xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision 5ecb687a5698d2d8ec1f3b3b5a7a16bceca3e29c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2017 Marvell International Ltd.
3  * Copyright(c) 2017 Semihalf.
4  * All rights reserved.
5  */
6 
7 #include <rte_string_fns.h>
8 #include <rte_ethdev_driver.h>
9 #include <rte_kvargs.h>
10 #include <rte_log.h>
11 #include <rte_malloc.h>
12 #include <rte_bus_vdev.h>
13 
14 #include <fcntl.h>
15 #include <linux/ethtool.h>
16 #include <linux/sockios.h>
17 #include <net/if.h>
18 #include <net/if_arp.h>
19 #include <sys/ioctl.h>
20 #include <sys/socket.h>
21 #include <sys/stat.h>
22 #include <sys/types.h>
23 
24 #include <rte_mvep_common.h>
25 #include "mrvl_ethdev.h"
26 #include "mrvl_qos.h"
27 #include "mrvl_flow.h"
28 #include "mrvl_mtr.h"
29 #include "mrvl_tm.h"
30 
31 /* bitmask with reserved hifs */
32 #define MRVL_MUSDK_HIFS_RESERVED 0x0F
33 /* bitmask with reserved bpools */
34 #define MRVL_MUSDK_BPOOLS_RESERVED 0x07
35 /* bitmask with reserved kernel RSS tables */
36 #define MRVL_MUSDK_RSS_RESERVED 0x01
37 /* maximum number of available hifs */
38 #define MRVL_MUSDK_HIFS_MAX 9
39 
40 /* prefetch shift */
41 #define MRVL_MUSDK_PREFETCH_SHIFT 2
42 
43 /* TCAM has 25 entries reserved for uc/mc filter entries */
44 #define MRVL_MAC_ADDRS_MAX 25
45 #define MRVL_MATCH_LEN 16
46 #define MRVL_PKT_EFFEC_OFFS (MRVL_PKT_OFFS + MV_MH_SIZE)
47 /* Maximum allowable packet size */
48 #define MRVL_PKT_SIZE_MAX (10240 - MV_MH_SIZE)
49 
50 #define MRVL_IFACE_NAME_ARG "iface"
51 #define MRVL_CFG_ARG "cfg"
52 
53 #define MRVL_BURST_SIZE 64
54 
55 #define MRVL_ARP_LENGTH 28
56 
57 #define MRVL_COOKIE_ADDR_INVALID ~0ULL
58 #define MRVL_COOKIE_HIGH_ADDR_MASK 0xffffff0000000000
59 
60 /** Port Rx offload capabilities */
61 #define MRVL_RX_OFFLOADS (DEV_RX_OFFLOAD_VLAN_FILTER | \
62 			  DEV_RX_OFFLOAD_JUMBO_FRAME | \
63 			  DEV_RX_OFFLOAD_CHECKSUM)
64 
65 /** Port Tx offloads capabilities */
66 #define MRVL_TX_OFFLOADS (DEV_TX_OFFLOAD_IPV4_CKSUM | \
67 			  DEV_TX_OFFLOAD_UDP_CKSUM | \
68 			  DEV_TX_OFFLOAD_TCP_CKSUM | \
69 			  DEV_TX_OFFLOAD_MULTI_SEGS)
70 
71 static const char * const valid_args[] = {
72 	MRVL_IFACE_NAME_ARG,
73 	MRVL_CFG_ARG,
74 	NULL
75 };
76 
77 static int used_hifs = MRVL_MUSDK_HIFS_RESERVED;
78 static struct pp2_hif *hifs[RTE_MAX_LCORE];
79 static int used_bpools[PP2_NUM_PKT_PROC] = {
80 	[0 ... PP2_NUM_PKT_PROC - 1] = MRVL_MUSDK_BPOOLS_RESERVED
81 };
82 
83 static struct pp2_bpool *mrvl_port_to_bpool_lookup[RTE_MAX_ETHPORTS];
84 static int mrvl_port_bpool_size[PP2_NUM_PKT_PROC][PP2_BPOOL_NUM_POOLS][RTE_MAX_LCORE];
85 static uint64_t cookie_addr_high = MRVL_COOKIE_ADDR_INVALID;
86 
87 int mrvl_logtype;
88 
89 struct mrvl_ifnames {
90 	const char *names[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
91 	int idx;
92 };
93 
94 /*
95  * To use buffer harvesting based on loopback port shadow queue structure
96  * was introduced for buffers information bookkeeping.
97  *
98  * Before sending the packet, related buffer information (pp2_buff_inf) is
99  * stored in shadow queue. After packet is transmitted no longer used
100  * packet buffer is released back to it's original hardware pool,
101  * on condition it originated from interface.
102  * In case it  was generated by application itself i.e: mbuf->port field is
103  * 0xff then its released to software mempool.
104  */
105 struct mrvl_shadow_txq {
106 	int head;           /* write index - used when sending buffers */
107 	int tail;           /* read index - used when releasing buffers */
108 	u16 size;           /* queue occupied size */
109 	u16 num_to_release; /* number of descriptors sent, that can be
110 			     * released
111 			     */
112 	struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
113 };
114 
115 struct mrvl_rxq {
116 	struct mrvl_priv *priv;
117 	struct rte_mempool *mp;
118 	int queue_id;
119 	int port_id;
120 	int cksum_enabled;
121 	uint64_t bytes_recv;
122 	uint64_t drop_mac;
123 };
124 
125 struct mrvl_txq {
126 	struct mrvl_priv *priv;
127 	int queue_id;
128 	int port_id;
129 	uint64_t bytes_sent;
130 	struct mrvl_shadow_txq shadow_txqs[RTE_MAX_LCORE];
131 	int tx_deferred_start;
132 };
133 
134 static int mrvl_lcore_first;
135 static int mrvl_lcore_last;
136 static int mrvl_dev_num;
137 
138 static int mrvl_fill_bpool(struct mrvl_rxq *rxq, int num);
139 static inline void mrvl_free_sent_buffers(struct pp2_ppio *ppio,
140 			struct pp2_hif *hif, unsigned int core_id,
141 			struct mrvl_shadow_txq *sq, int qid, int force);
142 
143 static uint16_t mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
144 				  uint16_t nb_pkts);
145 static uint16_t mrvl_tx_sg_pkt_burst(void *txq,	struct rte_mbuf **tx_pkts,
146 				     uint16_t nb_pkts);
147 
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 void
176 mrvl_fill_shadowq(struct mrvl_shadow_txq *sq, struct rte_mbuf *buf)
177 {
178 	sq->ent[sq->head].buff.cookie = (uint64_t)buf;
179 	sq->ent[sq->head].buff.addr = buf ?
180 		rte_mbuf_data_iova_default(buf) : 0;
181 
182 	sq->ent[sq->head].bpool =
183 		(unlikely(!buf || buf->port >= RTE_MAX_ETHPORTS ||
184 		 buf->refcnt > 1)) ? NULL :
185 		 mrvl_port_to_bpool_lookup[buf->port];
186 
187 	sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
188 	sq->size++;
189 }
190 
191 static inline void
192 mrvl_fill_desc(struct pp2_ppio_desc *desc, struct rte_mbuf *buf)
193 {
194 	pp2_ppio_outq_desc_reset(desc);
195 	pp2_ppio_outq_desc_set_phys_addr(desc, rte_pktmbuf_iova(buf));
196 	pp2_ppio_outq_desc_set_pkt_offset(desc, 0);
197 	pp2_ppio_outq_desc_set_pkt_len(desc, rte_pktmbuf_data_len(buf));
198 }
199 
200 static inline int
201 mrvl_get_bpool_size(int pp2_id, int pool_id)
202 {
203 	int i;
204 	int size = 0;
205 
206 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
207 		size += mrvl_port_bpool_size[pp2_id][pool_id][i];
208 
209 	return size;
210 }
211 
212 static inline int
213 mrvl_reserve_bit(int *bitmap, int max)
214 {
215 	int n = sizeof(*bitmap) * 8 - __builtin_clz(*bitmap);
216 
217 	if (n >= max)
218 		return -1;
219 
220 	*bitmap |= 1 << n;
221 
222 	return n;
223 }
224 
225 static int
226 mrvl_init_hif(int core_id)
227 {
228 	struct pp2_hif_params params;
229 	char match[MRVL_MATCH_LEN];
230 	int ret;
231 
232 	ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
233 	if (ret < 0) {
234 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
235 		return ret;
236 	}
237 
238 	snprintf(match, sizeof(match), "hif-%d", ret);
239 	memset(&params, 0, sizeof(params));
240 	params.match = match;
241 	params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
242 	ret = pp2_hif_init(&params, &hifs[core_id]);
243 	if (ret) {
244 		MRVL_LOG(ERR, "Failed to initialize hif %d", core_id);
245 		return ret;
246 	}
247 
248 	return 0;
249 }
250 
251 static inline struct pp2_hif*
252 mrvl_get_hif(struct mrvl_priv *priv, int core_id)
253 {
254 	int ret;
255 
256 	if (likely(hifs[core_id] != NULL))
257 		return hifs[core_id];
258 
259 	rte_spinlock_lock(&priv->lock);
260 
261 	ret = mrvl_init_hif(core_id);
262 	if (ret < 0) {
263 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
264 		goto out;
265 	}
266 
267 	if (core_id < mrvl_lcore_first)
268 		mrvl_lcore_first = core_id;
269 
270 	if (core_id > mrvl_lcore_last)
271 		mrvl_lcore_last = core_id;
272 out:
273 	rte_spinlock_unlock(&priv->lock);
274 
275 	return hifs[core_id];
276 }
277 
278 /**
279  * Set tx burst function according to offload flag
280  *
281  * @param dev
282  *   Pointer to Ethernet device structure.
283  */
284 static void
285 mrvl_set_tx_function(struct rte_eth_dev *dev)
286 {
287 	struct mrvl_priv *priv = dev->data->dev_private;
288 
289 	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
290 	if (priv->multiseg) {
291 		RTE_LOG(INFO, PMD, "Using multi-segment tx callback\n");
292 		dev->tx_pkt_burst = mrvl_tx_sg_pkt_burst;
293 	} else {
294 		RTE_LOG(INFO, PMD, "Using single-segment tx callback\n");
295 		dev->tx_pkt_burst = mrvl_tx_pkt_burst;
296 	}
297 }
298 
299 /**
300  * Configure rss based on dpdk rss configuration.
301  *
302  * @param priv
303  *   Pointer to private structure.
304  * @param rss_conf
305  *   Pointer to RSS configuration.
306  *
307  * @return
308  *   0 on success, negative error value otherwise.
309  */
310 static int
311 mrvl_configure_rss(struct mrvl_priv *priv, struct rte_eth_rss_conf *rss_conf)
312 {
313 	if (rss_conf->rss_key)
314 		MRVL_LOG(WARNING, "Changing hash key is not supported");
315 
316 	if (rss_conf->rss_hf == 0) {
317 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
318 	} else if (rss_conf->rss_hf & ETH_RSS_IPV4) {
319 		priv->ppio_params.inqs_params.hash_type =
320 			PP2_PPIO_HASH_T_2_TUPLE;
321 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP) {
322 		priv->ppio_params.inqs_params.hash_type =
323 			PP2_PPIO_HASH_T_5_TUPLE;
324 		priv->rss_hf_tcp = 1;
325 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) {
326 		priv->ppio_params.inqs_params.hash_type =
327 			PP2_PPIO_HASH_T_5_TUPLE;
328 		priv->rss_hf_tcp = 0;
329 	} else {
330 		return -EINVAL;
331 	}
332 
333 	return 0;
334 }
335 
336 /**
337  * Ethernet device configuration.
338  *
339  * Prepare the driver for a given number of TX and RX queues and
340  * configure RSS.
341  *
342  * @param dev
343  *   Pointer to Ethernet device structure.
344  *
345  * @return
346  *   0 on success, negative error value otherwise.
