version: 21.11-rc0
[dpdk.git] / drivers / net / mvpp2 / mrvl_ethdev.c
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  * + 1 for primary mac address
45  */
46 #define MRVL_MAC_ADDRS_MAX (1 + 25)
47 #define MRVL_MATCH_LEN 16
48 #define MRVL_PKT_EFFEC_OFFS (MRVL_PKT_OFFS + MV_MH_SIZE)
49 /* Maximum allowable packet size */
50 #define MRVL_PKT_SIZE_MAX (10240 - MV_MH_SIZE)
51
52 #define MRVL_IFACE_NAME_ARG "iface"
53 #define MRVL_CFG_ARG "cfg"
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_OFFLOAD_CHECKSUM (DEV_TX_OFFLOAD_IPV4_CKSUM | \
67                                   DEV_TX_OFFLOAD_UDP_CKSUM  | \
68                                   DEV_TX_OFFLOAD_TCP_CKSUM)
69 #define MRVL_TX_OFFLOADS (MRVL_TX_OFFLOAD_CHECKSUM | \
70                           DEV_TX_OFFLOAD_MULTI_SEGS)
71
72 #define MRVL_TX_PKT_OFFLOADS (PKT_TX_IP_CKSUM | \
73                               PKT_TX_TCP_CKSUM | \
74                               PKT_TX_UDP_CKSUM)
75
76 static const char * const valid_args[] = {
77         MRVL_IFACE_NAME_ARG,
78         MRVL_CFG_ARG,
79         NULL
80 };
81
82 static int used_hifs = MRVL_MUSDK_HIFS_RESERVED;
83 static struct pp2_hif *hifs[RTE_MAX_LCORE];
84 static int used_bpools[PP2_NUM_PKT_PROC] = {
85         [0 ... PP2_NUM_PKT_PROC - 1] = MRVL_MUSDK_BPOOLS_RESERVED
86 };
87
88 static struct pp2_bpool *mrvl_port_to_bpool_lookup[RTE_MAX_ETHPORTS];
89 static int mrvl_port_bpool_size[PP2_NUM_PKT_PROC][PP2_BPOOL_NUM_POOLS][RTE_MAX_LCORE];
90 static uint64_t cookie_addr_high = MRVL_COOKIE_ADDR_INVALID;
91 static int dummy_pool_id[PP2_NUM_PKT_PROC];
92 struct pp2_bpool *dummy_pool[PP2_NUM_PKT_PROC] = {0};
93
94 struct mrvl_ifnames {
95         const char *names[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
96         int idx;
97 };
98
99 /*
100  * To use buffer harvesting based on loopback port shadow queue structure
101  * was introduced for buffers information bookkeeping.
102  *
103  * Before sending the packet, related buffer information (pp2_buff_inf) is
104  * stored in shadow queue. After packet is transmitted no longer used
105  * packet buffer is released back to it's original hardware pool,
106  * on condition it originated from interface.
107  * In case it  was generated by application itself i.e: mbuf->port field is
108  * 0xff then its released to software mempool.
109  */
110 struct mrvl_shadow_txq {
111         int head;           /* write index - used when sending buffers */
112         int tail;           /* read index - used when releasing buffers */
113         u16 size;           /* queue occupied size */
114         u16 num_to_release; /* number of descriptors sent, that can be
115                              * released
116                              */
117         struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
118 };
119
120 struct mrvl_rxq {
121         struct mrvl_priv *priv;
122         struct rte_mempool *mp;
123         int queue_id;
124         int port_id;
125         int cksum_enabled;
126         uint64_t bytes_recv;
127         uint64_t drop_mac;
128 };
129
130 struct mrvl_txq {
131         struct mrvl_priv *priv;
132         int queue_id;
133         int port_id;
134         uint64_t bytes_sent;
135         struct mrvl_shadow_txq shadow_txqs[RTE_MAX_LCORE];
136         int tx_deferred_start;
137 };
138
139 static int mrvl_lcore_first;
140 static int mrvl_lcore_last;
141 static int mrvl_dev_num;
142
143 static int mrvl_fill_bpool(struct mrvl_rxq *rxq, int num);
144 static inline void mrvl_free_sent_buffers(struct pp2_ppio *ppio,
145                         struct pp2_hif *hif, unsigned int core_id,
146                         struct mrvl_shadow_txq *sq, int qid, int force);
147
148 static uint16_t mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
149                                   uint16_t nb_pkts);
150 static uint16_t mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
151                                      uint16_t nb_pkts);
152 static int rte_pmd_mrvl_remove(struct rte_vdev_device *vdev);
153 static void mrvl_deinit_pp2(void);
154 static void mrvl_deinit_hifs(void);
155
156 static int
157 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
158                   uint32_t index, uint32_t vmdq __rte_unused);
159 static int
160 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr);
161 static int
162 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on);
163 static int mrvl_promiscuous_enable(struct rte_eth_dev *dev);
164 static int mrvl_allmulticast_enable(struct rte_eth_dev *dev);
165 static int
166 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf);
167
168 #define MRVL_XSTATS_TBL_ENTRY(name) { \
169         #name, offsetof(struct pp2_ppio_statistics, name),      \
170         sizeof(((struct pp2_ppio_statistics *)0)->name)         \
171 }
172
173 /* Table with xstats data */
174 static struct {
175         const char *name;
176         unsigned int offset;
177         unsigned int size;
178 } mrvl_xstats_tbl[] = {
179         MRVL_XSTATS_TBL_ENTRY(rx_bytes),
180         MRVL_XSTATS_TBL_ENTRY(rx_packets),
181         MRVL_XSTATS_TBL_ENTRY(rx_unicast_packets),
182         MRVL_XSTATS_TBL_ENTRY(rx_errors),
183         MRVL_XSTATS_TBL_ENTRY(rx_fullq_dropped),
184         MRVL_XSTATS_TBL_ENTRY(rx_bm_dropped),
185         MRVL_XSTATS_TBL_ENTRY(rx_early_dropped),
186         MRVL_XSTATS_TBL_ENTRY(rx_fifo_dropped),
187         MRVL_XSTATS_TBL_ENTRY(rx_cls_dropped),
188         MRVL_XSTATS_TBL_ENTRY(tx_bytes),
189         MRVL_XSTATS_TBL_ENTRY(tx_packets),
190         MRVL_XSTATS_TBL_ENTRY(tx_unicast_packets),
191         MRVL_XSTATS_TBL_ENTRY(tx_errors)
192 };
193
194 static inline int
195 mrvl_reserve_bit(int *bitmap, int max)
196 {
197         int n = sizeof(*bitmap) * 8 - __builtin_clz(*bitmap);
198
199         if (n >= max)
200                 return -1;
201
202         *bitmap |= 1 << n;
203
204         return n;
205 }
206
207 static int
208 mrvl_pp2_fixup_init(void)
209 {
210         struct pp2_bpool_params bpool_params;
211         char                    name[15];
212         int                     err, i;
213
214         memset(dummy_pool, 0, sizeof(dummy_pool));
215         for (i = 0; i < pp2_get_num_inst(); i++) {
216                 dummy_pool_id[i] = mrvl_reserve_bit(&used_bpools[i],
217                                              PP2_BPOOL_NUM_POOLS);
218                 if (dummy_pool_id[i] < 0) {
219                         MRVL_LOG(ERR, "Can't find free pool\n");
220                         return -1;
221                 }
222
223                 memset(name, 0, sizeof(name));
224                 snprintf(name, sizeof(name), "pool-%d:%d", i, dummy_pool_id[i]);
225                 memset(&bpool_params, 0, sizeof(bpool_params));
226                 bpool_params.match = name;
227                 bpool_params.buff_len = MRVL_PKT_OFFS;
228                 bpool_params.dummy_short_pool = 1;
229                 err = pp2_bpool_init(&bpool_params, &dummy_pool[i]);
230                 if (err != 0 || !dummy_pool[i]) {
231                         MRVL_LOG(ERR, "BPool init failed!\n");
232                         used_bpools[i] &= ~(1 << dummy_pool_id[i]);
233                         return -1;
234                 }
235         }
236
237         return 0;
238 }
239
240 /**
241  * Initialize packet processor.
242  *
243  * @return
244  *   0 on success, negative error value otherwise.
245  */
246 static int
247 mrvl_init_pp2(void)
248 {
249         struct pp2_init_params  init_params;
250         int                     err;
251
252         memset(&init_params, 0, sizeof(init_params));
253         init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
254         init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
255         init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
256         if (mrvl_cfg && mrvl_cfg->pp2_cfg.prs_udfs.num_udfs)
257                 memcpy(&init_params.prs_udfs, &mrvl_cfg->pp2_cfg.prs_udfs,
258                        sizeof(struct pp2_parse_udfs));
259         err = pp2_init(&init_params);
260         if (err != 0) {
261                 MRVL_LOG(ERR, "PP2 init failed");
262                 return -1;
263         }
264
265         err = mrvl_pp2_fixup_init();
266         if (err != 0) {
267                 MRVL_LOG(ERR, "PP2 fixup init failed");
268                 return -1;
269         }
270
271         return 0;
272 }
273
274 static void
275 mrvl_pp2_fixup_deinit(void)
276 {
277         int i;
278
279         for (i = 0; i < PP2_NUM_PKT_PROC; i++) {
280                 if (!dummy_pool[i])
281                         continue;
282                 pp2_bpool_deinit(dummy_pool[i]);
283                 used_bpools[i] &= ~(1 << dummy_pool_id[i]);
284         }
285 }
286
287 /**
288  * Deinitialize packet processor.
289  *
290  * @return
291  *   0 on success, negative error value otherwise.
292  */
293 static void
294 mrvl_deinit_pp2(void)
295 {
296         mrvl_pp2_fixup_deinit();
297         pp2_deinit();
298 }
299
300 static inline void
301 mrvl_fill_shadowq(struct mrvl_shadow_txq *sq, struct rte_mbuf *buf)
302 {
303         sq->ent[sq->head].buff.cookie = (uint64_t)buf;
304         sq->ent[sq->head].buff.addr = buf ?
305                 rte_mbuf_data_iova_default(buf) : 0;
306
307         sq->ent[sq->head].bpool =
308                 (unlikely(!buf || buf->port >= RTE_MAX_ETHPORTS ||
309                  buf->refcnt > 1)) ? NULL :
310                  mrvl_port_to_bpool_lookup[buf->port];
311
312         sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
313         sq->size++;
314 }
315
316 /**
317  * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
318  */
319 static void
320 mrvl_deinit_hifs(void)
321 {
322         int i;
323
324         for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
325                 if (hifs[i])
326                         pp2_hif_deinit(hifs[i]);
327         }
328         used_hifs = MRVL_MUSDK_HIFS_RESERVED;
329         memset(hifs, 0, sizeof(hifs));
330 }
331
332 static inline void
333 mrvl_fill_desc(struct pp2_ppio_desc *desc, struct rte_mbuf *buf)
334 {
335         pp2_ppio_outq_desc_reset(desc);
336         pp2_ppio_outq_desc_set_phys_addr(desc, rte_pktmbuf_iova(buf));
337         pp2_ppio_outq_desc_set_pkt_offset(desc, 0);
338         pp2_ppio_outq_desc_set_pkt_len(desc, rte_pktmbuf_data_len(buf));
339 }
340
341 static inline int
342 mrvl_get_bpool_size(int pp2_id, int pool_id)
343 {
344         int i;
345         int size = 0;
346
347         for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
348                 size += mrvl_port_bpool_size[pp2_id][pool_id][i];
349
350         return size;
351 }
352
353 static int
354 mrvl_init_hif(int core_id)
355 {
356         struct pp2_hif_params params;
357         char match[MRVL_MATCH_LEN];
358         int ret;
359
360         ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
361         if (ret < 0) {
362                 MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
363                 return ret;
364         }
365
366         snprintf(match, sizeof(match), "hif-%d", ret);
367         memset(&params, 0, sizeof(params));
368         params.match = match;
369         params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
370         ret = pp2_hif_init(&params, &hifs[core_id]);
371         if (ret) {
372                 MRVL_LOG(ERR, "Failed to initialize hif %d", core_id);
373                 return ret;
374         }
375
376         return 0;
377 }
378
379 static inline struct pp2_hif*
380 mrvl_get_hif(struct mrvl_priv *priv, int core_id)
381 {
382         int ret;
383
384         if (likely(hifs[core_id] != NULL))
385                 return hifs[core_id];
386
387         rte_spinlock_lock(&priv->lock);
388
389         ret = mrvl_init_hif(core_id);
390         if (ret < 0) {
391                 MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
392                 goto out;
393         }
394
395         if (core_id < mrvl_lcore_first)
396                 mrvl_lcore_first = core_id;
397
398         if (core_id > mrvl_lcore_last)
399                 mrvl_lcore_last = core_id;
400 out:
401         rte_spinlock_unlock(&priv->lock);
402
403         return hifs[core_id];
404 }
405
406 /**
407  * Set tx burst function according to offload flag
408  *
409  * @param dev
410  *   Pointer to Ethernet device structure.
