@@ -262,6 +262,13 @@ CONFIG_RTE_LIBRTE_NFP_PMD=n
CONFIG_RTE_LIBRTE_NFP_DEBUG=n
#
+# Compile Marvell PMD driver
+#
+CONFIG_RTE_LIBRTE_MRVL_PMD=n
+CONFIG_RTE_LIBRTE_MRVL_DEBUG=n
+CONFIG_RTE_MRVL_MUSDK_DMA_MEMSIZE=41943040
+
+#
# Compile burst-oriented Broadcom BNXT PMD driver
#
CONFIG_RTE_LIBRTE_BNXT_PMD=y
@@ -73,6 +73,8 @@ DIRS-$(CONFIG_RTE_LIBRTE_MLX4_PMD) += mlx4
DEPDIRS-mlx4 = $(core-libs)
DIRS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += mlx5
DEPDIRS-mlx5 = $(core-libs)
+DIRS-$(CONFIG_RTE_LIBRTE_MRVL_PMD) += mrvl
+DEPDIRS-mrvl = $(core-libs)
DIRS-$(CONFIG_RTE_LIBRTE_NFP_PMD) += nfp
DEPDIRS-nfp = $(core-libs)
DIRS-$(CONFIG_RTE_LIBRTE_BNXT_PMD) += bnxt
new file mode 100644
@@ -0,0 +1,69 @@
+# BSD LICENSE
+#
+# Copyright(c) 2017 Semihalf. All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#
+# * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+# * Redistributions in binary form must reproduce the above copyright
+# notice, this list of conditions and the following disclaimer in
+# the documentation and/or other materials provided with the
+# distribution.
+# * Neither the name of Semihalf nor the names of its
+# contributors may be used to endorse or promote products derived
+# from this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+include $(RTE_SDK)/mk/rte.vars.mk
+
+ifneq ($(MAKECMDGOALS),clean)
+ifneq ($(MAKECMDGOALS),config)
+ifeq ($(LIBMUSDK_PATH),)
+$(error "Please define LIBMUSDK_PATH environment variable")
+endif
+ifeq ($(CONFIG_RTE_LIBRTE_CFGFILE),n)
+$(error "RTE_LIBRTE_CFGFILE must be enabled in configuration!")
+endif
+endif
+endif
+
+# library name
+LIB = librte_pmd_mrvl.a
+
+# library version
+LIBABIVER := 1
+
+# versioning export map
+EXPORT_MAP := rte_pmd_mrvl_version.map
+
+# external library dependencies
+CFLAGS += -I$(LIBMUSDK_PATH)/include
+CFLAGS += -DMVCONF_ARCH_DMA_ADDR_T_64BIT
+CFLAGS += -DCONF_PP2_BPOOL_COOKIE_SIZE=32
+CFLAGS += $(WERROR_FLAGS)
+CFLAGS += -O3
+LDLIBS += -L$(LIBMUSDK_PATH)/lib
+LDLIBS += -lmusdk
+
+# library source files
+SRCS-$(CONFIG_RTE_LIBRTE_MRVL_PMD) += mrvl_ethdev.c
+SRCS-$(CONFIG_RTE_LIBRTE_MRVL_PMD) += mrvl_qos.c
+
+# library dependencies
+DEPDIRS-$(CONFIG_RTE_LIBRTE_MRVL_PMD) += lib/librte_cfgfile
+
+include $(RTE_SDK)/mk/rte.lib.mk
new file mode 100644
@@ -0,0 +1,2277 @@
+/*-
+ * BSD LICENSE
+ *
+ * Copyright(c) 2017 Semihalf. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * * Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in
+ * the documentation and/or other materials provided with the
+ * distribution.
+ * * Neither the name of Semihalf nor the names of its
+ * contributors may be used to endorse or promote products derived
+ * from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <rte_ethdev.h>
+#include <rte_kvargs.h>
+#include <rte_log.h>
+#include <rte_malloc.h>
+#include <rte_vdev.h>
+#include <rte_cycles.h>
+
+/* Unluckily, container_of is defined by both DPDK and MUSDK,
+ * we'll declare only one version.
+ *
+ * Note that it is not used in this PMD anyway.
+ */
+#ifdef container_of
+#undef container_of
+#endif
+
+#include <drivers/mv_pp2.h>
+#include <drivers/mv_pp2_bpool.h>
+#include <drivers/mv_pp2_hif.h>
+
+#include <assert.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 "mrvl_ethdev.h"
+#include "mrvl_qos.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_SHIFT (sizeof(pp2_cookie_t) * 8)
+#define MRVL_COOKIE_HIGH_ADDR_MASK (~0ULL << MRVL_COOKIE_HIGH_ADDR_SHIFT)
+
+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] = {
+ MRVL_MUSDK_BPOOLS_RESERVED,
+ MRVL_MUSDK_BPOOLS_RESERVED
+};
+
+struct pp2_bpool *mrvl_port_to_bpool_lookup[RTE_MAX_ETHPORTS];
+int mrvl_port_bpool_size[PP2_NUM_PKT_PROC][PP2_BPOOL_NUM_POOLS][RTE_MAX_LCORE];
+uint64_t cookie_addr_high = MRVL_COOKIE_ADDR_INVALID;
+
+/*
+ * 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 buffers sent, that can be released */
+ struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
+};
+
+struct mrvl_rxq;
+struct mrvl_txq;
+
+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;
+};
+
+/*
+ * Every tx queue should have dedicated shadow tx queue.
+ *
+ * Ports assigned by DPDK might not start at zero or be continuous so
+ * as a workaround define shadow queues for each possible port so that
+ * we eventually fit somewhere.
