examples/l2fwd-jobstats: add new example
authorPawel Wodkowski <pawelx.wodkowski@intel.com>
Tue, 24 Feb 2015 16:33:24 +0000 (17:33 +0100)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Wed, 25 Feb 2015 09:23:31 +0000 (10:23 +0100)
This app demonstrate usage of new rte_jobstats library.
It is basically the orginal l2fwd with following modifications to met
library requirements:
- main_loop() was split into two jobs: forward job and flush job. Logic
for those jobs is almost the same as in original application.
- stats is moved to rte_alarm callback to not introduce overhead of
printing.
- stats are expanded to show rte_jobstats statistics.
- added new parameter '-l' to automatic thousands separator.

Comparing original l2fwd and l2fwd-jobstats apps will show approach what
is needed to properly write own application with rte_jobstats
measurements.

New available statistics:
- Total and % of fwd and flush execution time
- management time - overhead of rte_timer + overhead of rte_jobstats
library
- Idle time and % of time spent waiting for fwd or flush to be ready to
execute.
- per job execution time and period.

Fixes: 2caeb8c0141d ("examples/l2fwd-jobstats: new example")
[Thomas: files were missing in the previous commit]

Signed-off-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
examples/l2fwd-jobstats/Makefile [new file with mode: 0644]
examples/l2fwd-jobstats/main.c [new file with mode: 0644]

diff --git a/examples/l2fwd-jobstats/Makefile b/examples/l2fwd-jobstats/Makefile
new file mode 100644 (file)
index 0000000..ab089f6
--- /dev/null
@@ -0,0 +1,51 @@
+#   BSD LICENSE
+#
+#   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
+#   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 Intel Corporation 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.
+
+ifeq ($(RTE_SDK),)
+$(error "Please define RTE_SDK environment variable")
+endif
+
+# Default target, can be overridden by command line or environment
+RTE_TARGET ?= x86_64-native-linuxapp-gcc
+
+include $(RTE_SDK)/mk/rte.vars.mk
+
+# binary name
+APP = l2fwd-jobstats
+
+# all source are stored in SRCS-y
+SRCS-y := main.c
+
+
+CFLAGS += -O3
+CFLAGS += $(WERROR_FLAGS)
+
+include $(RTE_SDK)/mk/rte.extapp.mk
diff --git a/examples/l2fwd-jobstats/main.c b/examples/l2fwd-jobstats/main.c
new file mode 100644 (file)
index 0000000..a5a1aaa
--- /dev/null
@@ -0,0 +1,1040 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
+ *   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 Intel Corporation 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 <locale.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdint.h>
+#include <ctype.h>
+#include <getopt.h>
+
+#include <rte_alarm.h>
+#include <rte_common.h>
+#include <rte_log.h>
+#include <rte_memory.h>
+#include <rte_memcpy.h>
+#include <rte_memzone.h>
+#include <rte_tailq.h>
+#include <rte_eal.h>
+#include <rte_per_lcore.h>
+#include <rte_launch.h>
+#include <rte_atomic.h>
+#include <rte_cycles.h>
+#include <rte_prefetch.h>
+#include <rte_lcore.h>
+#include <rte_per_lcore.h>
+#include <rte_branch_prediction.h>
+#include <rte_interrupts.h>
+#include <rte_pci.h>
+#include <rte_debug.h>
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#include <rte_ring.h>
+#include <rte_mempool.h>
+#include <rte_mbuf.h>
+#include <rte_spinlock.h>
+
+#include <rte_errno.h>
+#include <rte_jobstats.h>
+#include <rte_timer.h>
+#include <rte_alarm.h>
+
+#define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
+
+#define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
+#define NB_MBUF   8192
+
+#define MAX_PKT_BURST 32
+#define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
+
+/*
+ * Configurable number of RX/TX ring descriptors
+ */
+#define RTE_TEST_RX_DESC_DEFAULT 128
+#define RTE_TEST_TX_DESC_DEFAULT 512
+static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
+static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
+
+/* ethernet addresses of ports */
+static struct ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
+
+/* mask of enabled ports */
+static uint32_t l2fwd_enabled_port_mask;
+
+/* list of enabled ports */
+static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
+
+#define UPDATE_STEP_UP 1
+#define UPDATE_STEP_DOWN 32
+
+static unsigned int l2fwd_rx_queue_per_lcore = 1;
+
+struct mbuf_table {
+       uint64_t next_flush_time;
+       unsigned len;
+       struct rte_mbuf *mbufs[MAX_PKT_BURST];
+};
+
+#define MAX_RX_QUEUE_PER_LCORE 16
+#define MAX_TX_QUEUE_PER_PORT 16
+struct lcore_queue_conf {
+       unsigned n_rx_port;
+       unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
+       struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
+
+       struct rte_timer rx_timers[MAX_RX_QUEUE_PER_LCORE];
+       struct rte_jobstats port_fwd_jobs[MAX_RX_QUEUE_PER_LCORE];
+
+       struct rte_timer flush_timer;
+       struct rte_jobstats flush_job;
+       struct rte_jobstats idle_job;