411  */
412 static void
413 mrvl_set_tx_function(struct rte_eth_dev *dev)
414 {
415         struct mrvl_priv *priv = dev->data->dev_private;
416
417         /* Use a simple Tx queue (no offloads, no multi segs) if possible */
418         if (priv->multiseg) {
419                 RTE_LOG(INFO, PMD, "Using multi-segment tx callback\n");
420                 dev->tx_pkt_burst = mrvl_tx_sg_pkt_burst;
421         } else {
422                 RTE_LOG(INFO, PMD, "Using single-segment tx callback\n");
423                 dev->tx_pkt_burst = mrvl_tx_pkt_burst;
424         }
425 }
426
427 /**
428  * Configure rss based on dpdk rss configuration.
429  *
430  * @param priv
431  *   Pointer to private structure.
432  * @param rss_conf
433  *   Pointer to RSS configuration.
434  *
435  * @return
436  *   0 on success, negative error value otherwise.
437  */
438 static int
439 mrvl_configure_rss(struct mrvl_priv *priv, struct rte_eth_rss_conf *rss_conf)
440 {
441         if (rss_conf->rss_key)
442                 MRVL_LOG(WARNING, "Changing hash key is not supported");
443
444         if (rss_conf->rss_hf == 0) {
445                 priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
446         } else if (rss_conf->rss_hf & ETH_RSS_IPV4) {
447                 priv->ppio_params.inqs_params.hash_type =
448                         PP2_PPIO_HASH_T_2_TUPLE;
449         } else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP) {
450                 priv->ppio_params.inqs_params.hash_type =
451                         PP2_PPIO_HASH_T_5_TUPLE;
452                 priv->rss_hf_tcp = 1;
453         } else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) {
454                 priv->ppio_params.inqs_params.hash_type =
455                         PP2_PPIO_HASH_T_5_TUPLE;
456                 priv->rss_hf_tcp = 0;
457         } else {
458                 return -EINVAL;
459         }
460
461         return 0;
462 }
463
464 /**
465  * Ethernet device configuration.
466  *
467  * Prepare the driver for a given number of TX and RX queues and
468  * configure RSS.
469  *
470  * @param dev
471  *   Pointer to Ethernet device structure.
472  *
473  * @return
474  *   0 on success, negative error value otherwise.
475  */
476 static int
477 mrvl_dev_configure(struct rte_eth_dev *dev)
478 {
479         struct mrvl_priv *priv = dev->data->dev_private;
480         int ret;
481
482         if (priv->ppio) {
483                 MRVL_LOG(INFO, "Device reconfiguration is not supported");
484                 return -EINVAL;
485         }
486
487         if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_NONE &&
488             dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
489                 MRVL_LOG(INFO, "Unsupported rx multi queue mode %d",
490                         dev->data->dev_conf.rxmode.mq_mode);
491                 return -EINVAL;
492         }
493
494         if (dev->data->dev_conf.rxmode.split_hdr_size) {
495                 MRVL_LOG(INFO, "Split headers not supported");
496                 return -EINVAL;
497         }
498
499         if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) {
500                 dev->data->mtu = dev->data->dev_conf.rxmode.max_rx_pkt_len -
501                                  MRVL_PP2_ETH_HDRS_LEN;
502                 if (dev->data->mtu > priv->max_mtu) {
503                         MRVL_LOG(ERR, "inherit MTU %u from max_rx_pkt_len %u is larger than max_mtu %u\n",
504                                  dev->data->mtu,
505                                  dev->data->dev_conf.rxmode.max_rx_pkt_len,
506                                  priv->max_mtu);
507                         return -EINVAL;
508                 }
509         }
510
511         if (dev->data->dev_conf.txmode.offloads & DEV_TX_OFFLOAD_MULTI_SEGS)
512                 priv->multiseg = 1;
513
514         ret = mrvl_configure_rxqs(priv, dev->data->port_id,
515                                   dev->data->nb_rx_queues);
516         if (ret < 0)
517                 return ret;
518
519         ret = mrvl_configure_txqs(priv, dev->data->port_id,
520                                   dev->data->nb_tx_queues);
521         if (ret < 0)
522                 return ret;
523
524         priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
525         priv->ppio_params.maintain_stats = 1;
526         priv->nb_rx_queues = dev->data->nb_rx_queues;
527
528         ret = mrvl_tm_init(dev);
529         if (ret < 0)
530                 return ret;
531
532         if (dev->data->nb_rx_queues == 1 &&
533             dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
534                 MRVL_LOG(WARNING, "Disabling hash for 1 rx queue");
535                 priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
536                 priv->configured = 1;
537                 return 0;
538         }
539
540         ret = mrvl_configure_rss(priv,
541                         &dev->data->dev_conf.rx_adv_conf.rss_conf);
542         if (ret < 0)
543                 return ret;
544
545         priv->configured = 1;
546
547         return 0;
548 }
549
550 /**
551  * DPDK callback to change the MTU.
552  *
553  * Setting the MTU affects hardware MRU (packets larger than the MRU
554  * will be dropped).
555  *
556  * @param dev
557  *   Pointer to Ethernet device structure.
558  * @param mtu
559  *   New MTU.
560  *
561  * @return
562  *   0 on success, negative error value otherwise.
563  */
564 static int
565 mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
566 {
567         struct mrvl_priv *priv = dev->data->dev_private;
568         uint16_t mru;
569         uint16_t mbuf_data_size = 0; /* SW buffer size */
570         int ret;
571
572         mru = MRVL_PP2_MTU_TO_MRU(mtu);
573         /*
574          * min_rx_buf_size is equal to mbuf data size
575          * if pmd didn't set it differently
576          */
577         mbuf_data_size = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
578         /* Prevent PMD from:
579          * - setting mru greater than the mbuf size resulting in
580          * hw and sw buffer size mismatch
581          * - setting mtu that requires the support of scattered packets
582          * when this feature has not been enabled/supported so far
583          * (TODO check scattered_rx flag here once scattered RX is supported).
584          */
585         if (mru - RTE_ETHER_CRC_LEN + MRVL_PKT_OFFS > mbuf_data_size) {
586                 mru = mbuf_data_size + RTE_ETHER_CRC_LEN - MRVL_PKT_OFFS;
587                 mtu = MRVL_PP2_MRU_TO_MTU(mru);
588                 MRVL_LOG(WARNING, "MTU too big, max MTU possible limitted "
589                         "by current mbuf size: %u. Set MTU to %u, MRU to %u",
590                         mbuf_data_size, mtu, mru);
591         }
592
593         if (mtu < RTE_ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX) {
594                 MRVL_LOG(ERR, "Invalid MTU [%u] or MRU [%u]", mtu, mru);
595                 return -EINVAL;
596         }
597
598         dev->data->mtu = mtu;
599         dev->data->dev_conf.rxmode.max_rx_pkt_len = mru - MV_MH_SIZE;
600
601         if (!priv->ppio)
602                 return 0;
603
604         ret = pp2_ppio_set_mru(priv->ppio, mru);
605         if (ret) {
606                 MRVL_LOG(ERR, "Failed to change MRU");
607                 return ret;
608         }
609
610         ret = pp2_ppio_set_mtu(priv->ppio, mtu);
611         if (ret) {
612                 MRVL_LOG(ERR, "Failed to change MTU");
613                 return ret;
614         }
615
616         return 0;
617 }
618
619 /**
620  * DPDK callback to bring the link up.
621  *
622  * @param dev
623  *   Pointer to Ethernet device structure.
624  *
625  * @return
626  *   0 on success, negative error value otherwise.
627  */
628 static int
629 mrvl_dev_set_link_up(struct rte_eth_dev *dev)
630 {
631         struct mrvl_priv *priv = dev->data->dev_private;
632         int ret;
633
634         if (!priv->ppio) {
635                 dev->data->dev_link.link_status = ETH_LINK_UP;
636                 return 0;
637         }
638
639         ret = pp2_ppio_enable(priv->ppio);
640         if (ret)
641                 return ret;
642
643         /*
644          * mtu/mru can be updated if pp2_ppio_enable() was called at least once
645          * as pp2_ppio_enable() changes port->t_mode from default 0 to
646          * PP2_TRAFFIC_INGRESS_EGRESS.
647          *
648          * Set mtu to default DPDK value here.
649          */
650         ret = mrvl_mtu_set(dev, dev->data->mtu);
651         if (ret) {
652                 pp2_ppio_disable(priv->ppio);
653                 return ret;
654         }
655
656         dev->data->dev_link.link_status = ETH_LINK_UP;
657         return 0;
658 }
659
660 /**
661  * DPDK callback to bring the link down.
662  *
663  * @param dev
664  *   Pointer to Ethernet device structure.
665  *
666  * @return
667  *   0 on success, negative error value otherwise.
668  */
669 static int
670 mrvl_dev_set_link_down(struct rte_eth_dev *dev)
671 {
672         struct mrvl_priv *priv = dev->data->dev_private;
673         int ret;
674
675         if (!priv->ppio) {
676                 dev->data->dev_link.link_status = ETH_LINK_DOWN;
677                 return 0;
678         }
679         ret = pp2_ppio_disable(priv->ppio);
680         if (ret)
681                 return ret;
682
683         dev->data->dev_link.link_status = ETH_LINK_DOWN;
684         return 0;
685 }
686
687 /**
688  * DPDK callback to start tx queue.
689  *
690  * @param dev
691  *   Pointer to Ethernet device structure.
692  * @param queue_id
693  *   Transmit queue index.
694  *
695  * @return
696  *   0 on success, negative error value otherwise.
697  */
698 static int
699 mrvl_tx_queue_start(struct rte_eth_dev *dev, uint16_t queue_id)
700 {
701         struct mrvl_priv *priv = dev->data->dev_private;
702         int ret;
703
704         if (!priv)
705                 return -EPERM;
706
707         /* passing 1 enables given tx queue */
708         ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 1);
709         if (ret) {
710                 MRVL_LOG(ERR, "Failed to start txq %d", queue_id);
711                 return ret;
712         }
713
714         dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
715
716         return 0;
717 }
718
719 /**
720  * DPDK callback to stop tx queue.
721  *
722  * @param dev
723  *   Pointer to Ethernet device structure.
724  * @param queue_id
725  *   Transmit queue index.
726  *
727  * @return
728  *   0 on success, negative error value otherwise.
729  */
730 static int
731 mrvl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t queue_id)
732 {
733         struct mrvl_priv *priv = dev->data->dev_private;
734         int ret;
735
736         if (!priv->ppio)
737                 return -EPERM;
738
739         /* passing 0 disables given tx queue */
740         ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 0);
741         if (ret) {
742                 MRVL_LOG(ERR, "Failed to stop txq %d", queue_id);
743                 return ret;
744         }
745
746         dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
747
748         return 0;
749 }
750
751 /**
752  * Populate VLAN Filter configuration.
753  *
754  * @param dev
755  *   Pointer to Ethernet device structure.
756  * @param on
757  *   Toggle filter.
758  *
759  * @return
760  *   0 on success, negative error value otherwise.
761  */
762 static int mrvl_populate_vlan_table(struct rte_eth_dev *dev, int on)
763 {
764         uint32_t j;
765         int ret;
766         struct rte_vlan_filter_conf *vfc;
767
768         vfc = &dev->data->vlan_filter_conf;
769         for (j = 0; j < RTE_DIM(vfc->ids); j++) {
770                 uint64_t vlan;
771                 uint64_t vbit;
772                 uint64_t ids = vfc->ids[j];
773
774                 if (ids == 0)
775                         continue;
776
777                 while (ids) {
778                         vlan = 64 * j;
779                         /* count trailing zeroes */
780                         vbit = ~ids & (ids - 1);
781                         /* clear least significant bit set */
782                         ids ^= (ids ^ (ids - 1)) ^ vbit;
783                         for (; vbit; vlan++)
784                                 vbit >>= 1;
785                         ret = mrvl_vlan_filter_set(dev, vlan, on);
786                         if (ret) {
787                                 MRVL_LOG(ERR, "Failed to setup VLAN filter\n");
788                                 return ret;
789                         }
790                 }
791         }
792
793         return 0;
794 }
795
796 /**
797  * DPDK callback to start the device.
798  *
799  * @param dev
800  *   Pointer to Ethernet device structure.
801  *
802  * @return
803  *   0 on success, negative errno value on failure.
804  */
805 static int
806 mrvl_dev_start(struct rte_eth_dev *dev)
807 {
808         struct mrvl_priv *priv = dev->data->dev_private;
809         char match[MRVL_MATCH_LEN];
810         int ret = 0, i, def_init_size;
811         struct rte_ether_addr *mac_addr;
812
813         if (priv->ppio)
814                 return mrvl_dev_set_link_up(dev);
815
816         snprintf(match, sizeof(match), "ppio-%d:%d",
817                  priv->pp_id, priv->ppio_id);
818         priv->ppio_params.match = match;
819         priv->ppio_params.eth_start_hdr = PP2_PPIO_HDR_ETH;
820         priv->forward_bad_frames = 0;
821         priv->fill_bpool_buffs = MRVL_BURST_SIZE;
822
823         if (mrvl_cfg) {
824                 priv->ppio_params.eth_start_hdr =
825                         mrvl_cfg->port[dev->data->port_id].eth_start_hdr;
826                 priv->forward_bad_frames =
827                         mrvl_cfg->port[dev->data->port_id].forward_bad_frames;
828                 priv->fill_bpool_buffs =
829                         mrvl_cfg->port[dev->data->port_id].fill_bpool_buffs;
830         }
831
832         /*
833          * Calculate the minimum bpool size for refill feature as follows:
834          * 2 default burst sizes multiply by number of rx queues.
