numam-dpdk/drivers/net/mvpp2/mrvl_ethdev.c

/* SPDX-License-Identifier: BSD-3-Clause
 * Copyright(c) 2017 Marvell International Ltd.
 * Copyright(c) 2017 Semihalf.
 * All rights reserved.
 */

#include <rte_string_fns.h>
#include <rte_ethdev_driver.h>
#include <rte_kvargs.h>
#include <rte_log.h>
#include <rte_malloc.h>
#include <rte_bus_vdev.h>

#include <fcntl.h>
#include <linux/ethtool.h>
#include <linux/sockios.h>
#include <net/if.h>
#include <net/if_arp.h>
#include <sys/ioctl.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/types.h>

#include <rte_mvep_common.h>
#include "mrvl_ethdev.h"
#include "mrvl_qos.h"
#include "mrvl_flow.h"
#include "mrvl_mtr.h"
#include "mrvl_tm.h"

/* bitmask with reserved hifs */
#define MRVL_MUSDK_HIFS_RESERVED 0x0F
/* bitmask with reserved bpools */
#define MRVL_MUSDK_BPOOLS_RESERVED 0x07
/* bitmask with reserved kernel RSS tables */
#define MRVL_MUSDK_RSS_RESERVED 0x01
/* maximum number of available hifs */
#define MRVL_MUSDK_HIFS_MAX 9

/* prefetch shift */
#define MRVL_MUSDK_PREFETCH_SHIFT 2

/* TCAM has 25 entries reserved for uc/mc filter entries */
#define MRVL_MAC_ADDRS_MAX 25
#define MRVL_MATCH_LEN 16
#define MRVL_PKT_EFFEC_OFFS (MRVL_PKT_OFFS + MV_MH_SIZE)
/* Maximum allowable packet size */
#define MRVL_PKT_SIZE_MAX (10240 - MV_MH_SIZE)

#define MRVL_IFACE_NAME_ARG "iface"
#define MRVL_CFG_ARG "cfg"

#define MRVL_BURST_SIZE 64

#define MRVL_ARP_LENGTH 28

#define MRVL_COOKIE_ADDR_INVALID ~0ULL
#define MRVL_COOKIE_HIGH_ADDR_MASK 0xffffff0000000000

/** Port Rx offload capabilities */
#define MRVL_RX_OFFLOADS (DEV_RX_OFFLOAD_VLAN_FILTER | \
			  DEV_RX_OFFLOAD_JUMBO_FRAME | \
			  DEV_RX_OFFLOAD_CHECKSUM)

/** Port Tx offloads capabilities */
#define MRVL_TX_OFFLOADS (DEV_TX_OFFLOAD_IPV4_CKSUM | \
			  DEV_TX_OFFLOAD_UDP_CKSUM | \
			  DEV_TX_OFFLOAD_TCP_CKSUM | \
			  DEV_TX_OFFLOAD_MULTI_SEGS)

static const char * const valid_args[] = {
	MRVL_IFACE_NAME_ARG,
	MRVL_CFG_ARG,
	NULL
};

static int used_hifs = MRVL_MUSDK_HIFS_RESERVED;
static struct pp2_hif *hifs[RTE_MAX_LCORE];
static int used_bpools[PP2_NUM_PKT_PROC] = {
	[0 ... PP2_NUM_PKT_PROC - 1] = MRVL_MUSDK_BPOOLS_RESERVED
};

static struct pp2_bpool *mrvl_port_to_bpool_lookup[RTE_MAX_ETHPORTS];
static int mrvl_port_bpool_size[PP2_NUM_PKT_PROC][PP2_BPOOL_NUM_POOLS][RTE_MAX_LCORE];
static uint64_t cookie_addr_high = MRVL_COOKIE_ADDR_INVALID;

struct mrvl_ifnames {
	const char *names[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
	int idx;
};

/*
 * To use buffer harvesting based on loopback port shadow queue structure
 * was introduced for buffers information bookkeeping.
 *
 * Before sending the packet, related buffer information (pp2_buff_inf) is
 * stored in shadow queue. After packet is transmitted no longer used
 * packet buffer is released back to it's original hardware pool,
 * on condition it originated from interface.
 * In case it  was generated by application itself i.e: mbuf->port field is
 * 0xff then its released to software mempool.
 */
struct mrvl_shadow_txq {
	int head;           /* write index - used when sending buffers */
	int tail;           /* read index - used when releasing buffers */
	u16 size;           /* queue occupied size */
	u16 num_to_release; /* number of descriptors sent, that can be
			     * released
			     */
	struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
};

struct mrvl_rxq {
	struct mrvl_priv *priv;
	struct rte_mempool *mp;
	int queue_id;
	int port_id;
	int cksum_enabled;
	uint64_t bytes_recv;
	uint64_t drop_mac;
};

struct mrvl_txq {
	struct mrvl_priv *priv;
	int queue_id;
	int port_id;
	uint64_t bytes_sent;
	struct mrvl_shadow_txq shadow_txqs[RTE_MAX_LCORE];
	int tx_deferred_start;
};

static int mrvl_lcore_first;
static int mrvl_lcore_last;
static int mrvl_dev_num;

static int mrvl_fill_bpool(struct mrvl_rxq *rxq, int num);
static inline void mrvl_free_sent_buffers(struct pp2_ppio *ppio,
			struct pp2_hif *hif, unsigned int core_id,
			struct mrvl_shadow_txq *sq, int qid, int force);

static uint16_t mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
				  uint16_t nb_pkts);
static uint16_t mrvl_tx_sg_pkt_burst(void *txq,	struct rte_mbuf **tx_pkts,
				     uint16_t nb_pkts);
static int rte_pmd_mrvl_remove(struct rte_vdev_device *vdev);
static void mrvl_deinit_pp2(void);
static void mrvl_deinit_hifs(void);


#define MRVL_XSTATS_TBL_ENTRY(name) { \
	#name, offsetof(struct pp2_ppio_statistics, name),	\
	sizeof(((struct pp2_ppio_statistics *)0)->name)		\
}

/* Table with xstats data */
static struct {
	const char *name;
	unsigned int offset;
	unsigned int size;
} mrvl_xstats_tbl[] = {
	MRVL_XSTATS_TBL_ENTRY(rx_bytes),
	MRVL_XSTATS_TBL_ENTRY(rx_packets),
	MRVL_XSTATS_TBL_ENTRY(rx_unicast_packets),
	MRVL_XSTATS_TBL_ENTRY(rx_errors),
	MRVL_XSTATS_TBL_ENTRY(rx_fullq_dropped),
	MRVL_XSTATS_TBL_ENTRY(rx_bm_dropped),
	MRVL_XSTATS_TBL_ENTRY(rx_early_dropped),
	MRVL_XSTATS_TBL_ENTRY(rx_fifo_dropped),
	MRVL_XSTATS_TBL_ENTRY(rx_cls_dropped),
	MRVL_XSTATS_TBL_ENTRY(tx_bytes),
	MRVL_XSTATS_TBL_ENTRY(tx_packets),
	MRVL_XSTATS_TBL_ENTRY(tx_unicast_packets),
	MRVL_XSTATS_TBL_ENTRY(tx_errors)
};

static inline void
mrvl_fill_shadowq(struct mrvl_shadow_txq *sq, struct rte_mbuf *buf)
{
	sq->ent[sq->head].buff.cookie = (uint64_t)buf;
	sq->ent[sq->head].buff.addr = buf ?
		rte_mbuf_data_iova_default(buf) : 0;

	sq->ent[sq->head].bpool =
		(unlikely(!buf || buf->port >= RTE_MAX_ETHPORTS ||
		 buf->refcnt > 1)) ? NULL :
		 mrvl_port_to_bpool_lookup[buf->port];

	sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
	sq->size++;
}

static inline void
mrvl_fill_desc(struct pp2_ppio_desc *desc, struct rte_mbuf *buf)
{
	pp2_ppio_outq_desc_reset(desc);
	pp2_ppio_outq_desc_set_phys_addr(desc, rte_pktmbuf_iova(buf));
	pp2_ppio_outq_desc_set_pkt_offset(desc, 0);
	pp2_ppio_outq_desc_set_pkt_len(desc, rte_pktmbuf_data_len(buf));
}

static inline int
mrvl_get_bpool_size(int pp2_id, int pool_id)
{
	int i;
	int size = 0;

	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
		size += mrvl_port_bpool_size[pp2_id][pool_id][i];

	return size;
}

static inline int
mrvl_reserve_bit(int *bitmap, int max)
{
	int n = sizeof(*bitmap) * 8 - __builtin_clz(*bitmap);

	if (n >= max)
		return -1;

	*bitmap |= 1 << n;

	return n;
}

static int
mrvl_init_hif(int core_id)
{
	struct pp2_hif_params params;
	char match[MRVL_MATCH_LEN];
	int ret;

	ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
	if (ret < 0) {
		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
		return ret;
	}

	snprintf(match, sizeof(match), "hif-%d", ret);
	memset(&params, 0, sizeof(params));
	params.match = match;
	params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
	ret = pp2_hif_init(&params, &hifs[core_id]);
	if (ret) {
		MRVL_LOG(ERR, "Failed to initialize hif %d", core_id);
		return ret;
	}

	return 0;
}

static inline struct pp2_hif*
mrvl_get_hif(struct mrvl_priv *priv, int core_id)
{
	int ret;

	if (likely(hifs[core_id] != NULL))
		return hifs[core_id];

	rte_spinlock_lock(&priv->lock);

	ret = mrvl_init_hif(core_id);
	if (ret < 0) {
		MRVL_LOG(ERR, "Failed to allocate hif %d", core_id);
		goto out;
	}

	if (core_id < mrvl_lcore_first)
		mrvl_lcore_first = core_id;

	if (core_id > mrvl_lcore_last)
		mrvl_lcore_last = core_id;
out:
	rte_spinlock_unlock(&priv->lock);

	return hifs[core_id];
}

/**
 * Set tx burst function according to offload flag
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 */
static void
mrvl_set_tx_function(struct rte_eth_dev *dev)
{
	struct mrvl_priv *priv = dev->data->dev_private;