347  */
348 static int
349 mrvl_dev_configure(struct rte_eth_dev *dev)
350 {
351 	struct mrvl_priv *priv = dev->data->dev_private;
352 	int ret;
353 
354 	if (priv->ppio) {
355 		MRVL_LOG(INFO, "Device reconfiguration is not supported");
356 		return -EINVAL;
357 	}
358 
359 	if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_NONE &&
360 	    dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
361 		MRVL_LOG(INFO, "Unsupported rx multi queue mode %d",
362 			dev->data->dev_conf.rxmode.mq_mode);
363 		return -EINVAL;
364 	}
365 
366 	if (dev->data->dev_conf.rxmode.split_hdr_size) {
367 		MRVL_LOG(INFO, "Split headers not supported");
368 		return -EINVAL;
369 	}
370 
371 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
372 		dev->data->mtu = dev->data->dev_conf.rxmode.max_rx_pkt_len -
373 				 MRVL_PP2_ETH_HDRS_LEN;
374 
375 	if (dev->data->dev_conf.txmode.offloads & DEV_TX_OFFLOAD_MULTI_SEGS)
376 		priv->multiseg = 1;
377 
378 	ret = mrvl_configure_rxqs(priv, dev->data->port_id,
379 				  dev->data->nb_rx_queues);
380 	if (ret < 0)
381 		return ret;
382 
383 	ret = mrvl_configure_txqs(priv, dev->data->port_id,
384 				  dev->data->nb_tx_queues);
385 	if (ret < 0)
386 		return ret;
387 
388 	priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
389 	priv->ppio_params.maintain_stats = 1;
390 	priv->nb_rx_queues = dev->data->nb_rx_queues;
391 
392 	ret = mrvl_tm_init(dev);
393 	if (ret < 0)
394 		return ret;
395 
396 	if (dev->data->nb_rx_queues == 1 &&
397 	    dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
398 		MRVL_LOG(WARNING, "Disabling hash for 1 rx queue");
399 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
400 
401 		return 0;
402 	}
403 
404 	return mrvl_configure_rss(priv,
405 				  &dev->data->dev_conf.rx_adv_conf.rss_conf);
406 }
407 
408 /**
409  * DPDK callback to change the MTU.
410  *
411  * Setting the MTU affects hardware MRU (packets larger than the MRU
412  * will be dropped).
413  *
414  * @param dev
415  *   Pointer to Ethernet device structure.
416  * @param mtu
417  *   New MTU.
418  *
419  * @return
420  *   0 on success, negative error value otherwise.
421  */
422 static int
423 mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
424 {
425 	struct mrvl_priv *priv = dev->data->dev_private;
426 	uint16_t mru;
427 	uint16_t mbuf_data_size = 0; /* SW buffer size */
428 	int ret;
429 
430 	mru = MRVL_PP2_MTU_TO_MRU(mtu);
431 	/*
432 	 * min_rx_buf_size is equal to mbuf data size
433 	 * if pmd didn't set it differently
434 	 */
435 	mbuf_data_size = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
436 	/* Prevent PMD from:
437 	 * - setting mru greater than the mbuf size resulting in
438 	 * hw and sw buffer size mismatch
439 	 * - setting mtu that requires the support of scattered packets
440 	 * when this feature has not been enabled/supported so far
441 	 * (TODO check scattered_rx flag here once scattered RX is supported).
442 	 */
443 	if (mru + MRVL_PKT_OFFS > mbuf_data_size) {
444 		mru = mbuf_data_size - MRVL_PKT_OFFS;
445 		mtu = MRVL_PP2_MRU_TO_MTU(mru);
446 		MRVL_LOG(WARNING, "MTU too big, max MTU possible limitted "
447 			"by current mbuf size: %u. Set MTU to %u, MRU to %u",
448 			mbuf_data_size, mtu, mru);
449 	}
450 
451 	if (mtu < ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX) {
452 		MRVL_LOG(ERR, "Invalid MTU [%u] or MRU [%u]", mtu, mru);
453 		return -EINVAL;
454 	}
455 
456 	dev->data->mtu = mtu;
457 	dev->data->dev_conf.rxmode.max_rx_pkt_len = mru - MV_MH_SIZE;
458 
459 	if (!priv->ppio)
460 		return 0;
461 
462 	ret = pp2_ppio_set_mru(priv->ppio, mru);
463 	if (ret) {
464 		MRVL_LOG(ERR, "Failed to change MRU");
465 		return ret;
466 	}
467 
468 	ret = pp2_ppio_set_mtu(priv->ppio, mtu);
469 	if (ret) {
470 		MRVL_LOG(ERR, "Failed to change MTU");
471 		return ret;
472 	}
473 
474 	return 0;
475 }
476 
477 /**
478  * DPDK callback to bring the link up.
479  *
480  * @param dev
481  *   Pointer to Ethernet device structure.
482  *
483  * @return
484  *   0 on success, negative error value otherwise.
485  */
486 static int
487 mrvl_dev_set_link_up(struct rte_eth_dev *dev)
488 {
489 	struct mrvl_priv *priv = dev->data->dev_private;
490 	int ret;
491 
492 	if (!priv->ppio)
493 		return -EPERM;
494 
495 	ret = pp2_ppio_enable(priv->ppio);
496 	if (ret)
497 		return ret;
498 
499 	/*
500 	 * mtu/mru can be updated if pp2_ppio_enable() was called at least once
501 	 * as pp2_ppio_enable() changes port->t_mode from default 0 to
502 	 * PP2_TRAFFIC_INGRESS_EGRESS.
503 	 *
504 	 * Set mtu to default DPDK value here.
505 	 */
506 	ret = mrvl_mtu_set(dev, dev->data->mtu);
507 	if (ret)
508 		pp2_ppio_disable(priv->ppio);
509 
510 	return ret;
511 }
512 
513 /**
514  * DPDK callback to bring the link down.
515  *
516  * @param dev
517  *   Pointer to Ethernet device structure.
518  *
519  * @return
520  *   0 on success, negative error value otherwise.
521  */
522 static int
523 mrvl_dev_set_link_down(struct rte_eth_dev *dev)
524 {
525 	struct mrvl_priv *priv = dev->data->dev_private;
526 
527 	if (!priv->ppio)
528 		return -EPERM;
529 
530 	return pp2_ppio_disable(priv->ppio);
531 }
532 
533 /**
534  * DPDK callback to start tx queue.
535  *
536  * @param dev
537  *   Pointer to Ethernet device structure.
538  * @param queue_id
539  *   Transmit queue index.
540  *
541  * @return
542  *   0 on success, negative error value otherwise.
543  */
544 static int
545 mrvl_tx_queue_start(struct rte_eth_dev *dev, uint16_t queue_id)
546 {
547 	struct mrvl_priv *priv = dev->data->dev_private;
548 	int ret;
549 
550 	if (!priv)
551 		return -EPERM;
552 
553 	/* passing 1 enables given tx queue */
554 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 1);
555 	if (ret) {
556 		MRVL_LOG(ERR, "Failed to start txq %d", queue_id);
557 		return ret;
558 	}
559 
560 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
561 
562 	return 0;
563 }
564 
565 /**
566  * DPDK callback to stop tx queue.
567  *
568  * @param dev
569  *   Pointer to Ethernet device structure.
570  * @param queue_id
571  *   Transmit queue index.
572  *
573  * @return
574  *   0 on success, negative error value otherwise.
575  */
576 static int
577 mrvl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t queue_id)
578 {
579 	struct mrvl_priv *priv = dev->data->dev_private;
580 	int ret;
581 
582 	if (!priv->ppio)
583 		return -EPERM;
584 
585 	/* passing 0 disables given tx queue */
586 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 0);
587 	if (ret) {
588 		MRVL_LOG(ERR, "Failed to stop txq %d", queue_id);
589 		return ret;
590 	}
591 
592 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
593 
594 	return 0;
595 }
596 
597 /**
598  * DPDK callback to start the device.
599  *
600  * @param dev
601  *   Pointer to Ethernet device structure.
602  *
603  * @return
604  *   0 on success, negative errno value on failure.
605  */
606 static int
607 mrvl_dev_start(struct rte_eth_dev *dev)
608 {
609 	struct mrvl_priv *priv = dev->data->dev_private;
610 	char match[MRVL_MATCH_LEN];
611 	int ret = 0, i, def_init_size;
612 
613 	if (priv->ppio)
614 		return mrvl_dev_set_link_up(dev);
615 
616 	snprintf(match, sizeof(match), "ppio-%d:%d",
617 		 priv->pp_id, priv->ppio_id);
618 	priv->ppio_params.match = match;
619 
620 	/*
621 	 * Calculate the minimum bpool size for refill feature as follows:
622 	 * 2 default burst sizes multiply by number of rx queues.
623 	 * If the bpool size will be below this value, new buffers will
624 	 * be added to the pool.
625 	 */
626 	priv->bpool_min_size = priv->nb_rx_queues * MRVL_BURST_SIZE * 2;
627 
628 	/* In case initial bpool size configured in queues setup is
629 	 * smaller than minimum size add more buffers
630 	 */
631 	def_init_size = priv->bpool_min_size + MRVL_BURST_SIZE * 2;
632 	if (priv->bpool_init_size < def_init_size) {
633 		int buffs_to_add = def_init_size - priv->bpool_init_size;
634 
635 		priv->bpool_init_size += buffs_to_add;
636 		ret = mrvl_fill_bpool(dev->data->rx_queues[0], buffs_to_add);
637 		if (ret)
638 			MRVL_LOG(ERR, "Failed to add buffers to bpool");
639 	}
640 
641 	/*
642 	 * Calculate the maximum bpool size for refill feature as follows:
643 	 * maximum number of descriptors in rx queue multiply by number
644 	 * of rx queues plus minimum bpool size.
645 	 * In case the bpool size will exceed this value, superfluous buffers
646 	 * will be removed
647 	 */
648 	priv->bpool_max_size = (priv->nb_rx_queues * MRVL_PP2_RXD_MAX) +
649 				priv->bpool_min_size;
650 
651 	ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
652 	if (ret) {
653 		MRVL_LOG(ERR, "Failed to init ppio");
654 		return ret;
655 	}
656 
657 	/*
658 	 * In case there are some some stale uc/mc mac addresses flush them
659 	 * here. It cannot be done during mrvl_dev_close() as port information
660 	 * is already gone at that point (due to pp2_ppio_deinit() in
661 	 * mrvl_dev_stop()).
662 	 */
663 	if (!priv->uc_mc_flushed) {
664 		ret = pp2_ppio_flush_mac_addrs(priv->ppio, 1, 1);
665 		if (ret) {
666 			MRVL_LOG(ERR,
667 				"Failed to flush uc/mc filter list");
668 			goto out;
669 		}
670 		priv->uc_mc_flushed = 1;
671 	}
672 
673 	if (!priv->vlan_flushed) {
674 		ret = pp2_ppio_flush_vlan(priv->ppio);
675 		if (ret) {
676 			MRVL_LOG(ERR, "Failed to flush vlan list");
677 			/*
678 			 * TODO
679 			 * once pp2_ppio_flush_vlan() is supported jump to out
680 			 * goto out;
681 			 */
682 		}
683 		priv->vlan_flushed = 1;
684 	}
685 	ret = mrvl_mtu_set(dev, dev->data->mtu);
686 	if (ret)
687 		MRVL_LOG(ERR, "Failed to set MTU to %d", dev->data->mtu);
688 
689 	/* For default QoS config, don't start classifier. */
690 	if (mrvl_qos_cfg  &&
691 	    mrvl_qos_cfg->port[dev->data->port_id].use_global_defaults == 0) {
692 		ret = mrvl_start_qos_mapping(priv);
693 		if (ret) {
694 			MRVL_LOG(ERR, "Failed to setup QoS mapping");
695 			goto out;
696 		}
697 	}
698 
699 	ret = mrvl_dev_set_link_up(dev);
700 	if (ret) {
701 		MRVL_LOG(ERR, "Failed to set link up");
702 		goto out;
703 	}
704 
705 	/* start tx queues */
706 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
707 		struct mrvl_txq *txq = dev->data->tx_queues[i];
708 
709 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
710 
711 		if (!txq->tx_deferred_start)
712 			continue;
713 
714 		/*
715 		 * All txqs are started by default. Stop them
716 		 * so that tx_deferred_start works as expected.
