drivers/dpaa: optimize thread local storage
[dpdk.git] / drivers / event / dpaa / dpaa_eventdev.c
index ba9b814..a3c138b 100644 (file)
@@ -1,5 +1,5 @@
 /*   SPDX-License-Identifier:        BSD-3-Clause
- *   Copyright 2017 NXP
+ *   Copyright 2017-2019 NXP
  */
 
 #include <assert.h>
 #include <rte_eventdev_pmd_vdev.h>
 #include <rte_ethdev.h>
 #include <rte_event_eth_rx_adapter.h>
+#include <rte_event_eth_tx_adapter.h>
+#include <rte_cryptodev.h>
 #include <rte_dpaa_bus.h>
 #include <rte_dpaa_logs.h>
-#include <rte_cycles_64.h>
+#include <rte_cycles.h>
+#include <rte_kvargs.h>
 
 #include <dpaa_ethdev.h>
+#include <dpaa_sec_event.h>
 #include "dpaa_eventdev.h"
 #include <dpaa_mempool.h>
 
  * Eventqueue = Channel Instance
  * 1 Eventdev can have N Eventqueue
  */
+RTE_LOG_REGISTER(dpaa_logtype_eventdev, pmd.event.dpaa, NOTICE);
+
+#define DISABLE_INTR_MODE "disable_intr"
 
 static int
 dpaa_event_dequeue_timeout_ticks(struct rte_eventdev *dev, uint64_t ns,
                                 uint64_t *timeout_ticks)
 {
-       uint64_t cycles_per_second;
-
-       EVENTDEV_DRV_FUNC_TRACE();
+       EVENTDEV_INIT_FUNC_TRACE();
 
        RTE_SET_USED(dev);
 
+       uint64_t cycles_per_second;
+
        cycles_per_second = rte_get_timer_hz();
-       *timeout_ticks = ns * (cycles_per_second / NS_PER_S);
+       *timeout_ticks = (ns * cycles_per_second) / NS_PER_S;
+
+       return 0;
+}
+
+static int
+dpaa_event_dequeue_timeout_ticks_intr(struct rte_eventdev *dev, uint64_t ns,
+                                uint64_t *timeout_ticks)
+{
+       RTE_SET_USED(dev);
 
