app/eventdev: support Tx adapter
authorPavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Mon, 24 Sep 2018 08:02:19 +0000 (13:32 +0530)
committerJerin Jacob <jerin.jacob@caviumnetworks.com>
Mon, 1 Oct 2018 14:54:00 +0000 (16:54 +0200)
Convert existing Tx service based pipeline to Tx adapter based APIs and
simplify worker functions.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
app/test-eventdev/test_pipeline_atq.c
app/test-eventdev/test_pipeline_common.c
app/test-eventdev/test_pipeline_common.h
app/test-eventdev/test_pipeline_queue.c

index f0b2f90..c60635b 100644 (file)
@@ -15,7 +15,7 @@ pipeline_atq_nb_event_queues(struct evt_options *opt)
        return rte_eth_dev_count_avail();
 }
 
-static int
+static __rte_noinline int
 pipeline_atq_worker_single_stage_tx(void *arg)
 {
        PIPELINE_WORKER_SINGLE_STAGE_INIT;
@@ -28,23 +28,18 @@ pipeline_atq_worker_single_stage_tx(void *arg)
                        continue;
                }
 
-               if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
-                       pipeline_tx_pkt(ev.mbuf);
-                       w->processed_pkts++;
-                       continue;
-               }
-               pipeline_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
-               pipeline_event_enqueue(dev, port, &ev);
+               pipeline_event_tx(dev, port, &ev);
+               w->processed_pkts++;
        }
 
        return 0;
 }
 
-static int
+static __rte_noinline int
 pipeline_atq_worker_single_stage_fwd(void *arg)
 {
        PIPELINE_WORKER_SINGLE_STAGE_INIT;
-       const uint8_t tx_queue = t->tx_service.queue_id;
+       const uint8_t *tx_queue = t->tx_evqueue_id;
 
        while (t->done == false) {
                uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
@@ -54,16 +49,16 @@ pipeline_atq_worker_single_stage_fwd(void *arg)
                        continue;
                }
 
-               w->processed_pkts++;
-               ev.queue_id = tx_queue;
+               ev.queue_id = tx_queue[ev.mbuf->port];
                pipeline_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
                pipeline_event_enqueue(dev, port, &ev);
+               w->processed_pkts++;
        }
 
        return 0;
 }
 
-static int
+static __rte_noinline int
 pipeline_atq_worker_single_stage_burst_tx(void *arg)
 {
        PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
@@ -79,27 +74,21 @@ pipeline_atq_worker_single_stage_burst_tx(void *arg)
 
                for (i = 0; i < nb_rx; i++) {
                        rte_prefetch0(ev[i + 1].mbuf);
-                       if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
-
-                               pipeline_tx_pkt(ev[i].mbuf);
-                               ev[i].op = RTE_EVENT_OP_RELEASE;
-                               w->processed_pkts++;
-                       } else
-                               pipeline_fwd_event(&ev[i],
-                                               RTE_SCHED_TYPE_ATOMIC);
+                       rte_event_eth_tx_adapter_txq_set(ev[i].mbuf, 0);
                }
 
-               pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+               pipeline_event_tx_burst(dev, port, ev, nb_rx);
+               w->processed_pkts += nb_rx;
        }
 
        return 0;
 }
 
-static int
+static __rte_noinline int
 pipeline_atq_worker_single_stage_burst_fwd(void *arg)
 {
        PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
-       const uint8_t tx_queue = t->tx_service.queue_id;
+       const uint8_t *tx_queue = t->tx_evqueue_id;
 
        while (t->done == false) {
                uint16_t nb_rx = rte_event_dequeue_burst(dev, port, ev,
@@ -112,23 +101,22 @@ pipeline_atq_worker_single_stage_burst_fwd(void *arg)
 
                for (i = 0; i < nb_rx; i++) {
                        rte_prefetch0(ev[i + 1].mbuf);
-                       ev[i].queue_id = tx_queue;
+                       rte_event_eth_tx_adapter_txq_set(ev[i].mbuf, 0);
+                       ev[i].queue_id = tx_queue[ev[i].mbuf->port];
                        pipeline_fwd_event(&ev[i], RTE_SCHED_TYPE_ATOMIC);
-                       w->processed_pkts++;
                }
 
                pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+               w->processed_pkts += nb_rx;
        }
 
        return 0;
 }
 
-static int
+static __rte_noinline int
 pipeline_atq_worker_multi_stage_tx(void *arg)
 {
        PIPELINE_WORKER_MULTI_STAGE_INIT;
-       const uint8_t nb_stages = t->opt->nb_stages;
-
 
        while (t->done == false) {
                uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
@@ -141,29 +129,24 @@ pipeline_atq_worker_multi_stage_tx(void *arg)
                cq_id = ev.sub_event_type % nb_stages;
 
                if (cq_id == last_queue) {
-                       if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
-
-                               pipeline_tx_pkt(ev.mbuf);
-                               w->processed_pkts++;
-                               continue;
-                       }
-                       pipeline_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
-               } else {
-                       ev.sub_event_type++;
-                       pipeline_fwd_event(&ev, sched_type_list[cq_id]);
+                       pipeline_event_tx(dev, port, &ev);
+                       w->processed_pkts++;
+                       continue;
                }
 
+               ev.sub_event_type++;
+               pipeline_fwd_event(&ev, sched_type_list[cq_id]);
                pipeline_event_enqueue(dev, port, &ev);
        }
+
        return 0;
 }
 
-static int
+static __rte_noinline int
 pipeline_atq_worker_multi_stage_fwd(void *arg)
 {
        PIPELINE_WORKER_MULTI_STAGE_INIT;
-       const uint8_t nb_stages = t->opt->nb_stages;
-       const uint8_t tx_queue = t->tx_service.queue_id;
+       const uint8_t *tx_queue = t->tx_evqueue_id;
 
        while (t->done == false) {
                uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
@@ -176,9 +159,9 @@ pipeline_atq_worker_multi_stage_fwd(void *arg)
                cq_id = ev.sub_event_type % nb_stages;
 
                if (cq_id == last_queue) {
-                       w->processed_pkts++;
-                       ev.queue_id = tx_queue;
+                       ev.queue_id = tx_queue[ev.mbuf->port];
                        pipeline_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
+                       w->processed_pkts++;
                } else {
                        ev.sub_event_type++;
                        pipeline_fwd_event(&ev, sched_type_list[cq_id]);
@@ -186,14 +169,14 @@ pipeline_atq_worker_multi_stage_fwd(void *arg)
 
                pipeline_event_enqueue(dev, port, &ev);
        }
+
        return 0;
 }
 
-static int
+static __rte_noinline int
 pipeline_atq_worker_multi_stage_burst_tx(void *arg)
 {
        PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
-       const uint8_t nb_stages = t->opt->nb_stages;
 
        while (t->done == false) {
                uint16_t nb_rx = rte_event_dequeue_burst(dev, port, ev,
@@ -209,34 +192,27 @@ pipeline_atq_worker_multi_stage_burst_tx(void *arg)
                        cq_id = ev[i].sub_event_type % nb_stages;
 
