xref: /dpdk/drivers/net/mvpp2/mrvl_ethdev.c (revision fc026401f48fc606dcab33c39e48d25fe891ec19)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2017-2021 Marvell International Ltd.
3  * Copyright(c) 2017-2021 Semihalf.
4  * All rights reserved.
5  */
6 
7 #include <rte_string_fns.h>
8 #include <ethdev_driver.h>
9 #include <rte_kvargs.h>
10 #include <rte_log.h>
11 #include <rte_malloc.h>
12 #include <rte_bus_vdev.h>
13 
14 #include <fcntl.h>
15 #include <linux/ethtool.h>
16 #include <linux/sockios.h>
17 #include <net/if.h>
18 #include <net/if_arp.h>
19 #include <sys/ioctl.h>
20 #include <sys/socket.h>
21 #include <sys/stat.h>
22 #include <sys/types.h>
23 
24 #include <rte_mvep_common.h>
25 #include "mrvl_ethdev.h"
26 #include "mrvl_qos.h"
27 #include "mrvl_flow.h"
28 #include "mrvl_mtr.h"
29 #include "mrvl_tm.h"
30 
31 /* bitmask with reserved hifs */
32 #define MRVL_MUSDK_HIFS_RESERVED 0x0F
33 /* bitmask with reserved bpools */
34 #define MRVL_MUSDK_BPOOLS_RESERVED 0x07
35 /* bitmask with reserved kernel RSS tables */
36 #define MRVL_MUSDK_RSS_RESERVED 0x0F
37 /* maximum number of available hifs */
38 #define MRVL_MUSDK_HIFS_MAX 9
39 
40 /* prefetch shift */
41 #define MRVL_MUSDK_PREFETCH_SHIFT 2
42 
43 /* TCAM has 25 entries reserved for uc/mc filter entries */
44 #define MRVL_MAC_ADDRS_MAX 25
45 #define MRVL_MATCH_LEN 16
46 #define MRVL_PKT_EFFEC_OFFS (MRVL_PKT_OFFS + MV_MH_SIZE)
47 /* Maximum allowable packet size */
48 #define MRVL_PKT_SIZE_MAX (10240 - MV_MH_SIZE)
49 
50 #define MRVL_IFACE_NAME_ARG "iface"
51 #define MRVL_CFG_ARG "cfg"
52 
53 #define MRVL_BURST_SIZE 64
54 
55 #define MRVL_ARP_LENGTH 28
56 
57 #define MRVL_COOKIE_ADDR_INVALID ~0ULL
58 #define MRVL_COOKIE_HIGH_ADDR_MASK 0xffffff0000000000
59 
60 /** Port Rx offload capabilities */
61 #define MRVL_RX_OFFLOADS (DEV_RX_OFFLOAD_VLAN_FILTER | \
62 			  DEV_RX_OFFLOAD_JUMBO_FRAME | \
63 			  DEV_RX_OFFLOAD_CHECKSUM)
64 
65 /** Port Tx offloads capabilities */
66 #define MRVL_TX_OFFLOADS (DEV_TX_OFFLOAD_IPV4_CKSUM | \
67 			  DEV_TX_OFFLOAD_UDP_CKSUM | \
68 			  DEV_TX_OFFLOAD_TCP_CKSUM | \
69 			  DEV_TX_OFFLOAD_MULTI_SEGS)
70 
71 static const char * const valid_args[] = {
72 	MRVL_IFACE_NAME_ARG,
73 	MRVL_CFG_ARG,
74 	NULL
75 };
76 
77 static int used_hifs = MRVL_MUSDK_HIFS_RESERVED;
78 static struct pp2_hif *hifs[RTE_MAX_LCORE];
79 static int used_bpools[PP2_NUM_PKT_PROC] = {
80 	[0 ... PP2_NUM_PKT_PROC - 1] = MRVL_MUSDK_BPOOLS_RESERVED
81 };
82 
83 static struct pp2_bpool *mrvl_port_to_bpool_lookup[RTE_MAX_ETHPORTS];
84 static int mrvl_port_bpool_size[PP2_NUM_PKT_PROC][PP2_BPOOL_NUM_POOLS][RTE_MAX_LCORE];
85 static uint64_t cookie_addr_high = MRVL_COOKIE_ADDR_INVALID;
86 
87 struct mrvl_ifnames {
88 	const char *names[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
89 	int idx;
90 };
91 
92 /*
93  * To use buffer harvesting based on loopback port shadow queue structure
94  * was introduced for buffers information bookkeeping.
95  *
96  * Before sending the packet, related buffer information (pp2_buff_inf) is
97  * stored in shadow queue. After packet is transmitted no longer used
98  * packet buffer is released back to it's original hardware pool,
99  * on condition it originated from interface.
100  * In case it  was generated by application itself i.e: mbuf->port field is
101  * 0xff then its released to software mempool.
102  */
103 struct mrvl_shadow_txq {
104 	int head;           /* write index - used when sending buffers */
105 	int tail;           /* read index - used when releasing buffers */
106 	u16 size;           /* queue occupied size */
107 	u16 num_to_release; /* number of descriptors sent, that can be
108 			     * released
109 			     */
110 	struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
111 };
112 
113 struct mrvl_rxq {
114 	struct mrvl_priv *priv;
115 	struct rte_mempool *mp;
116 	int queue_id;
117 	int port_id;
118 	int cksum_enabled;
119 	uint64_t bytes_recv;
120 	uint64_t drop_mac;
121 };
122 
123 struct mrvl_txq {
124 	struct mrvl_priv *priv;
125 	int queue_id;
126 	int port_id;
127 	uint64_t bytes_sent;
128 	struct mrvl_shadow_txq shadow_txqs[RTE_MAX_LCORE];
129 	int tx_deferred_start;
130 };
131 
132 static int mrvl_lcore_first;
133 static int mrvl_lcore_last;
134 static int mrvl_dev_num;
135 
136 static int mrvl_fill_bpool(struct mrvl_rxq *rxq, int num);
137 static inline void mrvl_free_sent_buffers(struct pp2_ppio *ppio,
138 			struct pp2_hif *hif, unsigned int core_id,
139 			struct mrvl_shadow_txq *sq, int qid, int force);
140 
141 static uint16_t mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
142 				  uint16_t nb_pkts);
143 static uint16_t mrvl_tx_sg_pkt_burst(void *txq,	struct rte_mbuf **tx_pkts,
144 				     uint16_t nb_pkts);
145 static int rte_pmd_mrvl_remove(struct rte_vdev_device *vdev);
146 static void mrvl_deinit_pp2(void);
147 static void mrvl_deinit_hifs(void);
148 
149 static int
150 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
151 		  uint32_t index, uint32_t vmdq __rte_unused);
152 static int
153 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr);
154 static int
155 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on);
156 static int mrvl_promiscuous_enable(struct rte_eth_dev *dev);
157 static int mrvl_allmulticast_enable(struct rte_eth_dev *dev);
158 
159 #define MRVL_XSTATS_TBL_ENTRY(name) { \
160 	#name, offsetof(struct pp2_ppio_statistics, name),	\
161 	sizeof(((struct pp2_ppio_statistics *)0)->name)		\
162 }
163 
164 /* Table with xstats data */
165 static struct {
166 	const char *name;
167 	unsigned int offset;
168 	unsigned int size;
169 } mrvl_xstats_tbl[] = {
170 	MRVL_XSTATS_TBL_ENTRY(rx_bytes),
171 	MRVL_XSTATS_TBL_ENTRY(rx_packets),
172 	MRVL_XSTATS_TBL_ENTRY(rx_unicast_packets),
173 	MRVL_XSTATS_TBL_ENTRY(rx_errors),
174 	MRVL_XSTATS_TBL_ENTRY(rx_fullq_dropped),
175 	MRVL_XSTATS_TBL_ENTRY(rx_bm_dropped),
176 	MRVL_XSTATS_TBL_ENTRY(rx_early_dropped),
177 	MRVL_XSTATS_TBL_ENTRY(rx_fifo_dropped),
178 	MRVL_XSTATS_TBL_ENTRY(rx_cls_dropped),
179 	MRVL_XSTATS_TBL_ENTRY(tx_bytes),
180 	MRVL_XSTATS_TBL_ENTRY(tx_packets),
181 	MRVL_XSTATS_TBL_ENTRY(tx_unicast_packets),
182 	MRVL_XSTATS_TBL_ENTRY(tx_errors)
183 };
184 
185 static inline void
186 mrvl_fill_shadowq(struct mrvl_shadow_txq *sq, struct rte_mbuf *buf)
187 {
188 	sq->ent[sq->head].buff.cookie = (uint64_t)buf;
189 	sq->ent[sq->head].buff.addr = buf ?
190 		rte_mbuf_data_iova_default(buf) : 0;
191 
192 	sq->ent[sq->head].bpool =
193 		(unlikely(!buf || buf->port >= RTE_MAX_ETHPORTS ||
194 		 buf->refcnt > 1)) ? NULL :
195 		 mrvl_port_to_bpool_lookup[buf->port];
196 
197 	sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
198 	sq->size++;
199 }
200 
201 static inline void
202 mrvl_fill_desc(struct pp2_ppio_desc *desc, struct rte_mbuf *buf)
203 {
204 	pp2_ppio_outq_desc_reset(desc);
205 	pp2_ppio_outq_desc_set_phys_addr(desc, rte_pktmbuf_iova(buf));
206 	pp2_ppio_outq_desc_set_pkt_offset(desc, 0);
207 	pp2_ppio_outq_desc_set_pkt_len(desc, rte_pktmbuf_data_len(buf));
208 }
209 
210 static inline int
211 mrvl_get_bpool_size(int pp2_id, int pool_id)
212 {
213 	int i;
214 	int size = 0;
215 
216 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
217 		size += mrvl_port_bpool_size[pp2_id][pool_id][i];
218 
219 	return size;
220 }
221 
222 static inline int
223 mrvl_reserve_bit(int *bitmap, int max)
224 {
225 	int n = sizeof(*bitmap) * 8 - __builtin_clz(*bitmap);
226 
227 	if (n >= max)
228 		return -1;
229 
230 	*bitmap |= 1 << n;
231 
232 	return n;
233 }
234 
235 static int
236 mrvl_init_hif(int core_id)
237 {
238 	struct pp2_hif_params params;
239 	char match[MRVL_MATCH_LEN];
240 	int ret;
241 
242 	ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
243 	if (ret < 0) {
244 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
245 		return ret;
246 	}
247 
248 	snprintf(match, sizeof(match), "hif-%d", ret);
249 	memset(&params, 0, sizeof(params));
250 	params.match = match;
251 	params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
252 	ret = pp2_hif_init(&params, &hifs[core_id]);
253 	if (ret) {
254 		MRVL_LOG(ERR, "Failed to initialize hif %d", core_id);
255 		return ret;
256 	}
257 
258 	return 0;
259 }
260 
261 static inline struct pp2_hif*
262 mrvl_get_hif(struct mrvl_priv *priv, int core_id)
263 {
264 	int ret;
265 
266 	if (likely(hifs[core_id] != NULL))
267 		return hifs[core_id];
268 
269 	rte_spinlock_lock(&priv->lock);
270 
271 	ret = mrvl_init_hif(core_id);
272 	if (ret < 0) {
273 		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
274 		goto out;
275 	}
276 
277 	if (core_id < mrvl_lcore_first)
278 		mrvl_lcore_first = core_id;
279 
280 	if (core_id > mrvl_lcore_last)
281 		mrvl_lcore_last = core_id;
282 out:
283 	rte_spinlock_unlock(&priv->lock);
284 
285 	return hifs[core_id];
286 }
287 
288 /**
289  * Set tx burst function according to offload flag
290  *
291  * @param dev
292  *   Pointer to Ethernet device structure.
293  */
294 static void
295 mrvl_set_tx_function(struct rte_eth_dev *dev)
296 {
297 	struct mrvl_priv *priv = dev->data->dev_private;
298 
299 	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
300 	if (priv->multiseg) {
301 		RTE_LOG(INFO, PMD, "Using multi-segment tx callback\n");
302 		dev->tx_pkt_burst = mrvl_tx_sg_pkt_burst;
303 	} else {
304 		RTE_LOG(INFO, PMD, "Using single-segment tx callback\n");
305 		dev->tx_pkt_burst = mrvl_tx_pkt_burst;
306 	}
307 }
308 
309 /**
310  * Configure rss based on dpdk rss configuration.
311  *
312  * @param priv
313  *   Pointer to private structure.