	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
	if (priv->multiseg) {
		RTE_LOG(INFO, PMD, "Using multi-segment tx callback\n");
		dev->tx_pkt_burst = mrvl_tx_sg_pkt_burst;
	} else {
		RTE_LOG(INFO, PMD, "Using single-segment tx callback\n");
		dev->tx_pkt_burst = mrvl_tx_pkt_burst;
	}
}

/**
 * Configure rss based on dpdk rss configuration.
 *
 * @param priv
 *   Pointer to private structure.
 * @param rss_conf
 *   Pointer to RSS configuration.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_configure_rss(struct mrvl_priv *priv, struct rte_eth_rss_conf *rss_conf)
{
	if (rss_conf->rss_key)
		MRVL_LOG(WARNING, "Changing hash key is not supported");

	if (rss_conf->rss_hf == 0) {
		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
	} else if (rss_conf->rss_hf & ETH_RSS_IPV4) {
		priv->ppio_params.inqs_params.hash_type =
			PP2_PPIO_HASH_T_2_TUPLE;
	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP) {
		priv->ppio_params.inqs_params.hash_type =
			PP2_PPIO_HASH_T_5_TUPLE;
		priv->rss_hf_tcp = 1;
	} else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) {
		priv->ppio_params.inqs_params.hash_type =
			PP2_PPIO_HASH_T_5_TUPLE;
		priv->rss_hf_tcp = 0;
	} else {
		return -EINVAL;
	}

	return 0;
}

/**
 * Ethernet device configuration.
 *
 * Prepare the driver for a given number of TX and RX queues and
 * configure RSS.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_dev_configure(struct rte_eth_dev *dev)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	int ret;

	if (priv->ppio) {
		MRVL_LOG(INFO, "Device reconfiguration is not supported");
		return -EINVAL;
	}

	if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_NONE &&
	    dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
		MRVL_LOG(INFO, "Unsupported rx multi queue mode %d",
			dev->data->dev_conf.rxmode.mq_mode);
		return -EINVAL;
	}

	if (dev->data->dev_conf.rxmode.split_hdr_size) {
		MRVL_LOG(INFO, "Split headers not supported");
		return -EINVAL;
	}

	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
		dev->data->mtu = dev->data->dev_conf.rxmode.max_rx_pkt_len -
				 MRVL_PP2_ETH_HDRS_LEN;

	if (dev->data->dev_conf.txmode.offloads & DEV_TX_OFFLOAD_MULTI_SEGS)
		priv->multiseg = 1;

	ret = mrvl_configure_rxqs(priv, dev->data->port_id,
				  dev->data->nb_rx_queues);
	if (ret < 0)
		return ret;

	ret = mrvl_configure_txqs(priv, dev->data->port_id,
				  dev->data->nb_tx_queues);
	if (ret < 0)
		return ret;

	priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
	priv->ppio_params.maintain_stats = 1;
	priv->nb_rx_queues = dev->data->nb_rx_queues;

	ret = mrvl_tm_init(dev);
	if (ret < 0)
		return ret;

	if (dev->data->nb_rx_queues == 1 &&
	    dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
		MRVL_LOG(WARNING, "Disabling hash for 1 rx queue");
		priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;

		return 0;
	}

	return mrvl_configure_rss(priv,
				  &dev->data->dev_conf.rx_adv_conf.rss_conf);
}

/**
 * DPDK callback to change the MTU.
 *
 * Setting the MTU affects hardware MRU (packets larger than the MRU
 * will be dropped).
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param mtu
 *   New MTU.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	uint16_t mru;
	uint16_t mbuf_data_size = 0; /* SW buffer size */
	int ret;

	mru = MRVL_PP2_MTU_TO_MRU(mtu);
	/*
	 * min_rx_buf_size is equal to mbuf data size
	 * if pmd didn't set it differently
	 */
	mbuf_data_size = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
	/* Prevent PMD from:
	 * - setting mru greater than the mbuf size resulting in
	 * hw and sw buffer size mismatch
	 * - setting mtu that requires the support of scattered packets
	 * when this feature has not been enabled/supported so far
	 * (TODO check scattered_rx flag here once scattered RX is supported).
	 */
	if (mru + MRVL_PKT_OFFS > mbuf_data_size) {
		mru = mbuf_data_size - MRVL_PKT_OFFS;
		mtu = MRVL_PP2_MRU_TO_MTU(mru);
		MRVL_LOG(WARNING, "MTU too big, max MTU possible limitted "
			"by current mbuf size: %u. Set MTU to %u, MRU to %u",
			mbuf_data_size, mtu, mru);
	}

	if (mtu < RTE_ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX) {
		MRVL_LOG(ERR, "Invalid MTU [%u] or MRU [%u]", mtu, mru);
		return -EINVAL;
	}

	dev->data->mtu = mtu;
	dev->data->dev_conf.rxmode.max_rx_pkt_len = mru - MV_MH_SIZE;

	if (!priv->ppio)
		return 0;

	ret = pp2_ppio_set_mru(priv->ppio, mru);
	if (ret) {
		MRVL_LOG(ERR, "Failed to change MRU");
		return ret;
	}

	ret = pp2_ppio_set_mtu(priv->ppio, mtu);
	if (ret) {
		MRVL_LOG(ERR, "Failed to change MTU");
		return ret;
	}

	return 0;
}

/**
 * DPDK callback to bring the link up.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_dev_set_link_up(struct rte_eth_dev *dev)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	int ret;

	if (!priv->ppio)
		return -EPERM;

	ret = pp2_ppio_enable(priv->ppio);
	if (ret)
		return ret;

	/*
	 * mtu/mru can be updated if pp2_ppio_enable() was called at least once
	 * as pp2_ppio_enable() changes port->t_mode from default 0 to
	 * PP2_TRAFFIC_INGRESS_EGRESS.
	 *
	 * Set mtu to default DPDK value here.
	 */
	ret = mrvl_mtu_set(dev, dev->data->mtu);
	if (ret)
		pp2_ppio_disable(priv->ppio);

	return ret;
}

/**
 * DPDK callback to bring the link down.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_dev_set_link_down(struct rte_eth_dev *dev)
{
	struct mrvl_priv *priv = dev->data->dev_private;

	if (!priv->ppio)
		return -EPERM;

	return pp2_ppio_disable(priv->ppio);
}

/**
 * DPDK callback to start tx queue.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param queue_id
 *   Transmit queue index.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_tx_queue_start(struct rte_eth_dev *dev, uint16_t queue_id)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	int ret;

	if (!priv)
		return -EPERM;

	/* passing 1 enables given tx queue */
	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 1);
	if (ret) {
		MRVL_LOG(ERR, "Failed to start txq %d", queue_id);
		return ret;
	}

	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STARTED;

	return 0;
}

/**
 * DPDK callback to stop tx queue.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param queue_id
 *   Transmit queue index.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t queue_id)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	int ret;

	if (!priv->ppio)
		return -EPERM;

	/* passing 0 disables given tx queue */
	ret = pp2_ppio_set_outq_state(priv->ppio, queue_id, 0);
	if (ret) {
		MRVL_LOG(ERR, "Failed to stop txq %d", queue_id);
		return ret;
	}

	dev->data->tx_queue_state[queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;

	return 0;
}

/**
 * DPDK callback to start the device.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 *
 * @return
 *   0 on success, negative errno value on failure.
 */
static int
mrvl_dev_start(struct rte_eth_dev *dev)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	char match[MRVL_MATCH_LEN];
	int ret = 0, i, def_init_size;

	if (priv->ppio)
		return mrvl_dev_set_link_up(dev);

	snprintf(match, sizeof(match), "ppio-%d:%d",
		 priv->pp_id, priv->ppio_id);
	priv->ppio_params.match = match;

	/*
	 * Calculate the minimum bpool size for refill feature as follows:
	 * 2 default burst sizes multiply by number of rx queues.
	 * If the bpool size will be below this value, new buffers will
	 * be added to the pool.
	 */
	priv->bpool_min_size = priv->nb_rx_queues * MRVL_BURST_SIZE * 2;

	/* In case initial bpool size configured in queues setup is
	 * smaller than minimum size add more buffers
	 */
	def_init_size = priv->bpool_min_size + MRVL_BURST_SIZE * 2;
	if (priv->bpool_init_size < def_init_size) {
		int buffs_to_add = def_init_size - priv->bpool_init_size;

		priv->bpool_init_size += buffs_to_add;
		ret = mrvl_fill_bpool(dev->data->rx_queues[0], buffs_to_add);
		if (ret)
			MRVL_LOG(ERR, "Failed to add buffers to bpool");
	}

	/*
	 * Calculate the maximum bpool size for refill feature as follows:
	 * maximum number of descriptors in rx queue multiply by number
	 * of rx queues plus minimum bpool size.
	 * In case the bpool size will exceed this value, superfluous buffers
	 * will be removed
	 */
	priv->bpool_max_size = (priv->nb_rx_queues * MRVL_PP2_RXD_MAX) +
				priv->bpool_min_size;

	ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
	if (ret) {
		MRVL_LOG(ERR, "Failed to init ppio");
		return ret;
	}

	/*
	 * In case there are some some stale uc/mc mac addresses flush them
	 * here. It cannot be done during mrvl_dev_close() as port information
	 * is already gone at that point (due to pp2_ppio_deinit() in
	 * mrvl_dev_stop()).
	 */
	if (!priv->uc_mc_flushed) {
		ret = pp2_ppio_flush_mac_addrs(priv->ppio, 1, 1);
		if (ret) {
			MRVL_LOG(ERR,
				"Failed to flush uc/mc filter list");
			goto out;
		}
		priv->uc_mc_flushed = 1;
	}

	if (!priv->vlan_flushed) {
		ret = pp2_ppio_flush_vlan(priv->ppio);
		if (ret) {
			MRVL_LOG(ERR, "Failed to flush vlan list");
			/*
			 * TODO
			 * once pp2_ppio_flush_vlan() is supported jump to out
			 * goto out;
			 */
		}
		priv->vlan_flushed = 1;
	}
	ret = mrvl_mtu_set(dev, dev->data->mtu);
	if (ret)
		MRVL_LOG(ERR, "Failed to set MTU to %d", dev->data->mtu);