+       struct rte_jobstats_context jobs_context;
+
+       rte_atomic16_t stats_read_pending;
+       rte_spinlock_t lock;
+} __rte_cache_aligned;
+struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
+
+static const struct rte_eth_conf port_conf = {
+       .rxmode = {
+               .split_hdr_size = 0,
+               .header_split   = 0, /**< Header Split disabled */
+               .hw_ip_checksum = 0, /**< IP checksum offload disabled */
+               .hw_vlan_filter = 0, /**< VLAN filtering disabled */
+               .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
+               .hw_strip_crc   = 0, /**< CRC stripped by hardware */
+       },
+       .txmode = {
+               .mq_mode = ETH_MQ_TX_NONE,
+       },
+};
+
+struct rte_mempool *l2fwd_pktmbuf_pool = NULL;
+
+/* Per-port statistics struct */
+struct l2fwd_port_statistics {
+       uint64_t tx;
+       uint64_t rx;
+       uint64_t dropped;
+} __rte_cache_aligned;
+struct l2fwd_port_statistics port_statistics[RTE_MAX_ETHPORTS];
+
+/* 1 day max */
+#define MAX_TIMER_PERIOD 86400
+/* default period is 10 seconds */
+static int64_t timer_period = 10;
+/* default timer frequency */
+static double hz;
+/* BURST_TX_DRAIN_US converted to cycles */
+uint64_t drain_tsc;
+/* Convert cycles to ns */
+static inline double
+cycles_to_ns(uint64_t cycles)
+{
+       double t = cycles;
+
+       t *= (double)NS_PER_S;
+       t /= hz;
+       return t;
+}
+
+static void
+show_lcore_stats(unsigned lcore_id)
+{
+       struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
+       struct rte_jobstats_context *ctx = &qconf->jobs_context;
+       struct rte_jobstats *job;
+       uint8_t i;
+
+       /* LCore statistics. */
+       uint64_t stats_period, loop_count;
+       uint64_t exec, exec_min, exec_max;
+       uint64_t management, management_min, management_max;
+       uint64_t busy, busy_min, busy_max;
+
+       /* Jobs statistics. */
+       const uint8_t port_cnt = qconf->n_rx_port;
+       uint64_t jobs_exec_cnt[port_cnt], jobs_period[port_cnt];
+       uint64_t jobs_exec[port_cnt], jobs_exec_min[port_cnt],
+                               jobs_exec_max[port_cnt];
+
+       uint64_t flush_exec_cnt, flush_period;
+       uint64_t flush_exec, flush_exec_min, flush_exec_max;
+
+       uint64_t idle_exec_cnt;
+       uint64_t idle_exec, idle_exec_min, idle_exec_max;
+       uint64_t collection_time = rte_get_timer_cycles();
+
+       /* Ask forwarding thread to give us stats. */
+       rte_atomic16_set(&qconf->stats_read_pending, 1);
+       rte_spinlock_lock(&qconf->lock);
+       rte_atomic16_set(&qconf->stats_read_pending, 0);
+
+       /* Collect context statistics. */
+       stats_period = ctx->state_time - ctx->start_time;
+       loop_count = ctx->loop_cnt;
+
+       exec = ctx->exec_time;
+       exec_min = ctx->min_exec_time;
+       exec_max = ctx->max_exec_time;
+
+       management = ctx->management_time;
+       management_min = ctx->min_management_time;
+       management_max = ctx->max_management_time;
+
+       rte_jobstats_context_reset(ctx);
+
+       for (i = 0; i < port_cnt; i++) {
+               job = &qconf->port_fwd_jobs[i];
+
+               jobs_exec_cnt[i] = job->exec_cnt;
+               jobs_period[i] = job->period;
+
+               jobs_exec[i] = job->exec_time;
+               jobs_exec_min[i] = job->min_exec_time;
+               jobs_exec_max[i] = job->max_exec_time;
+
+               rte_jobstats_reset(job);
+       }
+
+       flush_exec_cnt = qconf->flush_job.exec_cnt;
+       flush_period = qconf->flush_job.period;
+       flush_exec = qconf->flush_job.exec_time;
+       flush_exec_min = qconf->flush_job.min_exec_time;
+       flush_exec_max = qconf->flush_job.max_exec_time;
+       rte_jobstats_reset(&qconf->flush_job);
+
+       idle_exec_cnt = qconf->idle_job.exec_cnt;
+       idle_exec = qconf->idle_job.exec_time;
+       idle_exec_min = qconf->idle_job.min_exec_time;
+       idle_exec_max = qconf->idle_job.max_exec_time;
+       rte_jobstats_reset(&qconf->idle_job);
+
+       rte_spinlock_unlock(&qconf->lock);
+
+       exec -= idle_exec;
+       busy = exec + management;
+       busy_min = exec_min + management_min;
+       busy_max = exec_max + management_max;
+
+
+       collection_time = rte_get_timer_cycles() - collection_time;
+
+#define STAT_FMT "\n%-18s %'14.0f %6.1f%% %'10.0f %'10.0f %'10.0f"
+
+       printf("\n----------------"
+                       "\nLCore %3u: statistics (time in ns, collected in %'9.0f)"
+                       "\n%-18s %14s %7s %10s %10s %10s "
+                       "\n%-18s %'14.0f"
+                       "\n%-18s %'14" PRIu64
+                       STAT_FMT /* Exec */
+                       STAT_FMT /* Management */
+                       STAT_FMT /* Busy */
+                       STAT_FMT, /* Idle  */
+                       lcore_id, cycles_to_ns(collection_time),