717 		 */
718 		ret = mrvl_tx_queue_stop(dev, i);
719 		if (ret)
720 			goto out;
721 	}
722 
723 	mrvl_flow_init(dev);
724 	mrvl_mtr_init(dev);
725 	mrvl_set_tx_function(dev);
726 
727 	return 0;
728 out:
729 	MRVL_LOG(ERR, "Failed to start device");
730 	pp2_ppio_deinit(priv->ppio);
731 	return ret;
732 }
733 
734 /**
735  * Flush receive queues.
736  *
737  * @param dev
738  *   Pointer to Ethernet device structure.
739  */
740 static void
741 mrvl_flush_rx_queues(struct rte_eth_dev *dev)
742 {
743 	int i;
744 
745 	MRVL_LOG(INFO, "Flushing rx queues");
746 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
747 		int ret, num;
748 
749 		do {
750 			struct mrvl_rxq *q = dev->data->rx_queues[i];
751 			struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];
752 
753 			num = MRVL_PP2_RXD_MAX;
754 			ret = pp2_ppio_recv(q->priv->ppio,
755 					    q->priv->rxq_map[q->queue_id].tc,
756 					    q->priv->rxq_map[q->queue_id].inq,
757 					    descs, (uint16_t *)&num);
758 		} while (ret == 0 && num);
759 	}
760 }
761 
762 /**
763  * Flush transmit shadow queues.
764  *
765  * @param dev
766  *   Pointer to Ethernet device structure.
767  */
768 static void
769 mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
770 {
771 	int i, j;
772 	struct mrvl_txq *txq;
773 
774 	MRVL_LOG(INFO, "Flushing tx shadow queues");
775 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
776 		txq = (struct mrvl_txq *)dev->data->tx_queues[i];
777 
778 		for (j = 0; j < RTE_MAX_LCORE; j++) {
779 			struct mrvl_shadow_txq *sq;
780 
781 			if (!hifs[j])
782 				continue;
783 
784 			sq = &txq->shadow_txqs[j];
785 			mrvl_free_sent_buffers(txq->priv->ppio,
786 				hifs[j], j, sq, txq->queue_id, 1);
787 			while (sq->tail != sq->head) {
788 				uint64_t addr = cookie_addr_high |
789 					sq->ent[sq->tail].buff.cookie;
790 				rte_pktmbuf_free(
791 					(struct rte_mbuf *)addr);
792 				sq->tail = (sq->tail + 1) &
793 					    MRVL_PP2_TX_SHADOWQ_MASK;
794 			}
795 			memset(sq, 0, sizeof(*sq));
796 		}
797 	}
798 }
799 
800 /**
801  * Flush hardware bpool (buffer-pool).
802  *
803  * @param dev
804  *   Pointer to Ethernet device structure.
805  */
806 static void
807 mrvl_flush_bpool(struct rte_eth_dev *dev)
808 {
809 	struct mrvl_priv *priv = dev->data->dev_private;
810 	struct pp2_hif *hif;
811 	uint32_t num;
812 	int ret;
813 	unsigned int core_id = rte_lcore_id();
814 
815 	if (core_id == LCORE_ID_ANY)
816 		core_id = 0;
817 
818 	hif = mrvl_get_hif(priv, core_id);
819 
820 	ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
821 	if (ret) {
822 		MRVL_LOG(ERR, "Failed to get bpool buffers number");
823 		return;
824 	}
825 
826 	while (num--) {
827 		struct pp2_buff_inf inf;
828 		uint64_t addr;
829 
830 		ret = pp2_bpool_get_buff(hif, priv->bpool, &inf);
831 		if (ret)
832 			break;
833 
834 		addr = cookie_addr_high | inf.cookie;
835 		rte_pktmbuf_free((struct rte_mbuf *)addr);
836 	}
837 }
838 
839 /**
840  * DPDK callback to stop the device.
841  *
842  * @param dev
843  *   Pointer to Ethernet device structure.
844  */
845 static void
846 mrvl_dev_stop(struct rte_eth_dev *dev)
847 {
848 	mrvl_dev_set_link_down(dev);
849 }
850 
851 /**
852  * DPDK callback to close the device.
853  *
854  * @param dev
855  *   Pointer to Ethernet device structure.
856  */
857 static void
858 mrvl_dev_close(struct rte_eth_dev *dev)
859 {
860 	struct mrvl_priv *priv = dev->data->dev_private;
861 	size_t i;
862 
863 	mrvl_flush_rx_queues(dev);
864 	mrvl_flush_tx_shadow_queues(dev);
865 	mrvl_flow_deinit(dev);
866 	mrvl_mtr_deinit(dev);
867 
868 	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
869 		struct pp2_ppio_tc_params *tc_params =
870 			&priv->ppio_params.inqs_params.tcs_params[i];
871 
872 		if (tc_params->inqs_params) {
873 			rte_free(tc_params->inqs_params);
874 			tc_params->inqs_params = NULL;
875 		}
876 	}
877 
878 	if (priv->cls_tbl) {
879 		pp2_cls_tbl_deinit(priv->cls_tbl);
880 		priv->cls_tbl = NULL;
881 	}
882 
883 	if (priv->qos_tbl) {
884 		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
885 		priv->qos_tbl = NULL;
886 	}
887 
888 	mrvl_flush_bpool(dev);
889 	mrvl_tm_deinit(dev);
890 
891 	if (priv->ppio) {
892 		pp2_ppio_deinit(priv->ppio);
893 		priv->ppio = NULL;
894 	}
895 
896 	/* policer must be released after ppio deinitialization */
897 	if (priv->default_policer) {
898 		pp2_cls_plcr_deinit(priv->default_policer);
899 		priv->default_policer = NULL;
900 	}
901 }
902 
903 /**
904  * DPDK callback to retrieve physical link information.
905  *
906  * @param dev
907  *   Pointer to Ethernet device structure.
908  * @param wait_to_complete
909  *   Wait for request completion (ignored).
910  *
911  * @return
912  *   0 on success, negative error value otherwise.
913  */
914 static int
915 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
916 {
917 	/*
918 	 * TODO
919 	 * once MUSDK provides necessary API use it here
920 	 */
921 	struct mrvl_priv *priv = dev->data->dev_private;
922 	struct ethtool_cmd edata;
923 	struct ifreq req;
924 	int ret, fd, link_up;
925 
926 	if (!priv->ppio)
927 		return -EPERM;
928 
929 	edata.cmd = ETHTOOL_GSET;
930 
931 	strcpy(req.ifr_name, dev->data->name);
932 	req.ifr_data = (void *)&edata;
933 
934 	fd = socket(AF_INET, SOCK_DGRAM, 0);
935 	if (fd == -1)
936 		return -EFAULT;
937 
938 	ret = ioctl(fd, SIOCETHTOOL, &req);
939 	if (ret == -1) {
940 		close(fd);
941 		return -EFAULT;
942 	}
943 
944 	close(fd);
945 
946 	switch (ethtool_cmd_speed(&edata)) {
947 	case SPEED_10:
948 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
949 		break;
950 	case SPEED_100:
951 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
952 		break;
953 	case SPEED_1000:
954 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
955 		break;
956 	case SPEED_10000:
957 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
958 		break;
959 	default:
960 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
961 	}
962 
963 	dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
964 							 ETH_LINK_HALF_DUPLEX;
965 	dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
966 							   ETH_LINK_FIXED;
967 	pp2_ppio_get_link_state(priv->ppio, &link_up);
968 	dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
969 
970 	return 0;
971 }
972 
973 /**
974  * DPDK callback to enable promiscuous mode.
975  *
976  * @param dev
977  *   Pointer to Ethernet device structure.
978  */
979 static void
980 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
981 {
982 	struct mrvl_priv *priv = dev->data->dev_private;
983 	int ret;
984 
985 	if (!priv->ppio)
986 		return;
987 
988 	if (priv->isolated)
989 		return;
990 
991 	ret = pp2_ppio_set_promisc(priv->ppio, 1);
992 	if (ret)
993 		MRVL_LOG(ERR, "Failed to enable promiscuous mode");
994 }
995 
996 /**
997  * DPDK callback to enable allmulti mode.
998  *
999  * @param dev
1000  *   Pointer to Ethernet device structure.
1001  */
1002 static void
1003 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
1004 {
1005 	struct mrvl_priv *priv = dev->data->dev_private;
1006 	int ret;
1007 
1008 	if (!priv->ppio)
1009 		return;
1010 
1011 	if (priv->isolated)
1012 		return;
1013 
1014 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
1015 	if (ret)
1016 		MRVL_LOG(ERR, "Failed enable all-multicast mode");
1017 }
1018 
1019 /**
1020  * DPDK callback to disable promiscuous mode.
1021  *
1022  * @param dev
1023  *   Pointer to Ethernet device structure.
1024  */
1025 static void
1026 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
1027 {
1028 	struct mrvl_priv *priv = dev->data->dev_private;
1029 	int ret;
1030 
1031 	if (!priv->ppio)
1032 		return;
1033 
1034 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
1035 	if (ret)
1036 		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
1037 }
1038 
1039 /**
1040  * DPDK callback to disable allmulticast mode.
1041  *
1042  * @param dev
1043  *   Pointer to Ethernet device structure.
1044  */
1045 static void
1046 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
1047 {
1048 	struct mrvl_priv *priv = dev->data->dev_private;
1049 	int ret;
1050 
1051 	if (!priv->ppio)
1052 		return;
1053 
1054 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
1055 	if (ret)
1056 		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
1057 }
1058 
1059 /**
1060  * DPDK callback to remove a MAC address.
1061  *
1062  * @param dev
1063  *   Pointer to Ethernet device structure.
1064  * @param index
1065  *   MAC address index.
1066  */
1067 static void
1068 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1069 {
1070 	struct mrvl_priv *priv = dev->data->dev_private;
1071 	char buf[ETHER_ADDR_FMT_SIZE];
1072 	int ret;
1073 
1074 	if (!priv->ppio)
1075 		return;
1076 
1077 	if (priv->isolated)
1078 		return;
1079 
1080 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
1081 				       dev->data->mac_addrs[index].addr_bytes);
1082 	if (ret) {
1083 		ether_format_addr(buf, sizeof(buf),
1084 				  &dev->data->mac_addrs[index]);
1085 		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
1086 	}
1087 }
1088 
1089 /**
1090  * DPDK callback to add a MAC address.
1091  *
1092  * @param dev
1093  *   Pointer to Ethernet device structure.
1094  * @param mac_addr
1095  *   MAC address to register.
1096  * @param index
1097  *   MAC address index.
1098  * @param vmdq
1099  *   VMDq pool index to associate address with (unused).
1100  *
1101  * @return
1102  *   0 on success, negative error value otherwise.
1103  */
1104 static int
1105 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1106 		  uint32_t index, uint32_t vmdq __rte_unused)
1107 {
1108 	struct mrvl_priv *priv = dev->data->dev_private;
1109 	char buf[ETHER_ADDR_FMT_SIZE];
1110 	int ret;
1111 
1112 	if (priv->isolated)
1113 		return -ENOTSUP;
1114 
1115 	if (index == 0)
1116 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1117 		return -1;
1118 
1119 	if (!priv->ppio)
1120 		return 0;
1121 
1122 	/*
1123 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1124 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1125 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1126 	 * 21 respectively.
1127 	 *
1128 	 * If more than uc_filter_max uc addresses were added to filter list
1129 	 * then NIC will switch to promiscuous mode automatically.