835          * If the bpool size will be below this value, new buffers will
836          * be added to the pool.
837          */
838         priv->bpool_min_size = priv->nb_rx_queues * MRVL_BURST_SIZE * 2;
839
840         /* In case initial bpool size configured in queues setup is
841          * smaller than minimum size add more buffers
842          */
843         def_init_size = priv->bpool_min_size + MRVL_BURST_SIZE * 2;
844         if (priv->bpool_init_size < def_init_size) {
845                 int buffs_to_add = def_init_size - priv->bpool_init_size;
846
847                 priv->bpool_init_size += buffs_to_add;
848                 ret = mrvl_fill_bpool(dev->data->rx_queues[0], buffs_to_add);
849                 if (ret)
850                         MRVL_LOG(ERR, "Failed to add buffers to bpool");
851         }
852
853         /*
854          * Calculate the maximum bpool size for refill feature as follows:
855          * maximum number of descriptors in rx queue multiply by number
856          * of rx queues plus minimum bpool size.
857          * In case the bpool size will exceed this value, superfluous buffers
858          * will be removed
859          */
860         priv->bpool_max_size = (priv->nb_rx_queues * MRVL_PP2_RXD_MAX) +
861                                 priv->bpool_min_size;
862
863         ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
864         if (ret) {
865                 MRVL_LOG(ERR, "Failed to init ppio");
866                 return ret;
867         }
868
869         /*
870          * In case there are some some stale uc/mc mac addresses flush them
871          * here. It cannot be done during mrvl_dev_close() as port information
872          * is already gone at that point (due to pp2_ppio_deinit() in
873          * mrvl_dev_stop()).
874          */
875         if (!priv->uc_mc_flushed) {
876                 ret = pp2_ppio_flush_mac_addrs(priv->ppio, 1, 1);
877                 if (ret) {
878                         MRVL_LOG(ERR,
879                                 "Failed to flush uc/mc filter list");
880                         goto out;
881                 }
882                 priv->uc_mc_flushed = 1;
883         }
884
885         ret = mrvl_mtu_set(dev, dev->data->mtu);
886         if (ret)
887                 MRVL_LOG(ERR, "Failed to set MTU to %d", dev->data->mtu);
888
889         if (!rte_is_zero_ether_addr(&dev->data->mac_addrs[0]))
890                 mrvl_mac_addr_set(dev, &dev->data->mac_addrs[0]);
891
892         for (i = 1; i < MRVL_MAC_ADDRS_MAX; i++) {
893                 mac_addr = &dev->data->mac_addrs[i];
894
895                 /* skip zero address */
896                 if (rte_is_zero_ether_addr(mac_addr))
897                         continue;
898
899                 mrvl_mac_addr_add(dev, mac_addr, i, 0);
900         }
901
902         if (dev->data->all_multicast == 1)
903                 mrvl_allmulticast_enable(dev);
904
905         if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_VLAN_FILTER) {
906                 ret = mrvl_populate_vlan_table(dev, 1);
907                 if (ret) {
908                         MRVL_LOG(ERR, "Failed to populate VLAN table");
909                         goto out;
910                 }
911         }
912
913         /* For default QoS config, don't start classifier. */
914         if (mrvl_cfg  &&
915             mrvl_cfg->port[dev->data->port_id].use_qos_global_defaults == 0) {
916                 ret = mrvl_start_qos_mapping(priv);
917                 if (ret) {
918                         MRVL_LOG(ERR, "Failed to setup QoS mapping");
919                         goto out;
920                 }
921         }
922
923         ret = pp2_ppio_set_loopback(priv->ppio, dev->data->dev_conf.lpbk_mode);
924         if (ret) {
925                 MRVL_LOG(ERR, "Failed to set loopback");
926                 goto out;
927         }
928
929         if (dev->data->promiscuous == 1)
930                 mrvl_promiscuous_enable(dev);
931
932         if (priv->flow_ctrl) {
933                 ret = mrvl_flow_ctrl_set(dev, &priv->fc_conf);
934                 if (ret) {
935                         MRVL_LOG(ERR, "Failed to configure flow control");
936                         goto out;
937                 }
938                 priv->flow_ctrl = 0;
939         }
940
941         if (dev->data->dev_link.link_status == ETH_LINK_UP) {
942                 ret = mrvl_dev_set_link_up(dev);
943                 if (ret) {
944                         MRVL_LOG(ERR, "Failed to set link up");
945                         dev->data->dev_link.link_status = ETH_LINK_DOWN;
946                         goto out;
947                 }
948         }
949
950         /* start tx queues */
951         for (i = 0; i < dev->data->nb_tx_queues; i++) {
952                 struct mrvl_txq *txq = dev->data->tx_queues[i];
953
954                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
955
956                 if (!txq->tx_deferred_start)
957                         continue;
958
959                 /*
960                  * All txqs are started by default. Stop them
961                  * so that tx_deferred_start works as expected.
962                  */
963                 ret = mrvl_tx_queue_stop(dev, i);
964                 if (ret)
965                         goto out;
966         }
967
968         mrvl_flow_init(dev);
969         mrvl_mtr_init(dev);
970         mrvl_set_tx_function(dev);
971
972         return 0;
973 out:
974         MRVL_LOG(ERR, "Failed to start device");
975         pp2_ppio_deinit(priv->ppio);
976         return ret;
977 }
978
979 /**
980  * Flush receive queues.
981  *
982  * @param dev
983  *   Pointer to Ethernet device structure.
984  */
985 static void
986 mrvl_flush_rx_queues(struct rte_eth_dev *dev)
987 {
988         int i;
989
990         MRVL_LOG(INFO, "Flushing rx queues");
991         for (i = 0; i < dev->data->nb_rx_queues; i++) {
992                 int ret, num;
993
994                 do {
995                         struct mrvl_rxq *q = dev->data->rx_queues[i];
996                         struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];
997
998                         num = MRVL_PP2_RXD_MAX;
999                         ret = pp2_ppio_recv(q->priv->ppio,
1000                                             q->priv->rxq_map[q->queue_id].tc,
1001                                             q->priv->rxq_map[q->queue_id].inq,
1002                                             descs, (uint16_t *)&num);
1003                 } while (ret == 0 && num);
1004         }
1005 }
1006
1007 /**
1008  * Flush transmit shadow queues.
1009  *
1010  * @param dev
1011  *   Pointer to Ethernet device structure.
1012  */
1013 static void
1014 mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
1015 {
1016         int i, j;
1017         struct mrvl_txq *txq;
1018
1019         MRVL_LOG(INFO, "Flushing tx shadow queues");
1020         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1021                 txq = (struct mrvl_txq *)dev->data->tx_queues[i];
1022
1023                 for (j = 0; j < RTE_MAX_LCORE; j++) {
1024                         struct mrvl_shadow_txq *sq;
1025
1026                         if (!hifs[j])
1027                                 continue;
1028
1029                         sq = &txq->shadow_txqs[j];
1030                         mrvl_free_sent_buffers(txq->priv->ppio,
1031                                 hifs[j], j, sq, txq->queue_id, 1);
1032                         while (sq->tail != sq->head) {
1033                                 uint64_t addr = cookie_addr_high |
1034                                         sq->ent[sq->tail].buff.cookie;
1035                                 rte_pktmbuf_free(
1036                                         (struct rte_mbuf *)addr);
1037                                 sq->tail = (sq->tail + 1) &
1038                                             MRVL_PP2_TX_SHADOWQ_MASK;
1039                         }
1040                         memset(sq, 0, sizeof(*sq));
1041                 }
1042         }
1043 }
1044
1045 /**
1046  * Flush hardware bpool (buffer-pool).
1047  *
1048  * @param dev
1049  *   Pointer to Ethernet device structure.
1050  */
1051 static void
1052 mrvl_flush_bpool(struct rte_eth_dev *dev)
1053 {
1054         struct mrvl_priv *priv = dev->data->dev_private;
1055         struct pp2_hif *hif;
1056         uint32_t num;
1057         int ret;
1058         unsigned int core_id = rte_lcore_id();
1059
1060         if (core_id == LCORE_ID_ANY)
1061                 core_id = rte_get_main_lcore();
1062
1063         hif = mrvl_get_hif(priv, core_id);
1064
1065         ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
1066         if (ret) {
1067                 MRVL_LOG(ERR, "Failed to get bpool buffers number");
1068                 return;
1069         }
1070
1071         while (num--) {
1072                 struct pp2_buff_inf inf;
1073                 uint64_t addr;
1074
1075                 ret = pp2_bpool_get_buff(hif, priv->bpool, &inf);
1076                 if (ret)
1077                         break;
1078
1079                 addr = cookie_addr_high | inf.cookie;
1080                 rte_pktmbuf_free((struct rte_mbuf *)addr);
1081         }
1082 }
1083
1084 /**
1085  * DPDK callback to stop the device.
1086  *
1087  * @param dev
1088  *   Pointer to Ethernet device structure.
1089  */
1090 static int
1091 mrvl_dev_stop(struct rte_eth_dev *dev)
1092 {
1093         return mrvl_dev_set_link_down(dev);
1094 }
1095
1096 /**
1097  * DPDK callback to close the device.
1098  *
1099  * @param dev
1100  *   Pointer to Ethernet device structure.
1101  */
1102 static int
1103 mrvl_dev_close(struct rte_eth_dev *dev)
1104 {
1105         struct mrvl_priv *priv = dev->data->dev_private;
1106         size_t i;
1107
1108         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1109                 return 0;
1110
1111         mrvl_flush_rx_queues(dev);
1112         mrvl_flush_tx_shadow_queues(dev);
1113         mrvl_flow_deinit(dev);
1114         mrvl_mtr_deinit(dev);
1115
1116         for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
1117                 struct pp2_ppio_tc_params *tc_params =
1118                         &priv->ppio_params.inqs_params.tcs_params[i];
1119
1120                 if (tc_params->inqs_params) {
1121                         rte_free(tc_params->inqs_params);
1122                         tc_params->inqs_params = NULL;
1123                 }
1124         }
1125
1126         if (priv->cls_tbl) {
1127                 pp2_cls_tbl_deinit(priv->cls_tbl);
1128                 priv->cls_tbl = NULL;
1129         }
1130
1131         if (priv->qos_tbl) {
1132                 pp2_cls_qos_tbl_deinit(priv->qos_tbl);
1133                 priv->qos_tbl = NULL;
1134         }
1135
1136         mrvl_flush_bpool(dev);
1137         mrvl_tm_deinit(dev);
1138
1139         if (priv->ppio) {
1140                 pp2_ppio_deinit(priv->ppio);
1141                 priv->ppio = NULL;
1142         }
1143
1144         /* policer must be released after ppio deinitialization */
1145         if (priv->default_policer) {
1146                 pp2_cls_plcr_deinit(priv->default_policer);
1147                 priv->default_policer = NULL;
1148         }
1149
1150
1151         if (priv->bpool) {
1152                 pp2_bpool_deinit(priv->bpool);
1153                 used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
1154                 priv->bpool = NULL;
1155         }
1156
1157         mrvl_dev_num--;
1158
1159         if (mrvl_dev_num == 0) {
1160                 MRVL_LOG(INFO, "Perform MUSDK deinit");
1161                 mrvl_deinit_hifs();
1162                 mrvl_deinit_pp2();
1163                 rte_mvep_deinit(MVEP_MOD_T_PP2);
1164         }
1165
1166         return 0;
1167 }
1168
1169 /**
1170  * DPDK callback to retrieve physical link information.
1171  *
1172  * @param dev
1173  *   Pointer to Ethernet device structure.
1174  * @param wait_to_complete
1175  *   Wait for request completion (ignored).
1176  *
1177  * @return
1178  *   0 on success, negative error value otherwise.
1179  */
1180 static int
1181 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1182 {
1183         /*
1184          * TODO
1185          * once MUSDK provides necessary API use it here
1186          */
1187         struct mrvl_priv *priv = dev->data->dev_private;
1188         struct ethtool_cmd edata;
1189         struct ifreq req;
1190         int ret, fd, link_up;
1191
1192         if (!priv->ppio)
1193                 return -EPERM;
1194
1195         edata.cmd = ETHTOOL_GSET;
1196
1197         strcpy(req.ifr_name, dev->data->name);
1198         req.ifr_data = (void *)&edata;
1199
1200         fd = socket(AF_INET, SOCK_DGRAM, 0);
1201         if (fd == -1)
1202                 return -EFAULT;
1203
1204         ret = ioctl(fd, SIOCETHTOOL, &req);
1205         if (ret == -1) {
1206                 close(fd);
1207                 return -EFAULT;
1208         }
1209
1210         close(fd);
1211
1212         switch (ethtool_cmd_speed(&edata)) {
1213         case SPEED_10:
1214                 dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
1215                 break;
1216         case SPEED_100:
1217                 dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
1218                 break;
1219         case SPEED_1000:
1220                 dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
1221                 break;
1222         case SPEED_2500:
1223                 dev->data->dev_link.link_speed = ETH_SPEED_NUM_2_5G;
1224                 break;
1225         case SPEED_10000:
1226                 dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
1227                 break;
1228         default:
1229                 dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
1230         }
1231
1232         dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
1233                                                          ETH_LINK_HALF_DUPLEX;
1234         dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
1235                                                            ETH_LINK_FIXED;
1236         pp2_ppio_get_link_state(priv->ppio, &link_up);
1237         dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
1238
1239         return 0;
1240 }
1241
1242 /**
1243  * DPDK callback to enable promiscuous mode.