	/* For default QoS config, don't start classifier. */
	if (mrvl_qos_cfg  &&
	    mrvl_qos_cfg->port[dev->data->port_id].use_global_defaults == 0) {
		ret = mrvl_start_qos_mapping(priv);
		if (ret) {
			MRVL_LOG(ERR, "Failed to setup QoS mapping");
			goto out;
		}
	}

	ret = mrvl_dev_set_link_up(dev);
	if (ret) {
		MRVL_LOG(ERR, "Failed to set link up");
		goto out;
	}

	/* start tx queues */
	for (i = 0; i < dev->data->nb_tx_queues; i++) {
		struct mrvl_txq *txq = dev->data->tx_queues[i];

		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;

		if (!txq->tx_deferred_start)
			continue;

		/*
		 * All txqs are started by default. Stop them
		 * so that tx_deferred_start works as expected.
		 */
		ret = mrvl_tx_queue_stop(dev, i);
		if (ret)
			goto out;
	}

	mrvl_flow_init(dev);
	mrvl_mtr_init(dev);
	mrvl_set_tx_function(dev);

	return 0;
out:
	MRVL_LOG(ERR, "Failed to start device");
	pp2_ppio_deinit(priv->ppio);
	return ret;
}

/**
 * Flush receive queues.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 */
static void
mrvl_flush_rx_queues(struct rte_eth_dev *dev)
{
	int i;

	MRVL_LOG(INFO, "Flushing rx queues");
	for (i = 0; i < dev->data->nb_rx_queues; i++) {
		int ret, num;

		do {
			struct mrvl_rxq *q = dev->data->rx_queues[i];
			struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];

			num = MRVL_PP2_RXD_MAX;
			ret = pp2_ppio_recv(q->priv->ppio,
					    q->priv->rxq_map[q->queue_id].tc,
					    q->priv->rxq_map[q->queue_id].inq,
					    descs, (uint16_t *)&num);
		} while (ret == 0 && num);
	}
}

/**
 * Flush transmit shadow queues.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 */
static void
mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
{
	int i, j;
	struct mrvl_txq *txq;

	MRVL_LOG(INFO, "Flushing tx shadow queues");
	for (i = 0; i < dev->data->nb_tx_queues; i++) {
		txq = (struct mrvl_txq *)dev->data->tx_queues[i];

		for (j = 0; j < RTE_MAX_LCORE; j++) {
			struct mrvl_shadow_txq *sq;

			if (!hifs[j])
				continue;

			sq = &txq->shadow_txqs[j];
			mrvl_free_sent_buffers(txq->priv->ppio,
				hifs[j], j, sq, txq->queue_id, 1);
			while (sq->tail != sq->head) {
				uint64_t addr = cookie_addr_high |
					sq->ent[sq->tail].buff.cookie;
				rte_pktmbuf_free(
					(struct rte_mbuf *)addr);
				sq->tail = (sq->tail + 1) &
					    MRVL_PP2_TX_SHADOWQ_MASK;
			}
			memset(sq, 0, sizeof(*sq));
		}
	}
}

/**
 * Flush hardware bpool (buffer-pool).
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 */
static void
mrvl_flush_bpool(struct rte_eth_dev *dev)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	struct pp2_hif *hif;
	uint32_t num;
	int ret;
	unsigned int core_id = rte_lcore_id();

	if (core_id == LCORE_ID_ANY)
		core_id = rte_get_main_lcore();

	hif = mrvl_get_hif(priv, core_id);

	ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
	if (ret) {
		MRVL_LOG(ERR, "Failed to get bpool buffers number");
		return;
	}

	while (num--) {
		struct pp2_buff_inf inf;
		uint64_t addr;

		ret = pp2_bpool_get_buff(hif, priv->bpool, &inf);
		if (ret)
			break;

		addr = cookie_addr_high | inf.cookie;
		rte_pktmbuf_free((struct rte_mbuf *)addr);
	}
}

/**
 * DPDK callback to stop the device.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 */
static int
mrvl_dev_stop(struct rte_eth_dev *dev)
{
	return mrvl_dev_set_link_down(dev);
}

/**
 * DPDK callback to close the device.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 */
static int
mrvl_dev_close(struct rte_eth_dev *dev)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	size_t i;

	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
		return 0;

	mrvl_flush_rx_queues(dev);
	mrvl_flush_tx_shadow_queues(dev);
	mrvl_flow_deinit(dev);
	mrvl_mtr_deinit(dev);

	for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
		struct pp2_ppio_tc_params *tc_params =
			&priv->ppio_params.inqs_params.tcs_params[i];

		if (tc_params->inqs_params) {
			rte_free(tc_params->inqs_params);
			tc_params->inqs_params = NULL;
		}
	}

	if (priv->cls_tbl) {
		pp2_cls_tbl_deinit(priv->cls_tbl);
		priv->cls_tbl = NULL;
	}

	if (priv->qos_tbl) {
		pp2_cls_qos_tbl_deinit(priv->qos_tbl);
		priv->qos_tbl = NULL;
	}

	mrvl_flush_bpool(dev);
	mrvl_tm_deinit(dev);

	if (priv->ppio) {
		pp2_ppio_deinit(priv->ppio);
		priv->ppio = NULL;
	}

	/* policer must be released after ppio deinitialization */
	if (priv->default_policer) {
		pp2_cls_plcr_deinit(priv->default_policer);
		priv->default_policer = NULL;
	}


	if (priv->bpool) {
		pp2_bpool_deinit(priv->bpool);
		used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
		priv->bpool = NULL;
	}

	mrvl_dev_num--;

	if (mrvl_dev_num == 0) {
		MRVL_LOG(INFO, "Perform MUSDK deinit");
		mrvl_deinit_hifs();
		mrvl_deinit_pp2();
		rte_mvep_deinit(MVEP_MOD_T_PP2);
	}

	return 0;
}

/**
 * DPDK callback to retrieve physical link information.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param wait_to_complete
 *   Wait for request completion (ignored).
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
{
	/*
	 * TODO
	 * once MUSDK provides necessary API use it here
	 */
	struct mrvl_priv *priv = dev->data->dev_private;
	struct ethtool_cmd edata;
	struct ifreq req;
	int ret, fd, link_up;

	if (!priv->ppio)
		return -EPERM;

	edata.cmd = ETHTOOL_GSET;

	strcpy(req.ifr_name, dev->data->name);
	req.ifr_data = (void *)&edata;

	fd = socket(AF_INET, SOCK_DGRAM, 0);
	if (fd == -1)
		return -EFAULT;

	ret = ioctl(fd, SIOCETHTOOL, &req);
	if (ret == -1) {
		close(fd);
		return -EFAULT;
	}

	close(fd);

	switch (ethtool_cmd_speed(&edata)) {
	case SPEED_10:
		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
		break;
	case SPEED_100:
		dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
		break;
	case SPEED_1000:
		dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
		break;
	case SPEED_10000:
		dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
		break;
	default:
		dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
	}

	dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
							 ETH_LINK_HALF_DUPLEX;
	dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
							   ETH_LINK_FIXED;
	pp2_ppio_get_link_state(priv->ppio, &link_up);
	dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;

	return 0;
}

/**
 * DPDK callback to enable promiscuous mode.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_promiscuous_enable(struct rte_eth_dev *dev)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	int ret;

	if (!priv->ppio)
		return 0;

	if (priv->isolated)
		return 0;

	ret = pp2_ppio_set_promisc(priv->ppio, 1);
	if (ret) {
		MRVL_LOG(ERR, "Failed to enable promiscuous mode");
		return -EAGAIN;
	}

	return 0;
}

/**
 * DPDK callback to enable allmulti mode.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_allmulticast_enable(struct rte_eth_dev *dev)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	int ret;

	if (!priv->ppio)
		return 0;

	if (priv->isolated)
		return 0;

	ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
	if (ret) {
		MRVL_LOG(ERR, "Failed enable all-multicast mode");
		return -EAGAIN;
	}

	return 0;
}

/**
 * DPDK callback to disable promiscuous mode.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_promiscuous_disable(struct rte_eth_dev *dev)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	int ret;

	if (!priv->ppio)
		return 0;

	ret = pp2_ppio_set_promisc(priv->ppio, 0);
	if (ret) {
		MRVL_LOG(ERR, "Failed to disable promiscuous mode");
		return -EAGAIN;
	}

	return 0;
}

/**
 * DPDK callback to disable allmulticast mode.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_allmulticast_disable(struct rte_eth_dev *dev)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	int ret;

	if (!priv->ppio)
		return 0;

	ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
	if (ret) {
		MRVL_LOG(ERR, "Failed to disable all-multicast mode");
		return -EAGAIN;
	}

	return 0;
}

/**
 * DPDK callback to remove a MAC address.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param index
 *   MAC address index.
 */
static void
mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	char buf[RTE_ETHER_ADDR_FMT_SIZE];
	int ret;

	if (!priv->ppio)
		return;

	if (priv->isolated)
		return;

	ret = pp2_ppio_remove_mac_addr(priv->ppio,
				       dev->data->mac_addrs[index].addr_bytes);
	if (ret) {
		rte_ether_format_addr(buf, sizeof(buf),
				  &dev->data->mac_addrs[index]);
		MRVL_LOG(ERR, "Failed to remove mac %s", buf);
	}
}