1130 	 *
1131 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1132 	 * were added to filter list then NIC will switch to all-multicast mode
1133 	 * automatically.
1134 	 */
1135 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1136 	if (ret) {
1137 		ether_format_addr(buf, sizeof(buf), mac_addr);
1138 		MRVL_LOG(ERR, "Failed to add mac %s", buf);
1139 		return -1;
1140 	}
1141 
1142 	return 0;
1143 }
1144 
1145 /**
1146  * DPDK callback to set the primary MAC address.
1147  *
1148  * @param dev
1149  *   Pointer to Ethernet device structure.
1150  * @param mac_addr
1151  *   MAC address to register.
1152  *
1153  * @return
1154  *   0 on success, negative error value otherwise.
1155  */
1156 static int
1157 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1158 {
1159 	struct mrvl_priv *priv = dev->data->dev_private;
1160 	int ret;
1161 
1162 	if (!priv->ppio)
1163 		return 0;
1164 
1165 	if (priv->isolated)
1166 		return -ENOTSUP;
1167 
1168 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1169 	if (ret) {
1170 		char buf[ETHER_ADDR_FMT_SIZE];
1171 		ether_format_addr(buf, sizeof(buf), mac_addr);
1172 		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1173 	}
1174 
1175 	return ret;
1176 }
1177 
1178 /**
1179  * DPDK callback to get device statistics.
1180  *
1181  * @param dev
1182  *   Pointer to Ethernet device structure.
1183  * @param stats
1184  *   Stats structure output buffer.
1185  *
1186  * @return
1187  *   0 on success, negative error value otherwise.
1188  */
1189 static int
1190 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1191 {
1192 	struct mrvl_priv *priv = dev->data->dev_private;
1193 	struct pp2_ppio_statistics ppio_stats;
1194 	uint64_t drop_mac = 0;
1195 	unsigned int i, idx, ret;
1196 
1197 	if (!priv->ppio)
1198 		return -EPERM;
1199 
1200 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1201 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1202 		struct pp2_ppio_inq_statistics rx_stats;
1203 
1204 		if (!rxq)
1205 			continue;
1206 
1207 		idx = rxq->queue_id;
1208 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1209 			MRVL_LOG(ERR,
1210 				"rx queue %d stats out of range (0 - %d)",
1211 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1212 			continue;
1213 		}
1214 
1215 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1216 						  priv->rxq_map[idx].tc,
1217 						  priv->rxq_map[idx].inq,
1218 						  &rx_stats, 0);
1219 		if (unlikely(ret)) {
1220 			MRVL_LOG(ERR,
1221 				"Failed to update rx queue %d stats", idx);
1222 			break;
1223 		}
1224 
1225 		stats->q_ibytes[idx] = rxq->bytes_recv;
1226 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1227 		stats->q_errors[idx] = rx_stats.drop_early +
1228 				       rx_stats.drop_fullq +
1229 				       rx_stats.drop_bm +
1230 				       rxq->drop_mac;
1231 		stats->ibytes += rxq->bytes_recv;
1232 		drop_mac += rxq->drop_mac;
1233 	}
1234 
1235 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1236 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1237 		struct pp2_ppio_outq_statistics tx_stats;
1238 
1239 		if (!txq)
1240 			continue;
1241 
1242 		idx = txq->queue_id;
1243 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1244 			MRVL_LOG(ERR,
1245 				"tx queue %d stats out of range (0 - %d)",
1246 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1247 		}
1248 
1249 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1250 						   &tx_stats, 0);
1251 		if (unlikely(ret)) {
1252 			MRVL_LOG(ERR,
1253 				"Failed to update tx queue %d stats", idx);
1254 			break;
1255 		}
1256 
1257 		stats->q_opackets[idx] = tx_stats.deq_desc;
1258 		stats->q_obytes[idx] = txq->bytes_sent;
1259 		stats->obytes += txq->bytes_sent;
1260 	}
1261 
1262 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1263 	if (unlikely(ret)) {
1264 		MRVL_LOG(ERR, "Failed to update port statistics");
1265 		return ret;
1266 	}
1267 
1268 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1269 	stats->opackets += ppio_stats.tx_packets;
1270 	stats->imissed += ppio_stats.rx_fullq_dropped +
1271 			  ppio_stats.rx_bm_dropped +
1272 			  ppio_stats.rx_early_dropped +
1273 			  ppio_stats.rx_fifo_dropped +
1274 			  ppio_stats.rx_cls_dropped;
1275 	stats->ierrors = drop_mac;
1276 
1277 	return 0;
1278 }
1279 
1280 /**
1281  * DPDK callback to clear device statistics.
1282  *
1283  * @param dev
1284  *   Pointer to Ethernet device structure.
1285  */
1286 static void
1287 mrvl_stats_reset(struct rte_eth_dev *dev)
1288 {
1289 	struct mrvl_priv *priv = dev->data->dev_private;
1290 	int i;
1291 
1292 	if (!priv->ppio)
1293 		return;
1294 
1295 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1296 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1297 
1298 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1299 					    priv->rxq_map[i].inq, NULL, 1);
1300 		rxq->bytes_recv = 0;
1301 		rxq->drop_mac = 0;
1302 	}
1303 
1304 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1305 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1306 
1307 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1308 		txq->bytes_sent = 0;
1309 	}
1310 
1311 	pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1312 }
1313 
1314 /**
1315  * DPDK callback to get extended statistics.
1316  *
1317  * @param dev
1318  *   Pointer to Ethernet device structure.
1319  * @param stats
1320  *   Pointer to xstats table.
1321  * @param n
1322  *   Number of entries in xstats table.
1323  * @return
1324  *   Negative value on error, number of read xstats otherwise.
1325  */
1326 static int
1327 mrvl_xstats_get(struct rte_eth_dev *dev,
1328 		struct rte_eth_xstat *stats, unsigned int n)
1329 {
1330 	struct mrvl_priv *priv = dev->data->dev_private;
1331 	struct pp2_ppio_statistics ppio_stats;
1332 	unsigned int i;
1333 
1334 	if (!stats)
1335 		return 0;
1336 
1337 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1338 	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1339 		uint64_t val;
1340 
1341 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1342 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1343 					    mrvl_xstats_tbl[i].offset);
1344 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1345 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1346 					    mrvl_xstats_tbl[i].offset);
1347 		else
1348 			return -EINVAL;
1349 
1350 		stats[i].id = i;
1351 		stats[i].value = val;
1352 	}
1353 
1354 	return n;
1355 }
1356 
1357 /**
1358  * DPDK callback to reset extended statistics.
1359  *
1360  * @param dev
1361  *   Pointer to Ethernet device structure.
1362  */
1363 static void
1364 mrvl_xstats_reset(struct rte_eth_dev *dev)
1365 {
1366 	mrvl_stats_reset(dev);
1367 }
1368 
1369 /**
1370  * DPDK callback to get extended statistics names.
1371  *
1372  * @param dev (unused)
1373  *   Pointer to Ethernet device structure.
1374  * @param xstats_names
1375  *   Pointer to xstats names table.
1376  * @param size
1377  *   Size of the xstats names table.
1378  * @return
1379  *   Number of read names.
1380  */
1381 static int
1382 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1383 		      struct rte_eth_xstat_name *xstats_names,
1384 		      unsigned int size)
1385 {
1386 	unsigned int i;
1387 
1388 	if (!xstats_names)
1389 		return RTE_DIM(mrvl_xstats_tbl);
1390 
1391 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1392 		strlcpy(xstats_names[i].name, mrvl_xstats_tbl[i].name,
1393 			RTE_ETH_XSTATS_NAME_SIZE);
1394 
1395 	return size;
1396 }
1397 
1398 /**
1399  * DPDK callback to get information about the device.
1400  *
1401  * @param dev
1402  *   Pointer to Ethernet device structure (unused).
1403  * @param info
1404  *   Info structure output buffer.
1405  */
1406 static void
1407 mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
1408 		   struct rte_eth_dev_info *info)
1409 {
1410 	info->speed_capa = ETH_LINK_SPEED_10M |
1411 			   ETH_LINK_SPEED_100M |
1412 			   ETH_LINK_SPEED_1G |
1413 			   ETH_LINK_SPEED_10G;
1414 
1415 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1416 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1417 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1418 
1419 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1420 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1421 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1422 
1423 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1424 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1425 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1426 
1427 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1428 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1429 
1430 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1431 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1432 
1433 	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1434 				       ETH_RSS_NONFRAG_IPV4_TCP |
1435 				       ETH_RSS_NONFRAG_IPV4_UDP;
1436 
1437 	/* By default packets are dropped if no descriptors are available */
1438 	info->default_rxconf.rx_drop_en = 1;
1439 
1440 	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1441 }
1442 
1443 /**
1444  * Return supported packet types.
1445  *
1446  * @param dev
1447  *   Pointer to Ethernet device structure (unused).
1448  *
1449  * @return
1450  *   Const pointer to the table with supported packet types.
1451  */
1452 static const uint32_t *
1453 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1454 {
1455 	static const uint32_t ptypes[] = {
1456 		RTE_PTYPE_L2_ETHER,
1457 		RTE_PTYPE_L2_ETHER_VLAN,
1458 		RTE_PTYPE_L2_ETHER_QINQ,
1459 		RTE_PTYPE_L3_IPV4,
1460 		RTE_PTYPE_L3_IPV4_EXT,
1461 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1462 		RTE_PTYPE_L3_IPV6,
1463 		RTE_PTYPE_L3_IPV6_EXT,
1464 		RTE_PTYPE_L2_ETHER_ARP,
1465 		RTE_PTYPE_L4_TCP,
1466 		RTE_PTYPE_L4_UDP
1467 	};
1468 
1469 	return ptypes;
1470 }
1471 
1472 /**
1473  * DPDK callback to get information about specific receive queue.
1474  *
1475  * @param dev
1476  *   Pointer to Ethernet device structure.
1477  * @param rx_queue_id
1478  *   Receive queue index.
1479  * @param qinfo
1480  *   Receive queue information structure.
1481  */
1482 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1483 			      struct rte_eth_rxq_info *qinfo)
1484 {
1485 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1486 	struct mrvl_priv *priv = dev->data->dev_private;
1487 	int inq = priv->rxq_map[rx_queue_id].inq;
1488 	int tc = priv->rxq_map[rx_queue_id].tc;
1489 	struct pp2_ppio_tc_params *tc_params =
1490 		&priv->ppio_params.inqs_params.tcs_params[tc];
1491 
1492 	qinfo->mp = q->mp;
1493 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1494 }
1495 
1496 /**
1497  * DPDK callback to get information about specific transmit queue.
1498  *
1499  * @param dev
1500  *   Pointer to Ethernet device structure.
1501  * @param tx_queue_id
1502  *   Transmit queue index.
1503  * @param qinfo
1504  *   Transmit queue information structure.
1505  */
1506 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1507 			      struct rte_eth_txq_info *qinfo)
1508 {
1509 	struct mrvl_priv *priv = dev->data->dev_private;
1510 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1511 
1512 	qinfo->nb_desc =
1513 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1514 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1515 }
1516 
1517 /**
1518  * DPDK callback to Configure a VLAN filter.
1519  *
1520  * @param dev
1521  *   Pointer to Ethernet device structure.
1522  * @param vlan_id
1523  *   VLAN ID to filter.
1524  * @param on
1525  *   Toggle filter.
1526  *
1527  * @return
1528  *   0 on success, negative error value otherwise.
1529  */
1530 static int
1531 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1532 {
1533 	struct mrvl_priv *priv = dev->data->dev_private;
1534 
1535 	if (!priv->ppio)
1536 		return -EPERM;
1537 
1538 	if (priv->isolated)
1539 		return -ENOTSUP;
1540 
1541 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1542 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1543 }
1544 
1545 /**
1546  * Release buffers to hardware bpool (buffer-pool)
1547  *
1548  * @param rxq
1549  *   Receive queue pointer.