+                       "Stat type", "total", "%total", "avg", "min", "max",
+                       "Stats duration:", cycles_to_ns(stats_period),
+                       "Loop count:", loop_count,
+                       "Exec time",
+                       cycles_to_ns(exec), exec * 100.0 / stats_period,
+                       cycles_to_ns(loop_count  ? exec / loop_count : 0),
+                       cycles_to_ns(exec_min),
+                       cycles_to_ns(exec_max),
+                       "Management time",
+                       cycles_to_ns(management), management * 100.0 / stats_period,
+                       cycles_to_ns(loop_count  ? management / loop_count : 0),
+                       cycles_to_ns(management_min),
+                       cycles_to_ns(management_max),
+                       "Exec + management",
+                       cycles_to_ns(busy),  busy * 100.0 / stats_period,
+                       cycles_to_ns(loop_count ? busy / loop_count : 0),
+                       cycles_to_ns(busy_min),
+                       cycles_to_ns(busy_max),
+                       "Idle (job)",
+                       cycles_to_ns(idle_exec), idle_exec * 100.0 / stats_period,
+                       cycles_to_ns(idle_exec_cnt ? idle_exec / idle_exec_cnt : 0),
+                       cycles_to_ns(idle_exec_min),
+                       cycles_to_ns(idle_exec_max));
+
+       for (i = 0; i < qconf->n_rx_port; i++) {
+               job = &qconf->port_fwd_jobs[i];
+               printf("\n\nJob %" PRIu32 ": %-20s "
+                               "\n%-18s %'14" PRIu64
+                               "\n%-18s %'14.0f"
+                               STAT_FMT,
+                               i, job->name,
+                               "Exec count:", jobs_exec_cnt[i],
+                               "Exec period: ", cycles_to_ns(jobs_period[i]),
+                               "Exec time",
+                               cycles_to_ns(jobs_exec[i]), jobs_exec[i] * 100.0 / stats_period,
+                               cycles_to_ns(jobs_exec_cnt[i] ? jobs_exec[i] / jobs_exec_cnt[i]
+                                               : 0),
+                               cycles_to_ns(jobs_exec_min[i]),
+                               cycles_to_ns(jobs_exec_max[i]));
+       }
+
+       if (qconf->n_rx_port > 0) {
+               job = &qconf->flush_job;
+               printf("\n\nJob %" PRIu32 ": %-20s "
+                               "\n%-18s %'14" PRIu64
+                               "\n%-18s %'14.0f"
+                               STAT_FMT,
+                               i, job->name,
+                               "Exec count:", flush_exec_cnt,
+                               "Exec period: ", cycles_to_ns(flush_period),
+                               "Exec time",
+                               cycles_to_ns(flush_exec), flush_exec * 100.0 / stats_period,
+                               cycles_to_ns(flush_exec_cnt ? flush_exec / flush_exec_cnt : 0),
+                               cycles_to_ns(flush_exec_min),
+                               cycles_to_ns(flush_exec_max));
+       }
+}
+
+/* Print out statistics on packets dropped */
+static void
+show_stats_cb(__rte_unused void *param)
+{
+       uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
+       unsigned portid, lcore_id;
+
+       total_packets_dropped = 0;
+       total_packets_tx = 0;
+       total_packets_rx = 0;
+
+       const char clr[] = { 27, '[', '2', 'J', '\0' };
+       const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
+
+       /* Clear screen and move to top left */
+       printf("%s%s"
+                       "\nPort statistics ===================================",
+                       clr, topLeft);
+
+       for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
+               /* skip disabled ports */
+               if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
+                       continue;
+               printf("\nStatistics for port %u ------------------------------"
+                               "\nPackets sent: %24"PRIu64
+                               "\nPackets received: %20"PRIu64
+                               "\nPackets dropped: %21"PRIu64,
+                               portid,
+                               port_statistics[portid].tx,
+                               port_statistics[portid].rx,
+                               port_statistics[portid].dropped);
+
+               total_packets_dropped += port_statistics[portid].dropped;
+               total_packets_tx += port_statistics[portid].tx;
+               total_packets_rx += port_statistics[portid].rx;
+       }
+
+       printf("\nAggregate statistics ==============================="
+                       "\nTotal packets sent: %18"PRIu64
+                       "\nTotal packets received: %14"PRIu64