/**
 * DPDK callback to add a MAC address.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param mac_addr
 *   MAC address to register.
 * @param index
 *   MAC address index.
 * @param vmdq
 *   VMDq pool index to associate address with (unused).
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
		  uint32_t index, uint32_t vmdq __rte_unused)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	char buf[RTE_ETHER_ADDR_FMT_SIZE];
	int ret;

	if (priv->isolated)
		return -ENOTSUP;

	if (index == 0)
		/* For setting index 0, mrvl_mac_addr_set() should be used.*/
		return -1;

	if (!priv->ppio)
		return 0;

	/*
	 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
	 * parameter uc_filter_max. Maximum number of mc addresses is then
	 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
	 * 21 respectively.
	 *
	 * If more than uc_filter_max uc addresses were added to filter list
	 * then NIC will switch to promiscuous mode automatically.
	 *
	 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
	 * were added to filter list then NIC will switch to all-multicast mode
	 * automatically.
	 */
	ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
	if (ret) {
		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
		MRVL_LOG(ERR, "Failed to add mac %s", buf);
		return -1;
	}

	return 0;
}

/**
 * DPDK callback to set the primary MAC address.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param mac_addr
 *   MAC address to register.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	int ret;

	if (!priv->ppio)
		return 0;

	if (priv->isolated)
		return -ENOTSUP;

	ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
	if (ret) {
		char buf[RTE_ETHER_ADDR_FMT_SIZE];
		rte_ether_format_addr(buf, sizeof(buf), mac_addr);
		MRVL_LOG(ERR, "Failed to set mac to %s", buf);
	}

	return ret;
}

/**
 * DPDK callback to get device statistics.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param stats
 *   Stats structure output buffer.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	struct pp2_ppio_statistics ppio_stats;
	uint64_t drop_mac = 0;
	unsigned int i, idx, ret;

	if (!priv->ppio)
		return -EPERM;

	for (i = 0; i < dev->data->nb_rx_queues; i++) {
		struct mrvl_rxq *rxq = dev->data->rx_queues[i];
		struct pp2_ppio_inq_statistics rx_stats;

		if (!rxq)
			continue;

		idx = rxq->queue_id;
		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
			MRVL_LOG(ERR,
				"rx queue %d stats out of range (0 - %d)",
				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
			continue;
		}

		ret = pp2_ppio_inq_get_statistics(priv->ppio,
						  priv->rxq_map[idx].tc,
						  priv->rxq_map[idx].inq,
						  &rx_stats, 0);
		if (unlikely(ret)) {
			MRVL_LOG(ERR,
				"Failed to update rx queue %d stats", idx);
			break;
		}

		stats->q_ibytes[idx] = rxq->bytes_recv;
		stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
		stats->q_errors[idx] = rx_stats.drop_early +
				       rx_stats.drop_fullq +
				       rx_stats.drop_bm +
				       rxq->drop_mac;
		stats->ibytes += rxq->bytes_recv;
		drop_mac += rxq->drop_mac;
	}

	for (i = 0; i < dev->data->nb_tx_queues; i++) {
		struct mrvl_txq *txq = dev->data->tx_queues[i];
		struct pp2_ppio_outq_statistics tx_stats;

		if (!txq)
			continue;

		idx = txq->queue_id;
		if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
			MRVL_LOG(ERR,
				"tx queue %d stats out of range (0 - %d)",
				idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
		}

		ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
						   &tx_stats, 0);
		if (unlikely(ret)) {
			MRVL_LOG(ERR,
				"Failed to update tx queue %d stats", idx);
			break;
		}

		stats->q_opackets[idx] = tx_stats.deq_desc;
		stats->q_obytes[idx] = txq->bytes_sent;
		stats->obytes += txq->bytes_sent;
	}

	ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
	if (unlikely(ret)) {
		MRVL_LOG(ERR, "Failed to update port statistics");
		return ret;
	}

	stats->ipackets += ppio_stats.rx_packets - drop_mac;
	stats->opackets += ppio_stats.tx_packets;
	stats->imissed += ppio_stats.rx_fullq_dropped +
			  ppio_stats.rx_bm_dropped +
			  ppio_stats.rx_early_dropped +
			  ppio_stats.rx_fifo_dropped +
			  ppio_stats.rx_cls_dropped;
	stats->ierrors = drop_mac;

	return 0;
}

/**
 * DPDK callback to clear device statistics.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_stats_reset(struct rte_eth_dev *dev)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	int i;

	if (!priv->ppio)
		return 0;

	for (i = 0; i < dev->data->nb_rx_queues; i++) {
		struct mrvl_rxq *rxq = dev->data->rx_queues[i];

		pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
					    priv->rxq_map[i].inq, NULL, 1);
		rxq->bytes_recv = 0;
		rxq->drop_mac = 0;
	}

	for (i = 0; i < dev->data->nb_tx_queues; i++) {
		struct mrvl_txq *txq = dev->data->tx_queues[i];

		pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
		txq->bytes_sent = 0;
	}

	return pp2_ppio_get_statistics(priv->ppio, NULL, 1);
}

/**
 * DPDK callback to get extended statistics.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param stats
 *   Pointer to xstats table.
 * @param n
 *   Number of entries in xstats table.
 * @return
 *   Negative value on error, number of read xstats otherwise.
 */
static int
mrvl_xstats_get(struct rte_eth_dev *dev,
		struct rte_eth_xstat *stats, unsigned int n)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	struct pp2_ppio_statistics ppio_stats;
	unsigned int i;

	if (!stats)
		return 0;

	pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
	for (i = 0; i < n && i < RTE_DIM(mrvl_xstats_tbl); i++) {
		uint64_t val;

		if (mrvl_xstats_tbl[i].size == sizeof(uint32_t))
			val = *(uint32_t *)((uint8_t *)&ppio_stats +
					    mrvl_xstats_tbl[i].offset);
		else if (mrvl_xstats_tbl[i].size == sizeof(uint64_t))
			val = *(uint64_t *)((uint8_t *)&ppio_stats +
					    mrvl_xstats_tbl[i].offset);
		else
			return -EINVAL;

		stats[i].id = i;
		stats[i].value = val;
	}

	return n;
}

/**
 * DPDK callback to reset extended statistics.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_xstats_reset(struct rte_eth_dev *dev)
{
	return mrvl_stats_reset(dev);
}

/**
 * DPDK callback to get extended statistics names.
 *
 * @param dev (unused)
 *   Pointer to Ethernet device structure.
 * @param xstats_names
 *   Pointer to xstats names table.
 * @param size
 *   Size of the xstats names table.
 * @return
 *   Number of read names.
 */
static int
mrvl_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
		      struct rte_eth_xstat_name *xstats_names,
		      unsigned int size)
{
	unsigned int i;

	if (!xstats_names)
		return RTE_DIM(mrvl_xstats_tbl);

	for (i = 0; i < size && i < RTE_DIM(mrvl_xstats_tbl); i++)
		strlcpy(xstats_names[i].name, mrvl_xstats_tbl[i].name,
			RTE_ETH_XSTATS_NAME_SIZE);

	return size;
}

/**
 * DPDK callback to get information about the device.
 *
 * @param dev
 *   Pointer to Ethernet device structure (unused).
 * @param info
 *   Info structure output buffer.
 */
static int
mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
		   struct rte_eth_dev_info *info)
{
	info->speed_capa = ETH_LINK_SPEED_10M |
			   ETH_LINK_SPEED_100M |
			   ETH_LINK_SPEED_1G |
			   ETH_LINK_SPEED_10G;

	info->max_rx_queues = MRVL_PP2_RXQ_MAX;
	info->max_tx_queues = MRVL_PP2_TXQ_MAX;
	info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;

	info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
	info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
	info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;

	info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
	info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
	info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;

	info->rx_offload_capa = MRVL_RX_OFFLOADS;
	info->rx_queue_offload_capa = MRVL_RX_OFFLOADS;

	info->tx_offload_capa = MRVL_TX_OFFLOADS;
	info->tx_queue_offload_capa = MRVL_TX_OFFLOADS;

	info->flow_type_rss_offloads = ETH_RSS_IPV4 |
				       ETH_RSS_NONFRAG_IPV4_TCP |
				       ETH_RSS_NONFRAG_IPV4_UDP;

	/* By default packets are dropped if no descriptors are available */
	info->default_rxconf.rx_drop_en = 1;

	info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;

	return 0;
}

/**
 * Return supported packet types.
 *
 * @param dev
 *   Pointer to Ethernet device structure (unused).
 *
 * @return
 *   Const pointer to the table with supported packet types.
 */
static const uint32_t *
mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
{
	static const uint32_t ptypes[] = {
		RTE_PTYPE_L2_ETHER,
		RTE_PTYPE_L2_ETHER_VLAN,
		RTE_PTYPE_L2_ETHER_QINQ,
		RTE_PTYPE_L3_IPV4,
		RTE_PTYPE_L3_IPV4_EXT,
		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
		RTE_PTYPE_L3_IPV6,
		RTE_PTYPE_L3_IPV6_EXT,
		RTE_PTYPE_L2_ETHER_ARP,
		RTE_PTYPE_L4_TCP,
		RTE_PTYPE_L4_UDP
	};

	return ptypes;
}

/**
 * DPDK callback to get information about specific receive queue.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param rx_queue_id
 *   Receive queue index.
 * @param qinfo
 *   Receive queue information structure.
 */
static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
			      struct rte_eth_rxq_info *qinfo)
{
	struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
	struct mrvl_priv *priv = dev->data->dev_private;
	int inq = priv->rxq_map[rx_queue_id].inq;
	int tc = priv->rxq_map[rx_queue_id].tc;
	struct pp2_ppio_tc_params *tc_params =
		&priv->ppio_params.inqs_params.tcs_params[tc];

	qinfo->mp = q->mp;
	qinfo->nb_desc = tc_params->inqs_params[inq].size;
}

/**
 * DPDK callback to get information about specific transmit queue.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param tx_queue_id
 *   Transmit queue index.
 * @param qinfo
 *   Transmit queue information structure.
 */
static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
			      struct rte_eth_txq_info *qinfo)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	struct mrvl_txq *txq = dev->data->tx_queues[tx_queue_id];

	qinfo->nb_desc =
		priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
}

/**
 * DPDK callback to Configure a VLAN filter.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param vlan_id
 *   VLAN ID to filter.
 * @param on
 *   Toggle filter.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
{
	struct mrvl_priv *priv = dev->data->dev_private;

	if (!priv->ppio)
		return -EPERM;

	if (priv->isolated)
		return -ENOTSUP;

	return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
		    pp2_ppio_remove_vlan(priv->ppio, vlan_id);
}

