sq
2203  *   Pointer to the shadow queue.
2204  * @param qid
2205  *   Queue id number.
2206  * @param force
2207  *   Force releasing packets.
2208  */
2209 static inline void
2210 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2211 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2212 		       int qid, int force)
2213 {
2214 	struct buff_release_entry *entry;
2215 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2216 	int i;
2217 
2218 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2219 
2220 	sq->num_to_release += nb_done;
2221 
2222 	if (likely(!force &&
2223 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2224 		return;
2225 
2226 	nb_done = sq->num_to_release;
2227 	sq->num_to_release = 0;
2228 
2229 	for (i = 0; i < nb_done; i++) {
2230 		entry = &sq->ent[sq->tail + num];
2231 		if (unlikely(!entry->buff.addr)) {
2232 			RTE_LOG(ERR, PMD,
2233 				"Shadow memory @%d: cookie(%lx), pa(%lx)!\n",
2234 				sq->tail, (u64)entry->buff.cookie,
2235 				(u64)entry->buff.addr);
2236 			skip_bufs = 1;
2237 			goto skip;
2238 		}
2239 
2240 		if (unlikely(!entry->bpool)) {
2241 			struct rte_mbuf *mbuf;
2242 
2243 			mbuf = (struct rte_mbuf *)
2244 			       (cookie_addr_high | entry->buff.cookie);
2245 			rte_pktmbuf_free(mbuf);
2246 			skip_bufs = 1;
2247 			goto skip;
2248 		}
2249 
2250 		mrvl_port_bpool_size
2251 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2252 		num++;
2253 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2254 			goto skip;
2255 		continue;
2256 skip:
2257 		if (likely(num))
2258 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2259 		num += skip_bufs;
2260 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2261 		sq->size -= num;
2262 		num = 0;
2263 		skip_bufs = 0;
2264 	}
2265 
2266 	if (likely(num)) {
2267 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2268 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2269 		sq->size -= num;
2270 	}
2271 }
2272 
2273 /**
2274  * DPDK callback for transmit.
2275  *
2276  * @param txq
2277  *   Generic pointer transmit queue.
2278  * @param tx_pkts
2279  *   Packets to transmit.
2280  * @param nb_pkts
2281  *   Number of packets in array.
2282  *
2283  * @return
2284  *   Number of packets successfully transmitted.
2285  */
2286 static uint16_t
2287 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2288 {
2289 	struct mrvl_txq *q = txq;
2290 	struct mrvl_shadow_txq *sq;
2291 	struct pp2_hif *hif;
2292 	struct pp2_ppio_desc descs[nb_pkts];
2293 	unsigned int core_id = rte_lcore_id();
2294 	int i, ret, bytes_sent = 0;
2295 	uint16_t num, sq_free_size;
2296 	uint64_t addr;
2297 
2298 	hif = mrvl_get_hif(q->priv, core_id);
2299 	sq = &q->shadow_txqs[core_id];
2300 
2301 	if (unlikely(!q->priv->ppio || !hif))
2302 		return 0;
2303 
2304 	if (sq->size)
2305 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2306 				       sq, q->queue_id, 0);
2307 
2308 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2309 	if (unlikely(nb_pkts > sq_free_size)) {
2310 		RTE_LOG(DEBUG, PMD,
2311 			"No room in shadow queue for %d packets! %d packets will be sent.\n",
2312 			nb_pkts, sq_free_size);
2313 		nb_pkts = sq_free_size;
2314 	}
2315 
2316 	for (i = 0; i < nb_pkts; i++) {
2317 		struct rte_mbuf *mbuf = tx_pkts[i];
2318 		int gen_l3_cksum, gen_l4_cksum;
2319 		enum pp2_outq_l3_type l3_type;
2320 		enum pp2_outq_l4_type l4_type;
2321 
2322 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2323 			struct rte_mbuf *pref_pkt_hdr;
2324 
2325 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2326 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2327 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2328 		}
2329 
2330 		sq->ent[sq->head].buff.cookie = (pp2_cookie_t)(uint64_t)mbuf;
2331 		sq->ent[sq->head].buff.addr =
2332 			rte_mbuf_data_iova_default(mbuf);
2333 		sq->ent[sq->head].bpool =
2334 			(unlikely(mbuf->port >= RTE_MAX_ETHPORTS ||
2335 			 mbuf->refcnt > 1)) ? NULL :
2336 			 mrvl_port_to_bpool_lookup[mbuf->port];