1244  *
1245  * @param dev
1246  *   Pointer to Ethernet device structure.
1247  *
1248  * @return
1249  *   0 on success, negative error value otherwise.
1250  */
1251 static int
1252 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
1253 {
1254         struct mrvl_priv *priv = dev->data->dev_private;
1255         int ret;
1256
1257         if (priv->isolated)
1258                 return -ENOTSUP;
1259
1260         if (!priv->ppio)
1261                 return 0;
1262
1263         ret = pp2_ppio_set_promisc(priv->ppio, 1);
1264         if (ret) {
1265                 MRVL_LOG(ERR, "Failed to enable promiscuous mode");
1266                 return -EAGAIN;
1267         }
1268
1269         return 0;
1270 }
1271
1272 /**
1273  * DPDK callback to enable allmulti mode.
1274  *
1275  * @param dev
1276  *   Pointer to Ethernet device structure.
1277  *
1278  * @return
1279  *   0 on success, negative error value otherwise.
1280  */
1281 static int
1282 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
1283 {
1284         struct mrvl_priv *priv = dev->data->dev_private;
1285         int ret;
1286
1287         if (priv->isolated)
1288                 return -ENOTSUP;
1289
1290         if (!priv->ppio)
1291                 return 0;
1292
1293         ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
1294         if (ret) {
1295                 MRVL_LOG(ERR, "Failed enable all-multicast mode");
1296                 return -EAGAIN;
1297         }
1298
1299         return 0;
1300 }
1301
1302 /**
1303  * DPDK callback to disable promiscuous mode.
1304  *
1305  * @param dev
1306  *   Pointer to Ethernet device structure.
1307  *
1308  * @return
1309  *   0 on success, negative error value otherwise.
1310  */
1311 static int
1312 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
1313 {
1314         struct mrvl_priv *priv = dev->data->dev_private;
1315         int ret;
1316
1317         if (priv->isolated)
1318                 return -ENOTSUP;
1319
1320         if (!priv->ppio)
1321                 return 0;
1322
1323         ret = pp2_ppio_set_promisc(priv->ppio, 0);
1324         if (ret) {
1325                 MRVL_LOG(ERR, "Failed to disable promiscuous mode");
1326                 return -EAGAIN;
1327         }
1328
1329         return 0;
1330 }
1331
1332 /**
1333  * DPDK callback to disable allmulticast mode.
1334  *
1335  * @param dev
1336  *   Pointer to Ethernet device structure.
1337  *
1338  * @return
1339  *   0 on success, negative error value otherwise.
1340  */
1341 static int
1342 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
1343 {
1344         struct mrvl_priv *priv = dev->data->dev_private;
1345         int ret;
1346
1347         if (priv->isolated)
1348                 return -ENOTSUP;
1349
1350         if (!priv->ppio)
1351                 return 0;
1352
1353         ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
1354         if (ret) {
1355                 MRVL_LOG(ERR, "Failed to disable all-multicast mode");
1356                 return -EAGAIN;
1357         }
1358
1359         return 0;
1360 }
1361
1362 /**
1363  * DPDK callback to remove a MAC address.
1364  *
1365  * @param dev
1366  *   Pointer to Ethernet device structure.
1367  * @param index
1368  *   MAC address index.
1369  */
1370 static void
1371 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1372 {
1373         struct mrvl_priv *priv = dev->data->dev_private;
1374         char buf[RTE_ETHER_ADDR_FMT_SIZE];
1375         int ret;
1376
1377         if (priv->isolated)
1378                 return;
1379
1380         if (!priv->ppio)
1381                 return;
1382
1383         ret = pp2_ppio_remove_mac_addr(priv->ppio,
1384                                        dev->data->mac_addrs[index].addr_bytes);
1385         if (ret) {
1386                 rte_ether_format_addr(buf, sizeof(buf),
1387                                   &dev->data->mac_addrs[index]);
1388                 MRVL_LOG(ERR, "Failed to remove mac %s", buf);
1389         }
1390 }
1391
1392 /**
1393  * DPDK callback to add a MAC address.
1394  *
1395  * @param dev
1396  *   Pointer to Ethernet device structure.
1397  * @param mac_addr
1398  *   MAC address to register.
1399  * @param index
1400  *   MAC address index.
1401  * @param vmdq
1402  *   VMDq pool index to associate address with (unused).
1403  *
1404  * @return
1405  *   0 on success, negative error value otherwise.
1406  */
1407 static int
1408 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1409                   uint32_t index, uint32_t vmdq __rte_unused)
1410 {
1411         struct mrvl_priv *priv = dev->data->dev_private;
1412         char buf[RTE_ETHER_ADDR_FMT_SIZE];
1413         int ret;
1414
1415         if (priv->isolated)
1416                 return -ENOTSUP;
1417
1418         if (!priv->ppio)
1419                 return 0;
1420
1421         if (index == 0)
1422                 /* For setting index 0, mrvl_mac_addr_set() should be used.*/
1423                 return -1;
1424
1425         /*
1426          * Maximum number of uc addresses can be tuned via kernel module mvpp2x
1427          * parameter uc_filter_max. Maximum number of mc addresses is then
1428          * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
1429          * 21 respectively.
1430          *
1431          * If more than uc_filter_max uc addresses were added to filter list
1432          * then NIC will switch to promiscuous mode automatically.
1433          *
1434          * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
1435          * were added to filter list then NIC will switch to all-multicast mode
1436          * automatically.
1437          */
1438         ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
1439         if (ret) {
1440                 rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1441                 MRVL_LOG(ERR, "Failed to add mac %s", buf);
1442                 return -1;
1443         }
1444
1445         return 0;
1446 }
1447
1448 /**
1449  * DPDK callback to set the primary MAC address.
1450  *
1451  * @param dev
1452  *   Pointer to Ethernet device structure.
1453  * @param mac_addr
1454  *   MAC address to register.
1455  *
1456  * @return
1457  *   0 on success, negative error value otherwise.
1458  */
1459 static int
1460 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1461 {
1462         struct mrvl_priv *priv = dev->data->dev_private;
1463         int ret;
1464
1465         if (priv->isolated)
1466                 return -ENOTSUP;
1467
1468         if (!priv->ppio)
1469                 return 0;
1470
1471         ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
1472         if (ret) {
1473                 char buf[RTE_ETHER_ADDR_FMT_SIZE];
1474                 rte_ether_format_addr(buf, sizeof(buf), mac_addr);
1475                 MRVL_LOG(ERR, "Failed to set mac to %s", buf);
1476         }
1477
1478         return ret;
1479 }
1480
1481 /**
1482  * DPDK callback to get device statistics.
1483  *
1484  * @param dev
1485  *   Pointer to Ethernet device structure.
1486  * @param stats
1487  *   Stats structure output buffer.
1488  *
1489  * @return
1490  *   0 on success, negative error value otherwise.
1491  */
1492 static int
1493 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1494 {
1495         struct mrvl_priv *priv = dev->data->dev_private;
1496         struct pp2_ppio_statistics ppio_stats;
1497         uint64_t drop_mac = 0;
1498         unsigned int i, idx, ret;
1499
1500         if (!priv->ppio)
1501                 return -EPERM;
1502
1503         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1504                 struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1505                 struct pp2_ppio_inq_statistics rx_stats;
1506
1507                 if (!rxq)
1508                         continue;
1509
1510                 idx = rxq->queue_id;
1511                 if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1512                         MRVL_LOG(ERR,
1513                                 "rx queue %d stats out of range (0 - %d)",
1514                                 idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1515                         continue;
1516                 }
1517
1518                 ret = pp2_ppio_inq_get_statistics(priv->ppio,
1519                                                   priv->rxq_map[idx].tc,
1520                                                   priv->rxq_map[idx].inq,
1521                                                   &rx_stats, 0);
1522                 if (unlikely(ret)) {
1523                         MRVL_LOG(ERR,
1524                                 "Failed to update rx queue %d stats", idx);
1525                         break;
1526                 }
1527
1528                 stats->q_ibytes[idx] = rxq->bytes_recv;
1529                 stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
1530                 stats->q_errors[idx] = rx_stats.drop_early +
1531                                        rx_stats.drop_fullq +
1532                                        rx_stats.drop_bm +
1533                                        rxq->drop_mac;
1534                 stats->ibytes += rxq->bytes_recv;
1535                 drop_mac += rxq->drop_mac;
1536         }
1537
1538         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1539                 struct mrvl_txq *txq = dev->data->tx_queues[i];
1540                 struct pp2_ppio_outq_statistics tx_stats;
1541
1542                 if (!txq)
1543                         continue;
1544
1545                 idx = txq->queue_id;
1546                 if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
1547                         MRVL_LOG(ERR,
1548                                 "tx queue %d stats out of range (0 - %d)",
1549                                 idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
1550                 }
1551
1552                 ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
1553                                                    &tx_stats, 0);
1554                 if (unlikely(ret)) {
1555                         MRVL_LOG(ERR,
1556                                 "Failed to update tx queue %d stats", idx);
1557                         break;
1558                 }
1559
1560                 stats->q_opackets[idx] = tx_stats.deq_desc;
1561                 stats->q_obytes[idx] = txq->bytes_sent;
1562                 stats->obytes += txq->bytes_sent;
1563         }
1564
1565         ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1566         if (unlikely(ret)) {
1567                 MRVL_LOG(ERR, "Failed to update port statistics");
1568                 return ret;
1569         }
1570
1571         stats->ipackets += ppio_stats.rx_packets - drop_mac;
1572         stats->opackets += ppio_stats.tx_packets;
1573         stats->imissed += ppio_stats.rx_fullq_dropped +
1574                           ppio_stats.rx_bm_dropped +
1575                           ppio_stats.rx_early_dropped +
1576                           ppio_stats.rx_fifo_dropped +
1577                           ppio_stats.rx_cls_dropped;
1578         stats->ierrors = drop_mac;
1579
1580         return 0;
1581 }
1582
1583 /**
1584  * DPDK callback to clear device statistics.
1585  *
1586  * @param dev
1587  *   Pointer to Ethernet device structure.
1588  *
1589  * @return
1590  *   0 on success, negative error value otherwise.
1591  */
1592 static int
1593 mrvl_stats_reset(struct rte_eth_dev *dev)
1594 {
1595         struct mrvl_priv *priv = dev->data->dev_private;
1596         int i;
1597
1598         if (!priv->ppio)
1599                 return 0;
1600
1601         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1602                 struct mrvl_rxq *rxq = dev->data->rx_queues[i];
1603
1604                 pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
1605                                             priv->rxq_map[i].inq, NULL, 1);
1606                 rxq->bytes_recv = 0;
1607                 rxq->drop_mac = 0;
1608         }
1609
1610         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1611                 struct mrvl_txq *txq = dev->data->tx_queues[i];
1612
1613                 pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1614                 txq->bytes_sent = 0;
1615         }
1616
1617         return pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1618 }
1619
1620 /**
1621  * DPDK callback to get extended statistics.
1622  *
1623  * @param dev
1624  *   Pointer to Ethernet device structure.
1625  * @param stats
1626  *   Pointer to xstats table.
1627  * @param n
1628  *   Number of entries in xstats table.
1629  * @return
1630  *   Negative value on error, number of read xstats otherwise.
1631  */
1632 static int
1633 mrvl_xstats_get(struct rte_eth_dev *dev,
1634                 struct rte_eth_xstat *stats, unsigned int n)
1635 {
1636         struct mrvl_priv *priv = dev->data->dev_private;
1637         struct pp2_ppio_statistics ppio_stats;
1638         unsigned int i;
1639
1640         if (!stats)
1641                 return 0;
1642
1643         pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
1644         for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
1645                 uint64_t val;
1646
1647                 if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
1648                         val = *(uint32_t *)((uint8_t *)&ppio_stats +
1649                                             mrvl_xstats_tbl[i].offset);
1650                 else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
1651                         val = *(uint64_t *)((uint8_t *)&ppio_stats +
1652                                             mrvl_xstats_tbl[i].offset);
1653                 else
1654                         return -EINVAL;
1655
1656                 stats[i].id = i;
1657                 stats[i].value = val;
1658         }
1659
1660         return n;
1661 }
1662
1663 /**
1664  * DPDK callback to reset extended statistics.
1665  *
1666  * @param dev
1667  *   Pointer to Ethernet device structure.
1668  *
1669  * @return
1670  *   0 on success, negative error value otherwise.
1671  */
1672 static int
1673 mrvl_xstats_reset(struct rte_eth_dev *dev)
1674 {
1675         return mrvl_stats_reset(dev);
1676 }
1677
1678 /**
1679  * DPDK callback to get extended statistics names.
1680  *
1681  * @param dev (unused)
1682  *   Pointer to Ethernet device structure.
1683  * @param xstats_names
1684  *   Pointer to xstats names table.
1685  * @param size
1686  *   Size of the xstats names table.
1687  * @return
1688  *   Number of read names.