314  * @param rss_conf
315  *   Pointer to RSS configuration.
316  *
317  * @return
318  *   0 on success, negative error value otherwise.
319  */
320 static int
321 mrvl_configure_rss(struct mrvl_priv *priv, struct rte_eth_rss_conf *rss_conf)
322 {
323 	if (rss_conf->rss_key)
324 		MRVL_LOG(WARNING, "Changing hash key is not supported");
325 
326 	if (rss_conf->rss_hf == 0) {
327 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
328 	} else if (rss_conf->rss_hf & ETH_RSS_IPV4) {
329 		priv->ppio_params.inqs_params.hash_type =
330 			PP2_PPIO_HASH_T_2_TUPLE;
331 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP) {
332 		priv->ppio_params.inqs_params.hash_type =
333 			PP2_PPIO_HASH_T_5_TUPLE;
334 		priv->rss_hf_tcp = 1;
335 	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) {
336 		priv->ppio_params.inqs_params.hash_type =
337 			PP2_PPIO_HASH_T_5_TUPLE;
338 		priv->rss_hf_tcp = 0;
339 	} else {
340 		return -EINVAL;
341 	}
342 
343 	return 0;
344 }
345 
346 /**
347  * Ethernet device configuration.
348  *
349  * Prepare the driver for a given number of TX and RX queues and
350  * configure RSS.
351  *
352  * @param dev
353  *   Pointer to Ethernet device structure.
354  *
355  * @return
356  *   0 on success, negative error value otherwise.
357  */
358 static int
359 mrvl_dev_configure(struct rte_eth_dev *dev)
360 {
361 	struct mrvl_priv *priv = dev->data->dev_private;
362 	int ret;
363 
364 	if (priv->ppio) {
365 		MRVL_LOG(INFO, "Device reconfiguration is not supported");
366 		return -EINVAL;
367 	}
368 
369 	if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_NONE &&
370 	    dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
371 		MRVL_LOG(INFO, "Unsupported rx multi queue mode %d",
372 			dev->data->dev_conf.rxmode.mq_mode);
373 		return -EINVAL;
374 	}
375 
376 	if (dev->data->dev_conf.rxmode.split_hdr_size) {
377 		MRVL_LOG(INFO, "Split headers not supported");
378 		return -EINVAL;
379 	}
380 
381 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
382 		dev->data->mtu = dev->data->dev_conf.rxmode.max_rx_pkt_len -
383 				 MRVL_PP2_ETH_HDRS_LEN;
384 
385 	if (dev->data->dev_conf.txmode.offloads & DEV_TX_OFFLOAD_MULTI_SEGS)
386 		priv->multiseg = 1;
387 
388 	ret = mrvl_configure_rxqs(priv, dev->data->port_id,
389 				  dev->data->nb_rx_queues);
390 	if (ret < 0)
391 		return ret;
392 
393 	ret = mrvl_configure_txqs(priv, dev->data->port_id,
394 				  dev->data->nb_tx_queues);
395 	if (ret < 0)
396 		return ret;
397 
398 	priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
399 	priv->ppio_params.maintain_stats = 1;
400 	priv->nb_rx_queues = dev->data->nb_rx_queues;
401 
402 	ret = mrvl_tm_init(dev);
403 	if (ret < 0)
404 		return ret;
405 
406 	if (dev->data->nb_rx_queues == 1 &&
407 	    dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
408 		MRVL_LOG(WARNING, "Disabling hash for 1 rx queue");
409 		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
410 
411 		return 0;
412 	}
413 
414 	return mrvl_configure_rss(priv,
415 			&dev->data->dev_conf.rx_adv_conf.rss_conf);
416 }
417 
418 /**
419  * DPDK callback to change the MTU.
420  *
421  * Setting the MTU affects hardware MRU (packets larger than the MRU
422  * will be dropped).
423  *
424  * @param dev
425  *   Pointer to Ethernet device structure.
426  * @param mtu
427  *   New MTU.
428  *
429  * @return
430  *   0 on success, negative error value otherwise.
431  */
432 static int
433 mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
434 {
435 	struct mrvl_priv *priv = dev->data->dev_private;
436 	uint16_t mru;
437 	uint16_t mbuf_data_size = 0; /* SW buffer size */
438 	int ret;
439 
440 	mru = MRVL_PP2_MTU_TO_MRU(mtu);
441 	/*
442 	 * min_rx_buf_size is equal to mbuf data size
443 	 * if pmd didn't set it differently
444 	 */
445 	mbuf_data_size = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
446 	/* Prevent PMD from:
447 	 * - setting mru greater than the mbuf size resulting in
448 	 * hw and sw buffer size mismatch
449 	 * - setting mtu that requires the support of scattered packets
450 	 * when this feature has not been enabled/supported so far
451 	 * (TODO check scattered_rx flag here once scattered RX is supported).
452 	 */
453 	if (mru - RTE_ETHER_CRC_LEN + MRVL_PKT_OFFS > mbuf_data_size) {
454 		mru = mbuf_data_size + RTE_ETHER_CRC_LEN - MRVL_PKT_OFFS;
455 		mtu = MRVL_PP2_MRU_TO_MTU(mru);
456 		MRVL_LOG(WARNING, "MTU too big, max MTU possible limitted "
457 			"by current mbuf size: %u. Set MTU to %u, MRU to %u",
458 			mbuf_data_size, mtu, mru);
459 	}
460 
461 	if (mtu < RTE_ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX) {
462 		MRVL_LOG(ERR, "Invalid MTU [%u] or MRU [%u]", mtu, mru);
463 		return -EINVAL;
464 	}
465 
466 	dev->data->mtu = mtu;
467 	dev->data->dev_conf.rxmode.max_rx_pkt_len = mru - MV_MH_SIZE;
468 
469 	if (!priv->ppio)
470 		return 0;
471 
472 	ret = pp2_ppio_set_mru(priv->ppio, mru);
473 	if (ret) {
474 		MRVL_LOG(ERR, "Failed to change MRU");
475 		return ret;
476 	}
477 
478 	ret = pp2_ppio_set_mtu(priv->ppio, mtu);
479 	if (ret) {
480 		MRVL_LOG(ERR, "Failed to change MTU");
481 		return ret;
482 	}
483 
484 	return 0;
485 }
486 
487 /**
488  * DPDK callback to bring the link up.
489  *
490  * @param dev
491  *   Pointer to Ethernet device structure.
492  *
493  * @return
494  *   0 on success, negative error value otherwise.
495  */
496 static int
497 mrvl_dev_set_link_up(struct rte_eth_dev *dev)
498 {
499 	struct mrvl_priv *priv = dev->data->dev_private;
500 	int ret;
501 
502 	if (!priv->ppio) {
503 		dev->data->dev_link.link_status = ETH_LINK_UP;
504 		return 0;
505 	}
506 
507 	ret = pp2_ppio_enable(priv->ppio);
508 	if (ret)
509 		return ret;
510 
511 	/*
512 	 * mtu/mru can be updated if pp2_ppio_enable() was called at least once
513 	 * as pp2_ppio_enable() changes port->t_mode from default 0 to
514 	 * PP2_TRAFFIC_INGRESS_EGRESS.
515 	 *
516 	 * Set mtu to default DPDK value here.
517 	 */
518 	ret = mrvl_mtu_set(dev, dev->data->mtu);
519 	if (ret) {
520 		pp2_ppio_disable(priv->ppio);
521 		return ret;
522 	}
523 
524 	dev->data->dev_link.link_status = ETH_LINK_UP;
525 	return 0;
526 }
527 
528 /**
529  * DPDK callback to bring the link down.
530  *
531  * @param dev
532  *   Pointer to Ethernet device structure.
533  *
534  * @return
535  *   0 on success, negative error value otherwise.
536  */
537 static int
538 mrvl_dev_set_link_down(struct rte_eth_dev *dev)
539 {
540 	struct mrvl_priv *priv = dev->data->dev_private;
541 	int ret;
542 
543 	if (!priv->ppio) {
544 		dev->data->dev_link.link_status = ETH_LINK_DOWN;
545 		return 0;
546 	}
547 	ret = pp2_ppio_disable(priv->ppio);
548 	if (ret)
549 		return ret;
550 
551 	dev->data->dev_link.link_status = ETH_LINK_DOWN;
552 	return 0;
553 }
554 
555 /**
556  * DPDK callback to start tx queue.
557  *
558  * @param dev
559  *   Pointer to Ethernet device structure.
560  * @param queue_id
561  *   Transmit queue index.
562  *
563  * @return
564  *   0 on success, negative error value otherwise.
565  */
566 static int
567 mrvl_tx_queue_start(struct rte_eth_dev *dev, uint16_t queue_id)
568 {
569 	struct mrvl_priv *priv = dev->data->dev_private;
570 	int ret;
571 
572 	if (!priv)
573 		return -EPERM;
574 
575 	/* passing 1 enables given tx queue */
576 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 1);
577 	if (ret) {
578 		MRVL_LOG(ERR, "Failed to start txq %d", queue_id);
579 		return ret;
580 	}
581 
582 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
583 
584 	return 0;
585 }
586 
587 /**
588  * DPDK callback to stop tx queue.
589  *
590  * @param dev
591  *   Pointer to Ethernet device structure.
592  * @param queue_id
593  *   Transmit queue index.
594  *
595  * @return
596  *   0 on success, negative error value otherwise.
597  */
598 static int
599 mrvl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t queue_id)
600 {
601 	struct mrvl_priv *priv = dev->data->dev_private;
602 	int ret;
603 
604 	if (!priv->ppio)
605 		return -EPERM;
606 
607 	/* passing 0 disables given tx queue */
608 	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 0);
609 	if (ret) {
610 		MRVL_LOG(ERR, "Failed to stop txq %d", queue_id);
611 		return ret;
612 	}
613 
614 	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
615 
616 	return 0;
617 }
618 
619 /**
620  * DPDK callback to start the device.
621  *
622  * @param dev
623  *   Pointer to Ethernet device structure.
624  *
625  * @return
626  *   0 on success, negative errno value on failure.
627  */
628 static int
629 mrvl_dev_start(struct rte_eth_dev *dev)
630 {
631 	struct mrvl_priv *priv = dev->data->dev_private;
632 	char match[MRVL_MATCH_LEN];
633 	int ret = 0, i, def_init_size;
634 	uint32_t j;
635 	struct rte_vlan_filter_conf *vfc;
636 	struct rte_ether_addr *mac_addr;
637 
638 	if (priv->ppio)
639 		return mrvl_dev_set_link_up(dev);
640 
641 	snprintf(match, sizeof(match), "ppio-%d:%d",
642 		 priv->pp_id, priv->ppio_id);
643 	priv->ppio_params.match = match;
644 
645 	/*
646 	 * Calculate the minimum bpool size for refill feature as follows:
647 	 * 2 default burst sizes multiply by number of rx queues.
648 	 * If the bpool size will be below this value, new buffers will
649 	 * be added to the pool.
650 	 */
651 	priv->bpool_min_size = priv->nb_rx_queues * MRVL_BURST_SIZE * 2;
652 
653 	/* In case initial bpool size configured in queues setup is
654 	 * smaller than minimum size add more buffers
655 	 */
656 	def_init_size = priv->bpool_min_size + MRVL_BURST_SIZE * 2;
657 	if (priv->bpool_init_size < def_init_size) {
658 		int buffs_to_add = def_init_size - priv->bpool_init_size;
659 
660 		priv->bpool_init_size += buffs_to_add;
661 		ret = mrvl_fill_bpool(dev->data->rx_queues[0], buffs_to_add);
662 		if (ret)
663 			MRVL_LOG(ERR, "Failed to add buffers to bpool");
664 	}
665 
666 	/*
667 	 * Calculate the maximum bpool size for refill feature as follows:
668 	 * maximum number of descriptors in rx queue multiply by number
669 	 * of rx queues plus minimum bpool size.
670 	 * In case the bpool size will exceed this value, superfluous buffers
671 	 * will be removed
672 	 */
673 	priv->bpool_max_size = (priv->nb_rx_queues * MRVL_PP2_RXD_MAX) +
674 				priv->bpool_min_size;
675 
676 	ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
677 	if (ret) {
678 		MRVL_LOG(ERR, "Failed to init ppio");
679 		return ret;
680 	}
681 
682 	/*
683 	 * In case there are some some stale uc/mc mac addresses flush them
684 	 * here. It cannot be done during mrvl_dev_close() as port information
685 	 * is already gone at that point (due to pp2_ppio_deinit() in
686 	 * mrvl_dev_stop()).