+                       "\nTotal packets dropped: %15"PRIu64
+                       "\n====================================================",
+                       total_packets_tx,
+                       total_packets_rx,
+                       total_packets_dropped);
+
+       RTE_LCORE_FOREACH(lcore_id) {
+               if (lcore_queue_conf[lcore_id].n_rx_port > 0)
+                       show_lcore_stats(lcore_id);
+       }
+
+       printf("\n====================================================\n");
+       rte_eal_alarm_set(timer_period * US_PER_S, show_stats_cb, NULL);
+}
+
+/* Send the burst of packets on an output interface */
+static void
+l2fwd_send_burst(struct lcore_queue_conf *qconf, uint8_t port)
+{
+       struct mbuf_table *m_table;
+       uint16_t ret;
+       uint16_t queueid = 0;
+       uint16_t n;
+
+       m_table = &qconf->tx_mbufs[port];
+       n = m_table->len;
+
+       m_table->next_flush_time = rte_get_timer_cycles() + drain_tsc;
+       m_table->len = 0;
+
+       ret = rte_eth_tx_burst(port, queueid, m_table->mbufs, n);
+
+       port_statistics[port].tx += ret;
+       if (unlikely(ret < n)) {
+               port_statistics[port].dropped += (n - ret);
+               do {
+                       rte_pktmbuf_free(m_table->mbufs[ret]);
+               } while (++ret < n);
+       }
+}
+
+/* Enqueue packets for TX and prepare them to be sent */
+static int
+l2fwd_send_packet(struct rte_mbuf *m, uint8_t port)
+{
+       const unsigned lcore_id = rte_lcore_id();
+       struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
+       struct mbuf_table *m_table = &qconf->tx_mbufs[port];
+       uint16_t len = qconf->tx_mbufs[port].len;
+
+       m_table->mbufs[len] = m;
+
+       len++;
+       m_table->len = len;
+
+       /* Enough pkts to be sent. */
+       if (unlikely(len == MAX_PKT_BURST))
+               l2fwd_send_burst(qconf, port);
+
+       return 0;
+}
+
+static void
+l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
+{
+       struct ether_hdr *eth;
+       void *tmp;
+       unsigned dst_port;
+
+       dst_port = l2fwd_dst_ports[portid];
+       eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
+
+       /* 02:00:00:00:00:xx */
+       tmp = &eth->d_addr.addr_bytes[0];
+       *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
+
+       /* src addr */
+       ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], &eth->s_addr);
+
+       l2fwd_send_packet(m, (uint8_t) dst_port);
+}
+
+static void
+l2fwd_job_update_cb(struct rte_jobstats *job, int64_t result)
+{
+       int64_t err = job->target - result;
+       int64_t histeresis = job->target / 8;
+
+       if (err < -histeresis) {
+               if (job->min_period + UPDATE_STEP_DOWN < job->period)
+                       job->period -= UPDATE_STEP_DOWN;
+       } else if (err > histeresis) {
+               if (job->period + UPDATE_STEP_UP < job->max_period)
+                       job->period += UPDATE_STEP_UP;
+       }
+}
+
+static void
+l2fwd_fwd_job(__rte_unused struct rte_timer *timer, void *arg)
+{
+       struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
+       struct rte_mbuf *m;
+
+       const uint8_t port_idx = (uintptr_t) arg;
+       const unsigned lcore_id = rte_lcore_id();
+       struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
+       struct rte_jobstats *job = &qconf->port_fwd_jobs[port_idx];
+       const uint8_t portid = qconf->rx_port_list[port_idx];
+
+       uint8_t j;
+       uint16_t total_nb_rx;
+
+       rte_jobstats_start(&qconf->jobs_context, job);
+
+       /* Call rx burst 2 times. This allow rte_jobstats logic to see if this
+        * function must be called more frequently. */
+
+       total_nb_rx = rte_eth_rx_burst((uint8_t) portid, 0, pkts_burst,
+                       MAX_PKT_BURST);
+
+       for (j = 0; j < total_nb_rx; j++) {
+               m = pkts_burst[j];
+               rte_prefetch0(rte_pktmbuf_mtod(m, void *));
+               l2fwd_simple_forward(m, portid);
+       }
+
+       if (total_nb_rx == MAX_PKT_BURST) {
+               const uint16_t nb_rx = rte_eth_rx_burst((uint8_t) portid, 0, pkts_burst,
+                               MAX_PKT_BURST);
+
+               total_nb_rx += nb_rx;
+               for (j = 0; j < nb_rx; j++) {
+                       m = pkts_burst[j];
+                       rte_prefetch0(rte_pktmbuf_mtod(m, void *));
+                       l2fwd_simple_forward(m, portid);
+               }
+       }
+
+       port_statistics[portid].rx += total_nb_rx;
+
+       /* Adjust period time in which we are running here. */
+       if (rte_jobstats_finish(job, total_nb_rx) != 0) {
+               rte_timer_reset(&qconf->rx_timers[port_idx], job->period, PERIODICAL,
+                               lcore_id, l2fwd_fwd_job, arg);
+       }
+}
+
+static void
+l2fwd_flush_job(__rte_unused struct rte_timer *timer, __rte_unused void *arg)