+ */
+struct mrvl_shadow_txq shadow_txqs[RTE_MAX_ETHPORTS][RTE_MAX_LCORE];
+
+/** Number of ports configured. */
+int mrvl_ports_nb;
+static int mrvl_lcore_first;
+static int mrvl_lcore_last;
+
+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;
+}
+
+/**
+ * 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)
+ RTE_LOG(WARNING, PMD, "Changing hash key is not supported\n");
+
+ 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 (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_NONE &&
+ dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
+ RTE_LOG(INFO, PMD, "Unsupported rx multi queue mode %d\n",
+ dev->data->dev_conf.rxmode.mq_mode);
+ return -EINVAL;
+ }
+
+ if (!dev->data->dev_conf.rxmode.hw_strip_crc) {
+ RTE_LOG(INFO, PMD,
+ "L2 CRC stripping is always enabled in hw\n");
+ dev->data->dev_conf.rxmode.hw_strip_crc = 1;
+ }
+
+ if (dev->data->dev_conf.rxmode.hw_vlan_strip) {
+ RTE_LOG(INFO, PMD, "VLAN stripping not supported\n");
+ return -EINVAL;
+ }
+
+ if (dev->data->dev_conf.rxmode.split_hdr_size) {
+ RTE_LOG(INFO, PMD, "Split headers not supported\n");
+ return -EINVAL;
+ }
+
+ if (dev->data->dev_conf.rxmode.enable_scatter) {
+ RTE_LOG(INFO, PMD, "RX Scatter/Gather not supported\n");
+ return -EINVAL;
+ }
+
+ if (dev->data->dev_conf.rxmode.enable_lro) {
+ RTE_LOG(INFO, PMD, "LRO not supported\n");
+ return -EINVAL;
+ }
+
+ if (dev->data->dev_conf.rxmode.jumbo_frame)
+ dev->data->mtu = dev->data->dev_conf.rxmode.max_rx_pkt_len -
+ ETHER_HDR_LEN - ETHER_CRC_LEN;
+
+ ret = mrvl_configure_rxqs(priv, dev->data->port_id,
+ dev->data->nb_rx_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;
+
+ if (dev->data->nb_rx_queues == 1 &&
+ dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
+ RTE_LOG(WARNING, PMD, "Disabling hash for 1 rx queue\n");
+ 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;
+ /* extra MV_MH_SIZE bytes are required for Marvell tag */
+ uint16_t mru = mtu + MV_MH_SIZE + ETHER_HDR_LEN + ETHER_CRC_LEN;
+ int ret;
+
+ if (mtu < ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX)
+ return -EINVAL;
+
+ ret = pp2_ppio_set_mru(priv->ppio, mru);
+ if (ret)
+ return ret;
+
+ return pp2_ppio_set_mtu(priv->ppio, mtu);
+}
+
+/**
+ * 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;
+
+ 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);
+
+ dev->data->dev_link.link_status = ETH_LINK_UP;
+
+ 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;
+ int ret;
+
+ ret = pp2_ppio_disable(priv->ppio);
+ if (ret)
+ return ret;
+
+ dev->data->dev_link.link_status = ETH_LINK_DOWN;
+
+ return ret;
+}
+
+/**
+ * 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;
+
+ snprintf(match, sizeof(match), "ppio-%d:%d",
+ priv->pp_id, priv->ppio_id);
+ priv->ppio_params.match = match;
+
+ /*
+ * Calculate the maximum bpool size for refill feature to 1.5 of the
+ * configured size. In case the bpool size will exceed this value,
+ * superfluous buffers will be removed
+ */
+ priv->bpool_max_size = priv->bpool_init_size +
+ (priv->bpool_init_size >> 1);
+ /*
+ * 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;
+
+ ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
+ if (ret)
+ 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) {
+ RTE_LOG(ERR, PMD,
+ "Failed to flush uc/mc filter list\n");
+ goto out;
+ }
+ priv->uc_mc_flushed = 1;
+ }
+
+ if (!priv->vlan_flushed) {
+ ret = pp2_ppio_flush_vlan(priv->ppio);
+ if (ret) {
+ RTE_LOG(ERR, PMD, "Failed to flush vlan list\n");
+ /*
+ * TODO
+ * once pp2_ppio_flush_vlan() is supported jump to out
+ * goto out;
+ */
+ }
+ priv->vlan_flushed = 1;
+ }
+
+ /* For default QoS config, don't start classifier. */
+ if (mrvl_qos_cfg != NULL) {
+ ret = mrvl_start_qos_mapping(priv);
+ if (ret) {
+ pp2_ppio_deinit(priv->ppio);
+ return ret;
+ }
+ }
+
+ ret = mrvl_dev_set_link_up(dev);
+ if (ret)
+ goto out;
+
+ return 0;
+out:
+ 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;
+
+ RTE_LOG(INFO, PMD, "Flushing rx queues\n");
+ 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;
+
+ RTE_LOG(INFO, PMD, "Flushing tx shadow queues\n");
+ for (i = 0; i < RTE_MAX_LCORE; i++) {
+ struct mrvl_shadow_txq *sq =
+ &shadow_txqs[dev->data->port_id][i];
+
+ 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;
+ uint32_t num;
+ int ret;
+
+ ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
+ if (ret) {
+ RTE_LOG(ERR, PMD, "Failed to get bpool buffers number\n");
+ return;
+ }
+
+ while (num--) {
+ struct pp2_buff_inf inf;
+ uint64_t addr;
+
+ ret = pp2_bpool_get_buff(hifs[rte_lcore_id()], 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 void
+mrvl_dev_stop(struct rte_eth_dev *dev)
+{
+ struct mrvl_priv *priv = dev->data->dev_private;
+
+ mrvl_dev_set_link_down(dev);
+ mrvl_flush_rx_queues(dev);
+ mrvl_flush_tx_shadow_queues(dev);
+ if (priv->qos_tbl)
+ pp2_cls_qos_tbl_deinit(priv->qos_tbl);
+ pp2_ppio_deinit(priv->ppio);
+ priv->ppio = NULL;
+}
+
+/**
+ * DPDK callback to close the device.
+ *
+ * @param dev
+ * Pointer to Ethernet device structure.
+ */
+static void
+mrvl_dev_close(struct rte_eth_dev *dev)
+{
+ struct mrvl_priv *priv = dev->data->dev_private;
+ size_t i;
+
+ for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i)
+ if (priv->ppio_params.inqs_params.tcs_params[i].inqs_params) {
+ rte_free(priv->ppio_params.inqs_params.
+ tcs_params[i].inqs_params);
+ priv->ppio_params.inqs_params.
+ tcs_params[i].inqs_params = NULL;
+ }
+
+ mrvl_flush_bpool(dev);
+}
+
+/**
+ * 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 ethtool_cmd edata;
+ struct ifreq req;
+ int ret, fd;
+
+ 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;
+
+ return 0;
+}
+
+/**
+ * DPDK callback to enable promiscuous mode.
+ *
+ * @param dev
+ * Pointer to Ethernet device structure.
+ */
+static void
+mrvl_promiscuous_enable(struct rte_eth_dev *dev)
+{
+ struct mrvl_priv *priv = dev->data->dev_private;
+ int ret;
+
+ ret = pp2_ppio_set_uc_promisc(priv->ppio, 1);
+ if (ret)
+ RTE_LOG(ERR, PMD, "Failed to enable promiscuous mode\n");
+}
+
+/**
+ * DPDK callback to enable allmulti mode.
+ *
+ * @param dev
+ * Pointer to Ethernet device structure.
+ */
+static void
+mrvl_allmulticast_enable(struct rte_eth_dev *dev)
+{
+ struct mrvl_priv *priv = dev->data->dev_private;
+ int ret;
+
+ ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
+ if (ret)
+ RTE_LOG(ERR, PMD, "Failed enable all-multicast mode\n");
+}
+
+/**
+ * DPDK callback to disable promiscuous mode.
+ *
+ * @param dev
+ * Pointer to Ethernet device structure.
+ */
+static void
+mrvl_promiscuous_disable(struct rte_eth_dev *dev)
+{
+ struct mrvl_priv *priv = dev->data->dev_private;
+ int ret;
+
+ ret = pp2_ppio_set_uc_promisc(priv->ppio, 0);
+ if (ret)
+ RTE_LOG(ERR, PMD, "Failed to disable promiscuous mode\n");
+}
+
+/**
+ * DPDK callback to disable allmulticast mode.
+ *
+ * @param dev
+ * Pointer to Ethernet device structure.
+ */
+static void
+mrvl_allmulticast_disable(struct rte_eth_dev *dev)
+{
+ struct mrvl_priv *priv = dev->data->dev_private;
+ int ret;
+
+ ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
+ if (ret)
+ RTE_LOG(ERR, PMD, "Failed to disable all-multicast mode\n");
+}
+
+/**
+ * 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[ETHER_ADDR_FMT_SIZE];
+ int ret;
+
+ ret = pp2_ppio_remove_mac_addr(priv->ppio,
+ dev->data->mac_addrs[index].addr_bytes);
+ if (ret) {
+ ether_format_addr(buf, sizeof(buf),
+ &dev->data->mac_addrs[index]);
+ RTE_LOG(ERR, PMD, "Failed to remove mac %s\n", 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 ether_addr *mac_addr,
+ uint32_t index, uint32_t vmdq __rte_unused)
+{
+ struct mrvl_priv *priv = dev->data->dev_private;
+ char buf[ETHER_ADDR_FMT_SIZE];
+ int ret;
+
+ if (index == 0)
+ /* For setting index 0, mrvl_mac_addr_set() should be used.*/
+ return -1;
+
+ /*
+ * 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) {
+ ether_format_addr(buf, sizeof(buf), mac_addr);
+ RTE_LOG(ERR, PMD, "Failed to add mac %s\n", 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.