687 	 */
688 	if (!priv->uc_mc_flushed) {
689 		ret = pp2_ppio_flush_mac_addrs(priv->ppio, 1, 1);
690 		if (ret) {
691 			MRVL_LOG(ERR,
692 				"Failed to flush uc/mc filter list");
693 			goto out;
694 		}
695 		priv->uc_mc_flushed = 1;
696 	}
697 
698 	ret = mrvl_mtu_set(dev, dev->data->mtu);
699 	if (ret)
700 		MRVL_LOG(ERR, "Failed to set MTU to %d", dev->data->mtu);
701 
702 	if (!rte_is_zero_ether_addr(&dev->data->mac_addrs[0]))
703 		mrvl_mac_addr_set(dev, &dev->data->mac_addrs[0]);
704 
705 	for (i = 1; i < MRVL_MAC_ADDRS_MAX; i++) {
706 		mac_addr = &dev->data->mac_addrs[i];
707 
708 		/* skip zero address */
709 		if (rte_is_zero_ether_addr(mac_addr))
710 			continue;
711 
712 		mrvl_mac_addr_add(dev, mac_addr, i, 0);
713 	}
714 
715 	if (dev->data->all_multicast == 1)
716 		mrvl_allmulticast_enable(dev);
717 
718 	vfc = &dev->data->vlan_filter_conf;
719 	for (j = 0; j < RTE_DIM(vfc->ids); j++) {
720 		uint64_t vlan;
721 		uint64_t vbit;
722 		uint64_t ids = vfc->ids[j];
723 
724 		if (ids == 0)
725 			continue;
726 
727 		while (ids) {
728 			vlan = 64 * j;
729 			/* count trailing zeroes */
730 			vbit = ~ids & (ids - 1);
731 			/* clear least significant bit set */
732 			ids ^= (ids ^ (ids - 1)) ^ vbit;
733 			for (; vbit; vlan++)
734 				vbit >>= 1;
735 			ret = mrvl_vlan_filter_set(dev, vlan, 1);
736 			if (ret) {
737 				MRVL_LOG(ERR, "Failed to setup VLAN filter\n");
738 				goto out;
739 			}
740 		}
741 	}
742 
743 	/* For default QoS config, don't start classifier. */
744 	if (mrvl_qos_cfg  &&
745 	    mrvl_qos_cfg->port[dev->data->port_id].use_global_defaults == 0) {
746 		ret = mrvl_start_qos_mapping(priv);
747 		if (ret) {
748 			MRVL_LOG(ERR, "Failed to setup QoS mapping");
749 			goto out;
750 		}
751 	}
752 
753 	ret = pp2_ppio_set_loopback(priv->ppio, dev->data->dev_conf.lpbk_mode);
754 	if (ret) {
755 		MRVL_LOG(ERR, "Failed to set loopback");
756 		goto out;
757 	}
758 
759 	if (dev->data->promiscuous == 1)
760 		mrvl_promiscuous_enable(dev);
761 
762 	if (dev->data->dev_link.link_status == ETH_LINK_UP) {
763 		ret = mrvl_dev_set_link_up(dev);
764 		if (ret) {
765 			MRVL_LOG(ERR, "Failed to set link up");
766 			dev->data->dev_link.link_status = ETH_LINK_DOWN;
767 			goto out;
768 		}
769 	}
770 
771 	/* start tx queues */
772 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
773 		struct mrvl_txq *txq = dev->data->tx_queues[i];
774 
775 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
776 
777 		if (!txq->tx_deferred_start)
778 			continue;
779 
780 		/*
781 		 * All txqs are started by default. Stop them
782 		 * so that tx_deferred_start works as expected.
783 		 */
784 		ret = mrvl_tx_queue_stop(dev, i);
785 		if (ret)
786 			goto out;
787 	}
788 
789 	mrvl_flow_init(dev);
790 	mrvl_mtr_init(dev);
791 	mrvl_set_tx_function(dev);
792 
793 	return 0;
794 out:
795 	MRVL_LOG(ERR, "Failed to start device");
796 	pp2_ppio_deinit(priv->ppio);
797 	return ret;
798 }
799 
800 /**
801  * Flush receive queues.
802  *
803  * @param dev
804  *   Pointer to Ethernet device structure.
805  */
806 static void
807 mrvl_flush_rx_queues(struct rte_eth_dev *dev)
808 {
809 	int i;
810 
811 	MRVL_LOG(INFO, "Flushing rx queues");
812 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
813 		int ret, num;
814 
815 		do {
816 			struct mrvl_rxq *q = dev->data->rx_queues[i];
817 			struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];
818 
819 			num = MRVL_PP2_RXD_MAX;
820 			ret = pp2_ppio_recv(q->priv->ppio,
821 					    q->priv->rxq_map[q->queue_id].tc,
822 					    q->priv->rxq_map[q->queue_id].inq,
823 					    descs, (uint16_t *)&num);
824 		} while (ret == 0 && num);
825 	}
826 }
827 
828 /**
829  * Flush transmit shadow queues.
830  *
831  * @param dev
832  *   Pointer to Ethernet device structure.
833  */
834 static void
835 mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
836 {
837 	int i, j;
838 	struct mrvl_txq *txq;
839 
840 	MRVL_LOG(INFO, "Flushing tx shadow queues");
841 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
842 		txq = (struct mrvl_txq *)dev->data->tx_queues[i];
843 
844 		for (j = 0; j < RTE_MAX_LCORE; j++) {
845 			struct mrvl_shadow_txq *sq;
846 
847 			if (!hifs[j])
848 				continue;
849 
850 			sq = &txq->shadow_txqs[j];
851 			mrvl_free_sent_buffers(txq->priv->ppio,
852 				hifs[j], j, sq, txq->queue_id, 1);
853 			while (sq->tail != sq->head) {
854 				uint64_t addr = cookie_addr_high |
855 					sq->ent[sq->tail].buff.cookie;
856 				rte_pktmbuf_free(
857 					(struct rte_mbuf *)addr);
858 				sq->tail = (sq->tail + 1) &
859 					    MRVL_PP2_TX_SHADOWQ_MASK;
860 			}
861 			memset(sq, 0, sizeof(*sq));
862 		}
863 	}
864 }
865 
866 /**
867  * Flush hardware bpool (buffer-pool).
868  *
869  * @param dev
870  *   Pointer to Ethernet device structure.
871  */
872 static void
873 mrvl_flush_bpool(struct rte_eth_dev *dev)
874 {
875 	struct mrvl_priv *priv = dev->data->dev_private;
876 	struct pp2_hif *hif;
877 	uint32_t num;
878 	int ret;
879 	unsigned int core_id = rte_lcore_id();
880 
881 	if (core_id == LCORE_ID_ANY)
882 		core_id = rte_get_main_lcore();
883 
884 	hif = mrvl_get_hif(priv, core_id);
885 
886 	ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
887 	if (ret) {
888 		MRVL_LOG(ERR, "Failed to get bpool buffers number");
889 		return;
890 	}
891 
892 	while (num--) {
893 		struct pp2_buff_inf inf;
894 		uint64_t addr;
895 
896 		ret = pp2_bpool_get_buff(hif, priv->bpool, &inf);
897 		if (ret)
898 			break;
899 
900 		addr = cookie_addr_high | inf.cookie;
901 		rte_pktmbuf_free((struct rte_mbuf *)addr);
902 	}
903 }
904 
905 /**
906  * DPDK callback to stop the device.
907  *
908  * @param dev
909  *   Pointer to Ethernet device structure.
910  */
911 static int
912 mrvl_dev_stop(struct rte_eth_dev *dev)
913 {
914 	return mrvl_dev_set_link_down(dev);
915 }
916 
917 /**
918  * DPDK callback to close the device.
919  *
920  * @param dev
921  *   Pointer to Ethernet device structure.
922  */
923 static int
924 mrvl_dev_close(struct rte_eth_dev *dev)
925 {
926 	struct mrvl_priv *priv = dev->data->dev_private;
927 	size_t i;
928 
929 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
930 		return 0;
931 
932 	mrvl_flush_rx_queues(dev);
933 	mrvl_flush_tx_shadow_queues(dev);
934 	mrvl_flow_deinit(dev);
935 	mrvl_mtr_deinit(dev);
936 
937 	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
938 		struct pp2_ppio_tc_params *tc_params =
939 			&priv->ppio_params.inqs_params.tcs_params[i];
940 
941 		if (tc_params->inqs_params) {
942 			rte_free(tc_params->inqs_params);
943 			tc_params->inqs_params = NULL;
944 		}
945 	}
946 
947 	if (priv->cls_tbl) {
948 		pp2_cls_tbl_deinit(priv->cls_tbl);
949 		priv->cls_tbl = NULL;
950 	}
951 
952 	if (priv->qos_tbl) {
953 		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
954 		priv->qos_tbl = NULL;
955 	}
956 
957 	mrvl_flush_bpool(dev);
958 	mrvl_tm_deinit(dev);
959 
960 	if (priv->ppio) {
961 		pp2_ppio_deinit(priv->ppio);
962 		priv->ppio = NULL;
963 	}
964 
965 	/* policer must be released after ppio deinitialization */
966 	if (priv->default_policer) {
967 		pp2_cls_plcr_deinit(priv->default_policer);
968 		priv->default_policer = NULL;
969 	}
970 
971 
972 	if (priv->bpool) {
973 		pp2_bpool_deinit(priv->bpool);
974 		used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
975 		priv->bpool = NULL;
976 	}
977 
978 	mrvl_dev_num--;
979 
980 	if (mrvl_dev_num == 0) {
981 		MRVL_LOG(INFO, "Perform MUSDK deinit");
982 		mrvl_deinit_hifs();
983 		mrvl_deinit_pp2();
984 		rte_mvep_deinit(MVEP_MOD_T_PP2);
985 	}
986 
987 	return 0;
988 }
989 
990 /**
991  * DPDK callback to retrieve physical link information.
992  *
993  * @param dev
994  *   Pointer to Ethernet device structure.
995  * @param wait_to_complete
996  *   Wait for request completion (ignored).
997  *
998  * @return
999  *   0 on success, negative error value otherwise.
1000  */
1001 static int
1002 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1003 {
1004 	/*
1005 	 * TODO
1006 	 * once MUSDK provides necessary API use it here
1007 	 */
1008 	struct mrvl_priv *priv = dev->data->dev_private;
1009 	struct ethtool_cmd edata;
1010 	struct ifreq req;
1011 	int ret, fd, link_up;
1012 
1013 	if (!priv->ppio)
1014 		return -EPERM;
1015 
1016 	edata.cmd = ETHTOOL_GSET;
1017 
1018 	strcpy(req.ifr_name, dev->data->name);
1019 	req.ifr_data = (void *)&edata;
1020 
1021 	fd = socket(AF_INET, SOCK_DGRAM, 0);
1022 	if (fd == -1)
1023 		return -EFAULT;
1024 
1025 	ret = ioctl(fd, SIOCETHTOOL, &req);
1026 	if (ret == -1) {
1027 		close(fd);
1028 		return -EFAULT;
1029 	}
1030 
1031 	close(fd);
1032 
1033 	switch (ethtool_cmd_speed(&edata)) {
1034 	case SPEED_10:
1035 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
1036 		break;
1037 	case SPEED_100:
1038 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
1039 		break;
1040 	case SPEED_1000:
1041 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
1042 		break;
1043 	case SPEED_10000:
1044 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
1045 		break;
1046 	default:
1047 		dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
1048 	}
1049 
1050 	dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
1051 							 ETH_LINK_HALF_DUPLEX;
1052 	dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
1053 							   ETH_LINK_FIXED;
1054 	pp2_ppio_get_link_state(priv->ppio, &link_up);
1055 	dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
1056 
1057 	return 0;
1058 }
1059 
1060 /**
1061  * DPDK callback to enable promiscuous mode.
1062  *
1063  * @param dev
1064  *   Pointer to Ethernet device structure.
1065  *
1066  * @return
1067  *   0 on success, negative error value otherwise.
1068  */
1069 static int
1070 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
1071 {
1072 	struct mrvl_priv *priv = dev->data->dev_private;
1073 	int ret;
1074 
1075 	if (priv->isolated)
1076 		return -ENOTSUP;
1077 
1078 	if (!priv->ppio)
1079 		return 0;
1080 
1081 	ret = pp2_ppio_set_promisc(priv->ppio, 1);
1082 	if (ret) {
1083 		MRVL_LOG(ERR, "Failed to enable promiscuous mode");
1084 		return -EAGAIN;
1085 	}
1086 
1087 	return 0;
1088 }
1089 
1090 /**
1091  * DPDK callback to enable allmulti mode.