+{
+       uint64_t now;
+       unsigned lcore_id;
+       struct lcore_queue_conf *qconf;
+       struct mbuf_table *m_table;
+       uint8_t portid;
+
+       lcore_id = rte_lcore_id();
+       qconf = &lcore_queue_conf[lcore_id];
+
+       rte_jobstats_start(&qconf->jobs_context, &qconf->flush_job);
+
+       now = rte_get_timer_cycles();
+       lcore_id = rte_lcore_id();
+       qconf = &lcore_queue_conf[lcore_id];
+       for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
+               m_table = &qconf->tx_mbufs[portid];
+               if (m_table->len == 0 || m_table->next_flush_time <= now)
+                       continue;
+
+               l2fwd_send_burst(qconf, portid);
+       }
+
+
+       /* Pass target to indicate that this job is happy of time interwal
+        * in which it was called. */
+       rte_jobstats_finish(&qconf->flush_job, qconf->flush_job.target);
+}
+
+/* main processing loop */
+static void
+l2fwd_main_loop(void)
+{
+       unsigned lcore_id;
+       unsigned i, portid;
+       struct lcore_queue_conf *qconf;
+       uint8_t stats_read_pending = 0;
+       uint8_t need_manage;
+
+       lcore_id = rte_lcore_id();
+       qconf = &lcore_queue_conf[lcore_id];
+
+       if (qconf->n_rx_port == 0) {
+               RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
+               return;
+       }
+
+       RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
+
+       for (i = 0; i < qconf->n_rx_port; i++) {
+
+               portid = qconf->rx_port_list[i];
+               RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
+                       portid);
+       }
+
+       rte_jobstats_init(&qconf->idle_job, "idle", 0, 0, 0, 0);
+
+       for (;;) {
+               rte_spinlock_lock(&qconf->lock);
+
+               do {
+                       rte_jobstats_context_start(&qconf->jobs_context);
+
+                       /* Do the Idle job:
+                        * - Read stats_read_pending flag
+                        * - check if some real job need to be executed
+                        */
+                       rte_jobstats_start(&qconf->jobs_context, &qconf->idle_job);
+
+                       do {
+                               uint8_t i;
+                               uint64_t now = rte_get_timer_cycles();
+
+                               need_manage = qconf->flush_timer.expire < now;
+                               /* Check if we was esked to give a stats. */
+                               stats_read_pending =
+                                               rte_atomic16_read(&qconf->stats_read_pending);
+                               need_manage |= stats_read_pending;
+
+                               for (i = 0; i < qconf->n_rx_port && !need_manage; i++)
+                                       need_manage = qconf->rx_timers[i].expire < now;
+
+                       } while (!need_manage);
+                       rte_jobstats_finish(&qconf->idle_job, qconf->idle_job.target);
+
+                       rte_timer_manage();
+                       rte_jobstats_context_finish(&qconf->jobs_context);
+               } while (likely(stats_read_pending == 0));
+
+               rte_spinlock_unlock(&qconf->lock);
+               rte_pause();
+       }
+}
+
+static int
+l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
+{
+       l2fwd_main_loop();
+       return 0;
+}
+
+/* display usage */
+static void
+l2fwd_usage(const char *prgname)
+{
+       printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
+              "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
+              "  -q NQ: number of queue (=ports) per lcore (default is 1)\n"
+                  "  -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n"
+                  "  -l set system default locale instead of default (\"C\" locale) for thousands separator in stats.",
+              prgname);
+}
+
+static int
+l2fwd_parse_portmask(const char *portmask)
+{
+       char *end = NULL;
+       unsigned long pm;
+
+       /* parse hexadecimal string */
+       pm = strtoul(portmask, &end, 16);
+       if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
+               return -1;
+
+       if (pm == 0)
+               return -1;
+
+       return pm;
+}
+
+static unsigned int
+l2fwd_parse_nqueue(const char *q_arg)
+{
+       char *end = NULL;
+       unsigned long n;
+
+       /* parse hexadecimal string */
+       n = strtoul(q_arg, &end, 10);
+       if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
+               return 0;
+       if (n == 0)
+               return 0;
+       if (n >= MAX_RX_QUEUE_PER_LCORE)
+               return 0;
+
+       return n;
+}
+
+static int
+l2fwd_parse_timer_period(const char *q_arg)
+{
+       char *end = NULL;
+       int n;
+
+       /* parse number string */
+       n = strtol(q_arg, &end, 10);
+       if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