+ */
+static void
+mrvl_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
+{
+ struct mrvl_priv *priv = dev->data->dev_private;
+
+ pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
+ /*
+ * TODO
+ * Port stops sending packets if pp2_ppio_set_mac_addr()
+ * was called after pp2_ppio_enable(). As a quick fix issue
+ * enable port once again.
+ */
+ pp2_ppio_enable(priv->ppio);
+}
+
+/**
+ * DPDK callback to get device statistics.
+ *
+ * @param dev
+ * Pointer to Ethernet device structure.
+ * @param stats
+ * Stats structure output buffer.
+ */
+static void
+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;
+
+ 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)) {
+ RTE_LOG(ERR, PMD,
+ "rx queue %d stats out of range (0 - %d)\n",
+ 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)) {
+ RTE_LOG(ERR, PMD,
+ "Failed to update rx queue %d stats\n", 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)) {
+ RTE_LOG(ERR, PMD,
+ "tx queue %d stats out of range (0 - %d)\n",
+ idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
+ }
+
+ ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
+ &tx_stats, 0);
+ if (unlikely(ret)) {
+ RTE_LOG(ERR, PMD,
+ "Failed to update tx queue %d stats\n", 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)) {
+ RTE_LOG(ERR, PMD, "Failed to update port statistics\n");
+ return;
+ }
+
+ 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;
+}
+
+/**
+ * DPDK callback to clear device statistics.
+ *
+ * @param dev
+ * Pointer to Ethernet device structure.
+ */
+static void
+mrvl_stats_reset(struct rte_eth_dev *dev)
+{
+ struct mrvl_priv *priv = dev->data->dev_private;
+ int i;
+
+ 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;
+ }
+
+ pp2_ppio_get_statistics(priv->ppio, NULL, 1);
+}
+
+/**
+ * DPDK callback to get information about the device.
+ *
+ * @param dev
+ * Pointer to Ethernet device structure (unused).
+ * @param info
+ * Info structure output buffer.
+ */
+static void
+mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
+ struct rte_eth_dev_info *info)
+{
+ 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 = DEV_RX_OFFLOAD_IPV4_CKSUM |
+ DEV_RX_OFFLOAD_UDP_CKSUM |
+ DEV_RX_OFFLOAD_TCP_CKSUM;
+
+ info->tx_offload_capa = DEV_TX_OFFLOAD_IPV4_CKSUM |
+ DEV_TX_OFFLOAD_UDP_CKSUM |
+ DEV_TX_OFFLOAD_TCP_CKSUM;
+
+ 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 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_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;
+
+ qinfo->mp = q->mp;
+ qinfo->nb_desc = priv->ppio_params.inqs_params.
+ tcs_params[tc].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;
+
+ qinfo->nb_desc =
+ priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
+}
+
+/**
+ * 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;
+
+ 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_TXD_MAX];
+ struct rte_mbuf *mbufs[MRVL_PP2_TXD_MAX];
+ int i, ret;
+ unsigned int core_id = rte_lcore_id();
+ struct pp2_hif *hif = hifs[core_id];
+ struct pp2_bpool *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) {
+ RTE_LOG(ERR, PMD,
+ "mbuf virtual addr high 0x%lx out of range\n",
+ (uint64_t)mbufs[i] >> 32);
+ goto out;
+ }
+
+ entries[i].buff.addr =
+ rte_mbuf_data_dma_addr_default(mbufs[i]);
+ entries[i].buff.cookie = (pp2_cookie_t)(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 (unused_).
+ * @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 __rte_unused,
+ struct rte_mempool *mp)
+{
+ struct mrvl_priv *priv = dev->data->dev_private;
+ struct mrvl_rxq *rxq;
+ uint32_t min_size,
+ max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
+ int ret;
+
+ if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
+ /*
+ * Unknown TC mapping, mapping will not have a correct queue.
+ */
+ RTE_LOG(ERR, PMD, "Unknown TC mapping for queue %hu eth%hhu\n",
+ idx, priv->ppio_id);
+ return -EFAULT;
+ }
+
+ min_size = rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM -
+ MRVL_PKT_EFFEC_OFFS;
+ if (min_size < max_rx_pkt_len) {
+ RTE_LOG(ERR, PMD, "Mbuf size must be increased to %u bytes"
+ " to hold up to %u bytes of data.\n",
+ max_rx_pkt_len + RTE_PKTMBUF_HEADROOM +
+ MRVL_PKT_EFFEC_OFFS,
+ max_rx_pkt_len);
+ return -EINVAL;
+ }
+
+ 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 = dev->data->dev_conf.rxmode.hw_ip_checksum;
+ rxq->queue_id = idx;
+ rxq->port_id = dev->data->port_id;
+ mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
+
+ priv->ppio_params.inqs_params.
+ tcs_params[priv->rxq_map[rxq->queue_id].tc].
+ inqs_params[priv->rxq_map[rxq->queue_id].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;
+ int i, num;
+
+ if (!q)
+ return;
+
+ num = q->priv->ppio_params.inqs_params.
+ tcs_params[q->priv->rxq_map[q->queue_id].tc].
+ inqs_params[q->priv->rxq_map[q->queue_id].inq].size;
+ for (i = 0; i < num; i++) {
+ struct pp2_buff_inf inf;
+ uint64_t addr;
+
+ pp2_bpool_get_buff(hifs[rte_lcore_id()], 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
+ * Thresholds parameters (unused).
+ *
+ * @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 __rte_unused)
+{
+ 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;
+ dev->data->tx_queues[idx] = txq;
+
+ priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
+ priv->ppio_params.outqs_params.outqs_params[idx].weight = 1;
+
+ 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);
+}
+
+/**
+ * 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;
+
+ 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;
+}
+
+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,
+ .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,
+ .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,
+ .rss_hash_update = mrvl_rss_hash_update,
+ .rss_hash_conf_get = mrvl_rss_hash_conf_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 uint32_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;
+ 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);
+
+ packet_type = RTE_PTYPE_L2_ETHER;
+
+ 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:
+ RTE_LOG(DEBUG, PMD, "Failed to recognise l3 packet type\n");
+ 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:
+ RTE_LOG(DEBUG, PMD, "Failed to recognise l4 packet type\n");
+ 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;
+ unsigned int core_id = rte_lcore_id();
+
+ if (unlikely(!q->priv->ppio))
+ 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)) {
+ RTE_LOG(ERR, PMD, "Failed to receive packets\n");
+ 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_dma_addr_default(mbuf),
+ .cookie = (pp2_cookie_t)(uint64_t)mbuf,
+ };
+
+ pp2_bpool_put_buff(hifs[core_id], 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)
+ RTE_LOG(ERR, PMD, "Failed to fill bpool\n");
+ } 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;
+
+ RTE_LOG(DEBUG, PMD, "\nport-%d:%d: bpool %d oversize -"
+ " remove %d buffers (pool size: %d -> %d)\n",
+ 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++) {
+ pp2_bpool_get_buff(hifs[core_id], bpool, &buff);
+ mbuf = (struct rte_mbuf *)
+ (cookie_addr_high | buff.cookie);
+ rte_pktmbuf_free(mbuf);
+ }
+ mrvl_port_bpool_size
+ [bpool->pp2_id][bpool->id][core_id] -=
+ pkt_to_remove;
+ }
+ 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,
+ 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, core_id = rte_lcore_id();
+
+ 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)) {
+ RTE_LOG(ERR, PMD,
+ "Shadow memory @%d: cookie(%lx), pa(%lx)!\n",
+ 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;
+ }
+
+ 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 = &shadow_txqs[q->port_id][rte_lcore_id()];
+ struct pp2_hif *hif = hifs[rte_lcore_id()];
+ struct pp2_ppio_desc descs[nb_pkts];
+ int i, ret, bytes_sent = 0;
+ uint16_t num, sq_free_size;
+ uint64_t addr;
+
+ if (unlikely(!q->priv->ppio))
+ return 0;
+
+ if (sq->size)
+ mrvl_free_sent_buffers(q->priv->ppio, hif, sq, q->queue_id, 0);
+
+ sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
+ if (unlikely(nb_pkts > sq_free_size)) {
+ RTE_LOG(DEBUG, PMD,
+ "No room in shadow queue for %d packets!!!"