1092  *
1093  * @param dev
1094  *   Pointer to Ethernet device structure.
1095  *
1096  * @return
1097  *   0 on success, negative error value otherwise.
1098  */
1099 static int
1100 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
1101 {
1102 	struct mrvl_priv *priv = dev->data->dev_private;
1103 	int ret;
1104 
1105 	if (priv->isolated)
1106 		return -ENOTSUP;
1107 
1108 	if (!priv->ppio)
1109 		return 0;
1110 
1111 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
1112 	if (ret) {
1113 		MRVL_LOG(ERR, "Failed enable all-multicast mode");
1114 		return -EAGAIN;
1115 	}
1116 
1117 	return 0;
1118 }
1119 
1120 /**
1121  * DPDK callback to disable promiscuous mode.
1122  *
1123  * @param dev
1124  *   Pointer to Ethernet device structure.
1125  *
1126  * @return
1127  *   0 on success, negative error value otherwise.
1128  */
1129 static int
1130 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
1131 {
1132 	struct mrvl_priv *priv = dev->data->dev_private;
1133 	int ret;
1134 
1135 	if (priv->isolated)
1136 		return -ENOTSUP;
1137 
1138 	if (!priv->ppio)
1139 		return 0;
1140 
1141 	ret = pp2_ppio_set_promisc(priv->ppio, 0);
1142 	if (ret) {
1143 		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
1144 		return -EAGAIN;
1145 	}
1146 
1147 	return 0;
1148 }
1149 
1150 /**
1151  * DPDK callback to disable allmulticast mode.
1152  *
1153  * @param dev
1154  *   Pointer to Ethernet device structure.
1155  *
1156  * @return
1157  *   0 on success, negative error value otherwise.
1158  */
1159 static int
1160 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
1161 {
1162 	struct mrvl_priv *priv = dev->data->dev_private;
1163 	int ret;
1164 
1165 	if (priv->isolated)
1166 		return -ENOTSUP;
1167 
1168 	if (!priv->ppio)
1169 		return 0;
1170 
1171 	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
1172 	if (ret) {
1173 		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
1174 		return -EAGAIN;
1175 	}
1176 
1177 	return 0;
1178 }
1179 
1180 /**
1181  * DPDK callback to remove a MAC address.
1182  *
1183  * @param dev
1184  *   Pointer to Ethernet device structure.
1185  * @param index
1186  *   MAC address index.
1187  */
1188 static void
1189 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1190 {
1191 	struct mrvl_priv *priv = dev->data->dev_private;
1192 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1193 	int ret;
1194 
1195 	if (priv->isolated)
1196 		return;
1197 
1198 	if (!priv->ppio)
1199 		return;
1200 
1201 	ret = pp2_ppio_remove_mac_addr(priv->ppio,
1202 				       dev->data->mac_addrs[index].addr_bytes);
1203 	if (ret) {
1204 		rte_ether_format_addr(buf, sizeof(buf),
1205 				  &dev->data->mac_addrs[index]);
1206 		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
1207 	}
1208 }
1209 
1210 /**
1211  * DPDK callback to add a MAC address.
1212  *
1213  * @param dev
1214  *   Pointer to Ethernet device structure.
1215  * @param mac_addr
1216  *   MAC address to register.
1217  * @param index
1218  *   MAC address index.
1219  * @param vmdq
1220  *   VMDq pool index to associate address with (unused).
1221  *
1222  * @return
1223  *   0 on success, negative error value otherwise.
1224  */
1225 static int
1226 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1227 		  uint32_t index, uint32_t vmdq __rte_unused)
1228 {
1229 	struct mrvl_priv *priv = dev->data->dev_private;
1230 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1231 	int ret;
1232 
1233 	if (priv->isolated)
1234 		return -ENOTSUP;
1235 
1236 	if (!priv->ppio)
1237 		return 0;
1238 
1239 	if (index == 0)
1240 		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
1241 		return -1;
1242 
1243 	/*
1244 	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1245 	 * parameter uc_filter_max. Maximum number of mc addresses is then
1246 	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1247 	 * 21 respectively.
1248 	 *
1249 	 * If more than uc_filter_max uc addresses were added to filter list
1250 	 * then NIC will switch to promiscuous mode automatically.
1251 	 *
1252 	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1253 	 * were added to filter list then NIC will switch to all-multicast mode
1254 	 * automatically.
1255 	 */
1256 	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1257 	if (ret) {
1258 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1259 		MRVL_LOG(ERR, "Failed to add mac %s", buf);
1260 		return -1;
1261 	}
1262 
1263 	return 0;
1264 }
1265 
1266 /**
1267  * DPDK callback to set the primary MAC address.
1268  *
1269  * @param dev
1270  *   Pointer to Ethernet device structure.
1271  * @param mac_addr
1272  *   MAC address to register.
1273  *
1274  * @return
1275  *   0 on success, negative error value otherwise.
1276  */
1277 static int
1278 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1279 {
1280 	struct mrvl_priv *priv = dev->data->dev_private;
1281 	int ret;
1282 
1283 	if (priv->isolated)
1284 		return -ENOTSUP;
1285 
1286 	if (!priv->ppio)
1287 		return 0;
1288 
1289 	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1290 	if (ret) {
1291 		char buf[RTE_ETHER_ADDR_FMT_SIZE];
1292 		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1293 		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1294 	}
1295 
1296 	return ret;
1297 }
1298 
1299 /**
1300  * DPDK callback to get device statistics.
1301  *
1302  * @param dev
1303  *   Pointer to Ethernet device structure.
1304  * @param stats
1305  *   Stats structure output buffer.
1306  *
1307  * @return
1308  *   0 on success, negative error value otherwise.
1309  */
1310 static int
1311 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1312 {
1313 	struct mrvl_priv *priv = dev->data->dev_private;
1314 	struct pp2_ppio_statistics ppio_stats;
1315 	uint64_t drop_mac = 0;
1316 	unsigned int i, idx, ret;
1317 
1318 	if (!priv->ppio)
1319 		return -EPERM;
1320 
1321 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1322 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1323 		struct pp2_ppio_inq_statistics rx_stats;
1324 
1325 		if (!rxq)
1326 			continue;
1327 
1328 		idx = rxq->queue_id;
1329 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1330 			MRVL_LOG(ERR,
1331 				"rx queue %d stats out of range (0 - %d)",
1332 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1333 			continue;
1334 		}
1335 
1336 		ret = pp2_ppio_inq_get_statistics(priv->ppio,
1337 						  priv->rxq_map[idx].tc,
1338 						  priv->rxq_map[idx].inq,
1339 						  &rx_stats, 0);
1340 		if (unlikely(ret)) {
1341 			MRVL_LOG(ERR,
1342 				"Failed to update rx queue %d stats", idx);
1343 			break;
1344 		}
1345 
1346 		stats->q_ibytes[idx] = rxq->bytes_recv;
1347 		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1348 		stats->q_errors[idx] = rx_stats.drop_early +
1349 				       rx_stats.drop_fullq +
1350 				       rx_stats.drop_bm +
1351 				       rxq->drop_mac;
1352 		stats->ibytes += rxq->bytes_recv;
1353 		drop_mac += rxq->drop_mac;
1354 	}
1355 
1356 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1357 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1358 		struct pp2_ppio_outq_statistics tx_stats;
1359 
1360 		if (!txq)
1361 			continue;
1362 
1363 		idx = txq->queue_id;
1364 		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1365 			MRVL_LOG(ERR,
1366 				"tx queue %d stats out of range (0 - %d)",
1367 				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1368 		}
1369 
1370 		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1371 						   &tx_stats, 0);
1372 		if (unlikely(ret)) {
1373 			MRVL_LOG(ERR,
1374 				"Failed to update tx queue %d stats", idx);
1375 			break;
1376 		}
1377 
1378 		stats->q_opackets[idx] = tx_stats.deq_desc;
1379 		stats->q_obytes[idx] = txq->bytes_sent;
1380 		stats->obytes += txq->bytes_sent;
1381 	}
1382 
1383 	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1384 	if (unlikely(ret)) {
1385 		MRVL_LOG(ERR, "Failed to update port statistics");
1386 		return ret;
1387 	}
1388 
1389 	stats->ipackets += ppio_stats.rx_packets - drop_mac;
1390 	stats->opackets += ppio_stats.tx_packets;
1391 	stats->imissed += ppio_stats.rx_fullq_dropped +
1392 			  ppio_stats.rx_bm_dropped +
1393 			  ppio_stats.rx_early_dropped +
1394 			  ppio_stats.rx_fifo_dropped +
1395 			  ppio_stats.rx_cls_dropped;
1396 	stats->ierrors = drop_mac;
1397 
1398 	return 0;
1399 }
1400 
1401 /**
1402  * DPDK callback to clear device statistics.
1403  *
1404  * @param dev
1405  *   Pointer to Ethernet device structure.
1406  *
1407  * @return
1408  *   0 on success, negative error value otherwise.
1409  */
1410 static int
1411 mrvl_stats_reset(struct rte_eth_dev *dev)
1412 {
1413 	struct mrvl_priv *priv = dev->data->dev_private;
1414 	int i;
1415 
1416 	if (!priv->ppio)
1417 		return 0;
1418 
1419 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1420 		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1421 
1422 		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1423 					    priv->rxq_map[i].inq, NULL, 1);
1424 		rxq->bytes_recv = 0;
1425 		rxq->drop_mac = 0;
1426 	}
1427 
1428 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1429 		struct mrvl_txq *txq = dev->data->tx_queues[i];
1430 
1431 		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1432 		txq->bytes_sent = 0;
1433 	}
1434 
1435 	return pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1436 }
1437 
1438 /**
1439  * DPDK callback to get extended statistics.
1440  *
1441  * @param dev
1442  *   Pointer to Ethernet device structure.
1443  * @param stats
1444  *   Pointer to xstats table.
1445  * @param n
1446  *   Number of entries in xstats table.
1447  * @return
1448  *   Negative value on error, number of read xstats otherwise.
1449  */
1450 static int
1451 mrvl_xstats_get(struct rte_eth_dev *dev,
1452 		struct rte_eth_xstat *stats, unsigned int n)
1453 {
1454 	struct mrvl_priv *priv = dev->data->dev_private;
1455 	struct pp2_ppio_statistics ppio_stats;
1456 	unsigned int i;
1457 
1458 	if (!stats)
1459 		return 0;
1460 
1461 	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1462 	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1463 		uint64_t val;
1464 
1465 		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1466 			val = *(uint32_t *)((uint8_t *)&ppio_stats +
1467 					    mrvl_xstats_tbl[i].offset);
1468 		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1469 			val = *(uint64_t *)((uint8_t *)&ppio_stats +
1470 					    mrvl_xstats_tbl[i].offset);
1471 		else
1472 			return -EINVAL;
1473 
1474 		stats[i].id = i;
1475 		stats[i].value = val;
1476 	}
1477 
1478 	return n;
1479 }
1480 
1481 /**
1482  * DPDK callback to reset extended statistics.
1483  *
1484  * @param dev
1485  *   Pointer to Ethernet device structure.
1486  *
1487  * @return
1488  *   0 on success, negative error value otherwise.
1489  */
1490 static int
1491 mrvl_xstats_reset(struct rte_eth_dev *dev)
1492 {
1493 	return mrvl_stats_reset(dev);
1494 }
1495 
1496 /**
1497  * DPDK callback to get extended statistics names.
1498  *
1499  * @param dev (unused)
1500  *   Pointer to Ethernet device structure.
1501  * @param xstats_names
1502  *   Pointer to xstats names table.
1503  * @param size
1504  *   Size of the xstats names table.
1505  * @return
1506  *   Number of read names.
1507  */
1508 static int
1509 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1510 		      struct rte_eth_xstat_name *xstats_names,
1511 		      unsigned int size)
1512 {
1513 	unsigned int i;
1514 
1515 	if (!xstats_names)
1516 		return RTE_DIM(mrvl_xstats_tbl);
1517 
1518 	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1519 		strlcpy(xstats_names[i].name, mrvl_xstats_tbl[i].name,
1520 			RTE_ETH_XSTATS_NAME_SIZE);
1521 
1522 	return size;
1523 }
1524 
1525 /**
1526  * DPDK callback to get information about the device.
1527  *
1528  * @param dev
1529  *   Pointer to Ethernet device structure (unused).
1530  * @param info
1531  *   Info structure output buffer.