                        if (cq_id == last_queue) {
-                               if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
-
-                                       pipeline_tx_pkt(ev[i].mbuf);
-                                       ev[i].op = RTE_EVENT_OP_RELEASE;
-                                       w->processed_pkts++;
-                                       continue;
-                               }
-
-                               pipeline_fwd_event(&ev[i],
-                                               RTE_SCHED_TYPE_ATOMIC);
-                       } else {
-                               ev[i].sub_event_type++;
-                               pipeline_fwd_event(&ev[i],
-                                               sched_type_list[cq_id]);
+                               pipeline_event_tx(dev, port, &ev[i]);
+                               ev[i].op = RTE_EVENT_OP_RELEASE;
+                               w->processed_pkts++;
+                               continue;
                        }
+
+                       ev[i].sub_event_type++;
+                       pipeline_fwd_event(&ev[i], sched_type_list[cq_id]);
                }
 
                pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
        }
+
        return 0;
 }
 
-static int
+static __rte_noinline int
 pipeline_atq_worker_multi_stage_burst_fwd(void *arg)
 {
        PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
-       const uint8_t nb_stages = t->opt->nb_stages;
-       const uint8_t tx_queue = t->tx_service.queue_id;
+       const uint8_t *tx_queue = t->tx_evqueue_id;
 
        while (t->done == false) {
                uint16_t nb_rx = rte_event_dequeue_burst(dev, port, ev,
@@ -253,7 +229,7 @@ pipeline_atq_worker_multi_stage_burst_fwd(void *arg)
 
                        if (cq_id == last_queue) {
                                w->processed_pkts++;
-                               ev[i].queue_id = tx_queue;
+                               ev[i].queue_id = tx_queue[ev[i].mbuf->port];
                                pipeline_fwd_event(&ev[i],
                                                RTE_SCHED_TYPE_ATOMIC);
                        } else {
@@ -265,6 +241,7 @@ pipeline_atq_worker_multi_stage_burst_fwd(void *arg)
 
                pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
        }
+
        return 0;
 }
 
@@ -274,39 +251,36 @@ worker_wrapper(void *arg)
        struct worker_data *w  = arg;
        struct evt_options *opt = w->t->opt;
        const bool burst = evt_has_burst_mode(w->dev_id);
-       const bool mt_safe = !w->t->mt_unsafe;
+       const bool internal_port = w->t->internal_port;
        const uint8_t nb_stages = opt->nb_stages;
        RTE_SET_USED(opt);
 
        if (nb_stages == 1) {
-               if (!burst && mt_safe)
+               if (!burst && internal_port)
                        return pipeline_atq_worker_single_stage_tx(arg);
-               else if (!burst && !mt_safe)
+               else if (!burst && !internal_port)
                        return pipeline_atq_worker_single_stage_fwd(arg);
-               else if (burst && mt_safe)
+               else if (burst && internal_port)
                        return pipeline_atq_worker_single_stage_burst_tx(arg);
-               else if (burst && !mt_safe)
+               else if (burst && !internal_port)
                        return pipeline_atq_worker_single_stage_burst_fwd(arg);
        } else {
-               if (!burst && mt_safe)
+               if (!burst && internal_port)
                        return pipeline_atq_worker_multi_stage_tx(arg);
-               else if (!burst && !mt_safe)
+               else if (!burst && !internal_port)
                        return pipeline_atq_worker_multi_stage_fwd(arg);
-               if (burst && mt_safe)
+               if (burst && internal_port)
                        return pipeline_atq_worker_multi_stage_burst_tx(arg);
-               else if (burst && !mt_safe)
+               else if (burst && !internal_port)
                        return pipeline_atq_worker_multi_stage_burst_fwd(arg);
        }
+
        rte_panic("invalid worker\n");
 }
 
 static int
 pipeline_atq_launch_lcores(struct evt_test *test, struct evt_options *opt)
 {
-       struct test_pipeline *t = evt_test_priv(test);
-
-       if (t->mt_unsafe)
-               rte_service_component_runstate_set(t->tx_service.service_id, 1);
        return pipeline_launch_lcores(test, opt, worker_wrapper);
 }
 
@@ -317,34 +291,38 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
        int nb_ports;
        int nb_queues;
        uint8_t queue;
-       struct rte_event_dev_info info;
-       struct test_pipeline *t = evt_test_priv(test);
-       uint8_t tx_evqueue_id = 0;
+       uint8_t tx_evqueue_id[RTE_MAX_ETHPORTS];
        uint8_t queue_arr[RTE_EVENT_MAX_QUEUES_PER_DEV];
        uint8_t nb_worker_queues = 0;
+       uint8_t tx_evport_id = 0;
+       uint16_t prod = 0;
+       struct rte_event_dev_info info;
+       struct test_pipeline *t = evt_test_priv(test);
 
        nb_ports = evt_nr_active_lcores(opt->wlcores);
        nb_queues = rte_eth_dev_count_avail();
 
-       /* One extra port and queueu for Tx service */
-       if (t->mt_unsafe) {
-               tx_evqueue_id = nb_queues;
-               nb_ports++;
-               nb_queues++;
+       memset(tx_evqueue_id, 0, sizeof(uint8_t) * RTE_MAX_ETHPORTS);
+       memset(queue_arr, 0, sizeof(uint8_t) * RTE_EVENT_MAX_QUEUES_PER_DEV);
+       /* One queue for Tx adapter per port */
+       if (!t->internal_port) {
+               RTE_ETH_FOREACH_DEV(prod) {
+                       tx_evqueue_id[prod] = nb_queues;
+                       nb_queues++;
+               }
        }
 
-
        rte_event_dev_info_get(opt->dev_id, &info);
 
        const struct rte_event_dev_config config = {
-                       .nb_event_queues = nb_queues,
-                       .nb_event_ports = nb_ports,
-                       .nb_events_limit  = info.max_num_events,
-                       .nb_event_queue_flows = opt->nb_flows,
-                       .nb_event_port_dequeue_depth =
-                               info.max_event_port_dequeue_depth,
-                       .nb_event_port_enqueue_depth =
-                               info.max_event_port_enqueue_depth,
+               .nb_event_queues = nb_queues,
+               .nb_event_ports = nb_ports,
+               .nb_events_limit  = info.max_num_events,
+               .nb_event_queue_flows = opt->nb_flows,
+               .nb_event_port_dequeue_depth =
+                       info.max_event_port_dequeue_depth,
+               .nb_event_port_enqueue_depth =
+                       info.max_event_port_enqueue_depth,
        };
        ret = rte_event_dev_configure(opt->dev_id, &config);
        if (ret) {
@@ -353,21 +331,23 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
        }
 
        struct rte_event_queue_conf q_conf = {
-                       .priority = RTE_EVENT_DEV_PRIORITY_NORMAL,
-                       .nb_atomic_flows = opt->nb_flows,
-                       .nb_atomic_order_sequences = opt->nb_flows,
+               .priority = RTE_EVENT_DEV_PRIORITY_NORMAL,
+               .nb_atomic_flows = opt->nb_flows,
+               .nb_atomic_order_sequences = opt->nb_flows,
        };
        /* queue configurations */
        for (queue = 0; queue < nb_queues; queue++) {
                q_conf.event_queue_cfg = RTE_EVENT_QUEUE_CFG_ALL_TYPES;
 