+ "%d packets will be sent.\n",
+ 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);
+ }
+
+ sq->ent[sq->head].buff.cookie = (pp2_cookie_t)(uint64_t)mbuf;
+ sq->ent[sq->head].buff.addr =
+ rte_mbuf_data_dma_addr_default(mbuf);
+ sq->ent[sq->head].bpool =
+ (unlikely(mbuf->port == 0xff || mbuf->refcnt > 1)) ?
+ NULL : mrvl_port_to_bpool_lookup[mbuf->port];
+ sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
+ sq->size++;
+
+ pp2_ppio_outq_desc_reset(&descs[i]);
+ pp2_ppio_outq_desc_set_phys_addr(&descs[i],
+ rte_pktmbuf_mtophys(mbuf));
+ pp2_ppio_outq_desc_set_pkt_offset(&descs[i], 0);
+ pp2_ppio_outq_desc_set_pkt_len(&descs[i],
+ rte_pktmbuf_pkt_len(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;
+}
+
+/**
+ * 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_dev;
+ }
+
+ eth_dev->data->mac_addrs = rte_zmalloc("mac_addrs",
+ ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
+ if (!eth_dev->data->mac_addrs) {
+ RTE_LOG(ERR, PMD, "Failed to allocate space for eth addrs\n");
+ ret = -ENOMEM;
+ goto out_free_priv;
+ }
+
+ memset(&req, 0, sizeof(req));
+ strcpy(req.ifr_name, name);
+ ret = ioctl(fd, SIOCGIFHWADDR, &req);
+ if (ret)
+ goto out_free_mac;
+
+ memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
+ req.ifr_addr.sa_data, ETHER_ADDR_LEN);
+
+
+ eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
+ eth_dev->tx_pkt_burst = mrvl_tx_pkt_burst;
+ eth_dev->data->dev_private = priv;
+ eth_dev->device = &vdev->device;
+ eth_dev->dev_ops = &mrvl_ops;
+
+ return 0;
+out_free_mac:
+ rte_free(eth_dev->data->mac_addrs);
+out_free_dev:
+ rte_eth_dev_release_port(eth_dev);
+out_free_priv:
+ rte_free(priv);
+
+ return ret;
+}
+
+/**
+ * Cleanup previously created device representing Ethernet port.
+ *
+ * @param name
+ * Pointer to the port name.
+ */
+static void
+mrvl_eth_dev_destroy(const char *name)
+{
+ struct rte_eth_dev *eth_dev;
+ struct mrvl_priv *priv;
+
+ eth_dev = rte_eth_dev_allocated(name);
+ if (!eth_dev)
+ return;
+
+ priv = eth_dev->data->dev_private;
+ pp2_bpool_deinit(priv->bpool);
+ rte_free(priv);
+ rte_free(eth_dev->data->mac_addrs);
+ rte_eth_dev_release_port(eth_dev);
+}
+
+/**
+ * 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)
+{
+ const char **ifnames = extra_args;
+
+ ifnames[mrvl_ports_nb++] = value;
+
+ return 0;
+}
+
+/**
+ * Initialize per-lcore MUSDK hardware interfaces (hifs).
+ *
+ * @return
+ * 0 on success, negative error value otherwise.
+ */
+static int
+mrvl_init_hifs(void)
+{
+ struct pp2_hif_params params;
+ char match[MRVL_MATCH_LEN];
+ int i, ret;
+
+ RTE_LCORE_FOREACH(i) {
+ ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
+ if (ret < 0)
+ return ret;
+
+ snprintf(match, sizeof(match), "hif-%d", ret);
+ memset(¶ms, 0, sizeof(params));
+ params.match = match;
+ params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
+ ret = pp2_hif_init(¶ms, &hifs[i]);
+ if (ret) {
+ RTE_LOG(ERR, PMD, "Failed to initialize hif %d\n", i);
+ return ret;
+ }
+ }
+
+ return 0;
+}
+
+/**
+ * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
+ */
+static void
+mrvl_deinit_hifs(void)
+{
+ int i;
+
+ RTE_LCORE_FOREACH(i) {
+ if (hifs[i])
+ pp2_hif_deinit(hifs[i]);
+ }
+}
+
+static void mrvl_set_first_last_cores(int core_id)
+{
+ if (core_id < mrvl_lcore_first)
+ mrvl_lcore_first = core_id;
+
+ if (core_id > mrvl_lcore_last)
+ mrvl_lcore_last = core_id;
+}
+
+/**
+ * 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;
+ const char *ifnames[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
+ int ret = -EINVAL;
+ uint32_t i, ifnum, cfgnum, core_id;
+ 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))
+ goto out_free_kvlist;
+
+ rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
+ mrvl_get_ifnames, &ifnames);
+
+ cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
+ if (cfgnum > 1) {
+ RTE_LOG(ERR, PMD, "Cannot handle more than one config file!\n");
+ goto out_free_kvlist;
+ } else if (cfgnum == 1) {
+ rte_kvargs_process(kvlist, MRVL_CFG_ARG,
+ mrvl_get_qoscfg, &mrvl_qos_cfg);
+ }
+
+ /*
+ * ret == -EEXIST is correct, it means DMA
+ * has been already initialized (by another PMD).
+ */
+ ret = mv_sys_dma_mem_init(RTE_MRVL_MUSDK_DMA_MEMSIZE);
+ if (ret < 0 && (ret != -EEXIST))
+ goto out_free_kvlist;
+
+ ret = mrvl_init_pp2();
+ if (ret) {
+ RTE_LOG(ERR, PMD, "Failed to init PP!\n");
+ goto out_deinit_dma;
+ }
+
+ ret = mrvl_init_hifs();
+ if (ret)
+ goto out_deinit_hifs;
+
+ for (i = 0; i < ifnum; i++) {
+ RTE_LOG(INFO, PMD, "Creating %s\n", ifnames[i]);
+ ret = mrvl_eth_dev_create(vdev, ifnames[i]);
+ if (ret)
+ goto out_cleanup;
+ }
+
+ rte_kvargs_free(kvlist);
+
+ memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
+
+ mrvl_lcore_first = RTE_MAX_LCORE;
+ mrvl_lcore_last = 0;
+
+ RTE_LCORE_FOREACH(core_id) {
+ mrvl_set_first_last_cores(core_id);
+ }
+
+ return 0;
+out_cleanup:
+ for (; i > 0; i--)
+ mrvl_eth_dev_destroy(ifnames[i]);
+out_deinit_hifs:
+ mrvl_deinit_hifs();
+ mrvl_deinit_pp2();
+out_deinit_dma:
+ mv_sys_dma_mem_destroy();
+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)
+{
+ int i;
+ const char *name;
+
+ name = rte_vdev_device_name(vdev);
+ if (!name)
+ return -EINVAL;
+
+ RTE_LOG(INFO, PMD, "Removing %s\n", name);
+
+ for (i = 0; i < rte_eth_dev_count(); i++) {
+ char ifname[RTE_ETH_NAME_MAX_LEN];
+
+ rte_eth_dev_get_name_by_port(i, ifname);
+ mrvl_eth_dev_destroy(ifname);
+ }
+
+ mrvl_deinit_hifs();
+ mrvl_deinit_pp2();
+ mv_sys_dma_mem_destroy();
+
+ return 0;
+}
+
+static struct rte_vdev_driver pmd_mrvl_drv = {
+ .probe = rte_pmd_mrvl_probe,
+ .remove = rte_pmd_mrvl_remove,
+};
+
+RTE_PMD_REGISTER_VDEV(net_mrvl, pmd_mrvl_drv);
+RTE_PMD_REGISTER_ALIAS(net_mrvl, eth_mrvl);
new file mode 100644
@@ -0,0 +1,115 @@
+/*-
+ * BSD LICENSE
+ *
+ * Copyright(c) 2017 Semihalf. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * * Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in
+ * the documentation and/or other materials provided with the
+ * distribution.