2337 		sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
2338 		sq->size++;
2339 
2340 		pp2_ppio_outq_desc_reset(&descs[i]);
2341 		pp2_ppio_outq_desc_set_phys_addr(&descs[i],
2342 						 rte_pktmbuf_iova(mbuf));
2343 		pp2_ppio_outq_desc_set_pkt_offset(&descs[i], 0);
2344 		pp2_ppio_outq_desc_set_pkt_len(&descs[i],
2345 					       rte_pktmbuf_pkt_len(mbuf));
2346 
2347 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2348 		/*
2349 		 * in case unsupported ol_flags were passed
2350 		 * do not update descriptor offload information
2351 		 */
2352 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2353 					      &l3_type, &l4_type, &gen_l3_cksum,
2354 					      &gen_l4_cksum);
2355 		if (unlikely(ret))
2356 			continue;
2357 
2358 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2359 						  mbuf->l2_len,
2360 						  mbuf->l2_len + mbuf->l3_len,
2361 						  gen_l3_cksum, gen_l4_cksum);
2362 	}
2363 
2364 	num = nb_pkts;
2365 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2366 	/* number of packets that were not sent */
2367 	if (unlikely(num > nb_pkts)) {
2368 		for (i = nb_pkts; i < num; i++) {
2369 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2370 				MRVL_PP2_TX_SHADOWQ_MASK;
2371 			addr = cookie_addr_high | sq->ent[sq->head].buff.cookie;
2372 			bytes_sent -=
2373 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2374 		}
2375 		sq->size -= num - nb_pkts;
2376 	}
2377 
2378 	q->bytes_sent += bytes_sent;
2379 
2380 	return nb_pkts;
2381 }
2382 
2383 /**
2384  * Initialize packet processor.
2385  *
2386  * @return
2387  *   0 on success, negative error value otherwise.
2388  */
2389 static int
2390 mrvl_init_pp2(void)
2391 {
2392 	struct pp2_init_params init_params;
2393 
2394 	memset(&init_params, 0, sizeof(init_params));
2395 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
2396 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
2397 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
2398 
2399 	return pp2_init(&init_params);
2400 }
2401 
2402 /**
2403  * Deinitialize packet processor.
2404  *
2405  * @return
2406  *   0 on success, negative error value otherwise.
2407  */
2408 static void
2409 mrvl_deinit_pp2(void)
2410 {
2411 	pp2_deinit();
2412 }
2413 
2414 /**
2415  * Create private device structure.
2416  *
2417  * @param dev_name
2418  *   Pointer to the port name passed in the initialization parameters.
2419  *
2420  * @return
2421  *   Pointer to the newly allocated private device structure.
2422  */
2423 static struct mrvl_priv *
2424 mrvl_priv_create(const char *dev_name)
2425 {
2426 	struct pp2_bpool_params bpool_params;
2427 	char match[MRVL_MATCH_LEN];
2428 	struct mrvl_priv *priv;
2429 	int ret, bpool_bit;
2430 
2431 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
2432 	if (!priv)
2433 		return NULL;
2434 
2435 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
2436 				       &priv->pp_id, &priv->ppio_id);
2437 	if (ret)
2438 		goto out_free_priv;
2439 
2440 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
2441 				     PP2_BPOOL_NUM_POOLS);
2442 	if (bpool_bit < 0)
2443 		goto out_free_priv;
2444 	priv->bpool_bit = bpool_bit;
2445 
2446 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
2447 		 priv->bpool_bit);
2448 	memset(&bpool_params, 0, sizeof(bpool_params));
2449 	bpool_params.match = match;
2450 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
2451 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
2452 	if (ret)
2453 		goto out_clear_bpool_bit;
2454 
2455 	priv->ppio_params.type = PP2_PPIO_T_NIC;
2456 	rte_spinlock_init(&priv->lock);
2457 
2458 	return priv;
2459 out_clear_bpool_bit:
2460 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2461 out_free_priv:
2462 	rte_free(priv);
2463 	return NULL;
2464 }
2465 
2466 /**
2467  * Create device representing Ethernet port.
2468  *
2469  * @param name
2470  *   Pointer to the port's name.
2471  *
2472  * @return
2473  *   0 on success, negative error value otherwise.