/**
 * Release buffers to hardware bpool (buffer-pool)
 *
 * @param rxq
 *   Receive queue pointer.
 * @param num
 *   Number of buffers to release to bpool.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
{
	struct buff_release_entry entries[MRVL_PP2_RXD_MAX];
	struct rte_mbuf *mbufs[MRVL_PP2_RXD_MAX];
	int i, ret;
	unsigned int core_id;
	struct pp2_hif *hif;
	struct pp2_bpool *bpool;

	core_id = rte_lcore_id();
	if (core_id == LCORE_ID_ANY)
		core_id = rte_get_main_lcore();

	hif = mrvl_get_hif(rxq->priv, core_id);
	if (!hif)
		return -1;

	bpool = rxq->priv->bpool;

	ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
	if (ret)
		return ret;

	if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
		cookie_addr_high =
			(uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;

	for (i = 0; i < num; i++) {
		if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
			!= cookie_addr_high) {
			MRVL_LOG(ERR,
				"mbuf virtual addr high 0x%lx out of range",
				(uint64_t)mbufs[i] >> 32);
			goto out;
		}

		entries[i].buff.addr =
			rte_mbuf_data_iova_default(mbufs[i]);
		entries[i].buff.cookie = (uint64_t)mbufs[i];
		entries[i].bpool = bpool;
	}

	pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;

	if (i != num)
		goto out;

	return 0;
out:
	for (; i < num; i++)
		rte_pktmbuf_free(mbufs[i]);

	return -1;
}

/**
 * DPDK callback to configure the receive queue.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param idx
 *   RX queue index.
 * @param desc
 *   Number of descriptors to configure in queue.
 * @param socket
 *   NUMA socket on which memory must be allocated.
 * @param conf
 *   Thresholds parameters.
 * @param mp
 *   Memory pool for buffer allocations.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
		    unsigned int socket,
		    const struct rte_eth_rxconf *conf,
		    struct rte_mempool *mp)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	struct mrvl_rxq *rxq;
	uint32_t frame_size, buf_size = rte_pktmbuf_data_room_size(mp);
	uint32_t max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
	int ret, tc, inq;
	uint64_t offloads;

	offloads = conf->offloads | dev->data->dev_conf.rxmode.offloads;

	if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
		/*
		 * Unknown TC mapping, mapping will not have a correct queue.
		 */
		MRVL_LOG(ERR, "Unknown TC mapping for queue %hu eth%hhu",
			idx, priv->ppio_id);
		return -EFAULT;
	}

	frame_size = buf_size - RTE_PKTMBUF_HEADROOM - MRVL_PKT_EFFEC_OFFS;
	if (frame_size < max_rx_pkt_len) {
		MRVL_LOG(WARNING,
			"Mbuf size must be increased to %u bytes to hold up "
			"to %u bytes of data.",
			buf_size + max_rx_pkt_len - frame_size,
			max_rx_pkt_len);
		dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
		MRVL_LOG(INFO, "Setting max rx pkt len to %u",
			dev->data->dev_conf.rxmode.max_rx_pkt_len);
	}

	if (dev->data->rx_queues[idx]) {
		rte_free(dev->data->rx_queues[idx]);
		dev->data->rx_queues[idx] = NULL;
	}

	rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
	if (!rxq)
		return -ENOMEM;

	rxq->priv = priv;
	rxq->mp = mp;
	rxq->cksum_enabled = offloads & DEV_RX_OFFLOAD_IPV4_CKSUM;
	rxq->queue_id = idx;
	rxq->port_id = dev->data->port_id;
	mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;

	tc = priv->rxq_map[rxq->queue_id].tc,
	inq = priv->rxq_map[rxq->queue_id].inq;
	priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
		desc;

	ret = mrvl_fill_bpool(rxq, desc);
	if (ret) {
		rte_free(rxq);
		return ret;
	}

	priv->bpool_init_size += desc;

	dev->data->rx_queues[idx] = rxq;

	return 0;
}

/**
 * DPDK callback to release the receive queue.
 *
 * @param rxq
 *   Generic receive queue pointer.
 */
static void
mrvl_rx_queue_release(void *rxq)
{
	struct mrvl_rxq *q = rxq;
	struct pp2_ppio_tc_params *tc_params;
	int i, num, tc, inq;
	struct pp2_hif *hif;
	unsigned int core_id = rte_lcore_id();

	if (core_id == LCORE_ID_ANY)
		core_id = rte_get_main_lcore();

	if (!q)
		return;

	hif = mrvl_get_hif(q->priv, core_id);

	if (!hif)
		return;

	tc = q->priv->rxq_map[q->queue_id].tc;
	inq = q->priv->rxq_map[q->queue_id].inq;
	tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
	num = tc_params->inqs_params[inq].size;
	for (i = 0; i < num; i++) {
		struct pp2_buff_inf inf;
		uint64_t addr;

		pp2_bpool_get_buff(hif, q->priv->bpool, &inf);
		addr = cookie_addr_high | inf.cookie;
		rte_pktmbuf_free((struct rte_mbuf *)addr);
	}

	rte_free(q);
}

/**
 * DPDK callback to configure the transmit queue.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param idx
 *   Transmit queue index.
 * @param desc
 *   Number of descriptors to configure in the queue.
 * @param socket
 *   NUMA socket on which memory must be allocated.
 * @param conf
 *   Tx queue configuration parameters.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
		    unsigned int socket,
		    const struct rte_eth_txconf *conf)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	struct mrvl_txq *txq;

	if (dev->data->tx_queues[idx]) {
		rte_free(dev->data->tx_queues[idx]);
		dev->data->tx_queues[idx] = NULL;
	}

	txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
	if (!txq)
		return -ENOMEM;

	txq->priv = priv;
	txq->queue_id = idx;
	txq->port_id = dev->data->port_id;
	txq->tx_deferred_start = conf->tx_deferred_start;
	dev->data->tx_queues[idx] = txq;

	priv->ppio_params.outqs_params.outqs_params[idx].size = desc;

	return 0;
}

/**
 * DPDK callback to release the transmit queue.
 *
 * @param txq
 *   Generic transmit queue pointer.
 */
static void
mrvl_tx_queue_release(void *txq)
{
	struct mrvl_txq *q = txq;

	if (!q)
		return;

	rte_free(q);
}

/**
 * DPDK callback to get flow control configuration.
 *
 * @param dev
 *  Pointer to Ethernet device structure.
 * @param fc_conf
 *  Pointer to the flow control configuration.
 *
 * @return
 *  0 on success, negative error value otherwise.
 */
static int
mrvl_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	int ret, en;

	if (!priv)
		return -EPERM;

	ret = pp2_ppio_get_rx_pause(priv->ppio, &en);
	if (ret) {
		MRVL_LOG(ERR, "Failed to read rx pause state");
		return ret;
	}

	fc_conf->mode = en ? RTE_FC_RX_PAUSE : RTE_FC_NONE;

	return 0;
}

/**
 * DPDK callback to set flow control configuration.
 *
 * @param dev
 *  Pointer to Ethernet device structure.
 * @param fc_conf
 *  Pointer to the flow control configuration.
 *
 * @return
 *  0 on success, negative error value otherwise.
 */
static int
mrvl_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
{
	struct mrvl_priv *priv = dev->data->dev_private;

	if (!priv)
		return -EPERM;

	if (fc_conf->high_water ||
	    fc_conf->low_water ||
	    fc_conf->pause_time ||
	    fc_conf->mac_ctrl_frame_fwd ||
	    fc_conf->autoneg) {
		MRVL_LOG(ERR, "Flowctrl parameter is not supported");

		return -EINVAL;
	}

	if (fc_conf->mode == RTE_FC_NONE ||
	    fc_conf->mode == RTE_FC_RX_PAUSE) {
		int ret, en;

		en = fc_conf->mode == RTE_FC_NONE ? 0 : 1;
		ret = pp2_ppio_set_rx_pause(priv->ppio, en);
		if (ret)
			MRVL_LOG(ERR,
				"Failed to change flowctrl on RX side");

		return ret;
	}

	return 0;
}

/**
 * Update RSS hash configuration
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param rss_conf
 *   Pointer to RSS configuration.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_rss_hash_update(struct rte_eth_dev *dev,
		     struct rte_eth_rss_conf *rss_conf)
{
	struct mrvl_priv *priv = dev->data->dev_private;

	if (priv->isolated)
		return -ENOTSUP;

	return mrvl_configure_rss(priv, rss_conf);
}

/**
 * DPDK callback to get RSS hash configuration.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @rss_conf
 *   Pointer to RSS configuration.
 *
 * @return
 *   Always 0.
 */
static int
mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
		       struct rte_eth_rss_conf *rss_conf)
{
	struct mrvl_priv *priv = dev->data->dev_private;
	enum pp2_ppio_hash_type hash_type =
		priv->ppio_params.inqs_params.hash_type;

	rss_conf->rss_key = NULL;

	if (hash_type == PP2_PPIO_HASH_T_NONE)
		rss_conf->rss_hf = 0;
	else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
		rss_conf->rss_hf = ETH_RSS_IPV4;
	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
	else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
		rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;

	return 0;
}

/**
 * DPDK callback to get rte_flow callbacks.
 *
 * @param dev
 *   Pointer to the device structure.
 * @param filer_type
 *   Flow filter type.
 * @param filter_op
 *   Flow filter operation.
 * @param arg
 *   Pointer to pass the flow ops.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_eth_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
		     enum rte_filter_type filter_type,
		     enum rte_filter_op filter_op, void *arg)
{
	switch (filter_type) {
	case RTE_ETH_FILTER_GENERIC:
		if (filter_op != RTE_ETH_FILTER_GET)
			return -EINVAL;
		*(const void **)arg = &mrvl_flow_ops;
		return 0;
	default:
		MRVL_LOG(WARNING, "Filter type (%d) not supported",
				filter_type);
		return -EINVAL;
	}
}