1532  */
1533 static int
1534 mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
1535 		   struct rte_eth_dev_info *info)
1536 {
1537 	info->speed_capa = ETH_LINK_SPEED_10M |
1538 			   ETH_LINK_SPEED_100M |
1539 			   ETH_LINK_SPEED_1G |
1540 			   ETH_LINK_SPEED_10G;
1541 
1542 	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1543 	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1544 	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1545 
1546 	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1547 	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1548 	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1549 
1550 	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1551 	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1552 	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1553 
1554 	info->rx_offload_capa = MRVL_RX_OFFLOADS;
1555 	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1556 
1557 	info->tx_offload_capa = MRVL_TX_OFFLOADS;
1558 	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1559 
1560 	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1561 				       ETH_RSS_NONFRAG_IPV4_TCP |
1562 				       ETH_RSS_NONFRAG_IPV4_UDP;
1563 
1564 	/* By default packets are dropped if no descriptors are available */
1565 	info->default_rxconf.rx_drop_en = 1;
1566 
1567 	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1568 
1569 	return 0;
1570 }
1571 
1572 /**
1573  * Return supported packet types.
1574  *
1575  * @param dev
1576  *   Pointer to Ethernet device structure (unused).
1577  *
1578  * @return
1579  *   Const pointer to the table with supported packet types.
1580  */
1581 static const uint32_t *
1582 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1583 {
1584 	static const uint32_t ptypes[] = {
1585 		RTE_PTYPE_L2_ETHER,
1586 		RTE_PTYPE_L2_ETHER_VLAN,
1587 		RTE_PTYPE_L2_ETHER_QINQ,
1588 		RTE_PTYPE_L3_IPV4,
1589 		RTE_PTYPE_L3_IPV4_EXT,
1590 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1591 		RTE_PTYPE_L3_IPV6,
1592 		RTE_PTYPE_L3_IPV6_EXT,
1593 		RTE_PTYPE_L2_ETHER_ARP,
1594 		RTE_PTYPE_L4_TCP,
1595 		RTE_PTYPE_L4_UDP
1596 	};
1597 
1598 	return ptypes;
1599 }
1600 
1601 /**
1602  * DPDK callback to get information about specific receive queue.
1603  *
1604  * @param dev
1605  *   Pointer to Ethernet device structure.
1606  * @param rx_queue_id
1607  *   Receive queue index.
1608  * @param qinfo
1609  *   Receive queue information structure.
1610  */
1611 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1612 			      struct rte_eth_rxq_info *qinfo)
1613 {
1614 	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1615 	struct mrvl_priv *priv = dev->data->dev_private;
1616 	int inq = priv->rxq_map[rx_queue_id].inq;
1617 	int tc = priv->rxq_map[rx_queue_id].tc;
1618 	struct pp2_ppio_tc_params *tc_params =
1619 		&priv->ppio_params.inqs_params.tcs_params[tc];
1620 
1621 	qinfo->mp = q->mp;
1622 	qinfo->nb_desc = tc_params->inqs_params[inq].size;
1623 }
1624 
1625 /**
1626  * DPDK callback to get information about specific transmit queue.
1627  *
1628  * @param dev
1629  *   Pointer to Ethernet device structure.
1630  * @param tx_queue_id
1631  *   Transmit queue index.
1632  * @param qinfo
1633  *   Transmit queue information structure.
1634  */
1635 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1636 			      struct rte_eth_txq_info *qinfo)
1637 {
1638 	struct mrvl_priv *priv = dev->data->dev_private;
1639 	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1640 
1641 	qinfo->nb_desc =
1642 		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1643 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1644 }
1645 
1646 /**
1647  * DPDK callback to Configure a VLAN filter.
1648  *
1649  * @param dev
1650  *   Pointer to Ethernet device structure.
1651  * @param vlan_id
1652  *   VLAN ID to filter.
1653  * @param on
1654  *   Toggle filter.
1655  *
1656  * @return
1657  *   0 on success, negative error value otherwise.
1658  */
1659 static int
1660 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1661 {
1662 	struct mrvl_priv *priv = dev->data->dev_private;
1663 
1664 	if (priv->isolated)
1665 		return -ENOTSUP;
1666 
1667 	if (!priv->ppio)
1668 		return 0;
1669 
1670 	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1671 		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1672 }
1673 
1674 /**
1675  * Release buffers to hardware bpool (buffer-pool)
1676  *
1677  * @param rxq
1678  *   Receive queue pointer.
1679  * @param num
1680  *   Number of buffers to release to bpool.
1681  *
1682  * @return
1683  *   0 on success, negative error value otherwise.
1684  */
1685 static int
1686 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1687 {
1688 	struct buff_release_entry entries[num];
1689 	struct rte_mbuf *mbufs[num];
1690 	int i, ret;
1691 	unsigned int core_id;
1692 	struct pp2_hif *hif;
1693 	struct pp2_bpool *bpool;
1694 
1695 	core_id = rte_lcore_id();
1696 	if (core_id == LCORE_ID_ANY)
1697 		core_id = rte_get_main_lcore();
1698 
1699 	hif = mrvl_get_hif(rxq->priv, core_id);
1700 	if (!hif)
1701 		return -1;
1702 
1703 	bpool = rxq->priv->bpool;
1704 
1705 	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1706 	if (ret)
1707 		return ret;
1708 
1709 	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1710 		cookie_addr_high =
1711 			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1712 
1713 	for (i = 0; i < num; i++) {
1714 		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1715 			!= cookie_addr_high) {
1716 			MRVL_LOG(ERR,
1717 				"mbuf virtual addr high is out of range "
1718 				"0x%x instead of 0x%x\n",
1719 				(uint32_t)((uint64_t)mbufs[i] >> 32),
1720 				(uint32_t)(cookie_addr_high >> 32));
1721 			goto out;
1722 		}
1723 
1724 		entries[i].buff.addr =
1725 			rte_mbuf_data_iova_default(mbufs[i]);
1726 		entries[i].buff.cookie = (uintptr_t)mbufs[i];
1727 		entries[i].bpool = bpool;
1728 	}
1729 
1730 	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1731 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1732 
1733 	if (i != num)
1734 		goto out;
1735 
1736 	return 0;
1737 out:
1738 	for (; i < num; i++)
1739 		rte_pktmbuf_free(mbufs[i]);
1740 
1741 	return -1;
1742 }
1743 
1744 /**
1745  * DPDK callback to configure the receive queue.
1746  *
1747  * @param dev
1748  *   Pointer to Ethernet device structure.
1749  * @param idx
1750  *   RX queue index.
1751  * @param desc
1752  *   Number of descriptors to configure in queue.
1753  * @param socket
1754  *   NUMA socket on which memory must be allocated.
1755  * @param conf
1756  *   Thresholds parameters.
1757  * @param mp
1758  *   Memory pool for buffer allocations.
1759  *
1760  * @return
1761  *   0 on success, negative error value otherwise.
1762  */
1763 static int
1764 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1765 		    unsigned int socket,
1766 		    const struct rte_eth_rxconf *conf,
1767 		    struct rte_mempool *mp)
1768 {
1769 	struct mrvl_priv *priv = dev->data->dev_private;
1770 	struct mrvl_rxq *rxq;
1771 	uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
1772 	uint32_t max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1773 	int ret, tc, inq;
1774 	uint64_t offloads;
1775 
1776 	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
1777 
1778 	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1779 		/*
1780 		 * Unknown TC mapping, mapping will not have a correct queue.
1781 		 */
1782 		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
1783 			idx, priv->ppio_id);
1784 		return -EFAULT;
1785 	}
1786 
1787 	frame_size = buf_size - RTE_PKTMBUF_HEADROOM -
1788 		     MRVL_PKT_EFFEC_OFFS + RTE_ETHER_CRC_LEN;
1789 	if (frame_size < max_rx_pkt_len) {
1790 		MRVL_LOG(WARNING,
1791 			"Mbuf size must be increased to %u bytes to hold up "
1792 			"to %u bytes of data.",
1793 			buf_size + max_rx_pkt_len - frame_size,
1794 			max_rx_pkt_len);
1795 		dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
1796 		MRVL_LOG(INFO, "Setting max rx pkt len to %u",
1797 			dev->data->dev_conf.rxmode.max_rx_pkt_len);
1798 	}
1799 
1800 	if (dev->data->rx_queues[idx]) {
1801 		rte_free(dev->data->rx_queues[idx]);
1802 		dev->data->rx_queues[idx] = NULL;
1803 	}
1804 
1805 	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
1806 	if (!rxq)
1807 		return -ENOMEM;
1808 
1809 	rxq->priv = priv;
1810 	rxq->mp = mp;
1811 	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
1812 	rxq->queue_id = idx;
1813 	rxq->port_id = dev->data->port_id;
1814 	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
1815 
1816 	tc = priv->rxq_map[rxq->queue_id].tc,
1817 	inq = priv->rxq_map[rxq->queue_id].inq;
1818 	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
1819 		desc;
1820 
1821 	ret = mrvl_fill_bpool(rxq, desc);
1822 	if (ret) {
1823 		rte_free(rxq);
1824 		return ret;
1825 	}
1826 
1827 	priv->bpool_init_size += desc;
1828 
1829 	dev->data->rx_queues[idx] = rxq;
1830 
1831 	return 0;
1832 }
1833 
1834 /**
1835  * DPDK callback to release the receive queue.
1836  *
1837  * @param rxq
1838  *   Generic receive queue pointer.
1839  */
1840 static void
1841 mrvl_rx_queue_release(void *rxq)
1842 {
1843 	struct mrvl_rxq *q = rxq;
1844 	struct pp2_ppio_tc_params *tc_params;
1845 	int i, num, tc, inq;
1846 	struct pp2_hif *hif;
1847 	unsigned int core_id = rte_lcore_id();
1848 
1849 	if (core_id == LCORE_ID_ANY)
1850 		core_id = rte_get_main_lcore();
1851 
1852 	if (!q)
1853 		return;
1854 
1855 	hif = mrvl_get_hif(q->priv, core_id);
1856 
1857 	if (!hif)
1858 		return;
1859 
1860 	tc = q->priv->rxq_map[q->queue_id].tc;
1861 	inq = q->priv->rxq_map[q->queue_id].inq;
1862 	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
1863 	num = tc_params->inqs_params[inq].size;
1864 	for (i = 0; i < num; i++) {
1865 		struct pp2_buff_inf inf;
1866 		uint64_t addr;
1867 
1868 		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
1869 		addr = cookie_addr_high | inf.cookie;
1870 		rte_pktmbuf_free((struct rte_mbuf *)addr);
1871 	}
1872 
1873 	rte_free(q);
1874 }
1875 
1876 /**
1877  * DPDK callback to configure the transmit queue.
1878  *
1879  * @param dev
1880  *   Pointer to Ethernet device structure.
1881  * @param idx
1882  *   Transmit queue index.
1883  * @param desc
1884  *   Number of descriptors to configure in the queue.
1885  * @param socket
1886  *   NUMA socket on which memory must be allocated.
1887  * @param conf
1888  *   Tx queue configuration parameters.
1889  *
1890  * @return
1891  *   0 on success, negative error value otherwise.
1892  */
1893 static int
1894 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1895 		    unsigned int socket,
1896 		    const struct rte_eth_txconf *conf)
1897 {
1898 	struct mrvl_priv *priv = dev->data->dev_private;
1899 	struct mrvl_txq *txq;
1900 
1901 	if (dev->data->tx_queues[idx]) {
1902 		rte_free(dev->data->tx_queues[idx]);
1903 		dev->data->tx_queues[idx] = NULL;
1904 	}
1905 
1906 	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
1907 	if (!txq)
1908 		return -ENOMEM;
1909 
1910 	txq->priv = priv;
1911 	txq->queue_id = idx;
1912 	txq->port_id = dev->data->port_id;
1913 	txq->tx_deferred_start = conf->tx_deferred_start;
1914 	dev->data->tx_queues[idx] = txq;
1915 
1916 	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
1917 
1918 	return 0;
1919 }
1920 
1921 /**
1922  * DPDK callback to release the transmit queue.
1923  *
1924  * @param txq
1925  *   Generic transmit queue pointer.
1926  */
1927 static void
1928 mrvl_tx_queue_release(void *txq)
1929 {
1930 	struct mrvl_txq *q = txq;
1931 
1932 	if (!q)
1933 		return;
1934 
1935 	rte_free(q);
1936 }
1937 
1938 /**
1939  * DPDK callback to get flow control configuration.