+ * * Neither the name of Semihalf nor the names of its
+ * contributors may be used to endorse or promote products derived
+ * from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _MRVL_ETHDEV_H_
+#define _MRVL_ETHDEV_H_
+
+#include <rte_spinlock.h>
+#include <drivers/mv_pp2_cls.h>
+#include <drivers/mv_pp2_ppio.h>
+
+
+/** Maximum number of rx queues per port */
+#define MRVL_PP2_RXQ_MAX 32
+
+/** Maximum number of tx queues per port */
+#define MRVL_PP2_TXQ_MAX 8
+
+/** Minimum number of descriptors in tx queue */
+#define MRVL_PP2_TXD_MIN 16
+
+/** Maximum number of descriptors in tx queue */
+#define MRVL_PP2_TXD_MAX 2048
+
+/** Tx queue descriptors alignment */
+#define MRVL_PP2_TXD_ALIGN 16
+
+/** Minimum number of descriptors in rx queue */
+#define MRVL_PP2_RXD_MIN 16
+
+/** Maximum number of descriptors in rx queue */
+#define MRVL_PP2_RXD_MAX 2048
+
+/** Rx queue descriptors alignment */
+#define MRVL_PP2_RXD_ALIGN 16
+
+/** Maximum number of descriptors in tx aggregated queue */
+#define MRVL_PP2_AGGR_TXQD_MAX 2048
+
+/** Maximum number of Traffic Classes. */
+#define MRVL_PP2_TC_MAX 8
+
+/** Packet offset inside RX buffer. */
+#define MRVL_PKT_OFFS 64
+
+/** Maximum number of descriptors in shadow queue. Must be power of 2 */
+#define MRVL_PP2_TX_SHADOWQ_SIZE MRVL_PP2_TXD_MAX
+
+/** Shadow queue size mask (since shadow queue size is power of 2) */
+#define MRVL_PP2_TX_SHADOWQ_MASK (MRVL_PP2_TX_SHADOWQ_SIZE - 1)
+
+/** Minimum number of sent buffers to release from shadow queue to BM */
+#define MRVL_PP2_BUF_RELEASE_BURST_SIZE 64
+
+struct mrvl_priv {
+ /* Hot fields, used in fast path. */
+ struct pp2_bpool *bpool; /**< BPool pointer */
+ struct pp2_ppio *ppio; /**< Port handler pointer */
+ rte_spinlock_t lock; /**< Spinlock for checking bpool status */
+ uint16_t bpool_max_size; /**< BPool maximum size */
+ uint16_t bpool_min_size; /**< BPool minimum size */
+ uint16_t bpool_init_size; /**< Configured BPool size */
+
+ /** Mapping for DPDK rx queue->(TC, MRVL relative inq) */
+ struct {
+ uint8_t tc; /**< Traffic Class */
+ uint8_t inq; /**< Relative in-queue number */
+ } rxq_map[MRVL_PP2_RXQ_MAX] __rte_cache_aligned;
+
+ /* Configuration data, used sporadically. */
+ uint8_t pp_id;
+ uint8_t ppio_id;
+ uint8_t bpool_bit;
+ uint8_t rss_hf_tcp;
+ uint8_t uc_mc_flushed;
+ uint8_t vlan_flushed;
+
+ struct pp2_ppio_params ppio_params;
+ struct pp2_cls_qos_tbl_params qos_tbl_params;
+ struct pp2_cls_tbl *qos_tbl;
+ uint16_t nb_rx_queues;
+};
+
+/** Number of ports configured. */
+extern int mrvl_ports_nb;
+
+#endif /* _MRVL_ETHDEV_H_ */
new file mode 100644
@@ -0,0 +1,627 @@
+/*-
+ * BSD LICENSE
+ *
+ * Copyright(c) 2017 Semihalf. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * * Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in
+ * the documentation and/or other materials provided with the
+ * distribution.
+ * * Neither the name of Semihalf nor the names of its
+ * contributors may be used to endorse or promote products derived
+ * from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <rte_common.h>
+#include <rte_cfgfile.h>
+#include <rte_log.h>
+#include <rte_lcore.h>
+#include <rte_malloc.h>
+#include <rte_string_fns.h>
+
+/* Unluckily, container_of is defined by both DPDK and MUSDK,
+ * we'll declare only one version.
+ *
+ * Note that it is not used in this PMD anyway.
+ */
+#ifdef container_of
+#undef container_of
+#endif
+
+#include "mrvl_qos.h"
+
+/* Parsing tokens. Defined conveniently, so that any correction is easy. */
+#define MRVL_TOK_DEFAULT "default"
+#define MRVL_TOK_DEFAULT_TC "default_tc"
+#define MRVL_TOK_DSCP "dscp"
+#define MRVL_TOK_MAPPING_PRIORITY "mapping_priority"
+#define MRVL_TOK_IP "ip"
+#define MRVL_TOK_IP_VLAN "ip/vlan"
+#define MRVL_TOK_PCP "pcp"
+#define MRVL_TOK_PORT "port"
+#define MRVL_TOK_RXQ "rxq"
+#define MRVL_TOK_SP "SP"
+#define MRVL_TOK_TC "tc"
+#define MRVL_TOK_TXQ "txq"
+#define MRVL_TOK_VLAN "vlan"
+#define MRVL_TOK_VLAN_IP "vlan/ip"
+#define MRVL_TOK_WEIGHT "weight"
+
+/** Number of tokens in range a-b = 2. */
+#define MAX_RNG_TOKENS 2
+
+/** Maximum possible value of PCP. */
+#define MAX_PCP 7
+
+/** Maximum possible value of DSCP. */
+#define MAX_DSCP 63
+
+/** Global QoS configuration. */
+struct mrvl_qos_cfg *mrvl_qos_cfg;
+
+/**
+ * Convert string to uint32_t with extra checks for result correctness.
+ *
+ * @param string String to convert.
+ * @param val Conversion result.
+ * @returns 0 in case of success, negative value otherwise.
+ */
+static int
+get_val_securely(const char *string, uint32_t *val)
+{
+ char *endptr;
+ size_t len = strlen(string);
+
+ if (len == 0)
+ return -1;
+
+ *val = strtoul(string, &endptr, 0);
+ if (errno != 0 || (RTE_PTR_DIFF(endptr, string) != len))
+ return -2;
+
+ return 0;
+}
+
+/**
+ * Read out-queue configuration from file.
+ *
+ * @param file Path to the configuration file.
+ * @param port Port number.
+ * @param outq Out queue number.
+ * @param cfg Pointer to the Marvell QoS configuration structure.
+ * @returns 0 in case of success, negative value otherwise.
+ */
+static int
+get_outq_cfg(struct rte_cfgfile *file, int port, int outq,
+ struct mrvl_qos_cfg *cfg)
+{
+ char sec_name[32];
+ const char *entry;
+ uint32_t val;
+
+ snprintf(sec_name, sizeof(sec_name), "%s %d %s %d",
+ MRVL_TOK_PORT, port, MRVL_TOK_TXQ, outq);
+
+ /* Skip non-existing */
+ if (rte_cfgfile_num_sections(file, sec_name, strlen(sec_name)) <= 0)
+ return 0;
+
+ entry = rte_cfgfile_get_entry(file, sec_name,
+ MRVL_TOK_WEIGHT);
+ if (entry) {
+ if (get_val_securely(entry, &val) < 0)
+ return -1;
+ cfg->port[port].outq[outq].weight = (uint8_t)val;
+ }
+
+ return 0;
+}
+
+/**
+ * Gets multiple-entry values and places them in table.