/**
 * DPDK callback to get rte_mtr callbacks.
 *
 * @param dev
 *   Pointer to the device structure.
 * @param ops
 *   Pointer to pass the mtr ops.
 *
 * @return
 *   Always 0.
 */
static int
mrvl_mtr_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
{
	*(const void **)ops = &mrvl_mtr_ops;

	return 0;
}

/**
 * DPDK callback to get rte_tm callbacks.
 *
 * @param dev
 *   Pointer to the device structure.
 * @param ops
 *   Pointer to pass the tm ops.
 *
 * @return
 *   Always 0.
 */
static int
mrvl_tm_ops_get(struct rte_eth_dev *dev __rte_unused, void *ops)
{
	*(const void **)ops = &mrvl_tm_ops;

	return 0;
}

static const struct eth_dev_ops mrvl_ops = {
	.dev_configure = mrvl_dev_configure,
	.dev_start = mrvl_dev_start,
	.dev_stop = mrvl_dev_stop,
	.dev_set_link_up = mrvl_dev_set_link_up,
	.dev_set_link_down = mrvl_dev_set_link_down,
	.dev_close = mrvl_dev_close,
	.link_update = mrvl_link_update,
	.promiscuous_enable = mrvl_promiscuous_enable,
	.allmulticast_enable = mrvl_allmulticast_enable,
	.promiscuous_disable = mrvl_promiscuous_disable,
	.allmulticast_disable = mrvl_allmulticast_disable,
	.mac_addr_remove = mrvl_mac_addr_remove,
	.mac_addr_add = mrvl_mac_addr_add,
	.mac_addr_set = mrvl_mac_addr_set,
	.mtu_set = mrvl_mtu_set,
	.stats_get = mrvl_stats_get,
	.stats_reset = mrvl_stats_reset,
	.xstats_get = mrvl_xstats_get,
	.xstats_reset = mrvl_xstats_reset,
	.xstats_get_names = mrvl_xstats_get_names,
	.dev_infos_get = mrvl_dev_infos_get,
	.dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
	.rxq_info_get = mrvl_rxq_info_get,
	.txq_info_get = mrvl_txq_info_get,
	.vlan_filter_set = mrvl_vlan_filter_set,
	.tx_queue_start = mrvl_tx_queue_start,
	.tx_queue_stop = mrvl_tx_queue_stop,
	.rx_queue_setup = mrvl_rx_queue_setup,
	.rx_queue_release = mrvl_rx_queue_release,
	.tx_queue_setup = mrvl_tx_queue_setup,
	.tx_queue_release = mrvl_tx_queue_release,
	.flow_ctrl_get = mrvl_flow_ctrl_get,
	.flow_ctrl_set = mrvl_flow_ctrl_set,
	.rss_hash_update = mrvl_rss_hash_update,
	.rss_hash_conf_get = mrvl_rss_hash_conf_get,
	.filter_ctrl = mrvl_eth_filter_ctrl,
	.mtr_ops_get = mrvl_mtr_ops_get,
	.tm_ops_get = mrvl_tm_ops_get,
};

/**
 * Return packet type information and l3/l4 offsets.
 *
 * @param desc
 *   Pointer to the received packet descriptor.
 * @param l3_offset
 *   l3 packet offset.
 * @param l4_offset
 *   l4 packet offset.
 *
 * @return
 *   Packet type information.
 */
static inline uint64_t
mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
				    uint8_t *l3_offset, uint8_t *l4_offset)
{
	enum pp2_inq_l3_type l3_type;
	enum pp2_inq_l4_type l4_type;
	enum pp2_inq_vlan_tag vlan_tag;
	uint64_t packet_type;

	pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
	pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
	pp2_ppio_inq_desc_get_vlan_tag(desc, &vlan_tag);

	packet_type = RTE_PTYPE_L2_ETHER;

	switch (vlan_tag) {
	case PP2_INQ_VLAN_TAG_SINGLE:
		packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
		break;
	case PP2_INQ_VLAN_TAG_DOUBLE:
	case PP2_INQ_VLAN_TAG_TRIPLE:
		packet_type |= RTE_PTYPE_L2_ETHER_QINQ;
		break;
	default:
		break;
	}

	switch (l3_type) {
	case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
		packet_type |= RTE_PTYPE_L3_IPV4;
		break;
	case PP2_INQ_L3_TYPE_IPV4_OK:
		packet_type |= RTE_PTYPE_L3_IPV4_EXT;
		break;
	case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
		break;
	case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
		packet_type |= RTE_PTYPE_L3_IPV6;
		break;
	case PP2_INQ_L3_TYPE_IPV6_EXT:
		packet_type |= RTE_PTYPE_L3_IPV6_EXT;
		break;
	case PP2_INQ_L3_TYPE_ARP:
		packet_type |= RTE_PTYPE_L2_ETHER_ARP;
		/*
		 * In case of ARP l4_offset is set to wrong value.
		 * Set it to proper one so that later on mbuf->l3_len can be
		 * calculated subtracting l4_offset and l3_offset.
		 */
		*l4_offset = *l3_offset + MRVL_ARP_LENGTH;
		break;
	default:
		MRVL_LOG(DEBUG, "Failed to recognise l3 packet type");
		break;
	}

	switch (l4_type) {
	case PP2_INQ_L4_TYPE_TCP:
		packet_type |= RTE_PTYPE_L4_TCP;
		break;
	case PP2_INQ_L4_TYPE_UDP:
		packet_type |= RTE_PTYPE_L4_UDP;
		break;
	default:
		MRVL_LOG(DEBUG, "Failed to recognise l4 packet type");
		break;
	}

	return packet_type;
}

/**
 * Get offload information from the received packet descriptor.
 *
 * @param desc
 *   Pointer to the received packet descriptor.
 *
 * @return
 *   Mbuf offload flags.
 */
static inline uint64_t
mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc)
{
	uint64_t flags;
	enum pp2_inq_desc_status status;

	status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
	if (unlikely(status != PP2_DESC_ERR_OK))
		flags = PKT_RX_IP_CKSUM_BAD;
	else
		flags = PKT_RX_IP_CKSUM_GOOD;

	status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
	if (unlikely(status != PP2_DESC_ERR_OK))
		flags |= PKT_RX_L4_CKSUM_BAD;
	else
		flags |= PKT_RX_L4_CKSUM_GOOD;

	return flags;
}

/**
 * DPDK callback for receive.
 *
 * @param rxq
 *   Generic pointer to the receive queue.
 * @param rx_pkts
 *   Array to store received packets.
 * @param nb_pkts
 *   Maximum number of packets in array.
 *
 * @return
 *   Number of packets successfully received.
 */
static uint16_t
mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
{
	struct mrvl_rxq *q = rxq;
	struct pp2_ppio_desc descs[nb_pkts];
	struct pp2_bpool *bpool;
	int i, ret, rx_done = 0;
	int num;
	struct pp2_hif *hif;
	unsigned int core_id = rte_lcore_id();

	hif = mrvl_get_hif(q->priv, core_id);

	if (unlikely(!q->priv->ppio || !hif))
		return 0;

	bpool = q->priv->bpool;

	ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
			    q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
	if (unlikely(ret < 0)) {
		MRVL_LOG(ERR, "Failed to receive packets");
		return 0;
	}
	mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;

	for (i = 0; i < nb_pkts; i++) {
		struct rte_mbuf *mbuf;
		uint8_t l3_offset, l4_offset;
		enum pp2_inq_desc_status status;
		uint64_t addr;

		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
			struct pp2_ppio_desc *pref_desc;
			u64 pref_addr;

			pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
			pref_addr = cookie_addr_high |
				    pp2_ppio_inq_desc_get_cookie(pref_desc);
			rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
			rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
		}

		addr = cookie_addr_high |
		       pp2_ppio_inq_desc_get_cookie(&descs[i]);
		mbuf = (struct rte_mbuf *)addr;
		rte_pktmbuf_reset(mbuf);

		/* drop packet in case of mac, overrun or resource error */
		status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
		if (unlikely(status != PP2_DESC_ERR_OK)) {
			struct pp2_buff_inf binf = {
				.addr = rte_mbuf_data_iova_default(mbuf),
				.cookie = (uint64_t)mbuf,
			};

			pp2_bpool_put_buff(hif, bpool, &binf);
			mrvl_port_bpool_size
				[bpool->pp2_id][bpool->id][core_id]++;
			q->drop_mac++;
			continue;
		}

		mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
		mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
		mbuf->data_len = mbuf->pkt_len;
		mbuf->port = q->port_id;
		mbuf->packet_type =
			mrvl_desc_to_packet_type_and_offset(&descs[i],
							    &l3_offset,
							    &l4_offset);
		mbuf->l2_len = l3_offset;
		mbuf->l3_len = l4_offset - l3_offset;

		if (likely(q->cksum_enabled))
			mbuf->ol_flags = mrvl_desc_to_ol_flags(&descs[i]);

		rx_pkts[rx_done++] = mbuf;
		q->bytes_recv += mbuf->pkt_len;
	}

	if (rte_spinlock_trylock(&q->priv->lock) == 1) {
		num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);

		if (unlikely(num <= q->priv->bpool_min_size ||
			     (!rx_done && num < q->priv->bpool_init_size))) {
			ret = mrvl_fill_bpool(q, MRVL_BURST_SIZE);
			if (ret)
				MRVL_LOG(ERR, "Failed to fill bpool");
		} else if (unlikely(num > q->priv->bpool_max_size)) {
			int i;
			int pkt_to_remove = num - q->priv->bpool_init_size;
			struct rte_mbuf *mbuf;
			struct pp2_buff_inf buff;