1550  * @param num
1551  *   Number of buffers to release to bpool.
1552  *
1553  * @return
1554  *   0 on success, negative error value otherwise.
1555  */
1556 static int
1557 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1558 {
1559 	struct buff_release_entry entries[MRVL_PP2_RXD_MAX];
1560 	struct rte_mbuf *mbufs[MRVL_PP2_RXD_MAX];
1561 	int i, ret;
1562 	unsigned int core_id;
1563 	struct pp2_hif *hif;
1564 	struct pp2_bpool *bpool;
1565 
1566 	core_id = rte_lcore_id();
1567 	if (core_id == LCORE_ID_ANY)
1568 		core_id = 0;
1569 
1570 	hif = mrvl_get_hif(rxq->priv, core_id);
1571 	if (!hif)
1572 		return -1;
1573 
1574 	bpool = rxq->priv->bpool;
1575 
1576 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1577 	if (ret)
1578 		return ret;
1579 
1580 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1581 		cookie_addr_high =
1582 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1583 
1584 	for (i = 0; i < num; i++) {
1585 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1586 			!= cookie_addr_high) {
1587 			MRVL_LOG(ERR,
1588 				"mbuf virtual addr high 0x%lx out of range",
1589 				(uint64_t)mbufs[i] >> 32);
1590 			goto out;
1591 		}
1592 
1593 		entries[i].buff.addr =
1594 			rte_mbuf_data_iova_default(mbufs[i]);
1595 		entries[i].buff.cookie = (uint64_t)mbufs[i];
1596 		entries[i].bpool = bpool;
1597 	}
1598 
1599 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1600 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1601 
1602 	if (i != num)
1603 		goto out;
1604 
1605 	return 0;
1606 out:
1607 	for (; i < num; i++)
1608 		rte_pktmbuf_free(mbufs[i]);
1609 
1610 	return -1;
1611 }
1612 
1613 /**
1614  * DPDK callback to configure the receive queue.
1615  *
1616  * @param dev
1617  *   Pointer to Ethernet device structure.
1618  * @param idx
1619  *   RX queue index.
1620  * @param desc
1621  *   Number of descriptors to configure in queue.
1622  * @param socket
1623  *   NUMA socket on which memory must be allocated.
1624  * @param conf
1625  *   Thresholds parameters.
1626  * @param mp
1627  *   Memory pool for buffer allocations.
1628  *
1629  * @return
1630  *   0 on success, negative error value otherwise.
1631  */
1632 static int
1633 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1634 		    unsigned int socket,
1635 		    const struct rte_eth_rxconf *conf,
1636 		    struct rte_mempool *mp)
1637 {
1638 	struct mrvl_priv *priv = dev->data->dev_private;
1639 	struct mrvl_rxq *rxq;
1640 	uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
1641 	uint32_t max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1642 	int ret, tc, inq;
1643 	uint64_t offloads;
1644 
1645 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
1646 
1647 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1648 		/*
1649 		 * Unknown TC mapping, mapping will not have a correct queue.
1650 		 */
1651 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
1652 			idx, priv->ppio_id);
1653 		return -EFAULT;
1654 	}
1655 
1656 	frame_size = buf_size - RTE_PKTMBUF_HEADROOM - MRVL_PKT_EFFEC_OFFS;
1657 	if (frame_size < max_rx_pkt_len) {
1658 		MRVL_LOG(WARNING,
1659 			"Mbuf size must be increased to %u bytes to hold up "
1660 			"to %u bytes of data.",
1661 			buf_size + max_rx_pkt_len - frame_size,
1662 			max_rx_pkt_len);
1663 		dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
1664 		MRVL_LOG(INFO, "Setting max rx pkt len to %u",
1665 			dev->data->dev_conf.rxmode.max_rx_pkt_len);
1666 	}
1667 
1668 	if (dev->data->rx_queues[idx]) {
1669 		rte_free(dev->data->rx_queues[idx]);
1670 		dev->data->rx_queues[idx] = NULL;
1671 	}
1672 
1673 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
1674 	if (!rxq)
1675 		return -ENOMEM;
1676 
1677 	rxq->priv = priv;
1678 	rxq->mp = mp;
1679 	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
1680 	rxq->queue_id = idx;
1681 	rxq->port_id = dev->data->port_id;
1682 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
1683 
1684 	tc = priv->rxq_map[rxq->queue_id].tc,
1685 	inq = priv->rxq_map[rxq->queue_id].inq;
1686 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
1687 		desc;
1688 
1689 	ret = mrvl_fill_bpool(rxq, desc);
1690 	if (ret) {
1691 		rte_free(rxq);
1692 		return ret;
1693 	}
1694 
1695 	priv->bpool_init_size += desc;
1696 
1697 	dev->data->rx_queues[idx] = rxq;
1698 
1699 	return 0;
1700 }
1701 
1702 /**
1703  * DPDK callback to release the receive queue.
1704  *
1705  * @param rxq
1706  *   Generic receive queue pointer.
1707  */
1708 static void
1709 mrvl_rx_queue_release(void *rxq)
1710 {
1711 	struct mrvl_rxq *q = rxq;
1712 	struct pp2_ppio_tc_params *tc_params;
1713 	int i, num, tc, inq;
1714 	struct pp2_hif *hif;
1715 	unsigned int core_id = rte_lcore_id();
1716 
1717 	if (core_id == LCORE_ID_ANY)
1718 		core_id = 0;
1719 
1720 	if (!q)
1721 		return;
1722 
1723 	hif = mrvl_get_hif(q->priv, core_id);
1724 
1725 	if (!hif)
1726 		return;
1727 
1728 	tc = q->priv->rxq_map[q->queue_id].tc;
1729 	inq = q->priv->rxq_map[q->queue_id].inq;
1730 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
1731 	num = tc_params->inqs_params[inq].size;
1732 	for (i = 0; i < num; i++) {
1733 		struct pp2_buff_inf inf;
1734 		uint64_t addr;
1735 
1736 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
1737 		addr = cookie_addr_high | inf.cookie;
1738 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1739 	}
1740 
1741 	rte_free(q);
1742 }
1743 
1744 /**
1745  * DPDK callback to configure the transmit queue.
1746  *
1747  * @param dev
1748  *   Pointer to Ethernet device structure.
1749  * @param idx
1750  *   Transmit queue index.
1751  * @param desc
1752  *   Number of descriptors to configure in the queue.
1753  * @param socket
1754  *   NUMA socket on which memory must be allocated.
1755  * @param conf
1756  *   Tx queue configuration parameters.
1757  *
1758  * @return
1759  *   0 on success, negative error value otherwise.
1760  */
1761 static int
1762 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1763 		    unsigned int socket,
1764 		    const struct rte_eth_txconf *conf)
1765 {
1766 	struct mrvl_priv *priv = dev->data->dev_private;
1767 	struct mrvl_txq *txq;
1768 
1769 	if (dev->data->tx_queues[idx]) {
1770 		rte_free(dev->data->tx_queues[idx]);
1771 		dev->data->tx_queues[idx] = NULL;
1772 	}
1773 
1774 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
1775 	if (!txq)
1776 		return -ENOMEM;
1777 
1778 	txq->priv = priv;
1779 	txq->queue_id = idx;
1780 	txq->port_id = dev->data->port_id;
1781 	txq->tx_deferred_start = conf->tx_deferred_start;
1782 	dev->data->tx_queues[idx] = txq;
1783 
1784 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
1785 
1786 	return 0;
1787 }
1788 
1789 /**
1790  * DPDK callback to release the transmit queue.
1791  *
1792  * @param txq
1793  *   Generic transmit queue pointer.
1794  */
1795 static void
1796 mrvl_tx_queue_release(void *txq)
1797 {
1798 	struct mrvl_txq *q = txq;
1799 
1800 	if (!q)
1801 		return;
1802 
1803 	rte_free(q);
1804 }
1805 
1806 /**
1807  * DPDK callback to get flow control configuration.
1808  *
1809  * @param dev
1810  *  Pointer to Ethernet device structure.
1811  * @param fc_conf
1812  *  Pointer to the flow control configuration.
1813  *
1814  * @return
1815  *  0 on success, negative error value otherwise.
1816  */
1817 static int
1818 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1819 {
1820 	struct mrvl_priv *priv = dev->data->dev_private;
1821 	int ret, en;
1822 
1823 	if (!priv)
1824 		return -EPERM;
1825 
1826 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
1827 	if (ret) {
1828 		MRVL_LOG(ERR, "Failed to read rx pause state");
1829 		return ret;
1830 	}
1831 
1832 	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
1833 
1834 	return 0;
1835 }
1836 
1837 /**
1838  * DPDK callback to set flow control configuration.
1839  *
1840  * @param dev
1841  *  Pointer to Ethernet device structure.
1842  * @param fc_conf
1843  *  Pointer to the flow control configuration.
1844  *
1845  * @return
1846  *  0 on success, negative error value otherwise.
1847  */
1848 static int
1849 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1850 {
1851 	struct mrvl_priv *priv = dev->data->dev_private;
1852 
1853 	if (!priv)
1854 		return -EPERM;
1855 
1856 	if (fc_conf->high_water ||
1857 	    fc_conf->low_water ||
1858 	    fc_conf->pause_time ||
1859 	    fc_conf->mac_ctrl_frame_fwd ||
1860 	    fc_conf->autoneg) {
1861 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
1862 
1863 		return -EINVAL;
1864 	}
1865 
1866 	if (fc_conf->mode == RTE_FC_NONE ||
1867 	    fc_conf->mode == RTE_FC_RX_PAUSE) {
1868 		int ret, en;
1869 
1870 		en = fc_conf->mode == RTE_FC_NONE ? 0 : 1;
1871 		ret = pp2_ppio_set_rx_pause(priv->ppio, en);
1872 		if (ret)
1873 			MRVL_LOG(ERR,
1874 				"Failed to change flowctrl on RX side");
1875 
1876 		return ret;
1877 	}
1878 
1879 	return 0;
1880 }
1881 
1882 /**
1883  * Update RSS hash configuration
1884  *
1885  * @param dev
1886  *   Pointer to Ethernet device structure.
1887  * @param rss_conf
1888  *   Pointer to RSS configuration.
1889  *
1890  * @return
1891  *   0 on success, negative error value otherwise.
1892  */
1893 static int
1894 mrvl_rss_hash_update(struct rte_eth_dev *dev,
1895 		     struct rte_eth_rss_conf *rss_conf)
1896 {
1897 	struct mrvl_priv *priv = dev->data->dev_private;
1898 
1899 	if (priv->isolated)
1900 		return -ENOTSUP;
1901 
1902 	return mrvl_configure_rss(priv, rss_conf);
1903 }
1904 
1905 /**
1906  * DPDK callback to get RSS hash configuration.
1907  *
1908  * @param dev
1909  *   Pointer to Ethernet device structure.
1910  * @rss_conf
1911  *   Pointer to RSS configuration.
1912  *
1913  * @return
1914  *   Always 0.
1915  */
1916 static int
1917 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
1918 		       struct rte_eth_rss_conf *rss_conf)
1919 {
1920 	struct mrvl_priv *priv = dev->data->dev_private;
1921 	enum pp2_ppio_hash_type hash_type =
1922 		priv->ppio_params.inqs_params.hash_type;
1923 
1924 	rss_conf->rss_key = NULL;
1925 
1926 	if (hash_type == PP2_PPIO_HASH_T_NONE)
1927 		rss_conf->rss_hf = 0;
1928 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
1929 		rss_conf->rss_hf = ETH_RSS_IPV4;
1930 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
1931 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
1932 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
1933 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
1934 
1935 	return 0;
1936 }
1937 
1938 /**
1939  * DPDK callback to get rte_flow callbacks.
1940  *
1941  * @param dev
1942  *   Pointer to the device structure.