+ *
+ * Entry can be anything, e.g. "1 2-3 5 6 7-9". This needs to be converted to
+ * table entries, respectively: {1, 2, 3, 5, 6, 7, 8, 9}.
+ * As all result table's elements are always 1-byte long, we
+ * won't overcomplicate the function, but we'll keep API generic,
+ * check if someone hasn't changed element size and make it simple
+ * to extend to other sizes.
+ *
+ * This function is purely utilitary, it does not print any error, only returns
+ * different error numbers.
+ *
+ * @param entry[in] Values string to parse.
+ * @param tab[out] Results table.
+ * @param elem_sz[in] Element size (in bytes).
+ * @param max_elems[in] Number of results table elements available.
+ * @param max val[in] Maximum value allowed.
+ * @returns Number of correctly parsed elements in case of success.
+ * @retval -1 Wrong element size.
+ * @retval -2 More tokens than result table allows.
+ * @retval -3 Wrong range syntax.
+ * @retval -4 Wrong range values.
+ * @retval -5 Maximum value exceeded.
+ */
+static int
+get_entry_values(const char *entry, uint8_t *tab,
+ size_t elem_sz, uint8_t max_elems, uint8_t max_val)
+{
+ /* There should not be more tokens than max elements.
+ * Add 1 for error trap.
+ */
+ char *tokens[max_elems + 1];
+
+ /* Begin, End + error trap = 3. */
+ char *rng_tokens[MAX_RNG_TOKENS + 1];
+ long beg, end;
+ uint32_t token_val;
+ int nb_tokens, nb_rng_tokens;
+ int i;
+ int values = 0;
+ char val;
+ char entry_cpy[CFG_VALUE_LEN];
+
+ if (elem_sz != 1)
+ return -1;
+
+ /* Copy the entry to safely use rte_strsplit(). */
+ snprintf(entry_cpy, RTE_DIM(entry_cpy), "%s", entry);
+
+ /*
+ * If there are more tokens than array size, rte_strsplit will
+ * not return error, just array size.
+ */
+ nb_tokens = rte_strsplit(entry_cpy, strlen(entry_cpy),
+ tokens, max_elems + 1, ' ');
+
+ /* Quick check, will be refined later. */
+ if (nb_tokens > max_elems)
+ return -2;
+
+ for (i = 0; i < nb_tokens; ++i) {
+ if (strchr(tokens[i], '-') != NULL) {
+ /*
+ * Split to begin and end tokens.
+ * We want to catch error cases too, thus we leave
+ * option for number of tokens to be more than 2.
+ */
+ nb_rng_tokens = rte_strsplit(tokens[i],
+ strlen(tokens[i]), rng_tokens,
+ RTE_DIM(rng_tokens), '-');
+ if (nb_rng_tokens != 2)
+ return -3;
+
+ /* Range and sanity checks. */
+ if (get_val_securely(rng_tokens[0], &token_val) < 0)
+ return -4;
+ beg = (char)token_val;
+ if (get_val_securely(rng_tokens[1], &token_val) < 0)
+ return -4;
+ end = (char)token_val;
+ if (beg < 0 || beg > UCHAR_MAX ||
+ end < 0 || end > UCHAR_MAX || end < beg)
+ return -4;
+
+ for (val = beg; val <= end; ++val) {
+ if (val > max_val)
+ return -5;
+
+ *tab = val;
+ tab = RTE_PTR_ADD(tab, elem_sz);
+ ++values;
+ if (values >= max_elems)
+ return -2;
+ }
+ } else {
+ /* Single values. */
+ if (get_val_securely(tokens[i], &token_val) < 0)
+ return -5;
+ val = (char)token_val;
+ if (val > max_val)
+ return -5;
+
+ *tab = val;
+ tab = RTE_PTR_ADD(tab, elem_sz);
+ ++values;
+ if (values >= max_elems)
+ return -2;
+ }
+ }
+
+ return values;
+}
+
+/**
+ * Parse Traffic Class'es mapping configuration.
+ *
+ * @param file Config file handle.
+ * @param port Which port to look for.
+ * @param tc Which Traffic Class to look for.
+ * @param cfg[out] Parsing results.
+ * @returns 0 in case of success, negative value otherwise.
+ */
+static int
+parse_tc_cfg(struct rte_cfgfile *file, int port, int tc,
+ struct mrvl_qos_cfg *cfg)
+{
+ char sec_name[32];
+ const char *entry;
+ int n;
+
+ snprintf(sec_name, sizeof(sec_name), "%s %d %s %d",
+ MRVL_TOK_PORT, port, MRVL_TOK_TC, tc);
+
+ /* Skip non-existing */
+ if (rte_cfgfile_num_sections(file, sec_name, strlen(sec_name)) <= 0)
+ return 0;
+
+ entry = rte_cfgfile_get_entry(file, sec_name, MRVL_TOK_RXQ);
+ if (entry) {
+ n = get_entry_values(entry,
+ cfg->port[port].tc[tc].inq,
+ sizeof(cfg->port[port].tc[tc].inq[0]),
+ RTE_DIM(cfg->port[port].tc[tc].inq),
+ MRVL_PP2_RXQ_MAX);
+ if (n < 0) {
+ RTE_LOG(ERR, PMD, "Error %d while parsing: %s\n",
+ n, entry);
+ return n;
+ }
+ cfg->port[port].tc[tc].inqs = n;
+ }
+
+ entry = rte_cfgfile_get_entry(file, sec_name, MRVL_TOK_PCP);
+ if (entry) {
+ n = get_entry_values(entry,
+ cfg->port[port].tc[tc].pcp,
+ sizeof(cfg->port[port].tc[tc].pcp[0]),
+ RTE_DIM(cfg->port[port].tc[tc].pcp),
+ MAX_PCP);
+ if (n < 0) {
+ RTE_LOG(ERR, PMD, "Error %d while parsing: %s\n",
+ n, entry);
+ return n;
+ }
+ cfg->port[port].tc[tc].pcps = n;
+ }
+
+ entry = rte_cfgfile_get_entry(file, sec_name, MRVL_TOK_DSCP);
+ if (entry) {
+ n = get_entry_values(entry,
+ cfg->port[port].tc[tc].dscp,
+ sizeof(cfg->port[port].tc[tc].dscp[0]),
+ RTE_DIM(cfg->port[port].tc[tc].dscp),
+ MAX_DSCP);
+ if (n < 0) {
+ RTE_LOG(ERR, PMD, "Error %d while parsing: %s\n",
+ n, entry);
+ return n;
+ }
+ cfg->port[port].tc[tc].dscps = n;
+ }
+ return 0;
+}
+
+/**
+ * Parse QoS configuration - rte_kvargs_process handler.
+ *
+ * Opens configuration file and parses its content.
+ *
+ * @param key Unused.
+ * @param path Path to config file.
+ * @param extra_args Pointer to configuration structure.
+ * @returns 0 in case of success, exits otherwise.
+ */
+int
+mrvl_get_qoscfg(const char *key __rte_unused, const char *path,
+ void *extra_args)
+{
+ struct mrvl_qos_cfg **cfg = extra_args;
+ struct rte_cfgfile *file = rte_cfgfile_load(path, 0);
+ uint32_t val;
+ int n, i, ret;
+ const char *entry;
+ char sec_name[32];
+
+ if (file == NULL)
+ rte_exit(EXIT_FAILURE, "Cannot load configuration %s\n", path);
+
+ /* Create configuration. This is never accessed on the fast path,
+ * so we can ignore socket.