-               if (t->mt_unsafe) {
-                       if (queue == tx_evqueue_id) {
-                               q_conf.event_queue_cfg =
-                                       RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
-                       } else {
-                               queue_arr[nb_worker_queues] = queue;
-                               nb_worker_queues++;
+               if (!t->internal_port) {
+                       RTE_ETH_FOREACH_DEV(prod) {
+                               if (queue == tx_evqueue_id[prod]) {
+                                       q_conf.event_queue_cfg =
+                                               RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
+                               } else {
+                                       queue_arr[nb_worker_queues] = queue;
+                                       nb_worker_queues++;
+                               }
                        }
                }
 
@@ -383,20 +363,15 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 
        /* port configuration */
        const struct rte_event_port_conf p_conf = {
-                       .dequeue_depth = opt->wkr_deq_dep,
-                       .enqueue_depth = info.max_event_port_dequeue_depth,
-                       .new_event_threshold = info.max_num_events,
+               .dequeue_depth = opt->wkr_deq_dep,
+               .enqueue_depth = info.max_event_port_dequeue_depth,
+               .new_event_threshold = info.max_num_events,
        };
 
-       if (t->mt_unsafe) {
+       if (!t->internal_port)
                ret = pipeline_event_port_setup(test, opt, queue_arr,
                                nb_worker_queues, p_conf);
-               if (ret)
-                       return ret;
-
-               ret = pipeline_event_tx_service_setup(test, opt, tx_evqueue_id,
-                               nb_ports - 1, p_conf);
-       } else
+       else
                ret = pipeline_event_port_setup(test, opt, NULL, nb_queues,
                                p_conf);
 
@@ -408,30 +383,32 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
         *
         * eth_dev_count = 2, nb_stages = 2, atq mode
         *
-        * Multi thread safe :
+        * eth0, eth1 have Internal port capability :
         *      queues = 2
         *      stride = 1
         *
         *      event queue pipelines:
-        *      eth0 -> q0 ->tx
-        *      eth1 -> q1 ->tx
+        *      eth0 -> q0 ->Tx
+        *      eth1 -> q1 ->Tx
         *
         *      q0, q1 are configured as ATQ so, all the different stages can
         *      be enqueued on the same queue.
         *
-        * Multi thread unsafe :
-        *      queues = 3
+        * eth0, eth1 use Tx adapters service core :
+        *      queues = 4
         *      stride = 1
         *
         *      event queue pipelines:
-        *      eth0 -> q0
-        *                } (q3->tx) Tx service
-        *      eth1 -> q1
+        *      eth0 -> q0  -> q2 -> Tx
+        *      eth1 -> q1  -> q3 -> Tx
         *
-        *      q0,q1 are configured as stated above.
-        *      q3 configured as SINGLE_LINK|ATOMIC.
+        *      q0, q1 are configured as stated above.
+        *      q2, q3 configured as SINGLE_LINK.
         */
        ret = pipeline_event_rx_adapter_setup(opt, 1, p_conf);
+       if (ret)
+               return ret;
+       ret = pipeline_event_tx_adapter_setup(opt, p_conf);
        if (ret)
                return ret;
 
@@ -445,12 +422,58 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
                }
        }
 
+       /* Connect the tx_evqueue_id to the Tx adapter port */
+       if (!t->internal_port) {
+               RTE_ETH_FOREACH_DEV(prod) {
+                       ret = rte_event_eth_tx_adapter_event_port_get(prod,
+                                       &tx_evport_id);
+                       if (ret) {
+                               evt_err("Unable to get Tx adapter[%d]", prod);
+                               return ret;
+                       }
+
+                       if (rte_event_port_link(opt->dev_id, tx_evport_id,
+                                               &tx_evqueue_id[prod],
+                                               NULL, 1) != 1) {
+                               evt_err("Unable to link Tx adptr[%d] evprt[%d]",
+                                               prod, tx_evport_id);
+                               return ret;
+                       }
+               }
+       }
+
+       RTE_ETH_FOREACH_DEV(prod) {
+               ret = rte_eth_dev_start(prod);
+               if (ret) {
+                       evt_err("Ethernet dev [%d] failed to start."
+                                       " Using synthetic producer", prod);
+                       return ret;
+               }
+       }
+
        ret = rte_event_dev_start(opt->dev_id);
        if (ret) {
                evt_err("failed to start eventdev %d", opt->dev_id);
                return ret;
        }
 
+       RTE_ETH_FOREACH_DEV(prod) {
+               ret = rte_event_eth_rx_adapter_start(prod);
+               if (ret) {
+                       evt_err("Rx adapter[%d] start failed", prod);
+                       return ret;
+               }
+
+               ret = rte_event_eth_tx_adapter_start(prod);
+               if (ret) {
+                       evt_err("Tx adapter[%d] start failed", prod);
+                       return ret;
+               }
+       }
+
+       memcpy(t->tx_evqueue_id, tx_evqueue_id, sizeof(uint8_t) *
+                       RTE_MAX_ETHPORTS);
+
        return 0;
 }
 
index cdf9010..d07fa88 100644 (file)
@@ -5,58 +5,6 @@
 
 #include "test_pipeline_common.h"
 