1943  * @param filer_type
1944  *   Flow filter type.
1945  * @param filter_op
1946  *   Flow filter operation.
1947  * @param arg
1948  *   Pointer to pass the flow ops.
1949  *
1950  * @return
1951  *   0 on success, negative error value otherwise.
1952  */
1953 static int
1954 mrvl_eth_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
1955 		     enum rte_filter_type filter_type,
1956 		     enum rte_filter_op filter_op, void *arg)
1957 {
1958 	switch (filter_type) {
1959 	case RTE_ETH_FILTER_GENERIC:
1960 		if (filter_op != RTE_ETH_FILTER_GET)
1961 			return -EINVAL;
1962 		*(const void **)arg = &mrvl_flow_ops;
1963 		return 0;
1964 	default:
1965 		MRVL_LOG(WARNING, "Filter type (%d) not supported",
1966 				filter_type);
1967 		return -EINVAL;
1968 	}
1969 }
1970 
1971 /**
1972  * DPDK callback to get rte_mtr callbacks.
1973  *
1974  * @param dev
1975  *   Pointer to the device structure.
1976  * @param ops
1977  *   Pointer to pass the mtr ops.
1978  *
1979  * @return
1980  *   Always 0.
1981  */
1982 static int
1983 mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
1984 {
1985 	*(const void **)ops = &mrvl_mtr_ops;
1986 
1987 	return 0;
1988 }
1989 
1990 /**
1991  * DPDK callback to get rte_tm callbacks.
1992  *
1993  * @param dev
1994  *   Pointer to the device structure.
1995  * @param ops
1996  *   Pointer to pass the tm ops.
1997  *
1998  * @return
1999  *   Always 0.
2000  */
2001 static int
2002 mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2003 {
2004 	*(const void **)ops = &mrvl_tm_ops;
2005 
2006 	return 0;
2007 }
2008 
2009 static const struct eth_dev_ops mrvl_ops = {
2010 	.dev_configure = mrvl_dev_configure,
2011 	.dev_start = mrvl_dev_start,
2012 	.dev_stop = mrvl_dev_stop,
2013 	.dev_set_link_up = mrvl_dev_set_link_up,
2014 	.dev_set_link_down = mrvl_dev_set_link_down,
2015 	.dev_close = mrvl_dev_close,
2016 	.link_update = mrvl_link_update,
2017 	.promiscuous_enable = mrvl_promiscuous_enable,
2018 	.allmulticast_enable = mrvl_allmulticast_enable,
2019 	.promiscuous_disable = mrvl_promiscuous_disable,
2020 	.allmulticast_disable = mrvl_allmulticast_disable,
2021 	.mac_addr_remove = mrvl_mac_addr_remove,
2022 	.mac_addr_add = mrvl_mac_addr_add,
2023 	.mac_addr_set = mrvl_mac_addr_set,
2024 	.mtu_set = mrvl_mtu_set,
2025 	.stats_get = mrvl_stats_get,
2026 	.stats_reset = mrvl_stats_reset,
2027 	.xstats_get = mrvl_xstats_get,
2028 	.xstats_reset = mrvl_xstats_reset,
2029 	.xstats_get_names = mrvl_xstats_get_names,
2030 	.dev_infos_get = mrvl_dev_infos_get,
2031 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
2032 	.rxq_info_get = mrvl_rxq_info_get,
2033 	.txq_info_get = mrvl_txq_info_get,
2034 	.vlan_filter_set = mrvl_vlan_filter_set,
2035 	.tx_queue_start = mrvl_tx_queue_start,
2036 	.tx_queue_stop = mrvl_tx_queue_stop,
2037 	.rx_queue_setup = mrvl_rx_queue_setup,
2038 	.rx_queue_release = mrvl_rx_queue_release,
2039 	.tx_queue_setup = mrvl_tx_queue_setup,
2040 	.tx_queue_release = mrvl_tx_queue_release,
2041 	.flow_ctrl_get = mrvl_flow_ctrl_get,
2042 	.flow_ctrl_set = mrvl_flow_ctrl_set,
2043 	.rss_hash_update = mrvl_rss_hash_update,
2044 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
2045 	.filter_ctrl = mrvl_eth_filter_ctrl,
2046 	.mtr_ops_get = mrvl_mtr_ops_get,
2047 	.tm_ops_get = mrvl_tm_ops_get,
2048 };
2049 
2050 /**
2051  * Return packet type information and l3/l4 offsets.
2052  *
2053  * @param desc
2054  *   Pointer to the received packet descriptor.
2055  * @param l3_offset
2056  *   l3 packet offset.
2057  * @param l4_offset
2058  *   l4 packet offset.
2059  *
2060  * @return
2061  *   Packet type information.
2062  */
2063 static inline uint64_t
2064 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
2065 				    uint8_t *l3_offset, uint8_t *l4_offset)
2066 {
2067 	enum pp2_inq_l3_type l3_type;
2068 	enum pp2_inq_l4_type l4_type;
2069 	enum pp2_inq_vlan_tag vlan_tag;
2070 	uint64_t packet_type;
2071 
2072 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
2073 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
2074 	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
2075 
2076 	packet_type = RTE_PTYPE_L2_ETHER;
2077 
2078 	switch (vlan_tag) {
2079 	case PP2_INQ_VLAN_TAG_SINGLE:
2080 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
2081 		break;
2082 	case PP2_INQ_VLAN_TAG_DOUBLE:
2083 	case PP2_INQ_VLAN_TAG_TRIPLE:
2084 		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
2085 		break;
2086 	default:
2087 		break;
2088 	}
2089 
2090 	switch (l3_type) {
2091 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
2092 		packet_type |= RTE_PTYPE_L3_IPV4;
2093 		break;
2094 	case PP2_INQ_L3_TYPE_IPV4_OK:
2095 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
2096 		break;
2097 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
2098 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
2099 		break;
2100 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
2101 		packet_type |= RTE_PTYPE_L3_IPV6;
2102 		break;
2103 	case PP2_INQ_L3_TYPE_IPV6_EXT:
2104 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
2105 		break;
2106 	case PP2_INQ_L3_TYPE_ARP:
2107 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
2108 		/*
2109 		 * In case of ARP l4_offset is set to wrong value.
2110 		 * Set it to proper one so that later on mbuf->l3_len can be
2111 		 * calculated subtracting l4_offset and l3_offset.
2112 		 */
2113 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
2114 		break;
2115 	default:
2116 		MRVL_LOG(DEBUG, "Failed to recognise l3 packet type");
2117 		break;
2118 	}
2119 
2120 	switch (l4_type) {
2121 	case PP2_INQ_L4_TYPE_TCP:
2122 		packet_type |= RTE_PTYPE_L4_TCP;
2123 		break;
2124 	case PP2_INQ_L4_TYPE_UDP:
2125 		packet_type |= RTE_PTYPE_L4_UDP;
2126 		break;
2127 	default:
2128 		MRVL_LOG(DEBUG, "Failed to recognise l4 packet type");
2129 		break;
2130 	}
2131 
2132 	return packet_type;
2133 }
2134 
2135 /**
2136  * Get offload information from the received packet descriptor.
2137  *
2138  * @param desc
2139  *   Pointer to the received packet descriptor.
2140  *
2141  * @return
2142  *   Mbuf offload flags.
2143  */
2144 static inline uint64_t
2145 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc)
2146 {
2147 	uint64_t flags;
2148 	enum pp2_inq_desc_status status;
2149 
2150 	status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2151 	if (unlikely(status != PP2_DESC_ERR_OK))
2152 		flags = PKT_RX_IP_CKSUM_BAD;
2153 	else
2154 		flags = PKT_RX_IP_CKSUM_GOOD;
2155 
2156 	status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2157 	if (unlikely(status != PP2_DESC_ERR_OK))
2158 		flags |= PKT_RX_L4_CKSUM_BAD;
2159 	else
2160 		flags |= PKT_RX_L4_CKSUM_GOOD;
2161 
2162 	return flags;
2163 }
2164 
2165 /**
2166  * DPDK callback for receive.
2167  *
2168  * @param rxq
2169  *   Generic pointer to the receive queue.
2170  * @param rx_pkts
2171  *   Array to store received packets.
2172  * @param nb_pkts
2173  *   Maximum number of packets in array.
2174  *
2175  * @return
2176  *   Number of packets successfully received.
2177  */
2178 static uint16_t
2179 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2180 {
2181 	struct mrvl_rxq *q = rxq;
2182 	struct pp2_ppio_desc descs[nb_pkts];
2183 	struct pp2_bpool *bpool;
2184 	int i, ret, rx_done = 0;
2185 	int num;
2186 	struct pp2_hif *hif;
2187 	unsigned int core_id = rte_lcore_id();
2188 
2189 	hif = mrvl_get_hif(q->priv, core_id);
2190 
2191 	if (unlikely(!q->priv->ppio || !hif))
2192 		return 0;
2193 
2194 	bpool = q->priv->bpool;
2195 
2196 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2197 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2198 	if (unlikely(ret < 0)) {
2199 		MRVL_LOG(ERR, "Failed to receive packets");
2200 		return 0;
2201 	}
2202 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2203 
2204 	for (i = 0; i < nb_pkts; i++) {
2205 		struct rte_mbuf *mbuf;
2206 		uint8_t l3_offset, l4_offset;
2207 		enum pp2_inq_desc_status status;
2208 		uint64_t addr;
2209 
2210 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2211 			struct pp2_ppio_desc *pref_desc;
2212 			u64 pref_addr;
2213 
2214 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2215 			pref_addr = cookie_addr_high |
2216 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2217 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2218 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2219 		}
2220 
2221 		addr = cookie_addr_high |
2222 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2223 		mbuf = (struct rte_mbuf *)addr;
2224 		rte_pktmbuf_reset(mbuf);
2225 
2226 		/* drop packet in case of mac, overrun or resource error */
2227 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2228 		if (unlikely(status != PP2_DESC_ERR_OK)) {
2229 			struct pp2_buff_inf binf = {
2230 				.addr = rte_mbuf_data_iova_default(mbuf),
2231 				.cookie = (uint64_t)mbuf,
2232 			};
2233 
2234 			pp2_bpool_put_buff(hif, bpool, &binf);
2235 			mrvl_port_bpool_size
2236 				[bpool->pp2_id][bpool->id][core_id]++;
2237 			q->drop_mac++;
2238 			continue;
2239 		}
2240 
2241 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2242 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2243 		mbuf->data_len = mbuf->pkt_len;
2244 		mbuf->port = q->port_id;
2245 		mbuf->packet_type =
2246 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2247 							    &l3_offset,
2248 							    &l4_offset);
2249 		mbuf->l2_len = l3_offset;
2250 		mbuf->l3_len = l4_offset - l3_offset;
2251 
2252 		if (likely(q->cksum_enabled))
2253 			mbuf->ol_flags = mrvl_desc_to_ol_flags(&descs[i]);
2254 
2255 		rx_pkts[rx_done++] = mbuf;
2256 		q->bytes_recv += mbuf->pkt_len;
2257 	}
2258 
2259 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2260 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2261 
2262 		if (unlikely(num <= q->priv->bpool_min_size ||
2263 			     (!rx_done && num < q->priv->bpool_init_size))) {
2264 			ret = mrvl_fill_bpool(q, MRVL_BURST_SIZE);
2265 			if (ret)
2266 				MRVL_LOG(ERR, "Failed to fill bpool");
2267 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2268 			int i;
2269 			int pkt_to_remove = num - q->priv->bpool_init_size;
2270 			struct rte_mbuf *mbuf;
2271 			struct pp2_buff_inf buff;
2272 
2273 			MRVL_LOG(DEBUG,
2274 				"port-%d:%d: bpool %d oversize - remove %d buffers (pool size: %d -> %d)",
2275 				bpool->pp2_id, q->priv->ppio->port_id,
2276 				bpool->id, pkt_to_remove, num,
2277 				q->priv->bpool_init_size);
2278 
2279 			for (i = 0; i < pkt_to_remove; i++) {
2280 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2281 				if (ret)
2282 					break;
2283 				mbuf = (struct rte_mbuf *)
2284 					(cookie_addr_high | buff.cookie);
2285 				rte_pktmbuf_free(mbuf);
2286 			}
2287 			mrvl_port_bpool_size
2288 				[bpool->pp2_id][bpool->id][core_id] -= i;
2289 		}
2290 		rte_spinlock_unlock(&q->priv->lock);
2291 	}
2292 
2293 	return rx_done;
2294 }
2295 
2296 /**
2297  * Prepare offload information.