1689  */
1690 static int
1691 mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
1692                       struct rte_eth_xstat_name *xstats_names,
1693                       unsigned int size)
1694 {
1695         unsigned int i;
1696
1697         if (!xstats_names)
1698                 return RTE_DIM(mrvl_xstats_tbl);
1699
1700         for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
1701                 strlcpy(xstats_names[i].name, mrvl_xstats_tbl[i].name,
1702                         RTE_ETH_XSTATS_NAME_SIZE);
1703
1704         return size;
1705 }
1706
1707 /**
1708  * DPDK callback to get information about the device.
1709  *
1710  * @param dev
1711  *   Pointer to Ethernet device structure (unused).
1712  * @param info
1713  *   Info structure output buffer.
1714  */
1715 static int
1716 mrvl_dev_infos_get(struct rte_eth_dev *dev,
1717                    struct rte_eth_dev_info *info)
1718 {
1719         struct mrvl_priv *priv = dev->data->dev_private;
1720
1721         info->speed_capa = ETH_LINK_SPEED_10M |
1722                            ETH_LINK_SPEED_100M |
1723                            ETH_LINK_SPEED_1G |
1724                            ETH_LINK_SPEED_2_5G |
1725                            ETH_LINK_SPEED_10G;
1726
1727         info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1728         info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1729         info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1730
1731         info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1732         info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1733         info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1734
1735         info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1736         info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1737         info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1738
1739         info->rx_offload_capa = MRVL_RX_OFFLOADS;
1740         info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;
1741
1742         info->tx_offload_capa = MRVL_TX_OFFLOADS;
1743         info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;
1744
1745         info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1746                                        ETH_RSS_NONFRAG_IPV4_TCP |
1747                                        ETH_RSS_NONFRAG_IPV4_UDP;
1748
1749         /* By default packets are dropped if no descriptors are available */
1750         info->default_rxconf.rx_drop_en = 1;
1751
1752         info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1753         info->max_mtu = priv->max_mtu;
1754
1755         return 0;
1756 }
1757
1758 /**
1759  * Return supported packet types.
1760  *
1761  * @param dev
1762  *   Pointer to Ethernet device structure (unused).
1763  *
1764  * @return
1765  *   Const pointer to the table with supported packet types.
1766  */
1767 static const uint32_t *
1768 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1769 {
1770         static const uint32_t ptypes[] = {
1771                 RTE_PTYPE_L2_ETHER,
1772                 RTE_PTYPE_L2_ETHER_VLAN,
1773                 RTE_PTYPE_L2_ETHER_QINQ,
1774                 RTE_PTYPE_L3_IPV4,
1775                 RTE_PTYPE_L3_IPV4_EXT,
1776                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1777                 RTE_PTYPE_L3_IPV6,
1778                 RTE_PTYPE_L3_IPV6_EXT,
1779                 RTE_PTYPE_L2_ETHER_ARP,
1780                 RTE_PTYPE_L4_TCP,
1781                 RTE_PTYPE_L4_UDP
1782         };
1783
1784         return ptypes;
1785 }
1786
1787 /**
1788  * DPDK callback to get information about specific receive queue.
1789  *
1790  * @param dev
1791  *   Pointer to Ethernet device structure.
1792  * @param rx_queue_id
1793  *   Receive queue index.
1794  * @param qinfo
1795  *   Receive queue information structure.
1796  */
1797 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1798                               struct rte_eth_rxq_info *qinfo)
1799 {
1800         struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1801         struct mrvl_priv *priv = dev->data->dev_private;
1802         int inq = priv->rxq_map[rx_queue_id].inq;
1803         int tc = priv->rxq_map[rx_queue_id].tc;
1804         struct pp2_ppio_tc_params *tc_params =
1805                 &priv->ppio_params.inqs_params.tcs_params[tc];
1806
1807         qinfo->mp = q->mp;
1808         qinfo->nb_desc = tc_params->inqs_params[inq].size;
1809 }
1810
1811 /**
1812  * DPDK callback to get information about specific transmit queue.
1813  *
1814  * @param dev
1815  *   Pointer to Ethernet device structure.
1816  * @param tx_queue_id
1817  *   Transmit queue index.
1818  * @param qinfo
1819  *   Transmit queue information structure.
1820  */
1821 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1822                               struct rte_eth_txq_info *qinfo)
1823 {
1824         struct mrvl_priv *priv = dev->data->dev_private;
1825         struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];
1826
1827         qinfo->nb_desc =
1828                 priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1829         qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
1830 }
1831
1832 /**
1833  * DPDK callback to Configure a VLAN filter.
1834  *
1835  * @param dev
1836  *   Pointer to Ethernet device structure.
1837  * @param vlan_id
1838  *   VLAN ID to filter.
1839  * @param on
1840  *   Toggle filter.
1841  *
1842  * @return
1843  *   0 on success, negative error value otherwise.
1844  */
1845 static int
1846 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1847 {
1848         struct mrvl_priv *priv = dev->data->dev_private;
1849
1850         if (priv->isolated)
1851                 return -ENOTSUP;
1852
1853         if (!priv->ppio)
1854                 return 0;
1855
1856         return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1857                     pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1858 }
1859
1860 /**
1861  * DPDK callback to Configure VLAN offload.
1862  *
1863  * @param dev
1864  *   Pointer to Ethernet device structure.
1865  * @param mask
1866  *   VLAN offload mask.
1867  *
1868  * @return
1869  *   0 on success, negative error value otherwise.
1870  */
1871 static int mrvl_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1872 {
1873         uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
1874         int ret;
1875
1876         if (mask & ETH_VLAN_STRIP_MASK) {
1877                 MRVL_LOG(ERR, "VLAN stripping is not supported\n");
1878                 return -ENOTSUP;
1879         }
1880
1881         if (mask & ETH_VLAN_FILTER_MASK) {
1882                 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1883                         ret = mrvl_populate_vlan_table(dev, 1);
1884                 else
1885                         ret = mrvl_populate_vlan_table(dev, 0);
1886
1887                 if (ret)
1888                         return ret;
1889         }
1890
1891         if (mask & ETH_VLAN_EXTEND_MASK) {
1892                 MRVL_LOG(ERR, "Extend VLAN not supported\n");
1893                 return -ENOTSUP;
1894         }
1895
1896         return 0;
1897 }
1898
1899 /**
1900  * Release buffers to hardware bpool (buffer-pool)
1901  *
1902  * @param rxq
1903  *   Receive queue pointer.
1904  * @param num
1905  *   Number of buffers to release to bpool.
1906  *
1907  * @return
1908  *   0 on success, negative error value otherwise.
1909  */
1910 static int
1911 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1912 {
1913         struct buff_release_entry entries[num];
1914         struct rte_mbuf *mbufs[num];
1915         int i, ret;
1916         unsigned int core_id;
1917         struct pp2_hif *hif;
1918         struct pp2_bpool *bpool;
1919
1920         core_id = rte_lcore_id();
1921         if (core_id == LCORE_ID_ANY)
1922                 core_id = rte_get_main_lcore();
1923
1924         hif = mrvl_get_hif(rxq->priv, core_id);
1925         if (!hif)
1926                 return -1;
1927
1928         bpool = rxq->priv->bpool;
1929
1930         ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1931         if (ret)
1932                 return ret;
1933
1934         if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1935                 cookie_addr_high =
1936                         (uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1937
1938         for (i = 0; i < num; i++) {
1939                 if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1940                         != cookie_addr_high) {
1941                         MRVL_LOG(ERR,
1942                                 "mbuf virtual addr high is out of range "
1943                                 "0x%x instead of 0x%x\n",
1944                                 (uint32_t)((uint64_t)mbufs[i] >> 32),
1945                                 (uint32_t)(cookie_addr_high >> 32));
1946                         goto out;
1947                 }
1948
1949                 entries[i].buff.addr =
1950                         rte_mbuf_data_iova_default(mbufs[i]);
1951                 entries[i].buff.cookie = (uintptr_t)mbufs[i];
1952                 entries[i].bpool = bpool;
1953         }
1954
1955         pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1956         mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1957
1958         if (i != num)
1959                 goto out;
1960
1961         return 0;
1962 out:
1963         for (; i < num; i++)
1964                 rte_pktmbuf_free(mbufs[i]);
1965
1966         return -1;
1967 }
1968
1969 /**
1970  * DPDK callback to configure the receive queue.
1971  *
1972  * @param dev
1973  *   Pointer to Ethernet device structure.
1974  * @param idx
1975  *   RX queue index.
1976  * @param desc
1977  *   Number of descriptors to configure in queue.
1978  * @param socket
1979  *   NUMA socket on which memory must be allocated.
1980  * @param conf
1981  *   Thresholds parameters.
1982  * @param mp
1983  *   Memory pool for buffer allocations.
1984  *
1985  * @return
1986  *   0 on success, negative error value otherwise.
1987  */
1988 static int
1989 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1990                     unsigned int socket,
1991                     const struct rte_eth_rxconf *conf,
1992                     struct rte_mempool *mp)
1993 {
1994         struct mrvl_priv *priv = dev->data->dev_private;
1995         struct mrvl_rxq *rxq;
1996         uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
1997         uint32_t max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1998         int ret, tc, inq;
1999         uint64_t offloads;
2000
2001         offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;
2002
2003         if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
2004                 /*
2005                  * Unknown TC mapping, mapping will not have a correct queue.
2006                  */
2007                 MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
2008                         idx, priv->ppio_id);
2009                 return -EFAULT;
2010         }
2011
2012         frame_size = buf_size - RTE_PKTMBUF_HEADROOM -
2013                      MRVL_PKT_EFFEC_OFFS + RTE_ETHER_CRC_LEN;
2014         if (frame_size < max_rx_pkt_len) {
2015                 MRVL_LOG(WARNING,
2016                         "Mbuf size must be increased to %u bytes to hold up "
2017                         "to %u bytes of data.",
2018                         buf_size + max_rx_pkt_len - frame_size,
2019                         max_rx_pkt_len);
2020                 dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
2021                 MRVL_LOG(INFO, "Setting max rx pkt len to %u",
2022                         dev->data->dev_conf.rxmode.max_rx_pkt_len);
2023         }
2024
2025         if (dev->data->rx_queues[idx]) {
2026                 rte_free(dev->data->rx_queues[idx]);
2027                 dev->data->rx_queues[idx] = NULL;
2028         }
2029
2030         rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
2031         if (!rxq)
2032                 return -ENOMEM;
2033
2034         rxq->priv = priv;
2035         rxq->mp = mp;
2036         rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
2037         rxq->queue_id = idx;
2038         rxq->port_id = dev->data->port_id;
2039         mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
2040
2041         tc = priv->rxq_map[rxq->queue_id].tc,
2042         inq = priv->rxq_map[rxq->queue_id].inq;
2043         priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
2044                 desc;
2045
2046         ret = mrvl_fill_bpool(rxq, desc);
2047         if (ret) {
2048                 rte_free(rxq);
2049                 return ret;
2050         }
2051
2052         priv->bpool_init_size += desc;
2053
2054         dev->data->rx_queues[idx] = rxq;
2055
2056         return 0;
2057 }
2058
2059 /**
2060  * DPDK callback to release the receive queue.
2061  *
2062  * @param rxq
2063  *   Generic receive queue pointer.
2064  */
2065 static void
2066 mrvl_rx_queue_release(void *rxq)
2067 {
2068         struct mrvl_rxq *q = rxq;
2069         struct pp2_ppio_tc_params *tc_params;
2070         int i, num, tc, inq;
2071         struct pp2_hif *hif;
2072         unsigned int core_id = rte_lcore_id();
2073
2074         if (core_id == LCORE_ID_ANY)
2075                 core_id = rte_get_main_lcore();
2076
2077         if (!q)
2078                 return;
2079
2080         hif = mrvl_get_hif(q->priv, core_id);
2081
2082         if (!hif)
2083                 return;
2084
2085         tc = q->priv->rxq_map[q->queue_id].tc;
2086         inq = q->priv->rxq_map[q->queue_id].inq;
2087         tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
2088         num = tc_params->inqs_params[inq].size;
2089         for (i = 0; i < num; i++) {
2090                 struct pp2_buff_inf inf;
2091                 uint64_t addr;
2092
2093                 pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
2094                 addr = cookie_addr_high | inf.cookie;
2095                 rte_pktmbuf_free((struct rte_mbuf *)addr);
2096         }
2097
2098         rte_free(q);
2099 }
2100
2101 /**
2102  * DPDK callback to configure the transmit queue.
2103  *
2104  * @param dev
2105  *   Pointer to Ethernet device structure.
2106  * @param idx
2107  *   Transmit queue index.
2108  * @param desc
2109  *   Number of descriptors to configure in the queue.
2110  * @param socket
2111  *   NUMA socket on which memory must be allocated.
2112  * @param conf
2113  *   Tx queue configuration parameters.
2114  *
2115  * @return
2116  *   0 on success, negative error value otherwise.
2117  */
2118 static int
2119 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
2120                     unsigned int socket,
2121                     const struct rte_eth_txconf *conf)
2122 {
2123         struct mrvl_priv *priv = dev->data->dev_private;
2124         struct mrvl_txq *txq;
2125
2126         if (dev->data->tx_queues[idx]) {
2127                 rte_free(dev->data->tx_queues[idx]);
2128                 dev->data->tx_queues[idx] = NULL;
2129         }
2130
2131         txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
2132         if (!txq)
2133                 return -ENOMEM;
2134
2135         txq->priv = priv;
2136         txq->queue_id = idx;
2137         txq->port_id = dev->data->port_id;
2138         txq->tx_deferred_start = conf->tx_deferred_start;
2139         dev->data->tx_queues[idx] = txq;
2140
2141         priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
2142
2143         return 0;
2144 }
2145
2146 /**
2147  * DPDK callback to release the transmit queue.