-static int32_t
-pipeline_event_tx_burst_service_func(void *args)
-{
-
-       int i;
-       struct tx_service_data *tx = args;
-       const uint8_t dev = tx->dev_id;
-       const uint8_t port = tx->port_id;
-       struct rte_event ev[BURST_SIZE + 1];
-
-       uint16_t nb_rx = rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
-
-       if (!nb_rx) {
-               for (i = 0; i < tx->nb_ethports; i++)
-                       rte_eth_tx_buffer_flush(i, 0, tx->tx_buf[i]);
-               return 0;
-       }
-
-       for (i = 0; i < nb_rx; i++) {
-               struct rte_mbuf *m = ev[i].mbuf;
-               rte_eth_tx_buffer(m->port, 0, tx->tx_buf[m->port], m);
-       }
-       tx->processed_pkts += nb_rx;
-
-       return 0;
-}
-
-static int32_t
-pipeline_event_tx_service_func(void *args)
-{
-
-       int i;
-       struct tx_service_data *tx = args;
-       const uint8_t dev = tx->dev_id;
-       const uint8_t port = tx->port_id;
-       struct rte_event ev;
-
-       uint16_t nb_rx = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
-
-       if (!nb_rx) {
-               for (i = 0; i < tx->nb_ethports; i++)
-                       rte_eth_tx_buffer_flush(i, 0, tx->tx_buf[i]);
-               return 0;
-       }
-
-       struct rte_mbuf *m = ev.mbuf;
-       rte_eth_tx_buffer(m->port, 0, tx->tx_buf[m->port], m);
-       tx->processed_pkts++;
-
-       return 0;
-}
-
 int
 pipeline_test_result(struct evt_test *test, struct evt_options *opt)
 {
@@ -97,11 +45,8 @@ processed_pkts(struct test_pipeline *t)
        uint64_t total = 0;
 
        rte_smp_rmb();
-       if (t->mt_unsafe)
-               total = t->tx_service.processed_pkts;
-       else
-               for (i = 0; i < t->nb_workers; i++)
-                       total += t->worker[i].processed_pkts;
+       for (i = 0; i < t->nb_workers; i++)
+               total += t->worker[i].processed_pkts;
 
        return total;
 }
@@ -215,7 +160,6 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 {
        uint16_t i;
        uint8_t nb_queues = 1;
-       uint8_t mt_state = 0;
        struct test_pipeline *t = evt_test_priv(test);
        struct rte_eth_rxconf rx_conf;
        struct rte_eth_conf port_conf = {
@@ -237,13 +181,17 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
                return -ENODEV;
        }
 
+       t->internal_port = 1;
        RTE_ETH_FOREACH_DEV(i) {
                struct rte_eth_dev_info dev_info;
                struct rte_eth_conf local_port_conf = port_conf;
+               uint32_t caps = 0;
+
+               rte_event_eth_tx_adapter_caps_get(opt->dev_id, i, &caps);
+               if (!(caps & RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT))
+                       t->internal_port = 0;
 
                rte_eth_dev_info_get(i, &dev_info);
-               mt_state = !(dev_info.tx_offload_capa &
-                               DEV_TX_OFFLOAD_MT_LOCKFREE);
                rx_conf = dev_info.default_rxconf;
                rx_conf.offloads = port_conf.rxmode.offloads;
 
@@ -278,11 +226,6 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
                        return -EINVAL;
                }
 
-               t->mt_unsafe |= mt_state;
-               t->tx_service.tx_buf[i] =
-                       rte_malloc(NULL, RTE_ETH_TX_BUFFER_SIZE(BURST_SIZE), 0);
-               if (t->tx_service.tx_buf[i] == NULL)
-                       rte_panic("Unable to allocate Tx buffer memory.");
                rte_eth_promiscuous_enable(i);
        }
 
@@ -294,7 +237,6 @@ pipeline_event_port_setup(struct evt_test *test, struct evt_options *opt,
                uint8_t *queue_arr, uint8_t nb_queues,
                const struct rte_event_port_conf p_conf)
 {
-       int i;
        int ret;
        uint8_t port;
        struct test_pipeline *t = evt_test_priv(test);
@@ -315,23 +257,15 @@ pipeline_event_port_setup(struct evt_test *test, struct evt_options *opt,
                        return ret;
                }
 
-               if (queue_arr == NULL) {
-                       if (rte_event_port_link(opt->dev_id, port, NULL, NULL,
-                                               0) != nb_queues)
-                               goto link_fail;
-               } else {
-                       for (i = 0; i < nb_queues; i++) {
-                               if (rte_event_port_link(opt->dev_id, port,
-                                               &queue_arr[i], NULL, 1) != 1)
-                                       goto link_fail;
-                       }
-               }
+               if (rte_event_port_link(opt->dev_id, port, queue_arr, NULL,
+                                       nb_queues) != nb_queues)
+                       goto link_fail;
        }
 
        return 0;
 
 link_fail:
-       evt_err("failed to link all queues to port %d", port);
+       evt_err("failed to link queues to port %d", port);
        return -EINVAL;
 }
 
@@ -384,79 +318,64 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
                        }
                }
 
-               ret = rte_eth_dev_start(prod);
-               if (ret) {
-                       evt_err("Ethernet dev [%d] failed to start."
-                                       " Using synthetic producer", prod);
-                       return ret;
-               }
-
-               ret = rte_event_eth_rx_adapter_start(prod);
-               if (ret) {
-                       evt_err("Rx adapter[%d] start failed", prod);
-                       return ret;
-               }
-               evt_info("Port[%d] using Rx adapter[%d] started", prod, prod);
+               evt_info("Port[%d] using Rx adapter[%d] configured", prod,
+                               prod);
        }
 
        return ret;
 }
 
 int
-pipeline_event_tx_service_setup(struct evt_test *test, struct evt_options *opt,
-               uint8_t tx_queue_id, uint8_t tx_port_id,
-               const struct rte_event_port_conf p_conf)
+pipeline_event_tx_adapter_setup(struct evt_options *opt,
+               struct rte_event_port_conf port_conf)
 {
        int ret;
-       struct rte_service_spec serv;
-       struct test_pipeline *t = evt_test_priv(test);
-       struct tx_service_data *tx = &t->tx_service;
+       uint16_t consm;
 
-       ret = rte_event_port_setup(opt->dev_id, tx_port_id, &p_conf);
-       if (ret) {
-               evt_err("failed to setup port %d", tx_port_id);
-               return ret;
-       }
+       RTE_ETH_FOREACH_DEV(consm) {
+               uint32_t cap;
 
-       if (rte_event_port_link(opt->dev_id, tx_port_id, &tx_queue_id,
-                               NULL, 1) != 1) {
-               evt_err("failed to link queues to port %d", tx_port_id);
-               return -EINVAL;
-       }
+               ret = rte_event_eth_tx_adapter_caps_get(opt->dev_id,
+                               consm, &cap);
+               if (ret) {
+                       evt_err("failed to get event tx adapter[%d] caps",
+                                       consm);
+                       return ret;
+               }
 
-       tx->dev_id = opt->dev_id;
-       tx->queue_id = tx_queue_id;
-       tx->port_id = tx_port_id;
-       tx->nb_ethports = rte_eth_dev_count_avail();
-       tx->t = t;
-
-       /* Register Tx service */
-       memset(&serv, 0, sizeof(struct rte_service_spec));
-       snprintf(serv.name, sizeof(serv.name), "Tx_service");
-
-       if (evt_has_burst_mode(opt->dev_id))
-               serv.callback = pipeline_event_tx_burst_service_func;
-       else
-               serv.callback = pipeline_event_tx_service_func;
-
-       serv.callback_userdata = (void *)tx;
-       ret = rte_service_component_register(&serv, &tx->service_id);
-       if (ret) {
-               evt_err("failed to register Tx service");
-               return ret;
-       }
+               ret = rte_event_eth_tx_adapter_create(consm, opt->dev_id,
+                               &port_conf);
+               if (ret) {
+                       evt_err("failed to create tx adapter[%d]", consm);
+                       return ret;
+               }
 