1940  *
1941  * @param dev
1942  *  Pointer to Ethernet device structure.
1943  * @param fc_conf
1944  *  Pointer to the flow control configuration.
1945  *
1946  * @return
1947  *  0 on success, negative error value otherwise.
1948  */
1949 static int
1950 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1951 {
1952 	struct mrvl_priv *priv = dev->data->dev_private;
1953 	int ret, en;
1954 
1955 	if (!priv)
1956 		return -EPERM;
1957 
1958 	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
1959 	if (ret) {
1960 		MRVL_LOG(ERR, "Failed to read rx pause state");
1961 		return ret;
1962 	}
1963 
1964 	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
1965 
1966 	return 0;
1967 }
1968 
1969 /**
1970  * DPDK callback to set flow control configuration.
1971  *
1972  * @param dev
1973  *  Pointer to Ethernet device structure.
1974  * @param fc_conf
1975  *  Pointer to the flow control configuration.
1976  *
1977  * @return
1978  *  0 on success, negative error value otherwise.
1979  */
1980 static int
1981 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1982 {
1983 	struct mrvl_priv *priv = dev->data->dev_private;
1984 
1985 	if (!priv)
1986 		return -EPERM;
1987 
1988 	if (fc_conf->high_water ||
1989 	    fc_conf->low_water ||
1990 	    fc_conf->pause_time ||
1991 	    fc_conf->mac_ctrl_frame_fwd ||
1992 	    fc_conf->autoneg) {
1993 		MRVL_LOG(ERR, "Flowctrl parameter is not supported");
1994 
1995 		return -EINVAL;
1996 	}
1997 
1998 	if (fc_conf->mode == RTE_FC_NONE ||
1999 	    fc_conf->mode == RTE_FC_RX_PAUSE) {
2000 		int ret, en;
2001 
2002 		en = fc_conf->mode == RTE_FC_NONE ? 0 : 1;
2003 		ret = pp2_ppio_set_rx_pause(priv->ppio, en);
2004 		if (ret)
2005 			MRVL_LOG(ERR,
2006 				"Failed to change flowctrl on RX side");
2007 
2008 		return ret;
2009 	}
2010 
2011 	return 0;
2012 }
2013 
2014 /**
2015  * Update RSS hash configuration
2016  *
2017  * @param dev
2018  *   Pointer to Ethernet device structure.
2019  * @param rss_conf
2020  *   Pointer to RSS configuration.
2021  *
2022  * @return
2023  *   0 on success, negative error value otherwise.
2024  */
2025 static int
2026 mrvl_rss_hash_update(struct rte_eth_dev *dev,
2027 		     struct rte_eth_rss_conf *rss_conf)
2028 {
2029 	struct mrvl_priv *priv = dev->data->dev_private;
2030 
2031 	if (priv->isolated)
2032 		return -ENOTSUP;
2033 
2034 	return mrvl_configure_rss(priv, rss_conf);
2035 }
2036 
2037 /**
2038  * DPDK callback to get RSS hash configuration.
2039  *
2040  * @param dev
2041  *   Pointer to Ethernet device structure.
2042  * @rss_conf
2043  *   Pointer to RSS configuration.
2044  *
2045  * @return
2046  *   Always 0.
2047  */
2048 static int
2049 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
2050 		       struct rte_eth_rss_conf *rss_conf)
2051 {
2052 	struct mrvl_priv *priv = dev->data->dev_private;
2053 	enum pp2_ppio_hash_type hash_type =
2054 		priv->ppio_params.inqs_params.hash_type;
2055 
2056 	rss_conf->rss_key = NULL;
2057 
2058 	if (hash_type == PP2_PPIO_HASH_T_NONE)
2059 		rss_conf->rss_hf = 0;
2060 	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
2061 		rss_conf->rss_hf = ETH_RSS_IPV4;
2062 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
2063 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
2064 	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
2065 		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
2066 
2067 	return 0;
2068 }
2069 
2070 /**
2071  * DPDK callback to get rte_flow callbacks.
2072  *
2073  * @param dev
2074  *   Pointer to the device structure.
2075  * @param filer_type
2076  *   Flow filter type.
2077  * @param filter_op
2078  *   Flow filter operation.
2079  * @param arg
2080  *   Pointer to pass the flow ops.
2081  *
2082  * @return
2083  *   0 on success, negative error value otherwise.
2084  */
2085 static int
2086 mrvl_eth_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
2087 		     enum rte_filter_type filter_type,
2088 		     enum rte_filter_op filter_op, void *arg)
2089 {
2090 	switch (filter_type) {
2091 	case RTE_ETH_FILTER_GENERIC:
2092 		if (filter_op != RTE_ETH_FILTER_GET)
2093 			return -EINVAL;
2094 		*(const void **)arg = &mrvl_flow_ops;
2095 		return 0;
2096 	default:
2097 		MRVL_LOG(WARNING, "Filter type (%d) not supported",
2098 				filter_type);
2099 		return -EINVAL;
2100 	}
2101 }
2102 
2103 /**
2104  * DPDK callback to get rte_mtr callbacks.
2105  *
2106  * @param dev
2107  *   Pointer to the device structure.
2108  * @param ops
2109  *   Pointer to pass the mtr ops.
2110  *
2111  * @return
2112  *   Always 0.
2113  */
2114 static int
2115 mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2116 {
2117 	*(const void **)ops = &mrvl_mtr_ops;
2118 
2119 	return 0;
2120 }
2121 
2122 /**
2123  * DPDK callback to get rte_tm callbacks.
2124  *
2125  * @param dev
2126  *   Pointer to the device structure.
2127  * @param ops
2128  *   Pointer to pass the tm ops.
2129  *
2130  * @return
2131  *   Always 0.
2132  */
2133 static int
2134 mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2135 {
2136 	*(const void **)ops = &mrvl_tm_ops;
2137 
2138 	return 0;
2139 }
2140 
2141 static const struct eth_dev_ops mrvl_ops = {
2142 	.dev_configure = mrvl_dev_configure,
2143 	.dev_start = mrvl_dev_start,
2144 	.dev_stop = mrvl_dev_stop,
2145 	.dev_set_link_up = mrvl_dev_set_link_up,
2146 	.dev_set_link_down = mrvl_dev_set_link_down,
2147 	.dev_close = mrvl_dev_close,
2148 	.link_update = mrvl_link_update,
2149 	.promiscuous_enable = mrvl_promiscuous_enable,
2150 	.allmulticast_enable = mrvl_allmulticast_enable,
2151 	.promiscuous_disable = mrvl_promiscuous_disable,
2152 	.allmulticast_disable = mrvl_allmulticast_disable,
2153 	.mac_addr_remove = mrvl_mac_addr_remove,
2154 	.mac_addr_add = mrvl_mac_addr_add,
2155 	.mac_addr_set = mrvl_mac_addr_set,
2156 	.mtu_set = mrvl_mtu_set,
2157 	.stats_get = mrvl_stats_get,
2158 	.stats_reset = mrvl_stats_reset,
2159 	.xstats_get = mrvl_xstats_get,
2160 	.xstats_reset = mrvl_xstats_reset,
2161 	.xstats_get_names = mrvl_xstats_get_names,
2162 	.dev_infos_get = mrvl_dev_infos_get,
2163 	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
2164 	.rxq_info_get = mrvl_rxq_info_get,
2165 	.txq_info_get = mrvl_txq_info_get,
2166 	.vlan_filter_set = mrvl_vlan_filter_set,
2167 	.tx_queue_start = mrvl_tx_queue_start,
2168 	.tx_queue_stop = mrvl_tx_queue_stop,
2169 	.rx_queue_setup = mrvl_rx_queue_setup,
2170 	.rx_queue_release = mrvl_rx_queue_release,
2171 	.tx_queue_setup = mrvl_tx_queue_setup,
2172 	.tx_queue_release = mrvl_tx_queue_release,
2173 	.flow_ctrl_get = mrvl_flow_ctrl_get,
2174 	.flow_ctrl_set = mrvl_flow_ctrl_set,
2175 	.rss_hash_update = mrvl_rss_hash_update,
2176 	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
2177 	.filter_ctrl = mrvl_eth_filter_ctrl,
2178 	.mtr_ops_get = mrvl_mtr_ops_get,
2179 	.tm_ops_get = mrvl_tm_ops_get,
2180 };
2181 
2182 /**
2183  * Return packet type information and l3/l4 offsets.
2184  *
2185  * @param desc
2186  *   Pointer to the received packet descriptor.
2187  * @param l3_offset
2188  *   l3 packet offset.
2189  * @param l4_offset
2190  *   l4 packet offset.
2191  *
2192  * @return
2193  *   Packet type information.
2194  */
2195 static inline uint64_t
2196 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
2197 				    uint8_t *l3_offset, uint8_t *l4_offset)
2198 {
2199 	enum pp2_inq_l3_type l3_type;
2200 	enum pp2_inq_l4_type l4_type;
2201 	enum pp2_inq_vlan_tag vlan_tag;
2202 	uint64_t packet_type;
2203 
2204 	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
2205 	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
2206 	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
2207 
2208 	packet_type = RTE_PTYPE_L2_ETHER;
2209 
2210 	switch (vlan_tag) {
2211 	case PP2_INQ_VLAN_TAG_SINGLE:
2212 		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
2213 		break;
2214 	case PP2_INQ_VLAN_TAG_DOUBLE:
2215 	case PP2_INQ_VLAN_TAG_TRIPLE:
2216 		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
2217 		break;
2218 	default:
2219 		break;
2220 	}
2221 
2222 	switch (l3_type) {
2223 	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
2224 		packet_type |= RTE_PTYPE_L3_IPV4;
2225 		break;
2226 	case PP2_INQ_L3_TYPE_IPV4_OK:
2227 		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
2228 		break;
2229 	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
2230 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
2231 		break;
2232 	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
2233 		packet_type |= RTE_PTYPE_L3_IPV6;
2234 		break;
2235 	case PP2_INQ_L3_TYPE_IPV6_EXT:
2236 		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
2237 		break;
2238 	case PP2_INQ_L3_TYPE_ARP:
2239 		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
2240 		/*
2241 		 * In case of ARP l4_offset is set to wrong value.
2242 		 * Set it to proper one so that later on mbuf->l3_len can be
2243 		 * calculated subtracting l4_offset and l3_offset.
2244 		 */
2245 		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
2246 		break;
2247 	default:
2248 		break;
2249 	}
2250 
2251 	switch (l4_type) {
2252 	case PP2_INQ_L4_TYPE_TCP:
2253 		packet_type |= RTE_PTYPE_L4_TCP;
2254 		break;
2255 	case PP2_INQ_L4_TYPE_UDP:
2256 		packet_type |= RTE_PTYPE_L4_UDP;
2257 		break;
2258 	default:
2259 		break;
2260 	}
2261 
2262 	return packet_type;
2263 }
2264 
2265 /**
2266  * Get offload information from the received packet descriptor.
2267  *
2268  * @param desc
2269  *   Pointer to the received packet descriptor.
2270  *
2271  * @return
2272  *   Mbuf offload flags.
2273  */
2274 static inline uint64_t
2275 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc)
2276 {
2277 	uint64_t flags;
2278 	enum pp2_inq_desc_status status;
2279 
2280 	status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2281 	if (unlikely(status != PP2_DESC_ERR_OK))
2282 		flags = PKT_RX_IP_CKSUM_BAD;
2283 	else
2284 		flags = PKT_RX_IP_CKSUM_GOOD;
2285 
2286 	status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2287 	if (unlikely(status != PP2_DESC_ERR_OK))
2288 		flags |= PKT_RX_L4_CKSUM_BAD;
2289 	else
2290 		flags |= PKT_RX_L4_CKSUM_GOOD;
2291 
2292 	return flags;
2293 }
2294 
2295 /**
2296  * DPDK callback for receive.
2297  *
2298  * @param rxq
2299  *   Generic pointer to the receive queue.
2300  * @param rx_pkts
2301  *   Array to store received packets.
2302  * @param nb_pkts
2303  *   Maximum number of packets in array.
2304  *
2305  * @return
2306  *   Number of packets successfully received.