2148  *
2149  * @param txq
2150  *   Generic transmit queue pointer.
2151  */
2152 static void
2153 mrvl_tx_queue_release(void *txq)
2154 {
2155         struct mrvl_txq *q = txq;
2156
2157         if (!q)
2158                 return;
2159
2160         rte_free(q);
2161 }
2162
2163 /**
2164  * DPDK callback to get flow control configuration.
2165  *
2166  * @param dev
2167  *  Pointer to Ethernet device structure.
2168  * @param fc_conf
2169  *  Pointer to the flow control configuration.
2170  *
2171  * @return
2172  *  0 on success, negative error value otherwise.
2173  */
2174 static int
2175 mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2176 {
2177         struct mrvl_priv *priv = dev->data->dev_private;
2178         int ret, en;
2179
2180         if (!priv->ppio) {
2181                 memcpy(fc_conf, &priv->fc_conf, sizeof(struct rte_eth_fc_conf));
2182                 return 0;
2183         }
2184
2185         fc_conf->autoneg = 1;
2186         ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
2187         if (ret) {
2188                 MRVL_LOG(ERR, "Failed to read rx pause state");
2189                 return ret;
2190         }
2191
2192         fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;
2193
2194         ret = pp2_ppio_get_tx_pause(priv->ppio, &en);
2195         if (ret) {
2196                 MRVL_LOG(ERR, "Failed to read tx pause state");
2197                 return ret;
2198         }
2199
2200         if (en) {
2201                 if (fc_conf->mode == RTE_FC_NONE)
2202                         fc_conf->mode = RTE_FC_TX_PAUSE;
2203                 else
2204                         fc_conf->mode = RTE_FC_FULL;
2205         }
2206
2207         return 0;
2208 }
2209
2210 /**
2211  * DPDK callback to set flow control configuration.
2212  *
2213  * @param dev
2214  *  Pointer to Ethernet device structure.
2215  * @param fc_conf
2216  *  Pointer to the flow control configuration.
2217  *
2218  * @return
2219  *  0 on success, negative error value otherwise.
2220  */
2221 static int
2222 mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2223 {
2224         struct mrvl_priv *priv = dev->data->dev_private;
2225         struct pp2_ppio_tx_pause_params mrvl_pause_params;
2226         int ret;
2227         int rx_en, tx_en;
2228
2229         if (fc_conf->high_water ||
2230             fc_conf->low_water ||
2231             fc_conf->pause_time ||
2232             fc_conf->mac_ctrl_frame_fwd) {
2233                 MRVL_LOG(ERR, "Flowctrl parameter is not supported");
2234
2235                 return -EINVAL;
2236         }
2237
2238         if (fc_conf->autoneg == 0) {
2239                 MRVL_LOG(ERR, "Flowctrl Autoneg disable is not supported");
2240                 return -EINVAL;
2241         }
2242
2243         if (!priv->ppio) {
2244                 memcpy(&priv->fc_conf, fc_conf, sizeof(struct rte_eth_fc_conf));
2245                 priv->flow_ctrl = 1;
2246                 return 0;
2247         }
2248
2249         switch (fc_conf->mode) {
2250         case RTE_FC_FULL:
2251                 rx_en = 1;
2252                 tx_en = 1;
2253                 break;
2254         case RTE_FC_TX_PAUSE:
2255                 rx_en = 0;
2256                 tx_en = 1;
2257                 break;
2258         case RTE_FC_RX_PAUSE:
2259                 rx_en = 1;
2260                 tx_en = 0;
2261                 break;
2262         case RTE_FC_NONE:
2263                 rx_en = 0;
2264                 tx_en = 0;
2265                 break;
2266         default:
2267                 MRVL_LOG(ERR, "Incorrect Flow control flag (%d)",
2268                          fc_conf->mode);
2269                 return -EINVAL;
2270         }
2271
2272         /* Set RX flow control */
2273         ret = pp2_ppio_set_rx_pause(priv->ppio, rx_en);
2274         if (ret) {
2275                 MRVL_LOG(ERR, "Failed to change RX flowctrl");
2276                 return ret;
2277         }
2278
2279         /* Set TX flow control */
2280         mrvl_pause_params.en = tx_en;
2281         /* all inqs participate in xon/xoff decision */
2282         mrvl_pause_params.use_tc_pause_inqs = 0;
2283         ret = pp2_ppio_set_tx_pause(priv->ppio, &mrvl_pause_params);
2284         if (ret) {
2285                 MRVL_LOG(ERR, "Failed to change TX flowctrl");
2286                 return ret;
2287         }
2288
2289         return 0;
2290 }
2291
2292 /**
2293  * Update RSS hash configuration
2294  *
2295  * @param dev
2296  *   Pointer to Ethernet device structure.
2297  * @param rss_conf
2298  *   Pointer to RSS configuration.
2299  *
2300  * @return
2301  *   0 on success, negative error value otherwise.
2302  */
2303 static int
2304 mrvl_rss_hash_update(struct rte_eth_dev *dev,
2305                      struct rte_eth_rss_conf *rss_conf)
2306 {
2307         struct mrvl_priv *priv = dev->data->dev_private;
2308
2309         if (priv->isolated)
2310                 return -ENOTSUP;
2311
2312         return mrvl_configure_rss(priv, rss_conf);
2313 }
2314
2315 /**
2316  * DPDK callback to get RSS hash configuration.
2317  *
2318  * @param dev
2319  *   Pointer to Ethernet device structure.
2320  * @rss_conf
2321  *   Pointer to RSS configuration.
2322  *
2323  * @return
2324  *   Always 0.
2325  */
2326 static int
2327 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
2328                        struct rte_eth_rss_conf *rss_conf)
2329 {
2330         struct mrvl_priv *priv = dev->data->dev_private;
2331         enum pp2_ppio_hash_type hash_type =
2332                 priv->ppio_params.inqs_params.hash_type;
2333
2334         rss_conf->rss_key = NULL;
2335
2336         if (hash_type == PP2_PPIO_HASH_T_NONE)
2337                 rss_conf->rss_hf = 0;
2338         else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
2339                 rss_conf->rss_hf = ETH_RSS_IPV4;
2340         else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
2341                 rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
2342         else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
2343                 rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
2344
2345         return 0;
2346 }
2347
2348 /**
2349  * DPDK callback to get rte_flow callbacks.
2350  *
2351  * @param dev
2352  *   Pointer to the device structure.
2353  * @param ops
2354  *   Pointer to pass the flow ops.
2355  *
2356  * @return
2357  *   0 on success, negative error value otherwise.
2358  */
2359 static int
2360 mrvl_eth_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
2361                       const struct rte_flow_ops **ops)
2362 {
2363         *ops = &mrvl_flow_ops;
2364         return 0;
2365 }
2366
2367 /**
2368  * DPDK callback to get rte_mtr callbacks.
2369  *
2370  * @param dev
2371  *   Pointer to the device structure.
2372  * @param ops
2373  *   Pointer to pass the mtr ops.
2374  *
2375  * @return
2376  *   Always 0.
2377  */
2378 static int
2379 mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2380 {
2381         *(const void **)ops = &mrvl_mtr_ops;
2382
2383         return 0;
2384 }
2385
2386 /**
2387  * DPDK callback to get rte_tm callbacks.
2388  *
2389  * @param dev
2390  *   Pointer to the device structure.
2391  * @param ops
2392  *   Pointer to pass the tm ops.
2393  *
2394  * @return
2395  *   Always 0.
2396  */
2397 static int
2398 mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
2399 {
2400         *(const void **)ops = &mrvl_tm_ops;
2401
2402         return 0;
2403 }
2404
2405 static const struct eth_dev_ops mrvl_ops = {
2406         .dev_configure = mrvl_dev_configure,
2407         .dev_start = mrvl_dev_start,
2408         .dev_stop = mrvl_dev_stop,
2409         .dev_set_link_up = mrvl_dev_set_link_up,
2410         .dev_set_link_down = mrvl_dev_set_link_down,
2411         .dev_close = mrvl_dev_close,
2412         .link_update = mrvl_link_update,
2413         .promiscuous_enable = mrvl_promiscuous_enable,
2414         .allmulticast_enable = mrvl_allmulticast_enable,
2415         .promiscuous_disable = mrvl_promiscuous_disable,
2416         .allmulticast_disable = mrvl_allmulticast_disable,
2417         .mac_addr_remove = mrvl_mac_addr_remove,
2418         .mac_addr_add = mrvl_mac_addr_add,
2419         .mac_addr_set = mrvl_mac_addr_set,
2420         .mtu_set = mrvl_mtu_set,
2421         .stats_get = mrvl_stats_get,
2422         .stats_reset = mrvl_stats_reset,
2423         .xstats_get = mrvl_xstats_get,
2424         .xstats_reset = mrvl_xstats_reset,
2425         .xstats_get_names = mrvl_xstats_get_names,
2426         .dev_infos_get = mrvl_dev_infos_get,
2427         .dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
2428         .rxq_info_get = mrvl_rxq_info_get,
2429         .txq_info_get = mrvl_txq_info_get,
2430         .vlan_filter_set = mrvl_vlan_filter_set,
2431         .vlan_offload_set = mrvl_vlan_offload_set,
2432         .tx_queue_start = mrvl_tx_queue_start,
2433         .tx_queue_stop = mrvl_tx_queue_stop,
2434         .rx_queue_setup = mrvl_rx_queue_setup,
2435         .rx_queue_release = mrvl_rx_queue_release,
2436         .tx_queue_setup = mrvl_tx_queue_setup,
2437         .tx_queue_release = mrvl_tx_queue_release,
2438         .flow_ctrl_get = mrvl_flow_ctrl_get,
2439         .flow_ctrl_set = mrvl_flow_ctrl_set,
2440         .rss_hash_update = mrvl_rss_hash_update,
2441         .rss_hash_conf_get = mrvl_rss_hash_conf_get,
2442         .flow_ops_get = mrvl_eth_flow_ops_get,
2443         .mtr_ops_get = mrvl_mtr_ops_get,
2444         .tm_ops_get = mrvl_tm_ops_get,
2445 };
2446
2447 /**
2448  * Return packet type information and l3/l4 offsets.
2449  *
2450  * @param desc
2451  *   Pointer to the received packet descriptor.
2452  * @param l3_offset
2453  *   l3 packet offset.
2454  * @param l4_offset
2455  *   l4 packet offset.
2456  *
2457  * @return
2458  *   Packet type information.
2459  */
2460 static inline uint64_t
2461 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
2462                                     uint8_t *l3_offset, uint8_t *l4_offset)
2463 {
2464         enum pp2_inq_l3_type l3_type;
2465         enum pp2_inq_l4_type l4_type;
2466         enum pp2_inq_vlan_tag vlan_tag;
2467         uint64_t packet_type;
2468
2469         pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
2470         pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
2471         pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);
2472
2473         packet_type = RTE_PTYPE_L2_ETHER;
2474
2475         switch (vlan_tag) {
2476         case PP2_INQ_VLAN_TAG_SINGLE:
2477                 packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
2478                 break;
2479         case PP2_INQ_VLAN_TAG_DOUBLE:
2480         case PP2_INQ_VLAN_TAG_TRIPLE:
2481                 packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
2482                 break;
2483         default:
2484                 break;
2485         }
2486
2487         switch (l3_type) {
2488         case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
2489                 packet_type |= RTE_PTYPE_L3_IPV4;
2490                 break;
2491         case PP2_INQ_L3_TYPE_IPV4_OK:
2492                 packet_type |= RTE_PTYPE_L3_IPV4_EXT;
2493                 break;
2494         case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
2495                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
2496                 break;
2497         case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
2498                 packet_type |= RTE_PTYPE_L3_IPV6;
2499                 break;
2500         case PP2_INQ_L3_TYPE_IPV6_EXT:
2501                 packet_type |= RTE_PTYPE_L3_IPV6_EXT;
2502                 break;
2503         case PP2_INQ_L3_TYPE_ARP:
2504                 packet_type |= RTE_PTYPE_L2_ETHER_ARP;
2505                 /*
2506                  * In case of ARP l4_offset is set to wrong value.
2507                  * Set it to proper one so that later on mbuf->l3_len can be
2508                  * calculated subtracting l4_offset and l3_offset.
2509                  */
2510                 *l4_offset = *l3_offset + MRVL_ARP_LENGTH;
2511                 break;
2512         default:
2513                 break;
2514         }
2515
2516         switch (l4_type) {
2517         case PP2_INQ_L4_TYPE_TCP:
2518                 packet_type |= RTE_PTYPE_L4_TCP;
2519                 break;
2520         case PP2_INQ_L4_TYPE_UDP:
2521                 packet_type |= RTE_PTYPE_L4_UDP;
2522                 break;
2523         default:
2524                 break;
2525         }
2526
2527         return packet_type;
2528 }
2529
2530 /**
2531  * Get offload information from the received packet descriptor.
2532  *
2533  * @param desc
2534  *   Pointer to the received packet descriptor.