2298  *
2299  * @param ol_flags
2300  *   Offload flags.
2301  * @param packet_type
2302  *   Packet type bitfield.
2303  * @param l3_type
2304  *   Pointer to the pp2_ouq_l3_type structure.
2305  * @param l4_type
2306  *   Pointer to the pp2_outq_l4_type structure.
2307  * @param gen_l3_cksum
2308  *   Will be set to 1 in case l3 checksum is computed.
2309  * @param l4_cksum
2310  *   Will be set to 1 in case l4 checksum is computed.
2311  *
2312  * @return
2313  *   0 on success, negative error value otherwise.
2314  */
2315 static inline int
2316 mrvl_prepare_proto_info(uint64_t ol_flags, uint32_t packet_type,
2317 			enum pp2_outq_l3_type *l3_type,
2318 			enum pp2_outq_l4_type *l4_type,
2319 			int *gen_l3_cksum,
2320 			int *gen_l4_cksum)
2321 {
2322 	/*
2323 	 * Based on ol_flags prepare information
2324 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2325 	 * for offloading.
2326 	 */
2327 	if (ol_flags & PKT_TX_IPV4) {
2328 		*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2329 		*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2330 	} else if (ol_flags & PKT_TX_IPV6) {
2331 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2332 		/* no checksum for ipv6 header */
2333 		*gen_l3_cksum = 0;
2334 	} else {
2335 		/* if something different then stop processing */
2336 		return -1;
2337 	}
2338 
2339 	ol_flags &= PKT_TX_L4_MASK;
2340 	if ((packet_type & RTE_PTYPE_L4_TCP) &&
2341 	    ol_flags == PKT_TX_TCP_CKSUM) {
2342 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2343 		*gen_l4_cksum = 1;
2344 	} else if ((packet_type & RTE_PTYPE_L4_UDP) &&
2345 		   ol_flags == PKT_TX_UDP_CKSUM) {
2346 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2347 		*gen_l4_cksum = 1;
2348 	} else {
2349 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2350 		/* no checksum for other type */
2351 		*gen_l4_cksum = 0;
2352 	}
2353 
2354 	return 0;
2355 }
2356 
2357 /**
2358  * Release already sent buffers to bpool (buffer-pool).
2359  *
2360  * @param ppio
2361  *   Pointer to the port structure.
2362  * @param hif
2363  *   Pointer to the MUSDK hardware interface.
2364  * @param sq
2365  *   Pointer to the shadow queue.
2366  * @param qid
2367  *   Queue id number.
2368  * @param force
2369  *   Force releasing packets.
2370  */
2371 static inline void
2372 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2373 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2374 		       int qid, int force)
2375 {
2376 	struct buff_release_entry *entry;
2377 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2378 	int i;
2379 
2380 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2381 
2382 	sq->num_to_release += nb_done;
2383 
2384 	if (likely(!force &&
2385 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2386 		return;
2387 
2388 	nb_done = sq->num_to_release;
2389 	sq->num_to_release = 0;
2390 
2391 	for (i = 0; i < nb_done; i++) {
2392 		entry = &sq->ent[sq->tail + num];
2393 		if (unlikely(!entry->buff.addr)) {
2394 			MRVL_LOG(ERR,
2395 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2396 				sq->tail, (u64)entry->buff.cookie,
2397 				(u64)entry->buff.addr);
2398 			skip_bufs = 1;
2399 			goto skip;
2400 		}
2401 
2402 		if (unlikely(!entry->bpool)) {
2403 			struct rte_mbuf *mbuf;
2404 
2405 			mbuf = (struct rte_mbuf *)
2406 			       (cookie_addr_high | entry->buff.cookie);
2407 			rte_pktmbuf_free(mbuf);
2408 			skip_bufs = 1;
2409 			goto skip;
2410 		}
2411 
2412 		mrvl_port_bpool_size
2413 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2414 		num++;
2415 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2416 			goto skip;
2417 		continue;
2418 skip:
2419 		if (likely(num))
2420 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2421 		num += skip_bufs;
2422 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2423 		sq->size -= num;
2424 		num = 0;
2425 		skip_bufs = 0;
2426 	}
2427 
2428 	if (likely(num)) {
2429 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2430 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2431 		sq->size -= num;
2432 	}
2433 }
2434 
2435 /**
2436  * DPDK callback for transmit.
2437  *
2438  * @param txq
2439  *   Generic pointer transmit queue.
2440  * @param tx_pkts
2441  *   Packets to transmit.
2442  * @param nb_pkts
2443  *   Number of packets in array.
2444  *
2445  * @return
2446  *   Number of packets successfully transmitted.
2447  */
2448 static uint16_t
2449 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2450 {
2451 	struct mrvl_txq *q = txq;
2452 	struct mrvl_shadow_txq *sq;
2453 	struct pp2_hif *hif;
2454 	struct pp2_ppio_desc descs[nb_pkts];
2455 	unsigned int core_id = rte_lcore_id();
2456 	int i, ret, bytes_sent = 0;
2457 	uint16_t num, sq_free_size;
2458 	uint64_t addr;
2459 
2460 	hif = mrvl_get_hif(q->priv, core_id);
2461 	sq = &q->shadow_txqs[core_id];
2462 
2463 	if (unlikely(!q->priv->ppio || !hif))
2464 		return 0;
2465 
2466 	if (sq->size)
2467 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2468 				       sq, q->queue_id, 0);
2469 
2470 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2471 	if (unlikely(nb_pkts > sq_free_size)) {
2472 		MRVL_LOG(DEBUG,
2473 			"No room in shadow queue for %d packets! %d packets will be sent.",
2474 			nb_pkts, sq_free_size);
2475 		nb_pkts = sq_free_size;
2476 	}
2477 
2478 	for (i = 0; i < nb_pkts; i++) {
2479 		struct rte_mbuf *mbuf = tx_pkts[i];
2480 		int gen_l3_cksum, gen_l4_cksum;
2481 		enum pp2_outq_l3_type l3_type;
2482 		enum pp2_outq_l4_type l4_type;
2483 
2484 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2485 			struct rte_mbuf *pref_pkt_hdr;
2486 
2487 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2488 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2489 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2490 		}
2491 
2492 		mrvl_fill_shadowq(sq, mbuf);
2493 		mrvl_fill_desc(&descs[i], mbuf);
2494 
2495 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2496 		/*
2497 		 * in case unsupported ol_flags were passed
2498 		 * do not update descriptor offload information
2499 		 */
2500 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2501 					      &l3_type, &l4_type, &gen_l3_cksum,
2502 					      &gen_l4_cksum);
2503 		if (unlikely(ret))
2504 			continue;
2505 
2506 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2507 						  mbuf->l2_len,
2508 						  mbuf->l2_len + mbuf->l3_len,
2509 						  gen_l3_cksum, gen_l4_cksum);
2510 	}
2511 
2512 	num = nb_pkts;
2513 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2514 	/* number of packets that were not sent */
2515 	if (unlikely(num > nb_pkts)) {
2516 		for (i = nb_pkts; i < num; i++) {
2517 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2518 				MRVL_PP2_TX_SHADOWQ_MASK;
2519 			addr = cookie_addr_high | sq->ent[sq->head].buff.cookie;
2520 			bytes_sent -=
2521 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2522 		}
2523 		sq->size -= num - nb_pkts;
2524 	}
2525 
2526 	q->bytes_sent += bytes_sent;
2527 
2528 	return nb_pkts;
2529 }
2530 
2531 /** DPDK callback for S/G transmit.
2532  *
2533  * @param txq
2534  *   Generic pointer transmit queue.
2535  * @param tx_pkts
2536  *   Packets to transmit.
2537  * @param nb_pkts
2538  *   Number of packets in array.
2539  *
2540  * @return
2541  *   Number of packets successfully transmitted.
2542  */
2543 static uint16_t
2544 mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
2545 		     uint16_t nb_pkts)
2546 {
2547 	struct mrvl_txq *q = txq;
2548 	struct mrvl_shadow_txq *sq;
2549 	struct pp2_hif *hif;
2550 	struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
2551 	struct pp2_ppio_sg_pkts pkts;
2552 	uint8_t frags[nb_pkts];
2553 	unsigned int core_id = rte_lcore_id();
2554 	int i, j, ret, bytes_sent = 0;
2555 	int tail, tail_first;
2556 	uint16_t num, sq_free_size;
2557 	uint16_t nb_segs, total_descs = 0;
2558 	uint64_t addr;
2559 
2560 	hif = mrvl_get_hif(q->priv, core_id);
2561 	sq = &q->shadow_txqs[core_id];
2562 	pkts.frags = frags;
2563 	pkts.num = 0;
2564 
2565 	if (unlikely(!q->priv->ppio || !hif))
2566 		return 0;
2567 
2568 	if (sq->size)
2569 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2570 				       sq, q->queue_id, 0);
2571 
2572 	/* Save shadow queue free size */
2573 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2574 
2575 	tail = 0;
2576 	for (i = 0; i < nb_pkts; i++) {
2577 		struct rte_mbuf *mbuf = tx_pkts[i];
2578 		struct rte_mbuf *seg = NULL;
2579 		int gen_l3_cksum, gen_l4_cksum;
2580 		enum pp2_outq_l3_type l3_type;
2581 		enum pp2_outq_l4_type l4_type;
2582 
2583 		nb_segs = mbuf->nb_segs;
2584 		tail_first = tail;
2585 		total_descs += nb_segs;
2586 
2587 		/*
2588 		 * Check if total_descs does not exceed
2589 		 * shadow queue free size
2590 		 */
2591 		if (unlikely(total_descs > sq_free_size)) {
2592 			total_descs -= nb_segs;
2593 			RTE_LOG(DEBUG, PMD,
2594 				"No room in shadow queue for %d packets! "
2595 				"%d packets will be sent.\n",
2596 				nb_pkts, i);
2597 			break;
2598 		}
2599 
2600 		/* Check if nb_segs does not exceed the max nb of desc per
2601 		 * fragmented packet
2602 		 */
2603 		if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
2604 			total_descs -= nb_segs;
2605 			RTE_LOG(ERR, PMD,
2606 				"Too many segments. Packet won't be sent.\n");
2607 			break;
2608 		}
2609 
2610 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2611 			struct rte_mbuf *pref_pkt_hdr;
2612 
2613 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2614 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2615 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2616 		}
2617 
2618 		pkts.frags[pkts.num] = nb_segs;
2619 		pkts.num++;
2620 
2621 		seg = mbuf;
2622 		for (j = 0; j < nb_segs - 1; j++) {
2623 			/* For the subsequent segments, set shadow queue
2624 			 * buffer to NULL
2625 			 */
2626 			mrvl_fill_shadowq(sq, NULL);
2627 			mrvl_fill_desc(&descs[tail], seg);
2628 
2629 			tail++;
2630 			seg = seg->next;
2631 		}
2632 		/* Put first mbuf info in last shadow queue entry */
2633 		mrvl_fill_shadowq(sq, mbuf);
2634 		/* Update descriptor with last segment */
2635 		mrvl_fill_desc(&descs[tail++], seg);
2636 
2637 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2638 		/* In case unsupported ol_flags were passed
2639 		 * do not update descriptor offload information
2640 		 */
2641 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2642 					      &l3_type, &l4_type, &gen_l3_cksum,
2643 					      &gen_l4_cksum);
2644 		if (unlikely(ret))
2645 			continue;
2646 
2647 		pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
2648 						  l4_type, mbuf->l2_len,
2649 						  mbuf->l2_len + mbuf->l3_len,
2650 						  gen_l3_cksum, gen_l4_cksum);
2651 	}
2652 
2653 	num = total_descs;
2654 	pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
2655 			 &total_descs, &pkts);
2656 	/* number of packets that were not sent */
2657 	if (unlikely(num > total_descs)) {
2658 		for (i = total_descs; i < num; i++) {
2659 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2660 				MRVL_PP2_TX_SHADOWQ_MASK;
2661 
2662 			addr = sq->ent[sq->head].buff.cookie;
2663 			if (addr)
2664 				bytes_sent -=
2665 					rte_pktmbuf_pkt_len((struct rte_mbuf *)
2666 						(cookie_addr_high | addr));
2667 		}
2668 		sq->size -= num - total_descs;
2669 		nb_pkts = pkts.num;
2670 	}
2671 
2672 	q->bytes_sent += bytes_sent;
2673 
2674 	return nb_pkts;
2675 }
2676 
2677 /**
2678  * Initialize packet processor.