2474  */
2475 static int
2476 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
2477 {
2478 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
2479 	struct rte_eth_dev *eth_dev;
2480 	struct mrvl_priv *priv;
2481 	struct ifreq req;
2482 
2483 	eth_dev = rte_eth_dev_allocate(name);
2484 	if (!eth_dev)
2485 		return -ENOMEM;
2486 
2487 	priv = mrvl_priv_create(name);
2488 	if (!priv) {
2489 		ret = -ENOMEM;
2490 		goto out_free_dev;
2491 	}
2492 
2493 	eth_dev->data->mac_addrs =
2494 		rte_zmalloc("mac_addrs",
2495 			    ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
2496 	if (!eth_dev->data->mac_addrs) {
2497 		RTE_LOG(ERR, PMD, "Failed to allocate space for eth addrs\n");
2498 		ret = -ENOMEM;
2499 		goto out_free_priv;
2500 	}
2501 
2502 	memset(&req, 0, sizeof(req));
2503 	strcpy(req.ifr_name, name);
2504 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
2505 	if (ret)
2506 		goto out_free_mac;
2507 
2508 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
2509 	       req.ifr_addr.sa_data, ETHER_ADDR_LEN);
2510 
2511 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
2512 	eth_dev->tx_pkt_burst = mrvl_tx_pkt_burst;
2513 	eth_dev->data->kdrv = RTE_KDRV_NONE;
2514 	eth_dev->data->dev_private = priv;
2515 	eth_dev->device = &vdev->device;
2516 	eth_dev->dev_ops = &mrvl_ops;
2517 
2518 	rte_eth_dev_probing_finish(eth_dev);
2519 	return 0;
2520 out_free_mac:
2521 	rte_free(eth_dev->data->mac_addrs);
2522 out_free_dev:
2523 	rte_eth_dev_release_port(eth_dev);
2524 out_free_priv:
2525 	rte_free(priv);
2526 
2527 	return ret;
2528 }
2529 
2530 /**
2531  * Cleanup previously created device representing Ethernet port.
2532  *
2533  * @param name
2534  *   Pointer to the port name.
2535  */
2536 static void
2537 mrvl_eth_dev_destroy(const char *name)
2538 {
2539 	struct rte_eth_dev *eth_dev;
2540 	struct mrvl_priv *priv;
2541 
2542 	eth_dev = rte_eth_dev_allocated(name);
2543 	if (!eth_dev)
2544 		return;
2545 
2546 	priv = eth_dev->data->dev_private;
2547 	pp2_bpool_deinit(priv->bpool);
2548 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2549 	rte_free(priv);
2550 	rte_free(eth_dev->data->mac_addrs);
2551 	rte_eth_dev_release_port(eth_dev);
2552 }
2553 
2554 /**
2555  * Callback used by rte_kvargs_process() during argument parsing.
2556  *
2557  * @param key
2558  *   Pointer to the parsed key (unused).
2559  * @param value
2560  *   Pointer to the parsed value.
2561  * @param extra_args
2562  *   Pointer to the extra arguments which contains address of the
2563  *   table of pointers to parsed interface names.
2564  *
2565  * @return
2566  *   Always 0.
2567  */
2568 static int
2569 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
2570 		 void *extra_args)
2571 {
2572 	struct mrvl_ifnames *ifnames = extra_args;
2573 
2574 	ifnames->names[ifnames->idx++] = value;
2575 
2576 	return 0;
2577 }
2578 
2579 /**
2580  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
2581  */
2582 static void
2583 mrvl_deinit_hifs(void)
2584 {
2585 	int i;
2586 
2587 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
2588 		if (hifs[i])
2589 			pp2_hif_deinit(hifs[i]);
2590 	}
2591 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
2592 	memset(hifs, 0, sizeof(hifs));
2593 }
2594 
2595 /**
2596  * DPDK callback to register the virtual device.
2597  *
2598  * @param vdev
2599  *   Pointer to the virtual device.
2600  *
2601  * @return
2602  *   0 on success, negative error value otherwise.