2535  *
2536  * @return
2537  *   Mbuf offload flags.
2538  */
2539 static inline uint64_t
2540 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc, uint64_t packet_type)
2541 {
2542         uint64_t flags = 0;
2543         enum pp2_inq_desc_status status;
2544
2545         if (RTE_ETH_IS_IPV4_HDR(packet_type)) {
2546                 status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
2547                 if (unlikely(status != PP2_DESC_ERR_OK))
2548                         flags |= PKT_RX_IP_CKSUM_BAD;
2549                 else
2550                         flags |= PKT_RX_IP_CKSUM_GOOD;
2551         }
2552
2553         if (((packet_type & RTE_PTYPE_L4_UDP) == RTE_PTYPE_L4_UDP) ||
2554             ((packet_type & RTE_PTYPE_L4_TCP) == RTE_PTYPE_L4_TCP)) {
2555                 status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
2556                 if (unlikely(status != PP2_DESC_ERR_OK))
2557                         flags |= PKT_RX_L4_CKSUM_BAD;
2558                 else
2559                         flags |= PKT_RX_L4_CKSUM_GOOD;
2560         }
2561
2562         return flags;
2563 }
2564
2565 /**
2566  * DPDK callback for receive.
2567  *
2568  * @param rxq
2569  *   Generic pointer to the receive queue.
2570  * @param rx_pkts
2571  *   Array to store received packets.
2572  * @param nb_pkts
2573  *   Maximum number of packets in array.
2574  *
2575  * @return
2576  *   Number of packets successfully received.
2577  */
2578 static uint16_t
2579 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
2580 {
2581         struct mrvl_rxq *q = rxq;
2582         struct pp2_ppio_desc descs[nb_pkts];
2583         struct pp2_bpool *bpool;
2584         int i, ret, rx_done = 0;
2585         int num;
2586         struct pp2_hif *hif;
2587         unsigned int core_id = rte_lcore_id();
2588
2589         hif = mrvl_get_hif(q->priv, core_id);
2590
2591         if (unlikely(!q->priv->ppio || !hif))
2592                 return 0;
2593
2594         bpool = q->priv->bpool;
2595
2596         ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
2597                             q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
2598         if (unlikely(ret < 0))
2599                 return 0;
2600
2601         mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
2602
2603         for (i = 0; i < nb_pkts; i++) {
2604                 struct rte_mbuf *mbuf;
2605                 uint8_t l3_offset, l4_offset;
2606                 enum pp2_inq_desc_status status;
2607                 uint64_t addr;
2608
2609                 if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2610                         struct pp2_ppio_desc *pref_desc;
2611                         u64 pref_addr;
2612
2613                         pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
2614                         pref_addr = cookie_addr_high |
2615                                     pp2_ppio_inq_desc_get_cookie(pref_desc);
2616                         rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
2617                         rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
2618                 }
2619
2620                 addr = cookie_addr_high |
2621                        pp2_ppio_inq_desc_get_cookie(&descs[i]);
2622                 mbuf = (struct rte_mbuf *)addr;
2623                 rte_pktmbuf_reset(mbuf);
2624
2625                 /* drop packet in case of mac, overrun or resource error */
2626                 status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
2627                 if ((unlikely(status != PP2_DESC_ERR_OK)) &&
2628                         !(q->priv->forward_bad_frames)) {
2629                         struct pp2_buff_inf binf = {
2630                                 .addr = rte_mbuf_data_iova_default(mbuf),
2631                                 .cookie = (uint64_t)mbuf,
2632                         };
2633
2634                         pp2_bpool_put_buff(hif, bpool, &binf);
2635                         mrvl_port_bpool_size
2636                                 [bpool->pp2_id][bpool->id][core_id]++;
2637                         q->drop_mac++;
2638                         continue;
2639                 }
2640
2641                 mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
2642                 mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
2643                 mbuf->data_len = mbuf->pkt_len;
2644                 mbuf->port = q->port_id;
2645                 mbuf->packet_type =
2646                         mrvl_desc_to_packet_type_and_offset(&descs[i],
2647                                                             &l3_offset,
2648                                                             &l4_offset);
2649                 mbuf->l2_len = l3_offset;
2650                 mbuf->l3_len = l4_offset - l3_offset;
2651
2652                 if (likely(q->cksum_enabled))
2653                         mbuf->ol_flags =
2654                                 mrvl_desc_to_ol_flags(&descs[i],
2655                                                       mbuf->packet_type);
2656
2657                 rx_pkts[rx_done++] = mbuf;
2658                 q->bytes_recv += mbuf->pkt_len;
2659         }
2660
2661         if (rte_spinlock_trylock(&q->priv->lock) == 1) {
2662                 num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
2663
2664                 if (unlikely(num <= q->priv->bpool_min_size ||
2665                              (!rx_done && num < q->priv->bpool_init_size))) {
2666                         mrvl_fill_bpool(q, q->priv->fill_bpool_buffs);
2667                 } else if (unlikely(num > q->priv->bpool_max_size)) {
2668                         int i;
2669                         int pkt_to_remove = num - q->priv->bpool_init_size;
2670                         struct rte_mbuf *mbuf;
2671                         struct pp2_buff_inf buff;
2672
2673                         for (i = 0; i < pkt_to_remove; i++) {
2674                                 ret = pp2_bpool_get_buff(hif, bpool, &buff);
2675                                 if (ret)
2676                                         break;
2677                                 mbuf = (struct rte_mbuf *)
2678                                         (cookie_addr_high | buff.cookie);
2679                                 rte_pktmbuf_free(mbuf);
2680                         }
2681                         mrvl_port_bpool_size
2682                                 [bpool->pp2_id][bpool->id][core_id] -= i;
2683                 }
2684                 rte_spinlock_unlock(&q->priv->lock);
2685         }
2686
2687         return rx_done;
2688 }
2689
2690 /**
2691  * Prepare offload information.
2692  *
2693  * @param ol_flags
2694  *   Offload flags.
2695  * @param l3_type
2696  *   Pointer to the pp2_ouq_l3_type structure.
2697  * @param l4_type
2698  *   Pointer to the pp2_outq_l4_type structure.
2699  * @param gen_l3_cksum
2700  *   Will be set to 1 in case l3 checksum is computed.
2701  * @param l4_cksum
2702  *   Will be set to 1 in case l4 checksum is computed.
2703  */
2704 static inline void
2705 mrvl_prepare_proto_info(uint64_t ol_flags,
2706                         enum pp2_outq_l3_type *l3_type,
2707                         enum pp2_outq_l4_type *l4_type,
2708                         int *gen_l3_cksum,
2709                         int *gen_l4_cksum)
2710 {
2711         /*
2712          * Based on ol_flags prepare information
2713          * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
2714          * for offloading.
2715          * in most of the checksum cases ipv4 must be set, so this is the
2716          * default value
2717          */
2718         *l3_type = PP2_OUTQ_L3_TYPE_IPV4;
2719         *gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
2720
2721         if (ol_flags & PKT_TX_IPV6) {
2722                 *l3_type = PP2_OUTQ_L3_TYPE_IPV6;
2723                 /* no checksum for ipv6 header */
2724                 *gen_l3_cksum = 0;
2725         }
2726
2727         if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM) {
2728                 *l4_type = PP2_OUTQ_L4_TYPE_TCP;
2729                 *gen_l4_cksum = 1;
2730         } else if ((ol_flags & PKT_TX_L4_MASK) ==  PKT_TX_UDP_CKSUM) {
2731                 *l4_type = PP2_OUTQ_L4_TYPE_UDP;
2732                 *gen_l4_cksum = 1;
2733         } else {
2734                 *l4_type = PP2_OUTQ_L4_TYPE_OTHER;
2735                 /* no checksum for other type */
2736                 *gen_l4_cksum = 0;
2737         }
2738 }
2739
2740 /**
2741  * Release already sent buffers to bpool (buffer-pool).
2742  *
2743  * @param ppio
2744  *   Pointer to the port structure.
2745  * @param hif
2746  *   Pointer to the MUSDK hardware interface.
2747  * @param sq
2748  *   Pointer to the shadow queue.
2749  * @param qid
2750  *   Queue id number.
2751  * @param force
2752  *   Force releasing packets.
2753  */
2754 static inline void
2755 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
2756                        unsigned int core_id, struct mrvl_shadow_txq *sq,
2757                        int qid, int force)
2758 {
2759         struct buff_release_entry *entry;
2760         uint16_t nb_done = 0, num = 0, skip_bufs = 0;
2761         int i;
2762
2763         pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
2764
2765         sq->num_to_release += nb_done;
2766
2767         if (likely(!force &&
2768                    sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
2769                 return;
2770
2771         nb_done = sq->num_to_release;
2772         sq->num_to_release = 0;
2773
2774         for (i = 0; i < nb_done; i++) {
2775                 entry = &sq->ent[sq->tail + num];
2776                 if (unlikely(!entry->buff.addr)) {
2777                         MRVL_LOG(ERR,
2778                                 "Shadow memory @%d: cookie(%lx), pa(%lx)!",
2779                                 sq->tail, (u64)entry->buff.cookie,
2780                                 (u64)entry->buff.addr);
2781                         skip_bufs = 1;
2782                         goto skip;
2783                 }
2784
2785                 if (unlikely(!entry->bpool)) {
2786                         struct rte_mbuf *mbuf;
2787
2788                         mbuf = (struct rte_mbuf *)entry->buff.cookie;
2789                         rte_pktmbuf_free(mbuf);
2790                         skip_bufs = 1;
2791                         goto skip;
2792                 }
2793
2794                 mrvl_port_bpool_size
2795                         [entry->bpool->pp2_id][entry->bpool->id][core_id]++;
2796                 num++;
2797                 if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
2798                         goto skip;
2799                 continue;
2800 skip:
2801                 if (likely(num))
2802                         pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2803                 num += skip_bufs;
2804                 sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2805                 sq->size -= num;
2806                 num = 0;
2807                 skip_bufs = 0;
2808         }
2809
2810         if (likely(num)) {
2811                 pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
2812                 sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
2813                 sq->size -= num;
2814         }
2815 }
2816
2817 /**
2818  * DPDK callback for transmit.
2819  *
2820  * @param txq
2821  *   Generic pointer transmit queue.
2822  * @param tx_pkts
2823  *   Packets to transmit.
2824  * @param nb_pkts
2825  *   Number of packets in array.
2826  *
2827  * @return
2828  *   Number of packets successfully transmitted.
2829  */
2830 static uint16_t
2831 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
2832 {
2833         struct mrvl_txq *q = txq;
2834         struct mrvl_shadow_txq *sq;
2835         struct pp2_hif *hif;
2836         struct pp2_ppio_desc descs[nb_pkts];
2837         unsigned int core_id = rte_lcore_id();
2838         int i, bytes_sent = 0;
2839         uint16_t num, sq_free_size;
2840         uint64_t addr;
2841
2842         hif = mrvl_get_hif(q->priv, core_id);
2843         sq = &q->shadow_txqs[core_id];
2844
2845         if (unlikely(!q->priv->ppio || !hif))
2846                 return 0;
2847
2848         if (sq->size)
2849                 mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2850                                        sq, q->queue_id, 0);
2851
2852         sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2853         if (unlikely(nb_pkts > sq_free_size))
2854                 nb_pkts = sq_free_size;
2855
2856         for (i = 0; i < nb_pkts; i++) {
2857                 struct rte_mbuf *mbuf = tx_pkts[i];
2858                 int gen_l3_cksum, gen_l4_cksum;
2859                 enum pp2_outq_l3_type l3_type;
2860                 enum pp2_outq_l4_type l4_type;
2861
2862                 if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2863                         struct rte_mbuf *pref_pkt_hdr;
2864
2865                         pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2866                         rte_mbuf_prefetch_part1(pref_pkt_hdr);
2867                         rte_mbuf_prefetch_part2(pref_pkt_hdr);
2868                 }
2869
2870                 mrvl_fill_shadowq(sq, mbuf);
2871                 mrvl_fill_desc(&descs[i], mbuf);
2872
2873                 bytes_sent += rte_pktmbuf_pkt_len(mbuf);
2874                 /*
2875                  * in case unsupported ol_flags were passed
2876                  * do not update descriptor offload information
2877                  */
2878                 if (!(mbuf->ol_flags & MRVL_TX_PKT_OFFLOADS))
2879                         continue;
2880                 mrvl_prepare_proto_info(mbuf->ol_flags, &l3_type, &l4_type,
2881                                         &gen_l3_cksum, &gen_l4_cksum);
2882
2883                 pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
2884                                                   mbuf->l2_len,
2885                                                   mbuf->l2_len + mbuf->l3_len,
2886                                                   gen_l3_cksum, gen_l4_cksum);
2887         }
2888
2889         num = nb_pkts;
2890         pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
2891         /* number of packets that were not sent */
2892         if (unlikely(num > nb_pkts)) {
2893                 for (i = nb_pkts; i < num; i++) {
2894                         sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
2895                                 MRVL_PP2_TX_SHADOWQ_MASK;
2896                         addr = sq->ent[sq->head].buff.cookie;
2897                         bytes_sent -=
2898                                 rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
2899                 }
2900                 sq->size -= num - nb_pkts;
2901         }
2902
2903         q->bytes_sent += bytes_sent;
2904
2905         return nb_pkts;
2906 }
2907
2908 /** DPDK callback for S/G transmit.