+ */
+ *cfg = rte_zmalloc("mrvl_qos_cfg", sizeof(struct mrvl_qos_cfg), 0);
+ if (*cfg == NULL)
+ rte_exit(EXIT_FAILURE, "Cannot allocate configuration %s\n",
+ path);
+
+ n = rte_cfgfile_num_sections(file, MRVL_TOK_PORT,
+ sizeof(MRVL_TOK_PORT) - 1);
+
+ if (n == 0) {
+ /* This is weird, but not bad. */
+ RTE_LOG(WARNING, PMD, "Empty configuration file?\n");
+ return 0;
+ }
+
+ /* Use the number of ports given as vdev parameters. */
+ for (n = 0; n < mrvl_ports_nb; ++n) {
+ snprintf(sec_name, sizeof(sec_name), "%s %d %s",
+ MRVL_TOK_PORT, n, MRVL_TOK_DEFAULT);
+
+ /* Skip ports non-existing in configuration. */
+ if (rte_cfgfile_num_sections(file, sec_name,
+ strlen(sec_name)) <= 0) {
+ (*cfg)->port[n].use_global_defaults = 1;
+ (*cfg)->port[n].mapping_priority =
+ PP2_CLS_QOS_TBL_VLAN_IP_PRI;
+ continue;
+ }
+
+ entry = rte_cfgfile_get_entry(file, sec_name,
+ MRVL_TOK_DEFAULT_TC);
+ if (entry) {
+ if ((get_val_securely(entry, &val) < 0) ||
+ val > USHRT_MAX)
+ return -1;
+ (*cfg)->port[n].default_tc = (uint8_t)val;
+ } else {
+ RTE_LOG(ERR, PMD,
+ "Default Traffic Class required in custom configuration!\n");
+ return -1;
+ }
+
+ entry = rte_cfgfile_get_entry(file, sec_name,
+ MRVL_TOK_MAPPING_PRIORITY);
+ if (entry) {
+ if (!strncmp(entry, MRVL_TOK_VLAN_IP,
+ sizeof(MRVL_TOK_VLAN_IP)))
+ (*cfg)->port[n].mapping_priority =
+ PP2_CLS_QOS_TBL_VLAN_IP_PRI;
+ else if (!strncmp(entry, MRVL_TOK_IP_VLAN,
+ sizeof(MRVL_TOK_IP_VLAN)))
+ (*cfg)->port[n].mapping_priority =
+ PP2_CLS_QOS_TBL_IP_VLAN_PRI;
+ else if (!strncmp(entry, MRVL_TOK_IP,
+ sizeof(MRVL_TOK_IP)))
+ (*cfg)->port[n].mapping_priority =
+ PP2_CLS_QOS_TBL_IP_PRI;
+ else if (!strncmp(entry, MRVL_TOK_VLAN,
+ sizeof(MRVL_TOK_VLAN)))
+ (*cfg)->port[n].mapping_priority =
+ PP2_CLS_QOS_TBL_VLAN_PRI;
+ else
+ rte_exit(EXIT_FAILURE,
+ "Error in parsing %s value (%s)!\n",
+ MRVL_TOK_MAPPING_PRIORITY, entry);
+ } else {
+ (*cfg)->port[n].mapping_priority =
+ PP2_CLS_QOS_TBL_VLAN_IP_PRI;
+ }
+
+ for (i = 0; i < MRVL_PP2_RXQ_MAX; ++i) {
+ ret = get_outq_cfg(file, n, i, *cfg);
+ if (ret < 0)
+ rte_exit(EXIT_FAILURE,
+ "Error %d parsing port %d outq %d!\n",
+ ret, n, i);
+ }
+
+ for (i = 0; i < MRVL_PP2_TC_MAX; ++i) {
+ ret = parse_tc_cfg(file, n, i, *cfg);
+ if (ret < 0)
+ rte_exit(EXIT_FAILURE,
+ "Error %d parsing port %d tc %d!\n",
+ ret, n, i);
+ }
+ }
+
+ return 0;
+}
+
+/**
+ * Setup Traffic Class.
+ *
+ * Fill in TC parameters in single MUSDK TC config entry.
+ * @param param TC parameters entry.
+ * @param inqs Number of MUSDK in-queues in this TC.
+ * @param bpool Bpool for this TC.
+ * @returns 0 in case of success, exits otherwise.
+ */
+static int
+setup_tc(struct pp2_ppio_tc_params *param, uint8_t inqs,
+ struct pp2_bpool *bpool)
+{
+ struct pp2_ppio_inq_params *inq_params;
+
+ param->pkt_offset = MRVL_PKT_OFFS;
+ param->pools[0] = bpool;
+
+ inq_params = rte_zmalloc_socket("inq_params",
+ inqs * sizeof(*inq_params),
+ 0, rte_socket_id());
+ if (!inq_params)
+ return -ENOMEM;
+
+ param->num_in_qs = inqs;
+
+ /* Release old config if necessary. */
+ if (param->inqs_params)
+ rte_free(param->inqs_params);
+
+ param->inqs_params = inq_params;
+
+ return 0;
+}
+
+/**
+ * Configure RX Queues in a given port.
+ *
+ * Sets up RX queues, their Traffic Classes and DPDK rxq->(TC,inq) mapping.
+ *
+ * @param priv Port's private data
+ * @param portid DPDK port ID
+ * @param max_queues Maximum number of queues to configure.
+ * @returns 0 in case of success, negative value otherwise.
+ */
+int
+mrvl_configure_rxqs(struct mrvl_priv *priv, uint8_t portid,
+ uint16_t max_queues)
+{
+ size_t i, tc;
+
+ if (mrvl_qos_cfg == NULL ||
+ mrvl_qos_cfg->port[portid].use_global_defaults) {
+ /* No port configuration, use default: 1 TC, no QoS. */
+ priv->ppio_params.inqs_params.num_tcs = 1;
+ setup_tc(&priv->ppio_params.inqs_params.tcs_params[0],
+ max_queues, priv->bpool);
+
+ /* Direct mapping of queues i.e. 0->0, 1->1 etc. */
+ for (i = 0; i < max_queues; ++i) {
+ priv->rxq_map[i].tc = 0;
+ priv->rxq_map[i].inq = i;
+ }
+ return 0;
+ }
+
+ /* We need only a subset of configuration. */
+ struct port_cfg *port_cfg = &mrvl_qos_cfg->port[portid];
+
+ priv->qos_tbl_params.type = port_cfg->mapping_priority;
+
+ /*
+ * We need to reverse mapping, from tc->pcp (better from usability
+ * point of view) to pcp->tc (configurable in MUSDK).
+ * First, set all map elements to "default".
+ */
+ for (i = 0; i < RTE_DIM(priv->qos_tbl_params.pcp_cos_map); ++i)
+ priv->qos_tbl_params.pcp_cos_map[i].tc = port_cfg->default_tc;
+
+ /* Then, fill in all known values. */
+ for (tc = 0; tc < RTE_DIM(port_cfg->tc); ++tc) {
+ if (port_cfg->tc[tc].pcps > RTE_DIM(port_cfg->tc[0].pcp)) {
+ /* Better safe than sorry. */
+ RTE_LOG(ERR, PMD,
+ "Too many PCPs configured in TC %zu!\n", tc);
+ return -1;
+ }
+ for (i = 0; i < port_cfg->tc[tc].pcps; ++i) {
+ priv->qos_tbl_params.pcp_cos_map[
+ port_cfg->tc[tc].pcp[i]].tc = tc;
+ }
+ }
+
+ /*
+ * The same logic goes with DSCP.
+ * First, set all map elements to "default".