-       ret = evt_service_setup(tx->service_id);
-       if (ret) {
-               evt_err("Failed to setup service core for Tx service\n");
-               return ret;
-       }
+               ret = rte_event_eth_tx_adapter_queue_add(consm, consm, -1);
+               if (ret) {
+                       evt_err("failed to add tx queues to adapter[%d]",
+                                       consm);
+                       return ret;
+               }
 
-       rte_service_runstate_set(tx->service_id, 1);
+               if (!(cap & RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT)) {
+                       uint32_t service_id;
 
-       return 0;
-}
+                       rte_event_eth_tx_adapter_service_id_get(consm,
+                                       &service_id);
+                       ret = evt_service_setup(service_id);
+                       if (ret) {
+                               evt_err("Failed to setup service core"
+                                               " for Tx adapter\n");
+                               return ret;
+                       }
+               }
+
+               evt_info("Port[%d] using Tx adapter[%d] Configured", consm,
+                               consm);
+       }
 
+       return ret;
+}
 
 void
 pipeline_ethdev_destroy(struct evt_test *test, struct evt_options *opt)
@@ -464,16 +383,10 @@ pipeline_ethdev_destroy(struct evt_test *test, struct evt_options *opt)
        uint16_t i;
        RTE_SET_USED(test);
        RTE_SET_USED(opt);
-       struct test_pipeline *t = evt_test_priv(test);
-
-       if (t->mt_unsafe) {
-               rte_service_component_runstate_set(t->tx_service.service_id, 0);
-               rte_service_runstate_set(t->tx_service.service_id, 0);
-               rte_service_component_unregister(t->tx_service.service_id);
-       }
 
        RTE_ETH_FOREACH_DEV(i) {
                rte_event_eth_rx_adapter_stop(i);
+               rte_event_eth_tx_adapter_stop(i);
                rte_eth_dev_stop(i);
        }
 }
@@ -483,7 +396,6 @@ pipeline_eventdev_destroy(struct evt_test *test, struct evt_options *opt)
 {
        RTE_SET_USED(test);
 
-       rte_event_dev_stop(opt->dev_id);
        rte_event_dev_close(opt->dev_id);
 }
 
index 9cd6b90..0440b9e 100644 (file)
@@ -14,6 +14,7 @@
 #include <rte_ethdev.h>
 #include <rte_eventdev.h>
 #include <rte_event_eth_rx_adapter.h>
+#include <rte_event_eth_tx_adapter.h>
 #include <rte_lcore.h>
 #include <rte_malloc.h>
 #include <rte_mempool.h>
@@ -35,30 +36,19 @@ struct worker_data {
        struct test_pipeline *t;
 } __rte_cache_aligned;
 
-struct tx_service_data {
-       uint8_t dev_id;
-       uint8_t queue_id;
-       uint8_t port_id;
-       uint32_t service_id;
-       uint64_t processed_pkts;
-       uint16_t nb_ethports;
-       struct rte_eth_dev_tx_buffer *tx_buf[RTE_MAX_ETHPORTS];
-       struct test_pipeline *t;
-} __rte_cache_aligned;
-
 struct test_pipeline {
        /* Don't change the offset of "done". Signal handler use this memory
         * to terminate all lcores work.
         */
        int done;
        uint8_t nb_workers;
-       uint8_t mt_unsafe;
+       uint8_t internal_port;
+       uint8_t tx_evqueue_id[RTE_MAX_ETHPORTS];
        enum evt_test_result result;
        uint32_t nb_flows;
        uint64_t outstand_pkts;
        struct rte_mempool *pool;
        struct worker_data worker[EVT_MAX_PORTS];
-       struct tx_service_data tx_service;
        struct evt_options *opt;
        uint8_t sched_type_list[EVT_MAX_STAGES] __rte_cache_aligned;
 } __rte_cache_aligned;
@@ -70,7 +60,7 @@ struct test_pipeline {
        struct test_pipeline *t = w->t;   \
        const uint8_t dev = w->dev_id;    \
        const uint8_t port = w->port_id;  \
-       struct rte_event ev
+       struct rte_event ev __rte_cache_aligned
 
 #define PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT \
        int i;                                  \
@@ -78,7 +68,7 @@ struct test_pipeline {
        struct test_pipeline *t = w->t;         \
        const uint8_t dev = w->dev_id;          \
        const uint8_t port = w->port_id;        \
-       struct rte_event ev[BURST_SIZE + 1]
+       struct rte_event ev[BURST_SIZE + 1] __rte_cache_aligned
 
 #define PIPELINE_WORKER_MULTI_STAGE_INIT                         \
        struct worker_data *w  = arg;                            \
@@ -88,10 +78,11 @@ struct test_pipeline {
        const uint8_t port = w->port_id;                         \
        const uint8_t last_queue = t->opt->nb_stages - 1;        \
        uint8_t *const sched_type_list = &t->sched_type_list[0]; \
-       struct rte_event ev
+       const uint8_t nb_stages = t->opt->nb_stages + 1;         \
+       struct rte_event ev __rte_cache_aligned
 
 #define PIPELINE_WORKER_MULTI_STAGE_BURST_INIT                   \
-       int i;                                  \
+       int i;                                                   \
        struct worker_data *w  = arg;                            \
        struct test_pipeline *t = w->t;                          \
        uint8_t cq_id;                                           \
@@ -99,7 +90,8 @@ struct test_pipeline {
        const uint8_t port = w->port_id;                         \
        const uint8_t last_queue = t->opt->nb_stages - 1;        \
        uint8_t *const sched_type_list = &t->sched_type_list[0]; \
-       struct rte_event ev[BURST_SIZE + 1]
+       const uint8_t nb_stages = t->opt->nb_stages + 1;         \
+       struct rte_event ev[BURST_SIZE + 1] __rte_cache_aligned
 
 static __rte_always_inline void
 pipeline_fwd_event(struct rte_event *ev, uint8_t sched)
@@ -109,6 +101,28 @@ pipeline_fwd_event(struct rte_event *ev, uint8_t sched)
        ev->sched_type = sched;
 }
 