2909  *
2910  * @param txq
2911  *   Generic pointer transmit queue.
2912  * @param tx_pkts
2913  *   Packets to transmit.
2914  * @param nb_pkts
2915  *   Number of packets in array.
2916  *
2917  * @return
2918  *   Number of packets successfully transmitted.
2919  */
2920 static uint16_t
2921 mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
2922                      uint16_t nb_pkts)
2923 {
2924         struct mrvl_txq *q = txq;
2925         struct mrvl_shadow_txq *sq;
2926         struct pp2_hif *hif;
2927         struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
2928         struct pp2_ppio_sg_pkts pkts;
2929         uint8_t frags[nb_pkts];
2930         unsigned int core_id = rte_lcore_id();
2931         int i, j, bytes_sent = 0;
2932         int tail, tail_first;
2933         uint16_t num, sq_free_size;
2934         uint16_t nb_segs, total_descs = 0;
2935         uint64_t addr;
2936
2937         hif = mrvl_get_hif(q->priv, core_id);
2938         sq = &q->shadow_txqs[core_id];
2939         pkts.frags = frags;
2940         pkts.num = 0;
2941
2942         if (unlikely(!q->priv->ppio || !hif))
2943                 return 0;
2944
2945         if (sq->size)
2946                 mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
2947                                        sq, q->queue_id, 0);
2948
2949         /* Save shadow queue free size */
2950         sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
2951
2952         tail = 0;
2953         for (i = 0; i < nb_pkts; i++) {
2954                 struct rte_mbuf *mbuf = tx_pkts[i];
2955                 struct rte_mbuf *seg = NULL;
2956                 int gen_l3_cksum, gen_l4_cksum;
2957                 enum pp2_outq_l3_type l3_type;
2958                 enum pp2_outq_l4_type l4_type;
2959
2960                 nb_segs = mbuf->nb_segs;
2961                 tail_first = tail;
2962                 total_descs += nb_segs;
2963
2964                 /*
2965                  * Check if total_descs does not exceed
2966                  * shadow queue free size
2967                  */
2968                 if (unlikely(total_descs > sq_free_size)) {
2969                         total_descs -= nb_segs;
2970                         break;
2971                 }
2972
2973                 /* Check if nb_segs does not exceed the max nb of desc per
2974                  * fragmented packet
2975                  */
2976                 if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
2977                         total_descs -= nb_segs;
2978                         RTE_LOG(ERR, PMD,
2979                                 "Too many segments. Packet won't be sent.\n");
2980                         break;
2981                 }
2982
2983                 if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
2984                         struct rte_mbuf *pref_pkt_hdr;
2985
2986                         pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
2987                         rte_mbuf_prefetch_part1(pref_pkt_hdr);
2988                         rte_mbuf_prefetch_part2(pref_pkt_hdr);
2989                 }
2990
2991                 pkts.frags[pkts.num] = nb_segs;
2992                 pkts.num++;
2993
2994                 seg = mbuf;
2995                 for (j = 0; j < nb_segs - 1; j++) {
2996                         /* For the subsequent segments, set shadow queue
2997                          * buffer to NULL
2998                          */
2999                         mrvl_fill_shadowq(sq, NULL);
3000                         mrvl_fill_desc(&descs[tail], seg);
3001
3002                         tail++;
3003                         seg = seg->next;
3004                 }
3005                 /* Put first mbuf info in last shadow queue entry */
3006                 mrvl_fill_shadowq(sq, mbuf);
3007                 /* Update descriptor with last segment */
3008                 mrvl_fill_desc(&descs[tail++], seg);
3009
3010                 bytes_sent += rte_pktmbuf_pkt_len(mbuf);
3011                 /* In case unsupported ol_flags were passed
3012                  * do not update descriptor offload information
3013                  */
3014                 if (!(mbuf->ol_flags & MRVL_TX_PKT_OFFLOADS))
3015                         continue;
3016                 mrvl_prepare_proto_info(mbuf->ol_flags, &l3_type, &l4_type,
3017                                         &gen_l3_cksum, &gen_l4_cksum);
3018
3019                 pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
3020                                                   l4_type, mbuf->l2_len,
3021                                                   mbuf->l2_len + mbuf->l3_len,
3022                                                   gen_l3_cksum, gen_l4_cksum);
3023         }
3024
3025         num = total_descs;
3026         pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
3027                          &total_descs, &pkts);
3028         /* number of packets that were not sent */
3029         if (unlikely(num > total_descs)) {
3030                 for (i = total_descs; i < num; i++) {
3031                         sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
3032                                 MRVL_PP2_TX_SHADOWQ_MASK;
3033
3034                         addr = sq->ent[sq->head].buff.cookie;
3035                         if (addr)
3036                                 bytes_sent -=
3037                                         rte_pktmbuf_pkt_len((struct rte_mbuf *)
3038                                                 (cookie_addr_high | addr));
3039                 }
3040                 sq->size -= num - total_descs;
3041                 nb_pkts = pkts.num;
3042         }
3043
3044         q->bytes_sent += bytes_sent;
3045
3046         return nb_pkts;
3047 }
3048
3049 /**
3050  * Create private device structure.
3051  *
3052  * @param dev_name
3053  *   Pointer to the port name passed in the initialization parameters.
3054  *
3055  * @return
3056  *   Pointer to the newly allocated private device structure.
3057  */
3058 static struct mrvl_priv *
3059 mrvl_priv_create(const char *dev_name)
3060 {
3061         struct pp2_bpool_params bpool_params;
3062         char match[MRVL_MATCH_LEN];
3063         struct mrvl_priv *priv;
3064         uint16_t max_frame_size;
3065         int ret, bpool_bit;
3066
3067         priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
3068         if (!priv)
3069                 return NULL;
3070
3071         ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
3072                                        &priv->pp_id, &priv->ppio_id);
3073         if (ret)
3074                 goto out_free_priv;
3075
3076         ret = pp2_ppio_get_l4_cksum_max_frame_size(priv->pp_id, priv->ppio_id,
3077                                                    &max_frame_size);
3078         if (ret)
3079                 goto out_free_priv;
3080
3081         priv->max_mtu = max_frame_size + RTE_ETHER_CRC_LEN -
3082                 MRVL_PP2_ETH_HDRS_LEN;
3083
3084         bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
3085                                      PP2_BPOOL_NUM_POOLS);
3086         if (bpool_bit < 0)
3087                 goto out_free_priv;
3088         priv->bpool_bit = bpool_bit;
3089
3090         snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
3091                  priv->bpool_bit);
3092         memset(&bpool_params, 0, sizeof(bpool_params));
3093         bpool_params.match = match;
3094         bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
3095         ret = pp2_bpool_init(&bpool_params, &priv->bpool);
3096         if (ret)
3097                 goto out_clear_bpool_bit;
3098
3099         priv->ppio_params.type = PP2_PPIO_T_NIC;
3100         rte_spinlock_init(&priv->lock);
3101
3102         return priv;
3103 out_clear_bpool_bit:
3104         used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
3105 out_free_priv:
3106         rte_free(priv);
3107         return NULL;
3108 }
3109
3110 /**
3111  * Create device representing Ethernet port.
3112  *
3113  * @param name
3114  *   Pointer to the port's name.
3115  *
3116  * @return
3117  *   0 on success, negative error value otherwise.
3118  */
3119 static int
3120 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
3121 {
3122         int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
3123         struct rte_eth_dev *eth_dev;
3124         struct mrvl_priv *priv;
3125         struct ifreq req;
3126
3127         eth_dev = rte_eth_dev_allocate(name);
3128         if (!eth_dev)
3129                 return -ENOMEM;
3130
3131         priv = mrvl_priv_create(name);
3132         if (!priv) {
3133                 ret = -ENOMEM;
3134                 goto out_free;
3135         }
3136         eth_dev->data->dev_private = priv;
3137
3138         eth_dev->data->mac_addrs =
3139                 rte_zmalloc("mac_addrs",
3140                             RTE_ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
3141         if (!eth_dev->data->mac_addrs) {
3142                 MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
3143                 ret = -ENOMEM;
3144                 goto out_free;
3145         }
3146
3147         memset(&req, 0, sizeof(req));
3148         strcpy(req.ifr_name, name);
3149         ret = ioctl(fd, SIOCGIFHWADDR, &req);
3150         if (ret)
3151                 goto out_free;
3152
3153         memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
3154                req.ifr_addr.sa_data, RTE_ETHER_ADDR_LEN);
3155
3156         eth_dev->device = &vdev->device;
3157         eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
3158         mrvl_set_tx_function(eth_dev);
3159         eth_dev->dev_ops = &mrvl_ops;
3160         eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
3161
3162         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
3163
3164         rte_eth_dev_probing_finish(eth_dev);
3165         return 0;
3166 out_free:
3167         rte_eth_dev_release_port(eth_dev);
3168
3169         return ret;
3170 }
3171
3172 /**
3173  * Callback used by rte_kvargs_process() during argument parsing.
3174  *
3175  * @param key
3176  *   Pointer to the parsed key (unused).
3177  * @param value
3178  *   Pointer to the parsed value.
3179  * @param extra_args
3180  *   Pointer to the extra arguments which contains address of the
3181  *   table of pointers to parsed interface names.
3182  *
3183  * @return
3184  *   Always 0.
3185  */
3186 static int
3187 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
3188                  void *extra_args)
3189 {
3190         struct mrvl_ifnames *ifnames = extra_args;
3191
3192         ifnames->names[ifnames->idx++] = value;
3193
3194         return 0;
3195 }
3196
3197 /**
3198  * DPDK callback to register the virtual device.
3199  *
3200  * @param vdev
3201  *   Pointer to the virtual device.
3202  *
3203  * @return
3204  *   0 on success, negative error value otherwise.
3205  */
3206 static int
3207 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
3208 {
3209         struct rte_kvargs *kvlist;
3210         struct mrvl_ifnames ifnames;
3211         int ret = -EINVAL;
3212         uint32_t i, ifnum, cfgnum;
3213         const char *params;
3214
3215         params = rte_vdev_device_args(vdev);
3216         if (!params)
3217                 return -EINVAL;
3218
3219         kvlist = rte_kvargs_parse(params, valid_args);
3220         if (!kvlist)
3221                 return -EINVAL;
3222
3223         ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
3224         if (ifnum > RTE_DIM(ifnames.names))
3225                 goto out_free_kvlist;
3226
3227         ifnames.idx = 0;
3228         rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
3229                            mrvl_get_ifnames, &ifnames);
3230
3231
3232         /*
3233          * The below system initialization should be done only once,
3234          * on the first provided configuration file
3235          */
3236         if (!mrvl_cfg) {
3237                 cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
3238                 MRVL_LOG(INFO, "Parsing config file!");
3239                 if (cfgnum > 1) {
3240                         MRVL_LOG(ERR, "Cannot handle more than one config file!");
3241                         goto out_free_kvlist;
3242                 } else if (cfgnum == 1) {
3243                         rte_kvargs_process(kvlist, MRVL_CFG_ARG,
3244                                            mrvl_get_cfg, &mrvl_cfg);
3245                 }
3246         }
3247
3248         if (mrvl_dev_num)
3249                 goto init_devices;
3250
3251         MRVL_LOG(INFO, "Perform MUSDK initializations");
3252
3253         ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
3254         if (ret)
3255                 goto out_free_kvlist;
3256
3257         ret = mrvl_init_pp2();
3258         if (ret) {
3259                 MRVL_LOG(ERR, "Failed to init PP!");
3260                 rte_mvep_deinit(MVEP_MOD_T_PP2);
3261                 goto out_free_kvlist;
3262         }
3263
3264         memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
3265         memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));
3266
3267         mrvl_lcore_first = RTE_MAX_LCORE;
3268         mrvl_lcore_last = 0;
3269
3270 init_devices:
3271         for (i = 0; i < ifnum; i++) {
3272                 MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
3273                 ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
3274                 if (ret)
3275                         goto out_cleanup;
3276                 mrvl_dev_num++;
3277         }
3278
3279         rte_kvargs_free(kvlist);
3280
3281         return 0;
3282 out_cleanup:
3283         rte_pmd_mrvl_remove(vdev);
3284
3285 out_free_kvlist:
3286         rte_kvargs_free(kvlist);
3287
3288         return ret;
3289 }
3290
3291 /**
3292  * DPDK callback to remove virtual device.
3293  *
3294  * @param vdev
3295  *   Pointer to the removed virtual device.
3296  *
3297  * @return
3298  *   0 on success, negative error value otherwise.
3299  */
3300 static int
3301 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
3302 {
3303         uint16_t port_id;
3304         int ret = 0;
3305
3306         RTE_ETH_FOREACH_DEV(port_id) {
3307                 if (rte_eth_devices[port_id].device != &vdev->device)
3308                         continue;
3309                 ret |= rte_eth_dev_close(port_id);
3310         }
3311
3312         return ret == 0 ? 0 : -EIO;
3313 }
3314
3315 static struct rte_vdev_driver pmd_mrvl_drv = {
3316         .probe = rte_pmd_mrvl_probe,
3317         .remove = rte_pmd_mrvl_remove,
3318 };
3319
3320 RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
3321 RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
3322 RTE_LOG_REGISTER_DEFAULT(mrvl_logtype, NOTICE);