2307  */
2308 static uint16_t
2309 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2310 {
2311 	struct mrvl_rxq *q = rxq;
2312 	struct pp2_ppio_desc descs[nb_pkts];
2313 	struct pp2_bpool *bpool;
2314 	int i, ret, rx_done = 0;
2315 	int num;
2316 	struct pp2_hif *hif;
2317 	unsigned int core_id = rte_lcore_id();
2318 
2319 	hif = mrvl_get_hif(q->priv, core_id);
2320 
2321 	if (unlikely(!q->priv->ppio || !hif))
2322 		return 0;
2323 
2324 	bpool = q->priv->bpool;
2325 
2326 	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2327 			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2328 	if (unlikely(ret < 0))
2329 		return 0;
2330 
2331 	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2332 
2333 	for (i = 0; i < nb_pkts; i++) {
2334 		struct rte_mbuf *mbuf;
2335 		uint8_t l3_offset, l4_offset;
2336 		enum pp2_inq_desc_status status;
2337 		uint64_t addr;
2338 
2339 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2340 			struct pp2_ppio_desc *pref_desc;
2341 			u64 pref_addr;
2342 
2343 			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2344 			pref_addr = cookie_addr_high |
2345 				    pp2_ppio_inq_desc_get_cookie(pref_desc);
2346 			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2347 			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2348 		}
2349 
2350 		addr = cookie_addr_high |
2351 		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
2352 		mbuf = (struct rte_mbuf *)addr;
2353 		rte_pktmbuf_reset(mbuf);
2354 
2355 		/* drop packet in case of mac, overrun or resource error */
2356 		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2357 		if (unlikely(status != PP2_DESC_ERR_OK)) {
2358 			struct pp2_buff_inf binf = {
2359 				.addr = rte_mbuf_data_iova_default(mbuf),
2360 				.cookie = (uint64_t)mbuf,
2361 			};
2362 
2363 			pp2_bpool_put_buff(hif, bpool, &binf);
2364 			mrvl_port_bpool_size
2365 				[bpool->pp2_id][bpool->id][core_id]++;
2366 			q->drop_mac++;
2367 			continue;
2368 		}
2369 
2370 		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2371 		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2372 		mbuf->data_len = mbuf->pkt_len;
2373 		mbuf->port = q->port_id;
2374 		mbuf->packet_type =
2375 			mrvl_desc_to_packet_type_and_offset(&descs[i],
2376 							    &l3_offset,
2377 							    &l4_offset);
2378 		mbuf->l2_len = l3_offset;
2379 		mbuf->l3_len = l4_offset - l3_offset;
2380 
2381 		if (likely(q->cksum_enabled))
2382 			mbuf->ol_flags = mrvl_desc_to_ol_flags(&descs[i]);
2383 
2384 		rx_pkts[rx_done++] = mbuf;
2385 		q->bytes_recv += mbuf->pkt_len;
2386 	}
2387 
2388 	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2389 		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2390 
2391 		if (unlikely(num <= q->priv->bpool_min_size ||
2392 			     (!rx_done && num < q->priv->bpool_init_size))) {
2393 			mrvl_fill_bpool(q, MRVL_BURST_SIZE);
2394 		} else if (unlikely(num > q->priv->bpool_max_size)) {
2395 			int i;
2396 			int pkt_to_remove = num - q->priv->bpool_init_size;
2397 			struct rte_mbuf *mbuf;
2398 			struct pp2_buff_inf buff;
2399 
2400 			for (i = 0; i < pkt_to_remove; i++) {
2401 				ret = pp2_bpool_get_buff(hif, bpool, &buff);
2402 				if (ret)
2403 					break;
2404 				mbuf = (struct rte_mbuf *)
2405 					(cookie_addr_high | buff.cookie);
2406 				rte_pktmbuf_free(mbuf);
2407 			}
2408 			mrvl_port_bpool_size
2409 				[bpool->pp2_id][bpool->id][core_id] -= i;
2410 		}
2411 		rte_spinlock_unlock(&q->priv->lock);
2412 	}
2413 
2414 	return rx_done;
2415 }
2416 
2417 /**
2418  * Prepare offload information.
2419  *
2420  * @param ol_flags
2421  *   Offload flags.
2422  * @param packet_type
2423  *   Packet type bitfield.
2424  * @param l3_type
2425  *   Pointer to the pp2_ouq_l3_type structure.
2426  * @param l4_type
2427  *   Pointer to the pp2_outq_l4_type structure.
2428  * @param gen_l3_cksum
2429  *   Will be set to 1 in case l3 checksum is computed.
2430  * @param l4_cksum
2431  *   Will be set to 1 in case l4 checksum is computed.
2432  *
2433  * @return
2434  *   0 on success, negative error value otherwise.
2435  */
2436 static inline int
2437 mrvl_prepare_proto_info(uint64_t ol_flags, uint32_t packet_type,
2438 			enum pp2_outq_l3_type *l3_type,
2439 			enum pp2_outq_l4_type *l4_type,
2440 			int *gen_l3_cksum,
2441 			int *gen_l4_cksum)
2442 {
2443 	/*
2444 	 * Based on ol_flags prepare information
2445 	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2446 	 * for offloading.
2447 	 */
2448 	if (ol_flags & PKT_TX_IPV4) {
2449 		*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2450 		*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2451 	} else if (ol_flags & PKT_TX_IPV6) {
2452 		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2453 		/* no checksum for ipv6 header */
2454 		*gen_l3_cksum = 0;
2455 	} else {
2456 		/* if something different then stop processing */
2457 		return -1;
2458 	}
2459 
2460 	ol_flags &= PKT_TX_L4_MASK;
2461 	if ((packet_type & RTE_PTYPE_L4_TCP) &&
2462 	    ol_flags == PKT_TX_TCP_CKSUM) {
2463 		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
2464 		*gen_l4_cksum = 1;
2465 	} else if ((packet_type & RTE_PTYPE_L4_UDP) &&
2466 		   ol_flags == PKT_TX_UDP_CKSUM) {
2467 		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
2468 		*gen_l4_cksum = 1;
2469 	} else {
2470 		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2471 		/* no checksum for other type */
2472 		*gen_l4_cksum = 0;
2473 	}
2474 
2475 	return 0;
2476 }
2477 
2478 /**
2479  * Release already sent buffers to bpool (buffer-pool).
2480  *
2481  * @param ppio
2482  *   Pointer to the port structure.
2483  * @param hif
2484  *   Pointer to the MUSDK hardware interface.
2485  * @param sq
2486  *   Pointer to the shadow queue.
2487  * @param qid
2488  *   Queue id number.
2489  * @param force
2490  *   Force releasing packets.
2491  */
2492 static inline void
2493 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2494 		       unsigned int core_id, struct mrvl_shadow_txq *sq,
2495 		       int qid, int force)
2496 {
2497 	struct buff_release_entry *entry;
2498 	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2499 	int i;
2500 
2501 	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2502 
2503 	sq->num_to_release += nb_done;
2504 
2505 	if (likely(!force &&
2506 		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2507 		return;
2508 
2509 	nb_done = sq->num_to_release;
2510 	sq->num_to_release = 0;
2511 
2512 	for (i = 0; i < nb_done; i++) {
2513 		entry = &sq->ent[sq->tail + num];
2514 		if (unlikely(!entry->buff.addr)) {
2515 			MRVL_LOG(ERR,
2516 				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
2517 				sq->tail, (u64)entry->buff.cookie,
2518 				(u64)entry->buff.addr);
2519 			skip_bufs = 1;
2520 			goto skip;
2521 		}
2522 
2523 		if (unlikely(!entry->bpool)) {
2524 			struct rte_mbuf *mbuf;
2525 
2526 			mbuf = (struct rte_mbuf *)entry->buff.cookie;
2527 			rte_pktmbuf_free(mbuf);
2528 			skip_bufs = 1;
2529 			goto skip;
2530 		}
2531 
2532 		mrvl_port_bpool_size
2533 			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2534 		num++;
2535 		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2536 			goto skip;
2537 		continue;
2538 skip:
2539 		if (likely(num))
2540 			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2541 		num += skip_bufs;
2542 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2543 		sq->size -= num;
2544 		num = 0;
2545 		skip_bufs = 0;
2546 	}
2547 
2548 	if (likely(num)) {
2549 		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2550 		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2551 		sq->size -= num;
2552 	}
2553 }
2554 
2555 /**
2556  * DPDK callback for transmit.
2557  *
2558  * @param txq
2559  *   Generic pointer transmit queue.
2560  * @param tx_pkts
2561  *   Packets to transmit.
2562  * @param nb_pkts
2563  *   Number of packets in array.
2564  *
2565  * @return
2566  *   Number of packets successfully transmitted.
2567  */
2568 static uint16_t
2569 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2570 {
2571 	struct mrvl_txq *q = txq;
2572 	struct mrvl_shadow_txq *sq;
2573 	struct pp2_hif *hif;
2574 	struct pp2_ppio_desc descs[nb_pkts];
2575 	unsigned int core_id = rte_lcore_id();
2576 	int i, ret, bytes_sent = 0;
2577 	uint16_t num, sq_free_size;
2578 	uint64_t addr;
2579 
2580 	hif = mrvl_get_hif(q->priv, core_id);
2581 	sq = &q->shadow_txqs[core_id];
2582 
2583 	if (unlikely(!q->priv->ppio || !hif))
2584 		return 0;
2585 
2586 	if (sq->size)
2587 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2588 				       sq, q->queue_id, 0);
2589 
2590 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2591 	if (unlikely(nb_pkts > sq_free_size))
2592 		nb_pkts = sq_free_size;
2593 
2594 	for (i = 0; i < nb_pkts; i++) {
2595 		struct rte_mbuf *mbuf = tx_pkts[i];
2596 		int gen_l3_cksum, gen_l4_cksum;
2597 		enum pp2_outq_l3_type l3_type;
2598 		enum pp2_outq_l4_type l4_type;
2599 
2600 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2601 			struct rte_mbuf *pref_pkt_hdr;
2602 
2603 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2604 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2605 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2606 		}
2607 
2608 		mrvl_fill_shadowq(sq, mbuf);
2609 		mrvl_fill_desc(&descs[i], mbuf);
2610 
2611 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2612 		/*
2613 		 * in case unsupported ol_flags were passed
2614 		 * do not update descriptor offload information
2615 		 */
2616 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2617 					      &l3_type, &l4_type, &gen_l3_cksum,
2618 					      &gen_l4_cksum);
2619 		if (unlikely(ret))
2620 			continue;
2621 
2622 		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2623 						  mbuf->l2_len,
2624 						  mbuf->l2_len + mbuf->l3_len,
2625 						  gen_l3_cksum, gen_l4_cksum);
2626 	}
2627 
2628 	num = nb_pkts;
2629 	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2630 	/* number of packets that were not sent */
2631 	if (unlikely(num > nb_pkts)) {
2632 		for (i = nb_pkts; i < num; i++) {
2633 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2634 				MRVL_PP2_TX_SHADOWQ_MASK;
2635 			addr = sq->ent[sq->head].buff.cookie;
2636 			bytes_sent -=
2637 				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2638 		}
2639 		sq->size -= num - nb_pkts;
2640 	}
2641 
2642 	q->bytes_sent += bytes_sent;
2643 
2644 	return nb_pkts;
2645 }
2646 
2647 /** DPDK callback for S/G transmit.
2648  *
2649  * @param txq
2650  *   Generic pointer transmit queue.
2651  * @param tx_pkts
2652  *   Packets to transmit.
2653  * @param nb_pkts
2654  *   Number of packets in array.
2655  *
2656  * @return
2657  *   Number of packets successfully transmitted.