+static __rte_always_inline void
+pipeline_event_tx(const uint8_t dev, const uint8_t port,
+               struct rte_event * const ev)
+{
+       rte_event_eth_tx_adapter_txq_set(ev->mbuf, 0);
+       while (!rte_event_eth_tx_adapter_enqueue(dev, port, ev, 1))
+               rte_pause();
+}
+
+static __rte_always_inline void
+pipeline_event_tx_burst(const uint8_t dev, const uint8_t port,
+               struct rte_event *ev, const uint16_t nb_rx)
+{
+       uint16_t enq;
+
+       enq = rte_event_eth_tx_adapter_enqueue(dev, port, ev, nb_rx);
+       while (enq < nb_rx) {
+               enq += rte_event_eth_tx_adapter_enqueue(dev, port,
+                               ev + enq, nb_rx - enq);
+       }
+}
+
 static __rte_always_inline void
 pipeline_event_enqueue(const uint8_t dev, const uint8_t port,
                struct rte_event *ev)
@@ -130,13 +144,6 @@ pipeline_event_enqueue_burst(const uint8_t dev, const uint8_t port,
        }
 }
 
-static __rte_always_inline void
-pipeline_tx_pkt(struct rte_mbuf *mbuf)
-{
-       while (rte_eth_tx_burst(mbuf->port, 0, &mbuf, 1) != 1)
-               rte_pause();
-}
-
 static inline int
 pipeline_nb_event_ports(struct evt_options *opt)
 {
@@ -149,9 +156,8 @@ int pipeline_test_setup(struct evt_test *test, struct evt_options *opt);
 int pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt);
 int pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
                struct rte_event_port_conf prod_conf);
-int pipeline_event_tx_service_setup(struct evt_test *test,
-               struct evt_options *opt, uint8_t tx_queue_id,
-               uint8_t tx_port_id, const struct rte_event_port_conf p_conf);
+int pipeline_event_tx_adapter_setup(struct evt_options *opt,
+               struct rte_event_port_conf prod_conf);
 int pipeline_mempool_setup(struct evt_test *test, struct evt_options *opt);
 int pipeline_event_port_setup(struct evt_test *test, struct evt_options *opt,
                uint8_t *queue_arr, uint8_t nb_queues,
index 2e0d93d..2521700 100644 (file)
@@ -15,7 +15,7 @@ pipeline_queue_nb_event_queues(struct evt_options *opt)
        return (eth_count * opt->nb_stages) + eth_count;
 }
 
-static int
+static __rte_noinline int
 pipeline_queue_worker_single_stage_tx(void *arg)
 {
        PIPELINE_WORKER_SINGLE_STAGE_INIT;
@@ -29,7 +29,7 @@ pipeline_queue_worker_single_stage_tx(void *arg)
                }
 
                if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
-                       pipeline_tx_pkt(ev.mbuf);
+                       pipeline_event_tx(dev, port, &ev);
                        w->processed_pkts++;
                } else {
                        ev.queue_id++;
@@ -41,11 +41,11 @@ pipeline_queue_worker_single_stage_tx(void *arg)
        return 0;
 }
 
-static int
+static __rte_noinline int
 pipeline_queue_worker_single_stage_fwd(void *arg)
 {
        PIPELINE_WORKER_SINGLE_STAGE_INIT;
-       const uint8_t tx_queue = t->tx_service.queue_id;
+       const uint8_t *tx_queue = t->tx_evqueue_id;
 
        while (t->done == false) {
                uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
@@ -55,7 +55,8 @@ pipeline_queue_worker_single_stage_fwd(void *arg)
                        continue;
                }
 
-               ev.queue_id = tx_queue;
+               ev.queue_id = tx_queue[ev.mbuf->port];
+               rte_event_eth_tx_adapter_txq_set(ev.mbuf, 0);
                pipeline_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
                pipeline_event_enqueue(dev, port, &ev);
                w->processed_pkts++;
@@ -64,7 +65,7 @@ pipeline_queue_worker_single_stage_fwd(void *arg)
        return 0;
 }
 
-static int
+static __rte_noinline int
 pipeline_queue_worker_single_stage_burst_tx(void *arg)
 {
        PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
@@ -81,8 +82,7 @@ pipeline_queue_worker_single_stage_burst_tx(void *arg)
                for (i = 0; i < nb_rx; i++) {
                        rte_prefetch0(ev[i + 1].mbuf);
                        if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
-
-                               pipeline_tx_pkt(ev[i].mbuf);
+                               pipeline_event_tx(dev, port, &ev[i]);
                                ev[i].op = RTE_EVENT_OP_RELEASE;
                                w->processed_pkts++;
                        } else {
@@ -98,11 +98,11 @@ pipeline_queue_worker_single_stage_burst_tx(void *arg)
        return 0;
 }
 
-static int
+static __rte_noinline int
 pipeline_queue_worker_single_stage_burst_fwd(void *arg)
 {
        PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
-       const uint8_t tx_queue = t->tx_service.queue_id;
+       const uint8_t *tx_queue = t->tx_evqueue_id;
 
        while (t->done == false) {
                uint16_t nb_rx = rte_event_dequeue_burst(dev, port, ev,
@@ -115,23 +115,24 @@ pipeline_queue_worker_single_stage_burst_fwd(void *arg)
 
                for (i = 0; i < nb_rx; i++) {
                        rte_prefetch0(ev[i + 1].mbuf);
-                       ev[i].queue_id = tx_queue;
+                       ev[i].queue_id = tx_queue[ev[i].mbuf->port];
+                       rte_event_eth_tx_adapter_txq_set(ev[i].mbuf, 0);
                        pipeline_fwd_event(&ev[i], RTE_SCHED_TYPE_ATOMIC);
-                       w->processed_pkts++;
                }
 
                pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+               w->processed_pkts += nb_rx;
        }
 
        return 0;
 }
 
 
-static int
+static __rte_noinline int
 pipeline_queue_worker_multi_stage_tx(void *arg)
 {
        PIPELINE_WORKER_MULTI_STAGE_INIT;
-       const uint8_t nb_stages = t->opt->nb_stages + 1;
+       const uint8_t *tx_queue = t->tx_evqueue_id;
 
        while (t->done == false) {
                uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
@@ -143,31 +144,27 @@ pipeline_queue_worker_multi_stage_tx(void *arg)
 
                cq_id = ev.queue_id % nb_stages;
 
-               if (cq_id >= last_queue) {
-                       if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
-
-                               pipeline_tx_pkt(ev.mbuf);
-                               w->processed_pkts++;
-                               continue;
-                       }
-                       ev.queue_id += (cq_id == last_queue) ? 1 : 0;
-                       pipeline_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
-               } else {
-                       ev.queue_id++;
-                       pipeline_fwd_event(&ev, sched_type_list[cq_id]);
+               if (ev.queue_id == tx_queue[ev.mbuf->port]) {
+                       pipeline_event_tx(dev, port, &ev);
+                       w->processed_pkts++;
+                       continue;
                }
 
+               ev.queue_id++;
+               pipeline_fwd_event(&ev, cq_id != last_queue ?
+                               sched_type_list[cq_id] :
+                               RTE_SCHED_TYPE_ATOMIC);
                pipeline_event_enqueue(dev, port, &ev);
        }
+
        return 0;
 }
 