2603  */
2604 static int
2605 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
2606 {
2607 	struct rte_kvargs *kvlist;
2608 	struct mrvl_ifnames ifnames;
2609 	int ret = -EINVAL;
2610 	uint32_t i, ifnum, cfgnum;
2611 	const char *params;
2612 
2613 	params = rte_vdev_device_args(vdev);
2614 	if (!params)
2615 		return -EINVAL;
2616 
2617 	kvlist = rte_kvargs_parse(params, valid_args);
2618 	if (!kvlist)
2619 		return -EINVAL;
2620 
2621 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
2622 	if (ifnum > RTE_DIM(ifnames.names))
2623 		goto out_free_kvlist;
2624 
2625 	ifnames.idx = 0;
2626 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
2627 			   mrvl_get_ifnames, &ifnames);
2628 
2629 
2630 	/*
2631 	 * The below system initialization should be done only once,
2632 	 * on the first provided configuration file
2633 	 */
2634 	if (!mrvl_qos_cfg) {
2635 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
2636 		RTE_LOG(INFO, PMD, "Parsing config file!\n");
2637 		if (cfgnum > 1) {
2638 			RTE_LOG(ERR, PMD, "Cannot handle more than one config file!\n");
2639 			goto out_free_kvlist;
2640 		} else if (cfgnum == 1) {
2641 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
2642 					   mrvl_get_qoscfg, &mrvl_qos_cfg);
2643 		}
2644 	}
2645 
2646 	if (mrvl_dev_num)
2647 		goto init_devices;
2648 
2649 	RTE_LOG(INFO, PMD, "Perform MUSDK initializations\n");
2650 	/*
2651 	 * ret == -EEXIST is correct, it means DMA
2652 	 * has been already initialized (by another PMD).
2653 	 */
2654 	ret = mv_sys_dma_mem_init(MRVL_MUSDK_DMA_MEMSIZE);
2655 	if (ret < 0) {
2656 		if (ret != -EEXIST)
2657 			goto out_free_kvlist;
2658 		else
2659 			RTE_LOG(INFO, PMD,
2660 				"DMA memory has been already initialized by a different driver.\n");
2661 	}
2662 
2663 	ret = mrvl_init_pp2();
2664 	if (ret) {
2665 		RTE_LOG(ERR, PMD, "Failed to init PP!\n");
2666 		goto out_deinit_dma;
2667 	}
2668 
2669 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
2670 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
2671 
2672 	mrvl_lcore_first = RTE_MAX_LCORE;
2673 	mrvl_lcore_last = 0;
2674 
2675 init_devices:
2676 	for (i = 0; i < ifnum; i++) {
2677 		RTE_LOG(INFO, PMD, "Creating %s\n", ifnames.names[i]);
2678 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
2679 		if (ret)
2680 			goto out_cleanup;
2681 	}
2682 	mrvl_dev_num += ifnum;
2683 
2684 	rte_kvargs_free(kvlist);
2685 
2686 	return 0;
2687 out_cleanup:
2688 	for (; i > 0; i--)
2689 		mrvl_eth_dev_destroy(ifnames.names[i]);
2690 
2691 	if (mrvl_dev_num == 0)
2692 		mrvl_deinit_pp2();
2693 out_deinit_dma:
2694 	if (mrvl_dev_num == 0)
2695 		mv_sys_dma_mem_destroy();
2696 out_free_kvlist:
2697 	rte_kvargs_free(kvlist);
2698 
2699 	return ret;
2700 }
2701 
2702 /**
2703  * DPDK callback to remove virtual device.
2704  *
2705  * @param vdev
2706  *   Pointer to the removed virtual device.
2707  *
2708  * @return
2709  *   0 on success, negative error value otherwise.
2710  */
2711 static int
2712 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
2713 {
2714 	int i;
2715 	const char *name;
2716 
2717 	name = rte_vdev_device_name(vdev);
2718 	if (!name)
2719 		return -EINVAL;
2720 
2721 	RTE_LOG(INFO, PMD, "Removing %s\n", name);
2722 
2723 	RTE_ETH_FOREACH_DEV(i) { /* FIXME: removing all devices! */
2724 		char ifname[RTE_ETH_NAME_MAX_LEN];
2725 
2726 		rte_eth_dev_get_name_by_port(i, ifname);
2727 		mrvl_eth_dev_destroy(ifname);
2728 		mrvl_dev_num--;
2729 	}
2730 
2731 	if (mrvl_dev_num == 0) {
2732 		RTE_LOG(INFO, PMD, "Perform MUSDK deinit\n");
2733 		mrvl_deinit_hifs();
2734 		mrvl_deinit_pp2();
2735 		mv_sys_dma_mem_destroy();
2736 	}
2737 
2738 	return 0;
2739 }
2740 
2741 static struct rte_vdev_driver pmd_mrvl_drv = {
2742 	.probe = rte_pmd_mrvl_probe,
2743 	.remove = rte_pmd_mrvl_remove,
2744 };
2745 
2746 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
2747 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
2748