+               return -1;
+       if (n >= MAX_TIMER_PERIOD)
+               return -1;
+
+       return n;
+}
+
+/* Parse the argument given in the command line of the application */
+static int
+l2fwd_parse_args(int argc, char **argv)
+{
+       int opt, ret;
+       char **argvopt;
+       int option_index;
+       char *prgname = argv[0];
+       static struct option lgopts[] = {
+               {NULL, 0, 0, 0}
+       };
+
+       argvopt = argv;
+
+       while ((opt = getopt_long(argc, argvopt, "p:q:T:l",
+                                 lgopts, &option_index)) != EOF) {
+
+               switch (opt) {
+               /* portmask */
+               case 'p':
+                       l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
+                       if (l2fwd_enabled_port_mask == 0) {
+                               printf("invalid portmask\n");
+                               l2fwd_usage(prgname);
+                               return -1;
+                       }
+                       break;
+
+               /* nqueue */
+               case 'q':
+                       l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
+                       if (l2fwd_rx_queue_per_lcore == 0) {
+                               printf("invalid queue number\n");
+                               l2fwd_usage(prgname);
+                               return -1;
+                       }
+                       break;
+
+               /* timer period */
+               case 'T':
+                       timer_period = l2fwd_parse_timer_period(optarg);
+                       if (timer_period < 0) {
+                               printf("invalid timer period\n");
+                               l2fwd_usage(prgname);
+                               return -1;
+                       }
+                       break;
+
+               /* For thousands separator in printf. */
+               case 'l':
+                       setlocale(LC_ALL, "");
+                       break;
+
+               /* long options */
+               case 0:
+                       l2fwd_usage(prgname);
+                       return -1;
+
+               default:
+                       l2fwd_usage(prgname);
+                       return -1;
+               }
+       }
+
+       if (optind >= 0)
+               argv[optind-1] = prgname;
+
+       ret = optind-1;
+       optind = 0; /* reset getopt lib */
+       return ret;
+}
+
+/* Check the link status of all ports in up to 9s, and print them finally */
+static void
+check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
+{
+#define CHECK_INTERVAL 100 /* 100ms */
+#define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
+       uint8_t portid, count, all_ports_up, print_flag = 0;
+       struct rte_eth_link link;
+
+       printf("\nChecking link status");
+       fflush(stdout);
+       for (count = 0; count <= MAX_CHECK_TIME; count++) {
+               all_ports_up = 1;
+               for (portid = 0; portid < port_num; portid++) {
+                       if ((port_mask & (1 << portid)) == 0)
+                               continue;
+                       memset(&link, 0, sizeof(link));
+                       rte_eth_link_get_nowait(portid, &link);
+                       /* print link status if flag set */
+                       if (print_flag == 1) {
+                               if (link.link_status)
+                                       printf("Port %d Link Up - speed %u "
+                                               "Mbps - %s\n", (uint8_t)portid,
+                                               (unsigned)link.link_speed,
+                               (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
+                                       ("full-duplex") : ("half-duplex\n"));
+                               else
+                                       printf("Port %d Link Down\n",
+                                               (uint8_t)portid);
+                               continue;
+                       }
+                       /* clear all_ports_up flag if any link down */
+                       if (link.link_status == 0) {
+                               all_ports_up = 0;
+                               break;
+                       }
+               }
+               /* after finally printing all link status, get out */
+               if (print_flag == 1)
+                       break;
+
+               if (all_ports_up == 0) {
+                       printf(".");
+                       fflush(stdout);
+                       rte_delay_ms(CHECK_INTERVAL);
+               }
+
+               /* set the print_flag if all ports up or timeout */
+               if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
+                       print_flag = 1;
+                       printf("done\n");
+               }
+       }
+}
+