			MRVL_LOG(DEBUG,
				"port-%d:%d: bpool %d oversize - remove %d buffers (pool size: %d -> %d)",
				bpool->pp2_id, q->priv->ppio->port_id,
				bpool->id, pkt_to_remove, num,
				q->priv->bpool_init_size);

			for (i = 0; i < pkt_to_remove; i++) {
				ret = pp2_bpool_get_buff(hif, bpool, &buff);
				if (ret)
					break;
				mbuf = (struct rte_mbuf *)
					(cookie_addr_high | buff.cookie);
				rte_pktmbuf_free(mbuf);
			}
			mrvl_port_bpool_size
				[bpool->pp2_id][bpool->id][core_id] -= i;
		}
		rte_spinlock_unlock(&q->priv->lock);
	}

	return rx_done;
}

/**
 * Prepare offload information.
 *
 * @param ol_flags
 *   Offload flags.
 * @param packet_type
 *   Packet type bitfield.
 * @param l3_type
 *   Pointer to the pp2_ouq_l3_type structure.
 * @param l4_type
 *   Pointer to the pp2_outq_l4_type structure.
 * @param gen_l3_cksum
 *   Will be set to 1 in case l3 checksum is computed.
 * @param l4_cksum
 *   Will be set to 1 in case l4 checksum is computed.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static inline int
mrvl_prepare_proto_info(uint64_t ol_flags, uint32_t packet_type,
			enum pp2_outq_l3_type *l3_type,
			enum pp2_outq_l4_type *l4_type,
			int *gen_l3_cksum,
			int *gen_l4_cksum)
{
	/*
	 * Based on ol_flags prepare information
	 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
	 * for offloading.
	 */
	if (ol_flags & PKT_TX_IPV4) {
		*l3_type = PP2_OUTQ_L3_TYPE_IPV4;
		*gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
	} else if (ol_flags & PKT_TX_IPV6) {
		*l3_type = PP2_OUTQ_L3_TYPE_IPV6;
		/* no checksum for ipv6 header */
		*gen_l3_cksum = 0;
	} else {
		/* if something different then stop processing */
		return -1;
	}

	ol_flags &= PKT_TX_L4_MASK;
	if ((packet_type & RTE_PTYPE_L4_TCP) &&
	    ol_flags == PKT_TX_TCP_CKSUM) {
		*l4_type = PP2_OUTQ_L4_TYPE_TCP;
		*gen_l4_cksum = 1;
	} else if ((packet_type & RTE_PTYPE_L4_UDP) &&
		   ol_flags == PKT_TX_UDP_CKSUM) {
		*l4_type = PP2_OUTQ_L4_TYPE_UDP;
		*gen_l4_cksum = 1;
	} else {
		*l4_type = PP2_OUTQ_L4_TYPE_OTHER;
		/* no checksum for other type */
		*gen_l4_cksum = 0;
	}

	return 0;
}

/**
 * Release already sent buffers to bpool (buffer-pool).
 *
 * @param ppio
 *   Pointer to the port structure.
 * @param hif
 *   Pointer to the MUSDK hardware interface.
 * @param sq
 *   Pointer to the shadow queue.
 * @param qid
 *   Queue id number.
 * @param force
 *   Force releasing packets.
 */
static inline void
mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
		       unsigned int core_id, struct mrvl_shadow_txq *sq,
		       int qid, int force)
{
	struct buff_release_entry *entry;
	uint16_t nb_done = 0, num = 0, skip_bufs = 0;
	int i;

	pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);

	sq->num_to_release += nb_done;

	if (likely(!force &&
		   sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
		return;

	nb_done = sq->num_to_release;
	sq->num_to_release = 0;

	for (i = 0; i < nb_done; i++) {
		entry = &sq->ent[sq->tail + num];
		if (unlikely(!entry->buff.addr)) {
			MRVL_LOG(ERR,
				"Shadow memory @%d: cookie(%lx), pa(%lx)!",
				sq->tail, (u64)entry->buff.cookie,
				(u64)entry->buff.addr);
			skip_bufs = 1;
			goto skip;
		}

		if (unlikely(!entry->bpool)) {
			struct rte_mbuf *mbuf;

			mbuf = (struct rte_mbuf *)
			       (cookie_addr_high | entry->buff.cookie);
			rte_pktmbuf_free(mbuf);
			skip_bufs = 1;
			goto skip;
		}

		mrvl_port_bpool_size
			[entry->bpool->pp2_id][entry->bpool->id][core_id]++;
		num++;
		if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
			goto skip;
		continue;
skip:
		if (likely(num))
			pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
		num += skip_bufs;
		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
		sq->size -= num;
		num = 0;
		skip_bufs = 0;
	}

	if (likely(num)) {
		pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
		sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
		sq->size -= num;
	}
}

/**
 * DPDK callback for transmit.
 *
 * @param txq
 *   Generic pointer transmit queue.
 * @param tx_pkts
 *   Packets to transmit.
 * @param nb_pkts
 *   Number of packets in array.
 *
 * @return
 *   Number of packets successfully transmitted.
 */
static uint16_t
mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
{
	struct mrvl_txq *q = txq;
	struct mrvl_shadow_txq *sq;
	struct pp2_hif *hif;
	struct pp2_ppio_desc descs[nb_pkts];
	unsigned int core_id = rte_lcore_id();
	int i, ret, bytes_sent = 0;
	uint16_t num, sq_free_size;
	uint64_t addr;

	hif = mrvl_get_hif(q->priv, core_id);
	sq = &q->shadow_txqs[core_id];

	if (unlikely(!q->priv->ppio || !hif))
		return 0;

	if (sq->size)
		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
				       sq, q->queue_id, 0);

	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
	if (unlikely(nb_pkts > sq_free_size)) {
		MRVL_LOG(DEBUG,
			"No room in shadow queue for %d packets! %d packets will be sent.",
			nb_pkts, sq_free_size);
		nb_pkts = sq_free_size;
	}

	for (i = 0; i < nb_pkts; i++) {
		struct rte_mbuf *mbuf = tx_pkts[i];
		int gen_l3_cksum, gen_l4_cksum;
		enum pp2_outq_l3_type l3_type;
		enum pp2_outq_l4_type l4_type;

		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
			struct rte_mbuf *pref_pkt_hdr;

			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
			rte_mbuf_prefetch_part1(pref_pkt_hdr);
			rte_mbuf_prefetch_part2(pref_pkt_hdr);
		}

		mrvl_fill_shadowq(sq, mbuf);
		mrvl_fill_desc(&descs[i], mbuf);

		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
		/*
		 * in case unsupported ol_flags were passed
		 * do not update descriptor offload information
		 */
		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
					      &l3_type, &l4_type, &gen_l3_cksum,
					      &gen_l4_cksum);
		if (unlikely(ret))
			continue;

		pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
						  mbuf->l2_len,
						  mbuf->l2_len + mbuf->l3_len,
						  gen_l3_cksum, gen_l4_cksum);
	}

	num = nb_pkts;
	pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
	/* number of packets that were not sent */
	if (unlikely(num > nb_pkts)) {
		for (i = nb_pkts; i < num; i++) {
			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
				MRVL_PP2_TX_SHADOWQ_MASK;
			addr = cookie_addr_high | sq->ent[sq->head].buff.cookie;
			bytes_sent -=
				rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
		}
		sq->size -= num - nb_pkts;
	}

	q->bytes_sent += bytes_sent;

	return nb_pkts;
}

/** DPDK callback for S/G transmit.
 *
 * @param txq
 *   Generic pointer transmit queue.
 * @param tx_pkts
 *   Packets to transmit.
 * @param nb_pkts
 *   Number of packets in array.
 *
 * @return
 *   Number of packets successfully transmitted.
 */
static uint16_t
mrvl_tx_sg_pkt_burst(void *txq, struct rte_mbuf **tx_pkts,
		     uint16_t nb_pkts)
{
	struct mrvl_txq *q = txq;
	struct mrvl_shadow_txq *sq;
	struct pp2_hif *hif;
	struct pp2_ppio_desc descs[nb_pkts * PP2_PPIO_DESC_NUM_FRAGS];
	struct pp2_ppio_sg_pkts pkts;
	uint8_t frags[nb_pkts];
	unsigned int core_id = rte_lcore_id();
	int i, j, ret, bytes_sent = 0;
	int tail, tail_first;
	uint16_t num, sq_free_size;
	uint16_t nb_segs, total_descs = 0;
	uint64_t addr;

	hif = mrvl_get_hif(q->priv, core_id);
	sq = &q->shadow_txqs[core_id];
	pkts.frags = frags;
	pkts.num = 0;

	if (unlikely(!q->priv->ppio || !hif))
		return 0;

	if (sq->size)
		mrvl_free_sent_buffers(q->priv->ppio, hif, core_id,
				       sq, q->queue_id, 0);

	/* Save shadow queue free size */
	sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;

	tail = 0;
	for (i = 0; i < nb_pkts; i++) {
		struct rte_mbuf *mbuf = tx_pkts[i];
		struct rte_mbuf *seg = NULL;
		int gen_l3_cksum, gen_l4_cksum;
		enum pp2_outq_l3_type l3_type;
		enum pp2_outq_l4_type l4_type;

		nb_segs = mbuf->nb_segs;
		tail_first = tail;
		total_descs += nb_segs;

		/*
		 * Check if total_descs does not exceed
		 * shadow queue free size
		 */
		if (unlikely(total_descs > sq_free_size)) {
			total_descs -= nb_segs;
			RTE_LOG(DEBUG, PMD,
				"No room in shadow queue for %d packets! "
				"%d packets will be sent.\n",
				nb_pkts, i);
			break;
		}