2679  *
2680  * @return
2681  *   0 on success, negative error value otherwise.
2682  */
2683 static int
2684 mrvl_init_pp2(void)
2685 {
2686 	struct pp2_init_params init_params;
2687 
2688 	memset(&init_params, 0, sizeof(init_params));
2689 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
2690 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
2691 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
2692 
2693 	return pp2_init(&init_params);
2694 }
2695 
2696 /**
2697  * Deinitialize packet processor.
2698  *
2699  * @return
2700  *   0 on success, negative error value otherwise.
2701  */
2702 static void
2703 mrvl_deinit_pp2(void)
2704 {
2705 	pp2_deinit();
2706 }
2707 
2708 /**
2709  * Create private device structure.
2710  *
2711  * @param dev_name
2712  *   Pointer to the port name passed in the initialization parameters.
2713  *
2714  * @return
2715  *   Pointer to the newly allocated private device structure.
2716  */
2717 static struct mrvl_priv *
2718 mrvl_priv_create(const char *dev_name)
2719 {
2720 	struct pp2_bpool_params bpool_params;
2721 	char match[MRVL_MATCH_LEN];
2722 	struct mrvl_priv *priv;
2723 	int ret, bpool_bit;
2724 
2725 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
2726 	if (!priv)
2727 		return NULL;
2728 
2729 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
2730 				       &priv->pp_id, &priv->ppio_id);
2731 	if (ret)
2732 		goto out_free_priv;
2733 
2734 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
2735 				     PP2_BPOOL_NUM_POOLS);
2736 	if (bpool_bit < 0)
2737 		goto out_free_priv;
2738 	priv->bpool_bit = bpool_bit;
2739 
2740 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
2741 		 priv->bpool_bit);
2742 	memset(&bpool_params, 0, sizeof(bpool_params));
2743 	bpool_params.match = match;
2744 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
2745 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
2746 	if (ret)
2747 		goto out_clear_bpool_bit;
2748 
2749 	priv->ppio_params.type = PP2_PPIO_T_NIC;
2750 	rte_spinlock_init(&priv->lock);
2751 
2752 	return priv;
2753 out_clear_bpool_bit:
2754 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2755 out_free_priv:
2756 	rte_free(priv);
2757 	return NULL;
2758 }
2759 
2760 /**
2761  * Create device representing Ethernet port.
2762  *
2763  * @param name
2764  *   Pointer to the port's name.
2765  *
2766  * @return
2767  *   0 on success, negative error value otherwise.
2768  */
2769 static int
2770 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
2771 {
2772 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
2773 	struct rte_eth_dev *eth_dev;
2774 	struct mrvl_priv *priv;
2775 	struct ifreq req;
2776 
2777 	eth_dev = rte_eth_dev_allocate(name);
2778 	if (!eth_dev)
2779 		return -ENOMEM;
2780 
2781 	priv = mrvl_priv_create(name);
2782 	if (!priv) {
2783 		ret = -ENOMEM;
2784 		goto out_free;
2785 	}
2786 	eth_dev->data->dev_private = priv;
2787 
2788 	eth_dev->data->mac_addrs =
2789 		rte_zmalloc("mac_addrs",
2790 			    ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
2791 	if (!eth_dev->data->mac_addrs) {
2792 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
2793 		ret = -ENOMEM;
2794 		goto out_free;
2795 	}
2796 
2797 	memset(&req, 0, sizeof(req));
2798 	strcpy(req.ifr_name, name);
2799 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
2800 	if (ret)
2801 		goto out_free;
2802 
2803 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
2804 	       req.ifr_addr.sa_data, ETHER_ADDR_LEN);
2805 
2806 	eth_dev->data->kdrv = RTE_KDRV_NONE;
2807 	eth_dev->device = &vdev->device;
2808 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
2809 	mrvl_set_tx_function(eth_dev);
2810 	eth_dev->dev_ops = &mrvl_ops;
2811 
2812 	rte_eth_dev_probing_finish(eth_dev);
2813 	return 0;
2814 out_free:
2815 	rte_eth_dev_release_port(eth_dev);
2816 
2817 	return ret;
2818 }
2819 
2820 /**
2821  * Cleanup previously created device representing Ethernet port.
2822  *
2823  * @param name
2824  *   Pointer to the port name.
2825  */
2826 static void
2827 mrvl_eth_dev_destroy(const char *name)
2828 {
2829 	struct rte_eth_dev *eth_dev;
2830 	struct mrvl_priv *priv;
2831 
2832 	eth_dev = rte_eth_dev_allocated(name);
2833 	if (!eth_dev)
2834 		return;
2835 
2836 	priv = eth_dev->data->dev_private;
2837 	pp2_bpool_deinit(priv->bpool);
2838 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2839 	rte_eth_dev_release_port(eth_dev);
2840 }
2841 
2842 /**
2843  * Callback used by rte_kvargs_process() during argument parsing.
2844  *
2845  * @param key
2846  *   Pointer to the parsed key (unused).
2847  * @param value
2848  *   Pointer to the parsed value.
2849  * @param extra_args
2850  *   Pointer to the extra arguments which contains address of the
2851  *   table of pointers to parsed interface names.
2852  *
2853  * @return
2854  *   Always 0.
2855  */
2856 static int
2857 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
2858 		 void *extra_args)
2859 {
2860 	struct mrvl_ifnames *ifnames = extra_args;
2861 
2862 	ifnames->names[ifnames->idx++] = value;
2863 
2864 	return 0;
2865 }
2866 
2867 /**
2868  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
2869  */
2870 static void
2871 mrvl_deinit_hifs(void)
2872 {
2873 	int i;
2874 
2875 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
2876 		if (hifs[i])
2877 			pp2_hif_deinit(hifs[i]);
2878 	}
2879 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
2880 	memset(hifs, 0, sizeof(hifs));
2881 }
2882 
2883 /**
2884  * DPDK callback to register the virtual device.
2885  *
2886  * @param vdev
2887  *   Pointer to the virtual device.
2888  *
2889  * @return
2890  *   0 on success, negative error value otherwise.
2891  */
2892 static int
2893 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
2894 {
2895 	struct rte_kvargs *kvlist;
2896 	struct mrvl_ifnames ifnames;
2897 	int ret = -EINVAL;
2898 	uint32_t i, ifnum, cfgnum;
2899 	const char *params;
2900 
2901 	params = rte_vdev_device_args(vdev);
2902 	if (!params)
2903 		return -EINVAL;
2904 
2905 	kvlist = rte_kvargs_parse(params, valid_args);
2906 	if (!kvlist)
2907 		return -EINVAL;
2908 
2909 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
2910 	if (ifnum > RTE_DIM(ifnames.names))
2911 		goto out_free_kvlist;
2912 
2913 	ifnames.idx = 0;
2914 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
2915 			   mrvl_get_ifnames, &ifnames);
2916 
2917 
2918 	/*
2919 	 * The below system initialization should be done only once,
2920 	 * on the first provided configuration file
2921 	 */
2922 	if (!mrvl_qos_cfg) {
2923 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
2924 		MRVL_LOG(INFO, "Parsing config file!");
2925 		if (cfgnum > 1) {
2926 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
2927 			goto out_free_kvlist;
2928 		} else if (cfgnum == 1) {
2929 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
2930 					   mrvl_get_qoscfg, &mrvl_qos_cfg);
2931 		}
2932 	}
2933 
2934 	if (mrvl_dev_num)
2935 		goto init_devices;
2936 
2937 	MRVL_LOG(INFO, "Perform MUSDK initializations");
2938 
2939 	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
2940 	if (ret)
2941 		goto out_free_kvlist;
2942 
2943 	ret = mrvl_init_pp2();
2944 	if (ret) {
2945 		MRVL_LOG(ERR, "Failed to init PP!");
2946 		rte_mvep_deinit(MVEP_MOD_T_PP2);
2947 		goto out_free_kvlist;
2948 	}
2949 
2950 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
2951 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
2952 
2953 	mrvl_lcore_first = RTE_MAX_LCORE;
2954 	mrvl_lcore_last = 0;
2955 
2956 init_devices:
2957 	for (i = 0; i < ifnum; i++) {
2958 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
2959 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
2960 		if (ret)
2961 			goto out_cleanup;
2962 	}
2963 	mrvl_dev_num += ifnum;
2964 
2965 	rte_kvargs_free(kvlist);
2966 
2967 	return 0;
2968 out_cleanup:
2969 	for (; i > 0; i--)
2970 		mrvl_eth_dev_destroy(ifnames.names[i]);
2971 
2972 	if (mrvl_dev_num == 0) {
2973 		mrvl_deinit_pp2();
2974 		rte_mvep_deinit(MVEP_MOD_T_PP2);
2975 	}
2976 out_free_kvlist:
2977 	rte_kvargs_free(kvlist);
2978 
2979 	return ret;
2980 }
2981 
2982 /**
2983  * DPDK callback to remove virtual device.
2984  *
2985  * @param vdev
2986  *   Pointer to the removed virtual device.
2987  *
2988  * @return
2989  *   0 on success, negative error value otherwise.
2990  */
2991 static int
2992 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
2993 {
2994 	int i;
2995 	const char *name;
2996 
2997 	name = rte_vdev_device_name(vdev);
2998 	if (!name)
2999 		return -EINVAL;
3000 
3001 	MRVL_LOG(INFO, "Removing %s", name);
3002 
3003 	RTE_ETH_FOREACH_DEV(i) { /* FIXME: removing all devices! */
3004 		char ifname[RTE_ETH_NAME_MAX_LEN];
3005 
3006 		rte_eth_dev_get_name_by_port(i, ifname);
3007 		mrvl_eth_dev_destroy(ifname);
3008 		mrvl_dev_num--;
3009 	}
3010 
3011 	if (mrvl_dev_num == 0) {
3012 		MRVL_LOG(INFO, "Perform MUSDK deinit");
3013 		mrvl_deinit_hifs();
3014 		mrvl_deinit_pp2();
3015 		rte_mvep_deinit(MVEP_MOD_T_PP2);
3016 	}
3017 
3018 	return 0;
3019 }
3020 
3021 static struct rte_vdev_driver pmd_mrvl_drv = {
3022 	.probe = rte_pmd_mrvl_probe,
3023 	.remove = rte_pmd_mrvl_remove,
3024 };
3025 
3026 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
3027 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
3028 
3029 RTE_INIT(mrvl_init_log)
3030 {
3031 	mrvl_logtype = rte_log_register("pmd.net.mvpp2");
3032 	if (mrvl_logtype >= 0)
3033 		rte_log_set_level(mrvl_logtype, RTE_LOG_NOTICE);
3034 }
3035