+int
+main(int argc, char **argv)
+{
+       struct lcore_queue_conf *qconf;
+       struct rte_eth_dev_info dev_info;
+       unsigned lcore_id, rx_lcore_id;
+       unsigned nb_ports_in_mask = 0;
+       int ret;
+       char name[RTE_JOBSTATS_NAMESIZE];
+       uint8_t nb_ports;
+       uint8_t nb_ports_available;
+       uint8_t portid, last_port;
+       uint8_t i;
+
+       /* init EAL */
+       ret = rte_eal_init(argc, argv);
+       if (ret < 0)
+               rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
+       argc -= ret;
+       argv += ret;
+
+       /* parse application arguments (after the EAL ones) */
+       ret = l2fwd_parse_args(argc, argv);
+       if (ret < 0)
+               rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
+
+       rte_timer_subsystem_init();
+
+       /* fetch default timer frequency. */
+       hz = rte_get_timer_hz();
+
+       /* create the mbuf pool */
+       l2fwd_pktmbuf_pool =
+               rte_mempool_create("mbuf_pool", NB_MBUF,
+                                  MBUF_SIZE, 32,
+                                  sizeof(struct rte_pktmbuf_pool_private),
+                                  rte_pktmbuf_pool_init, NULL,
+                                  rte_pktmbuf_init, NULL,
+                                  rte_socket_id(), 0);
+       if (l2fwd_pktmbuf_pool == NULL)
+               rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
+
+       nb_ports = rte_eth_dev_count();
+       if (nb_ports == 0)
+               rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
+
+       if (nb_ports > RTE_MAX_ETHPORTS)
+               nb_ports = RTE_MAX_ETHPORTS;
+
+       /* reset l2fwd_dst_ports */
+       for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
+               l2fwd_dst_ports[portid] = 0;
+       last_port = 0;
+
+       /*
+        * Each logical core is assigned a dedicated TX queue on each port.
+        */
+       for (portid = 0; portid < nb_ports; portid++) {
+               /* skip ports that are not enabled */
+               if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
+                       continue;
+
+               if (nb_ports_in_mask % 2) {
+                       l2fwd_dst_ports[portid] = last_port;
+                       l2fwd_dst_ports[last_port] = portid;
+               } else
+                       last_port = portid;
+
+               nb_ports_in_mask++;
+
+               rte_eth_dev_info_get(portid, &dev_info);
+       }
+       if (nb_ports_in_mask % 2) {
+               printf("Notice: odd number of ports in portmask.\n");
+               l2fwd_dst_ports[last_port] = last_port;
+       }
+
+       rx_lcore_id = 0;
+       qconf = NULL;
+
+       /* Initialize the port/queue configuration of each logical core */
+       for (portid = 0; portid < nb_ports; portid++) {
+               /* skip ports that are not enabled */
+               if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
+                       continue;
+
+               /* get the lcore_id for this port */
+               while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
+                      lcore_queue_conf[rx_lcore_id].n_rx_port ==
+                      l2fwd_rx_queue_per_lcore) {
+                       rx_lcore_id++;
+                       if (rx_lcore_id >= RTE_MAX_LCORE)
+                               rte_exit(EXIT_FAILURE, "Not enough cores\n");
+               }
+
+               if (qconf != &lcore_queue_conf[rx_lcore_id])
+                       /* Assigned a new logical core in the loop above. */
+                       qconf = &lcore_queue_conf[rx_lcore_id];
+
+               qconf->rx_port_list[qconf->n_rx_port] = portid;
+               qconf->n_rx_port++;
+               printf("Lcore %u: RX port %u\n", rx_lcore_id, (unsigned) portid);
+       }
+
+       nb_ports_available = nb_ports;
+
+       /* Initialise each port */
+       for (portid = 0; portid < nb_ports; portid++) {
+               /* skip ports that are not enabled */
+               if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
+                       printf("Skipping disabled port %u\n", (unsigned) portid);
+                       nb_ports_available--;
+                       continue;
+               }
+               /* init port */
+               printf("Initializing port %u... ", (unsigned) portid);
+               fflush(stdout);
+               ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
+               if (ret < 0)
+                       rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
+                                 ret, (unsigned) portid);
+
+               rte_eth_macaddr_get(portid, &l2fwd_ports_eth_addr[portid]);
+
+               /* init one RX queue */
+               fflush(stdout);
+               ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
+                                            rte_eth_dev_socket_id(portid),