		/* Check if nb_segs does not exceed the max nb of desc per
		 * fragmented packet
		 */
		if (nb_segs > PP2_PPIO_DESC_NUM_FRAGS) {
			total_descs -= nb_segs;
			RTE_LOG(ERR, PMD,
				"Too many segments. Packet won't be sent.\n");
			break;
		}

		if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
			struct rte_mbuf *pref_pkt_hdr;

			pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
			rte_mbuf_prefetch_part1(pref_pkt_hdr);
			rte_mbuf_prefetch_part2(pref_pkt_hdr);
		}

		pkts.frags[pkts.num] = nb_segs;
		pkts.num++;

		seg = mbuf;
		for (j = 0; j < nb_segs - 1; j++) {
			/* For the subsequent segments, set shadow queue
			 * buffer to NULL
			 */
			mrvl_fill_shadowq(sq, NULL);
			mrvl_fill_desc(&descs[tail], seg);

			tail++;
			seg = seg->next;
		}
		/* Put first mbuf info in last shadow queue entry */
		mrvl_fill_shadowq(sq, mbuf);
		/* Update descriptor with last segment */
		mrvl_fill_desc(&descs[tail++], seg);

		bytes_sent += rte_pktmbuf_pkt_len(mbuf);
		/* In case unsupported ol_flags were passed
		 * do not update descriptor offload information
		 */
		ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
					      &l3_type, &l4_type, &gen_l3_cksum,
					      &gen_l4_cksum);
		if (unlikely(ret))
			continue;

		pp2_ppio_outq_desc_set_proto_info(&descs[tail_first], l3_type,
						  l4_type, mbuf->l2_len,
						  mbuf->l2_len + mbuf->l3_len,
						  gen_l3_cksum, gen_l4_cksum);
	}

	num = total_descs;
	pp2_ppio_send_sg(q->priv->ppio, hif, q->queue_id, descs,
			 &total_descs, &pkts);
	/* number of packets that were not sent */
	if (unlikely(num > total_descs)) {
		for (i = total_descs; i < num; i++) {
			sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
				MRVL_PP2_TX_SHADOWQ_MASK;

			addr = sq->ent[sq->head].buff.cookie;
			if (addr)
				bytes_sent -=
					rte_pktmbuf_pkt_len((struct rte_mbuf *)
						(cookie_addr_high | addr));
		}
		sq->size -= num - total_descs;
		nb_pkts = pkts.num;
	}

	q->bytes_sent += bytes_sent;

	return nb_pkts;
}

/**
 * Initialize packet processor.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_init_pp2(void)
{
	struct pp2_init_params init_params;

	memset(&init_params, 0, sizeof(init_params));
	init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
	init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
	init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;

	return pp2_init(&init_params);
}

/**
 * Deinitialize packet processor.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static void
mrvl_deinit_pp2(void)
{
	pp2_deinit();
}

/**
 * Create private device structure.
 *
 * @param dev_name
 *   Pointer to the port name passed in the initialization parameters.
 *
 * @return
 *   Pointer to the newly allocated private device structure.
 */
static struct mrvl_priv *
mrvl_priv_create(const char *dev_name)
{
	struct pp2_bpool_params bpool_params;
	char match[MRVL_MATCH_LEN];
	struct mrvl_priv *priv;
	int ret, bpool_bit;

	priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
	if (!priv)
		return NULL;

	ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
				       &priv->pp_id, &priv->ppio_id);
	if (ret)
		goto out_free_priv;

	bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
				     PP2_BPOOL_NUM_POOLS);
	if (bpool_bit < 0)
		goto out_free_priv;
	priv->bpool_bit = bpool_bit;

	snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
		 priv->bpool_bit);
	memset(&bpool_params, 0, sizeof(bpool_params));
	bpool_params.match = match;
	bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
	ret = pp2_bpool_init(&bpool_params, &priv->bpool);
	if (ret)
		goto out_clear_bpool_bit;

	priv->ppio_params.type = PP2_PPIO_T_NIC;
	rte_spinlock_init(&priv->lock);

	return priv;
out_clear_bpool_bit:
	used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
out_free_priv:
	rte_free(priv);
	return NULL;
}

/**
 * Create device representing Ethernet port.
 *
 * @param name
 *   Pointer to the port's name.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
{
	int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
	struct rte_eth_dev *eth_dev;
	struct mrvl_priv *priv;
	struct ifreq req;

	eth_dev = rte_eth_dev_allocate(name);
	if (!eth_dev)
		return -ENOMEM;

	priv = mrvl_priv_create(name);
	if (!priv) {
		ret = -ENOMEM;
		goto out_free;
	}
	eth_dev->data->dev_private = priv;

	eth_dev->data->mac_addrs =
		rte_zmalloc("mac_addrs",
			    RTE_ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
	if (!eth_dev->data->mac_addrs) {
		MRVL_LOG(ERR, "Failed to allocate space for eth addrs");
		ret = -ENOMEM;
		goto out_free;
	}

	memset(&req, 0, sizeof(req));
	strcpy(req.ifr_name, name);
	ret = ioctl(fd, SIOCGIFHWADDR, &req);
	if (ret)
		goto out_free;

	memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
	       req.ifr_addr.sa_data, RTE_ETHER_ADDR_LEN);

	eth_dev->device = &vdev->device;
	eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
	mrvl_set_tx_function(eth_dev);
	eth_dev->dev_ops = &mrvl_ops;
	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;

	rte_eth_dev_probing_finish(eth_dev);
	return 0;
out_free:
	rte_eth_dev_release_port(eth_dev);

	return ret;
}

/**
 * Callback used by rte_kvargs_process() during argument parsing.
 *
 * @param key
 *   Pointer to the parsed key (unused).
 * @param value
 *   Pointer to the parsed value.
 * @param extra_args
 *   Pointer to the extra arguments which contains address of the
 *   table of pointers to parsed interface names.
 *
 * @return
 *   Always 0.
 */
static int
mrvl_get_ifnames(const char *key __rte_unused, const char *value,
		 void *extra_args)
{
	struct mrvl_ifnames *ifnames = extra_args;

	ifnames->names[ifnames->idx++] = value;

	return 0;
}

/**
 * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
 */
static void
mrvl_deinit_hifs(void)
{
	int i;

	for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++) {
		if (hifs[i])
			pp2_hif_deinit(hifs[i]);
	}
	used_hifs = MRVL_MUSDK_HIFS_RESERVED;
	memset(hifs, 0, sizeof(hifs));
}

/**
 * DPDK callback to register the virtual device.
 *
 * @param vdev
 *   Pointer to the virtual device.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
{
	struct rte_kvargs *kvlist;
	struct mrvl_ifnames ifnames;
	int ret = -EINVAL;
	uint32_t i, ifnum, cfgnum;
	const char *params;

	params = rte_vdev_device_args(vdev);
	if (!params)
		return -EINVAL;

	kvlist = rte_kvargs_parse(params, valid_args);
	if (!kvlist)
		return -EINVAL;

	ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
	if (ifnum > RTE_DIM(ifnames.names))
		goto out_free_kvlist;

	ifnames.idx = 0;
	rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
			   mrvl_get_ifnames, &ifnames);


	/*
	 * The below system initialization should be done only once,
	 * on the first provided configuration file
	 */
	if (!mrvl_qos_cfg) {
		cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
		MRVL_LOG(INFO, "Parsing config file!");
		if (cfgnum > 1) {
			MRVL_LOG(ERR, "Cannot handle more than one config file!");
			goto out_free_kvlist;
		} else if (cfgnum == 1) {
			rte_kvargs_process(kvlist, MRVL_CFG_ARG,
					   mrvl_get_qoscfg, &mrvl_qos_cfg);
		}
	}

	if (mrvl_dev_num)
		goto init_devices;

	MRVL_LOG(INFO, "Perform MUSDK initializations");

	ret = rte_mvep_init(MVEP_MOD_T_PP2, kvlist);
	if (ret)
		goto out_free_kvlist;

	ret = mrvl_init_pp2();
	if (ret) {
		MRVL_LOG(ERR, "Failed to init PP!");
		rte_mvep_deinit(MVEP_MOD_T_PP2);
		goto out_free_kvlist;
	}

	memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
	memset(mrvl_port_to_bpool_lookup, 0, sizeof(mrvl_port_to_bpool_lookup));

	mrvl_lcore_first = RTE_MAX_LCORE;
	mrvl_lcore_last = 0;

init_devices:
	for (i = 0; i < ifnum; i++) {
		MRVL_LOG(INFO, "Creating %s", ifnames.names[i]);
		ret = mrvl_eth_dev_create(vdev, ifnames.names[i]);
		if (ret)
			goto out_cleanup;
		mrvl_dev_num++;
	}

	rte_kvargs_free(kvlist);

	return 0;
out_cleanup:
	rte_pmd_mrvl_remove(vdev);

out_free_kvlist:
	rte_kvargs_free(kvlist);

	return ret;
}

/**
 * DPDK callback to remove virtual device.
 *
 * @param vdev
 *   Pointer to the removed virtual device.
 *
 * @return
 *   0 on success, negative error value otherwise.
 */
static int
rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
{
	uint16_t port_id;
	int ret = 0;

	RTE_ETH_FOREACH_DEV(port_id) {
		if (rte_eth_devices[port_id].device != &vdev->device)
			continue;
		ret |= rte_eth_dev_close(port_id);
	}

	return ret == 0 ? 0 : -EIO;
}

static struct rte_vdev_driver pmd_mrvl_drv = {
	.probe = rte_pmd_mrvl_probe,
	.remove = rte_pmd_mrvl_remove,
};

RTE_PMD_REGISTER_VDEV(net_mvpp2, pmd_mrvl_drv);
RTE_PMD_REGISTER_ALIAS(net_mvpp2, eth_mvpp2);
RTE_LOG_REGISTER(mrvl_logtype, pmd.net.mvpp2, NOTICE);