+ */
+ for (i = 0; i < RTE_DIM(priv->qos_tbl_params.dscp_cos_map); ++i)
+ priv->qos_tbl_params.dscp_cos_map[i].tc =
+ port_cfg->default_tc;
+
+ /* Fill in all known values. */
+ for (tc = 0; tc < RTE_DIM(port_cfg->tc); ++tc) {
+ if (port_cfg->tc[tc].dscps > RTE_DIM(port_cfg->tc[0].dscp)) {
+ /* Better safe than sorry. */
+ RTE_LOG(ERR, PMD,
+ "Too many DSCPs configured in TC %zu!\n", tc);
+ return -1;
+ }
+ for (i = 0; i < port_cfg->tc[tc].dscps; ++i) {
+ priv->qos_tbl_params.dscp_cos_map[
+ port_cfg->tc[tc].dscp[i]].tc = tc;
+ }
+ }
+
+ /*
+ * Surprisingly, similar logic goes with queue mapping.
+ * We need only to store qid->tc mapping,
+ * to know TC when queue is read.
+ */
+ for (i = 0; i < RTE_DIM(priv->rxq_map); ++i)
+ priv->rxq_map[i].tc = MRVL_UNKNOWN_TC;
+
+ /* Set up DPDKq->(TC,inq) mapping. */
+ for (tc = 0; tc < RTE_DIM(port_cfg->tc); ++tc) {
+ if (port_cfg->tc[tc].inqs > RTE_DIM(port_cfg->tc[0].inq)) {
+ /* Overflow. */
+ RTE_LOG(ERR, PMD,
+ "Too many RX queues configured per TC %zu!\n",
+ tc);
+ return -1;
+ }
+ for (i = 0; i < port_cfg->tc[tc].inqs; ++i) {
+ uint8_t idx = port_cfg->tc[tc].inq[i];
+ priv->rxq_map[idx].tc = tc;
+ priv->rxq_map[idx].inq = i;
+ }
+ }
+
+ /*
+ * Set up TC configuration. TCs need to be sequenced: 0, 1, 2
+ * with no gaps. Empty TC means end of processing.
+ */
+ for (i = 0; i < MRVL_PP2_TC_MAX; ++i) {
+ if (port_cfg->tc[i].inqs == 0)
+ break;
+ setup_tc(&priv->ppio_params.inqs_params.tcs_params[i],
+ port_cfg->tc[i].inqs,
+ priv->bpool);
+ }
+
+ priv->ppio_params.inqs_params.num_tcs = i;
+
+ return 0;
+}
+
+/**
+ * Start QoS mapping.
+ *
+ * Finalize QoS table configuration and initialize it in SDK. It can be done
+ * only after port is started, so we have a valid ppio reference.
+ *
+ * @param priv Port's private (configuration) data.
+ * @returns 0 in case of success, exits otherwise.
+ */
+int
+mrvl_start_qos_mapping(struct mrvl_priv *priv)
+{
+ size_t i;
+
+ if (priv->ppio == NULL) {
+ RTE_LOG(ERR, PMD, "ppio must not be NULL here!\n");
+ return -1;
+ }
+
+ for (i = 0; i < RTE_DIM(priv->qos_tbl_params.pcp_cos_map); ++i)
+ priv->qos_tbl_params.pcp_cos_map[i].ppio = priv->ppio;
+
+ for (i = 0; i < RTE_DIM(priv->qos_tbl_params.dscp_cos_map); ++i)
+ priv->qos_tbl_params.dscp_cos_map[i].ppio = priv->ppio;
+
+ /* Initialize Classifier QoS table. */
+
+ return pp2_cls_qos_tbl_init(&priv->qos_tbl_params, &priv->qos_tbl);
+}
new file mode 100644
@@ -0,0 +1,112 @@
+/*-
+ * BSD LICENSE
+ *
+ * Copyright(c) 2017 Semihalf. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * * Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in
+ * the documentation and/or other materials provided with the
+ * distribution.
+ * * Neither the name of Semihalf nor the names of its
+ * contributors may be used to endorse or promote products derived
+ * from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _MRVL_QOS_H_
+#define _MRVL_QOS_H_
+
+#include <rte_common.h>
+#include <rte_config.h>
+
+#include "mrvl_ethdev.h"
+
+/** Code Points per Traffic Class. Equals max(DSCP, PCP). */
+#define MRVL_CP_PER_TC (64)
+
+/** Value used as "unknown". */
+#define MRVL_UNKNOWN_TC (0xFF)
+
+/* QoS config. */
+struct mrvl_qos_cfg {
+ struct port_cfg {
+ struct {
+ uint8_t inq[MRVL_PP2_RXQ_MAX];
+ uint8_t dscp[MRVL_CP_PER_TC];
+ uint8_t pcp[MRVL_CP_PER_TC];
+ uint8_t inqs;
+ uint8_t dscps;
+ uint8_t pcps;
+ } tc[MRVL_PP2_TC_MAX];
+ struct {
+ uint8_t weight;
+ } outq[MRVL_PP2_RXQ_MAX];
+ enum pp2_cls_qos_tbl_type mapping_priority;
+ uint16_t inqs;
+ uint16_t outqs;
+ uint8_t default_tc;
+ uint8_t use_global_defaults;
+ } port[RTE_MAX_ETHPORTS];
+};
+
+/** Global QoS configuration. */
+extern struct mrvl_qos_cfg *mrvl_qos_cfg;
+
+/**
+ * Parse QoS configuration - rte_kvargs_process handler.
+ *
+ * Opens configuration file and parses its content.
+ *
+ * @param key Unused.
+ * @param path Path to config file.
+ * @param extra_args Pointer to configuration structure.
+ * @returns 0 in case of success, exits otherwise.
+ */
+int
+mrvl_get_qoscfg(const char *key __rte_unused, const char *path,
+ void *extra_args);
+
+/**
+ * Configure RX Queues in a given port.
+ *
+ * Sets up RX queues, their Traffic Classes and DPDK rxq->(TC,inq) mapping.
+ *
+ * @param priv Port's private data
+ * @param portid DPDK port ID
+ * @param max_queues Maximum number of queues to configure.
+ * @returns 0 in case of success, negative value otherwise.
+ */
+int
+mrvl_configure_rxqs(struct mrvl_priv *priv, uint8_t portid,
+ uint16_t max_queues);
+
+/**
+ * Start QoS mapping.
+ *
+ * Finalize QoS table configuration and initialize it in SDK. It can be done
+ * only after port is started, so we have a valid ppio reference.
+ *
+ * @param priv Port's private (configuration) data.
+ * @returns 0 in case of success, exits otherwise.
+ */
+int
+mrvl_start_qos_mapping(struct mrvl_priv *priv);
+
+#endif /* _MRVL_QOS_H_ */
new file mode 100644
@@ -0,0 +1,3 @@
+DPDK_17.11 {
+ local: *;
+};
@@ -130,6 +130,7 @@ endif
_LDLIBS-$(CONFIG_RTE_LIBRTE_LIO_PMD) += -lrte_pmd_lio
_LDLIBS-$(CONFIG_RTE_LIBRTE_MLX4_PMD) += -lrte_pmd_mlx4 -libverbs
_LDLIBS-$(CONFIG_RTE_LIBRTE_MLX5_PMD) += -lrte_pmd_mlx5 -libverbs
+_LDLIBS-$(CONFIG_RTE_LIBRTE_MRVL_PMD) += -lrte_pmd_mrvl -L$(LIBMUSDK_PATH)/lib -lmusdk
_LDLIBS-$(CONFIG_RTE_LIBRTE_NFP_PMD) += -lrte_pmd_nfp
_LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_NULL) += -lrte_pmd_null
_LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_PCAP) += -lrte_pmd_pcap -lpcap