-static int
+static __rte_noinline int
 pipeline_queue_worker_multi_stage_fwd(void *arg)
 {
        PIPELINE_WORKER_MULTI_STAGE_INIT;
-       const uint8_t nb_stages = t->opt->nb_stages + 1;
-       const uint8_t tx_queue = t->tx_service.queue_id;
+       const uint8_t *tx_queue = t->tx_evqueue_id;
 
        while (t->done == false) {
                uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
@@ -180,7 +177,8 @@ pipeline_queue_worker_multi_stage_fwd(void *arg)
                cq_id = ev.queue_id % nb_stages;
 
                if (cq_id == last_queue) {
-                       ev.queue_id = tx_queue;
+                       ev.queue_id = tx_queue[ev.mbuf->port];
+                       rte_event_eth_tx_adapter_txq_set(ev.mbuf, 0);
                        pipeline_fwd_event(&ev, RTE_SCHED_TYPE_ATOMIC);
                        w->processed_pkts++;
                } else {
@@ -190,14 +188,15 @@ pipeline_queue_worker_multi_stage_fwd(void *arg)
 
                pipeline_event_enqueue(dev, port, &ev);
        }
+
        return 0;
 }
 
-static int
+static __rte_noinline int
 pipeline_queue_worker_multi_stage_burst_tx(void *arg)
 {
        PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
-       const uint8_t nb_stages = t->opt->nb_stages + 1;
+       const uint8_t *tx_queue = t->tx_evqueue_id;
 
        while (t->done == false) {
                uint16_t nb_rx = rte_event_dequeue_burst(dev, port, ev,
@@ -212,37 +211,30 @@ pipeline_queue_worker_multi_stage_burst_tx(void *arg)
                        rte_prefetch0(ev[i + 1].mbuf);
                        cq_id = ev[i].queue_id % nb_stages;
 
-                       if (cq_id >= last_queue) {
-                               if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
-
-                                       pipeline_tx_pkt(ev[i].mbuf);
-                                       ev[i].op = RTE_EVENT_OP_RELEASE;
-                                       w->processed_pkts++;
-                                       continue;
-                               }
-
-                               ev[i].queue_id += (cq_id == last_queue) ? 1 : 0;
-                               pipeline_fwd_event(&ev[i],
-                                               RTE_SCHED_TYPE_ATOMIC);
-                       } else {
-                               ev[i].queue_id++;
-                               pipeline_fwd_event(&ev[i],
-                                               sched_type_list[cq_id]);
+                       if (ev[i].queue_id == tx_queue[ev[i].mbuf->port]) {
+                               pipeline_event_tx(dev, port, &ev[i]);
+                               ev[i].op = RTE_EVENT_OP_RELEASE;
+                               w->processed_pkts++;
+                               continue;
                        }
 
+                       ev[i].queue_id++;
+                       pipeline_fwd_event(&ev[i], cq_id != last_queue ?
+                                       sched_type_list[cq_id] :
+                                       RTE_SCHED_TYPE_ATOMIC);
                }
 
                pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
        }
+
        return 0;
 }
 
-static int
+static __rte_noinline int
 pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
 {
        PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
-       const uint8_t nb_stages = t->opt->nb_stages + 1;
-       const uint8_t tx_queue = t->tx_service.queue_id;
+       const uint8_t *tx_queue = t->tx_evqueue_id;
 
        while (t->done == false) {
                uint16_t nb_rx = rte_event_dequeue_burst(dev, port, ev,
@@ -258,7 +250,8 @@ pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
                        cq_id = ev[i].queue_id % nb_stages;
 
                        if (cq_id == last_queue) {
-                               ev[i].queue_id = tx_queue;
+                               ev[i].queue_id = tx_queue[ev[i].mbuf->port];
+                               rte_event_eth_tx_adapter_txq_set(ev[i].mbuf, 0);
                                pipeline_fwd_event(&ev[i],
                                                RTE_SCHED_TYPE_ATOMIC);
                                w->processed_pkts++;
@@ -271,6 +264,7 @@ pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
 
                pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
        }
+
        return 0;
 }
 
@@ -280,28 +274,28 @@ worker_wrapper(void *arg)
        struct worker_data *w  = arg;
        struct evt_options *opt = w->t->opt;
        const bool burst = evt_has_burst_mode(w->dev_id);
-       const bool mt_safe = !w->t->mt_unsafe;
+       const bool internal_port = w->t->internal_port;
        const uint8_t nb_stages = opt->nb_stages;
        RTE_SET_USED(opt);
 
        if (nb_stages == 1) {
-               if (!burst && mt_safe)
+               if (!burst && internal_port)
                        return pipeline_queue_worker_single_stage_tx(arg);
-               else if (!burst && !mt_safe)
+               else if (!burst && !internal_port)
                        return pipeline_queue_worker_single_stage_fwd(arg);
-               else if (burst && mt_safe)
+               else if (burst && internal_port)
                        return pipeline_queue_worker_single_stage_burst_tx(arg);
-               else if (burst && !mt_safe)
+               else if (burst && !internal_port)
                        return pipeline_queue_worker_single_stage_burst_fwd(
                                        arg);
        } else {
-               if (!burst && mt_safe)
+               if (!burst && internal_port)
                        return pipeline_queue_worker_multi_stage_tx(arg);
-               else if (!burst && !mt_safe)
+               else if (!burst && !internal_port)
                        return pipeline_queue_worker_multi_stage_fwd(arg);
-               else if (burst && mt_safe)
+               else if (burst && internal_port)
                        return pipeline_queue_worker_multi_stage_burst_tx(arg);
-               else if (burst && !mt_safe)
+               else if (burst && !internal_port)
                        return pipeline_queue_worker_multi_stage_burst_fwd(arg);
 
        }
@@ -311,10 +305,6 @@ worker_wrapper(void *arg)
 static int
 pipeline_queue_launch_lcores(struct evt_test *test, struct evt_options *opt)
 {
-       struct test_pipeline *t = evt_test_priv(test);
-
-       if (t->mt_unsafe)
-               rte_service_component_runstate_set(t->tx_service.service_id, 1);
        return pipeline_launch_lcores(test, opt, worker_wrapper);
 }
 
@@ -326,25 +316,24 @@ pipeline_queue_eventdev_setup(struct evt_test *test, struct evt_options *opt)
        int nb_queues;
        int nb_stages = opt->nb_stages;
        uint8_t queue;
-       struct rte_event_dev_info info;
-       struct test_pipeline *t = evt_test_priv(test);
-       uint8_t tx_evqueue_id = 0;
+       uint8_t tx_evport_id = 0;
+       uint8_t tx_evqueue_id[RTE_MAX_ETHPORTS];
        uint8_t queue_arr[RTE_EVENT_MAX_QUEUES_PER_DEV];
        uint8_t nb_worker_queues = 0;
+       uint16_t prod = 0;
+       struct rte_event_dev_info info;
+       struct test_pipeline *t = evt_test_priv(test);
 
        nb_ports = evt_nr_active_lcores(opt->wlcores);
        nb_queues = rte_eth_dev_count_avail() * (nb_stages);
 