2658  */
2659 static uint16_t
2660 mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
2661 		     uint16_t nb_pkts)
2662 {
2663 	struct mrvl_txq *q = txq;
2664 	struct mrvl_shadow_txq *sq;
2665 	struct pp2_hif *hif;
2666 	struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
2667 	struct pp2_ppio_sg_pkts pkts;
2668 	uint8_t frags[nb_pkts];
2669 	unsigned int core_id = rte_lcore_id();
2670 	int i, j, ret, bytes_sent = 0;
2671 	int tail, tail_first;
2672 	uint16_t num, sq_free_size;
2673 	uint16_t nb_segs, total_descs = 0;
2674 	uint64_t addr;
2675 
2676 	hif = mrvl_get_hif(q->priv, core_id);
2677 	sq = &q->shadow_txqs[core_id];
2678 	pkts.frags = frags;
2679 	pkts.num = 0;
2680 
2681 	if (unlikely(!q->priv->ppio || !hif))
2682 		return 0;
2683 
2684 	if (sq->size)
2685 		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2686 				       sq, q->queue_id, 0);
2687 
2688 	/* Save shadow queue free size */
2689 	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2690 
2691 	tail = 0;
2692 	for (i = 0; i < nb_pkts; i++) {
2693 		struct rte_mbuf *mbuf = tx_pkts[i];
2694 		struct rte_mbuf *seg = NULL;
2695 		int gen_l3_cksum, gen_l4_cksum;
2696 		enum pp2_outq_l3_type l3_type;
2697 		enum pp2_outq_l4_type l4_type;
2698 
2699 		nb_segs = mbuf->nb_segs;
2700 		tail_first = tail;
2701 		total_descs += nb_segs;
2702 
2703 		/*
2704 		 * Check if total_descs does not exceed
2705 		 * shadow queue free size
2706 		 */
2707 		if (unlikely(total_descs > sq_free_size)) {
2708 			total_descs -= nb_segs;
2709 			break;
2710 		}
2711 
2712 		/* Check if nb_segs does not exceed the max nb of desc per
2713 		 * fragmented packet
2714 		 */
2715 		if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
2716 			total_descs -= nb_segs;
2717 			RTE_LOG(ERR, PMD,
2718 				"Too many segments. Packet won't be sent.\n");
2719 			break;
2720 		}
2721 
2722 		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2723 			struct rte_mbuf *pref_pkt_hdr;
2724 
2725 			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2726 			rte_mbuf_prefetch_part1(pref_pkt_hdr);
2727 			rte_mbuf_prefetch_part2(pref_pkt_hdr);
2728 		}
2729 
2730 		pkts.frags[pkts.num] = nb_segs;
2731 		pkts.num++;
2732 
2733 		seg = mbuf;
2734 		for (j = 0; j < nb_segs - 1; j++) {
2735 			/* For the subsequent segments, set shadow queue
2736 			 * buffer to NULL
2737 			 */
2738 			mrvl_fill_shadowq(sq, NULL);
2739 			mrvl_fill_desc(&descs[tail], seg);
2740 
2741 			tail++;
2742 			seg = seg->next;
2743 		}
2744 		/* Put first mbuf info in last shadow queue entry */
2745 		mrvl_fill_shadowq(sq, mbuf);
2746 		/* Update descriptor with last segment */
2747 		mrvl_fill_desc(&descs[tail++], seg);
2748 
2749 		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2750 		/* In case unsupported ol_flags were passed
2751 		 * do not update descriptor offload information
2752 		 */
2753 		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
2754 					      &l3_type, &l4_type, &gen_l3_cksum,
2755 					      &gen_l4_cksum);
2756 		if (unlikely(ret))
2757 			continue;
2758 
2759 		pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
2760 						  l4_type, mbuf->l2_len,
2761 						  mbuf->l2_len + mbuf->l3_len,
2762 						  gen_l3_cksum, gen_l4_cksum);
2763 	}
2764 
2765 	num = total_descs;
2766 	pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
2767 			 &total_descs, &pkts);
2768 	/* number of packets that were not sent */
2769 	if (unlikely(num > total_descs)) {
2770 		for (i = total_descs; i < num; i++) {
2771 			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2772 				MRVL_PP2_TX_SHADOWQ_MASK;
2773 
2774 			addr = sq->ent[sq->head].buff.cookie;
2775 			if (addr)
2776 				bytes_sent -=
2777 					rte_pktmbuf_pkt_len((struct rte_mbuf *)
2778 						(cookie_addr_high | addr));
2779 		}
2780 		sq->size -= num - total_descs;
2781 		nb_pkts = pkts.num;
2782 	}
2783 
2784 	q->bytes_sent += bytes_sent;
2785 
2786 	return nb_pkts;
2787 }
2788 
2789 /**
2790  * Initialize packet processor.
2791  *
2792  * @return
2793  *   0 on success, negative error value otherwise.
2794  */
2795 static int
2796 mrvl_init_pp2(void)
2797 {
2798 	struct pp2_init_params init_params;
2799 
2800 	memset(&init_params, 0, sizeof(init_params));
2801 	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
2802 	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
2803 	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
2804 
2805 	return pp2_init(&init_params);
2806 }
2807 
2808 /**
2809  * Deinitialize packet processor.
2810  *
2811  * @return
2812  *   0 on success, negative error value otherwise.
2813  */
2814 static void
2815 mrvl_deinit_pp2(void)
2816 {
2817 	pp2_deinit();
2818 }
2819 
2820 /**
2821  * Create private device structure.
2822  *
2823  * @param dev_name
2824  *   Pointer to the port name passed in the initialization parameters.
2825  *
2826  * @return
2827  *   Pointer to the newly allocated private device structure.
2828  */
2829 static struct mrvl_priv *
2830 mrvl_priv_create(const char *dev_name)
2831 {
2832 	struct pp2_bpool_params bpool_params;
2833 	char match[MRVL_MATCH_LEN];
2834 	struct mrvl_priv *priv;
2835 	int ret, bpool_bit;
2836 
2837 	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
2838 	if (!priv)
2839 		return NULL;
2840 
2841 	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
2842 				       &priv->pp_id, &priv->ppio_id);
2843 	if (ret)
2844 		goto out_free_priv;
2845 
2846 	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
2847 				     PP2_BPOOL_NUM_POOLS);
2848 	if (bpool_bit < 0)
2849 		goto out_free_priv;
2850 	priv->bpool_bit = bpool_bit;
2851 
2852 	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
2853 		 priv->bpool_bit);
2854 	memset(&bpool_params, 0, sizeof(bpool_params));
2855 	bpool_params.match = match;
2856 	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
2857 	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
2858 	if (ret)
2859 		goto out_clear_bpool_bit;
2860 
2861 	priv->ppio_params.type = PP2_PPIO_T_NIC;
2862 	rte_spinlock_init(&priv->lock);
2863 
2864 	return priv;
2865 out_clear_bpool_bit:
2866 	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2867 out_free_priv:
2868 	rte_free(priv);
2869 	return NULL;
2870 }
2871 
2872 /**
2873  * Create device representing Ethernet port.
2874  *
2875  * @param name
2876  *   Pointer to the port's name.
2877  *
2878  * @return
2879  *   0 on success, negative error value otherwise.
2880  */
2881 static int
2882 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
2883 {
2884 	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
2885 	struct rte_eth_dev *eth_dev;
2886 	struct mrvl_priv *priv;
2887 	struct ifreq req;
2888 
2889 	eth_dev = rte_eth_dev_allocate(name);
2890 	if (!eth_dev)
2891 		return -ENOMEM;
2892 
2893 	priv = mrvl_priv_create(name);
2894 	if (!priv) {
2895 		ret = -ENOMEM;
2896 		goto out_free;
2897 	}
2898 	eth_dev->data->dev_private = priv;
2899 
2900 	eth_dev->data->mac_addrs =
2901 		rte_zmalloc("mac_addrs",
2902 			    RTE_ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
2903 	if (!eth_dev->data->mac_addrs) {
2904 		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
2905 		ret = -ENOMEM;
2906 		goto out_free;
2907 	}
2908 
2909 	memset(&req, 0, sizeof(req));
2910 	strcpy(req.ifr_name, name);
2911 	ret = ioctl(fd, SIOCGIFHWADDR, &req);
2912 	if (ret)
2913 		goto out_free;
2914 
2915 	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
2916 	       req.ifr_addr.sa_data, RTE_ETHER_ADDR_LEN);
2917 
2918 	eth_dev->device = &vdev->device;
2919 	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
2920 	mrvl_set_tx_function(eth_dev);
2921 	eth_dev->dev_ops = &mrvl_ops;
2922 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2923 
2924 	eth_dev->data->dev_link.link_status = ETH_LINK_UP;
2925 
2926 	rte_eth_dev_probing_finish(eth_dev);
2927 	return 0;
2928 out_free:
2929 	rte_eth_dev_release_port(eth_dev);
2930 
2931 	return ret;
2932 }
2933 
2934 /**
2935  * Callback used by rte_kvargs_process() during argument parsing.
2936  *
2937  * @param key
2938  *   Pointer to the parsed key (unused).
2939  * @param value
2940  *   Pointer to the parsed value.
2941  * @param extra_args
2942  *   Pointer to the extra arguments which contains address of the
2943  *   table of pointers to parsed interface names.
2944  *
2945  * @return
2946  *   Always 0.
2947  */
2948 static int
2949 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
2950 		 void *extra_args)
2951 {
2952 	struct mrvl_ifnames *ifnames = extra_args;
2953 
2954 	ifnames->names[ifnames->idx++] = value;
2955 
2956 	return 0;
2957 }
2958 
2959 /**
2960  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
2961  */
2962 static void
2963 mrvl_deinit_hifs(void)
2964 {
2965 	int i;
2966 
2967 	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
2968 		if (hifs[i])
2969 			pp2_hif_deinit(hifs[i]);
2970 	}
2971 	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
2972 	memset(hifs, 0, sizeof(hifs));
2973 }
2974 
2975 /**
2976  * DPDK callback to register the virtual device.
2977  *
2978  * @param vdev
2979  *   Pointer to the virtual device.
2980  *
2981  * @return
2982  *   0 on success, negative error value otherwise.
2983  */
2984 static int
2985 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
2986 {
2987 	struct rte_kvargs *kvlist;
2988 	struct mrvl_ifnames ifnames;
2989 	int ret = -EINVAL;
2990 	uint32_t i, ifnum, cfgnum;
2991 	const char *params;
2992 
2993 	params = rte_vdev_device_args(vdev);
2994 	if (!params)
2995 		return -EINVAL;
2996 
2997 	kvlist = rte_kvargs_parse(params, valid_args);
2998 	if (!kvlist)
2999 		return -EINVAL;
3000 
3001 	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
3002 	if (ifnum > RTE_DIM(ifnames.names))
3003 		goto out_free_kvlist;
3004 
3005 	ifnames.idx = 0;
3006 	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
3007 			   mrvl_get_ifnames, &ifnames);
3008 
3009 
3010 	/*
3011 	 * The below system initialization should be done only once,
3012 	 * on the first provided configuration file
3013 	 */
3014 	if (!mrvl_qos_cfg) {
3015 		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
3016 		MRVL_LOG(INFO, "Parsing config file!");
3017 		if (cfgnum > 1) {
3018 			MRVL_LOG(ERR, "Cannot handle more than one config file!");
3019 			goto out_free_kvlist;
3020 		} else if (cfgnum == 1) {
3021 			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
3022 					   mrvl_get_qoscfg, &mrvl_qos_cfg);
3023 		}
3024 	}
3025 
3026 	if (mrvl_dev_num)
3027 		goto init_devices;
3028 
3029 	MRVL_LOG(INFO, "Perform MUSDK initializations");
3030 
3031 	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
3032 	if (ret)
3033 		goto out_free_kvlist;
3034 
3035 	ret = mrvl_init_pp2();
3036 	if (ret) {
3037 		MRVL_LOG(ERR, "Failed to init PP!");
3038 		rte_mvep_deinit(MVEP_MOD_T_PP2);
3039 		goto out_free_kvlist;
3040 	}
3041 
3042 	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
3043 	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
3044 
3045 	mrvl_lcore_first = RTE_MAX_LCORE;
3046 	mrvl_lcore_last = 0;
3047 
3048 init_devices:
3049 	for (i = 0; i < ifnum; i++) {
3050 		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
3051 		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
3052 		if (ret)
3053 			goto out_cleanup;
3054 		mrvl_dev_num++;
3055 	}
3056 
3057 	rte_kvargs_free(kvlist);
3058 
3059 	return 0;
3060 out_cleanup:
3061 	rte_pmd_mrvl_remove(vdev);
3062 
3063 out_free_kvlist:
3064 	rte_kvargs_free(kvlist);
3065 
3066 	return ret;
3067 }
3068 
3069 /**
3070  * DPDK callback to remove virtual device.
3071  *
3072  * @param vdev
3073  *   Pointer to the removed virtual device.
3074  *
3075  * @return
3076  *   0 on success, negative error value otherwise.
3077  */
3078 static int
3079 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
3080 {
3081 	uint16_t port_id;
3082 	int ret = 0;
3083 
3084 	RTE_ETH_FOREACH_DEV(port_id) {
3085 		if (rte_eth_devices[port_id].device != &vdev->device)
3086 			continue;
3087 		ret |= rte_eth_dev_close(port_id);
3088 	}
3089 
3090 	return ret == 0 ? 0 : -EIO;
3091 }
3092 
3093 static struct rte_vdev_driver pmd_mrvl_drv = {
3094 	.probe = rte_pmd_mrvl_probe,
3095 	.remove = rte_pmd_mrvl_remove,
3096 };
3097 
3098 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
3099 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
3100 RTE_LOG_REGISTER(mrvl_logtype, pmd.net.mvpp2, NOTICE);
3101