+                                            NULL,
+                                            l2fwd_pktmbuf_pool);
+               if (ret < 0)
+                       rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
+                                 ret, (unsigned) portid);
+
+               /* init one TX queue on each port */
+               fflush(stdout);
+               ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
+                               rte_eth_dev_socket_id(portid),
+                               NULL);
+               if (ret < 0)
+                       rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n",
+                               ret, (unsigned) portid);
+
+               /* Start device */
+               ret = rte_eth_dev_start(portid);
+               if (ret < 0)
+                       rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
+                                 ret, (unsigned) portid);
+
+               printf("done:\n");
+
+               rte_eth_promiscuous_enable(portid);
+
+               printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
+                               (unsigned) portid,
+                               l2fwd_ports_eth_addr[portid].addr_bytes[0],
+                               l2fwd_ports_eth_addr[portid].addr_bytes[1],
+                               l2fwd_ports_eth_addr[portid].addr_bytes[2],
+                               l2fwd_ports_eth_addr[portid].addr_bytes[3],
+                               l2fwd_ports_eth_addr[portid].addr_bytes[4],
+                               l2fwd_ports_eth_addr[portid].addr_bytes[5]);
+
+               /* initialize port stats */
+               memset(&port_statistics, 0, sizeof(port_statistics));
+       }
+
+       if (!nb_ports_available) {
+               rte_exit(EXIT_FAILURE,
+                       "All available ports are disabled. Please set portmask.\n");
+       }
+
+       check_all_ports_link_status(nb_ports, l2fwd_enabled_port_mask);
+
+       drain_tsc = (hz + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
+
+       RTE_LCORE_FOREACH(lcore_id) {
+               qconf = &lcore_queue_conf[lcore_id];
+
+               rte_spinlock_init(&qconf->lock);
+
+               if (rte_jobstats_context_init(&qconf->jobs_context) != 0)
+                       rte_panic("Jobs stats context for core %u init failed\n", lcore_id);
+
+               if (qconf->n_rx_port == 0) {
+                       RTE_LOG(INFO, L2FWD,
+                               "lcore %u: no ports so no jobs stats context initialization\n",
+                               lcore_id);
+                       continue;
+               }
+               /* Add flush job.
+                * Set fixed period by setting min = max = initial period. Set target to
+                * zero as it is irrelevant for this job. */
+               rte_jobstats_init(&qconf->flush_job, "flush", drain_tsc, drain_tsc,
+                               drain_tsc, 0);
+
+               rte_timer_init(&qconf->flush_timer);
+               rte_timer_reset(&qconf->flush_timer, drain_tsc, PERIODICAL, lcore_id,
+                               &l2fwd_flush_job, NULL);
+
+               if (ret < 0) {
+                       rte_exit(1, "Failed to add flush job for lcore %u: %s",
+                                       lcore_id, rte_strerror(-ret));
+               }
+
+               for (i = 0; i < qconf->n_rx_port; i++) {
+                       struct rte_jobstats *job = &qconf->port_fwd_jobs[i];
+
+                       portid = qconf->rx_port_list[i];
+                       printf("Setting forward jon for port %u\n", portid);
+
+                       snprintf(name, RTE_DIM(name), "port %u fwd", portid);
+                       /* Setup forward job.
+                        * Set min, max and initial period. Set target to MAX_PKT_BURST as
+                        * this is desired optimal RX/TX burst size. */
+                       rte_jobstats_init(job, name, 0, drain_tsc, 0, MAX_PKT_BURST);
+                       rte_jobstats_set_update_period_function(job, l2fwd_job_update_cb);
+
+                       rte_timer_init(&qconf->rx_timers[i]);
+                       rte_timer_reset(&qconf->rx_timers[i], 0, PERIODICAL, lcore_id,
+                                       &l2fwd_fwd_job, (void *)(uintptr_t)i);
+               }
+       }
+
+       if (timer_period)
+               rte_eal_alarm_set(timer_period * MS_PER_S, show_stats_cb, NULL);
+       else
+               RTE_LOG(INFO, L2FWD, "Stats display disabled\n");
+
+       /* launch per-lcore init on every lcore */
+       rte_eal_mp_remote_launch(l2fwd_launch_one_lcore, NULL, CALL_MASTER);
+       RTE_LCORE_FOREACH_SLAVE(lcore_id) {
+               if (rte_eal_wait_lcore(lcore_id) < 0)
+                       return -1;
+       }
+
+       return 0;
+}