-       /* Extra port for Tx service. */
-       if (t->mt_unsafe) {
-               tx_evqueue_id = nb_queues;
-               nb_ports++;
-               nb_queues++;
-       } else
-               nb_queues += rte_eth_dev_count_avail();
+       /* One queue for Tx adapter per port */
+       nb_queues += rte_eth_dev_count_avail();
 
-       rte_event_dev_info_get(opt->dev_id, &info);
+       memset(tx_evqueue_id, 0, sizeof(uint8_t) * RTE_MAX_ETHPORTS);
+       memset(queue_arr, 0, sizeof(uint8_t) * RTE_EVENT_MAX_QUEUES_PER_DEV);
 
+       rte_event_dev_info_get(opt->dev_id, &info);
        const struct rte_event_dev_config config = {
                        .nb_event_queues = nb_queues,
                        .nb_event_ports = nb_ports,
@@ -370,24 +359,19 @@ pipeline_queue_eventdev_setup(struct evt_test *test, struct evt_options *opt)
        for (queue = 0; queue < nb_queues; queue++) {
                uint8_t slot;
 
-               if (!t->mt_unsafe) {
-                       slot = queue % (nb_stages + 1);
-                       q_conf.schedule_type = slot == nb_stages ?
-                               RTE_SCHED_TYPE_ATOMIC :
-                               opt->sched_type_list[slot];
-               } else {
-                       slot = queue % nb_stages;
-
-                       if (queue == tx_evqueue_id) {
-                               q_conf.schedule_type = RTE_SCHED_TYPE_ATOMIC;
+               q_conf.event_queue_cfg = 0;
+               slot = queue % (nb_stages + 1);
+               if (slot == nb_stages) {
+                       q_conf.schedule_type = RTE_SCHED_TYPE_ATOMIC;
+                       if (!t->internal_port) {
                                q_conf.event_queue_cfg =
                                        RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
-                       } else {
-                               q_conf.schedule_type =
-                                       opt->sched_type_list[slot];
-                               queue_arr[nb_worker_queues] = queue;
-                               nb_worker_queues++;
                        }
+                       tx_evqueue_id[prod++] = queue;
+               } else {
+                       q_conf.schedule_type = opt->sched_type_list[slot];
+                       queue_arr[nb_worker_queues] = queue;
+                       nb_worker_queues++;
                }
 
                ret = rte_event_queue_setup(opt->dev_id, queue, &q_conf);
@@ -407,19 +391,11 @@ pipeline_queue_eventdev_setup(struct evt_test *test, struct evt_options *opt)
                        .new_event_threshold = info.max_num_events,
        };
 
-       /*
-        * If tx is multi thread safe then allow workers to do Tx else use Tx
-        * service to Tx packets.
-        */
-       if (t->mt_unsafe) {
+       if (!t->internal_port) {
                ret = pipeline_event_port_setup(test, opt, queue_arr,
                                nb_worker_queues, p_conf);
                if (ret)
                        return ret;
-
-               ret = pipeline_event_tx_service_setup(test, opt, tx_evqueue_id,
-                               nb_ports - 1, p_conf);
-
        } else
                ret = pipeline_event_port_setup(test, opt, NULL, nb_queues,
                                p_conf);
@@ -431,7 +407,6 @@ pipeline_queue_eventdev_setup(struct evt_test *test, struct evt_options *opt)
         *
         * eth_dev_count = 2, nb_stages = 2.
         *
-        * Multi thread safe :
         *      queues = 6
         *      stride = 3
         *
@@ -439,21 +414,14 @@ pipeline_queue_eventdev_setup(struct evt_test *test, struct evt_options *opt)
         *      eth0 -> q0 -> q1 -> (q2->tx)
         *      eth1 -> q3 -> q4 -> (q5->tx)
         *
-        *      q2, q5 configured as ATOMIC
-        *
-        * Multi thread unsafe :
-        *      queues = 5
-        *      stride = 2
-        *
-        *      event queue pipelines:
-        *      eth0 -> q0 -> q1
-        *                      } (q4->tx) Tx service
-        *      eth1 -> q2 -> q3
+        *      q2, q5 configured as ATOMIC | SINGLE_LINK
         *
-        *      q4 configured as SINGLE_LINK|ATOMIC
         */
-       ret = pipeline_event_rx_adapter_setup(opt,
-                       t->mt_unsafe ? nb_stages : nb_stages + 1, p_conf);
+       ret = pipeline_event_rx_adapter_setup(opt, nb_stages + 1, p_conf);
+       if (ret)
+               return ret;
+
+       ret = pipeline_event_tx_adapter_setup(opt, p_conf);
        if (ret)
                return ret;
 
@@ -467,12 +435,60 @@ pipeline_queue_eventdev_setup(struct evt_test *test, struct evt_options *opt)
                }
        }
 
+       /* Connect the tx_evqueue_id to the Tx adapter port */
+       if (!t->internal_port) {
+               RTE_ETH_FOREACH_DEV(prod) {
+                       ret = rte_event_eth_tx_adapter_event_port_get(prod,
+                                       &tx_evport_id);
+                       if (ret) {
+                               evt_err("Unable to get Tx adptr[%d] evprt[%d]",
+                                               prod, tx_evport_id);
+                               return ret;
+                       }
+
+                       if (rte_event_port_link(opt->dev_id, tx_evport_id,
+                                               &tx_evqueue_id[prod],
+                                               NULL, 1) != 1) {
+                               evt_err("Unable to link Tx adptr[%d] evprt[%d]",
+                                               prod, tx_evport_id);
+                               return ret;
+                       }
+               }
+       }
+
+       RTE_ETH_FOREACH_DEV(prod) {
+               ret = rte_eth_dev_start(prod);
+               if (ret) {
+                       evt_err("Ethernet dev [%d] failed to start."
+                                       " Using synthetic producer", prod);
+                       return ret;
+               }
+
+       }
+
        ret = rte_event_dev_start(opt->dev_id);
        if (ret) {
                evt_err("failed to start eventdev %d", opt->dev_id);
                return ret;
        }
 
+       RTE_ETH_FOREACH_DEV(prod) {
+               ret = rte_event_eth_rx_adapter_start(prod);
+               if (ret) {
+                       evt_err("Rx adapter[%d] start failed", prod);
+                       return ret;
+               }
+
+               ret = rte_event_eth_tx_adapter_start(prod);
+               if (ret) {
+                       evt_err("Tx adapter[%d] start failed", prod);
+                       return ret;
+               }
+       }
+
+       memcpy(t->tx_evqueue_id, tx_evqueue_id, sizeof(uint8_t) *
+                       RTE_MAX_ETHPORTS);
+
        return 0;
 }