+       *timeout_ticks = ns/1000;
        return 0;
 }
 
+static void
+dpaa_eventq_portal_add(u16 ch_id)
+{
+       uint32_t sdqcr;
+
+       sdqcr = QM_SDQCR_CHANNELS_POOL_CONV(ch_id);
+       qman_static_dequeue_add(sdqcr, NULL);
+}
+
+static uint16_t
+dpaa_event_enqueue_burst(void *port, const struct rte_event ev[],
+                        uint16_t nb_events)
+{
+       uint16_t i;
+       struct rte_mbuf *mbuf;
+
+       RTE_SET_USED(port);
+       /*Release all the contexts saved previously*/
+       for (i = 0; i < nb_events; i++) {
+               switch (ev[i].op) {
+               case RTE_EVENT_OP_RELEASE:
+                       qman_dca_index(ev[i].impl_opaque, 0);
+                       mbuf = DPAA_PER_LCORE_DQRR_MBUF(i);
+                       mbuf->seqn = DPAA_INVALID_MBUF_SEQN;
+                       DPAA_PER_LCORE_DQRR_HELD &= ~(1 << i);
+                       DPAA_PER_LCORE_DQRR_SIZE--;
+                       break;
+               default:
+                       break;
+               }
+       }
+
+       return nb_events;
+}
+
+static uint16_t
+dpaa_event_enqueue(void *port, const struct rte_event *ev)
+{
+       return dpaa_event_enqueue_burst(port, ev, 1);
+}
+
+static void drain_4_bytes(int fd, fd_set *fdset)
+{
+       if (FD_ISSET(fd, fdset)) {
+               /* drain 4 bytes */
+               uint32_t junk;
+               ssize_t sjunk = read(qman_thread_fd(), &junk, sizeof(junk));
+               if (sjunk != sizeof(junk))
+                       DPAA_EVENTDEV_ERR("UIO irq read error");
+       }
+}
+
+static inline int
+dpaa_event_dequeue_wait(uint64_t timeout_ticks)
+{
+       int fd_qman, nfds;
+       int ret;
+       fd_set readset;
+
+       /* Go into (and back out of) IRQ mode for each select,
+        * it simplifies exit-path considerations and other
+        * potential nastiness.
+        */
+       struct timeval tv = {
+               .tv_sec = timeout_ticks / 1000000,
+               .tv_usec = timeout_ticks % 1000000
+       };
+
+       fd_qman = qman_thread_fd();
+       nfds = fd_qman + 1;
+       FD_ZERO(&readset);
+       FD_SET(fd_qman, &readset);
+
+       qman_irqsource_add(QM_PIRQ_DQRI);
+
+       ret = select(nfds, &readset, NULL, NULL, &tv);
+       if (ret < 0)
+               return ret;
+       /* Calling irqsource_remove() prior to thread_irq()
+        * means thread_irq() will not process whatever caused
+        * the interrupts, however it does ensure that, once
+        * thread_irq() re-enables interrupts, they won't fire
+        * again immediately.
+        */
+       qman_irqsource_remove(~0);
+       drain_4_bytes(fd_qman, &readset);
+       qman_thread_irq();
+
+       return ret;
+}
+
+static uint16_t
+dpaa_event_dequeue_burst(void *port, struct rte_event ev[],
+                        uint16_t nb_events, uint64_t timeout_ticks)
+{
+       int ret;
+       u16 ch_id;
+       void *buffers[8];
+       u32 num_frames, i, irq = 0;
+       uint64_t cur_ticks = 0, wait_time_ticks = 0;
+       struct dpaa_port *portal = (struct dpaa_port *)port;
+       struct rte_mbuf *mbuf;
+
+       if (unlikely(!DPAA_PER_LCORE_PORTAL)) {
+               /* Affine current thread context to a qman portal */
+               ret = rte_dpaa_portal_init((void *)0);
+               if (ret) {
+                       DPAA_EVENTDEV_ERR("Unable to initialize portal");
+                       return ret;
+               }
+       }
+
+       if (unlikely(!portal->is_port_linked)) {
+               /*
+                * Affine event queue for current thread context
+                * to a qman portal.
+                */
+               for (i = 0; i < portal->num_linked_evq; i++) {
+                       ch_id = portal->evq_info[i].ch_id;
+                       dpaa_eventq_portal_add(ch_id);
+               }
+               portal->is_port_linked = true;
+       }
+
+       /* Check if there are atomic contexts to be released */
+       i = 0;
+       while (DPAA_PER_LCORE_DQRR_SIZE) {
+               if (DPAA_PER_LCORE_DQRR_HELD & (1 << i)) {
+                       qman_dca_index(i, 0);
+                       mbuf = DPAA_PER_LCORE_DQRR_MBUF(i);
+                       mbuf->seqn = DPAA_INVALID_MBUF_SEQN;
+                       DPAA_PER_LCORE_DQRR_HELD &= ~(1 << i);
+                       DPAA_PER_LCORE_DQRR_SIZE--;
+               }
+               i++;
+       }
+       DPAA_PER_LCORE_DQRR_HELD = 0;
+
+       if (timeout_ticks)
+               wait_time_ticks = timeout_ticks;
+       else
+               wait_time_ticks = portal->timeout_us;
+
+       wait_time_ticks += rte_get_timer_cycles();
+       do {
+               /* Lets dequeue the frames */
+               num_frames = qman_portal_dequeue(ev, nb_events, buffers);
+               if (irq)
+                       irq = 0;
+               if (num_frames)
+                       break;
+               cur_ticks = rte_get_timer_cycles();
+       } while (cur_ticks < wait_time_ticks);
+
+       return num_frames;
+}
+
+static uint16_t
+dpaa_event_dequeue(void *port, struct rte_event *ev, uint64_t timeout_ticks)
+{
+       return dpaa_event_dequeue_burst(port, ev, 1, timeout_ticks);
+}
+
+static uint16_t
+dpaa_event_dequeue_burst_intr(void *port, struct rte_event ev[],
+                             uint16_t nb_events, uint64_t timeout_ticks)
+{
+       int ret;
+       u16 ch_id;
+       void *buffers[8];
+       u32 num_frames, i, irq = 0;
+       uint64_t cur_ticks = 0, wait_time_ticks = 0;
+       struct dpaa_port *portal = (struct dpaa_port *)port;
+       struct rte_mbuf *mbuf;
+
+       if (unlikely(!DPAA_PER_LCORE_PORTAL)) {
+               /* Affine current thread context to a qman portal */
+               ret = rte_dpaa_portal_init((void *)0);
+               if (ret) {
+                       DPAA_EVENTDEV_ERR("Unable to initialize portal");
+                       return ret;
+               }
+       }
+
+       if (unlikely(!portal->is_port_linked)) {
+               /*
+                * Affine event queue for current thread context
+                * to a qman portal.
+                */
+               for (i = 0; i < portal->num_linked_evq; i++) {
+                       ch_id = portal->evq_info[i].ch_id;
+                       dpaa_eventq_portal_add(ch_id);
+               }
+               portal->is_port_linked = true;
+       }
+
+       /* Check if there are atomic contexts to be released */
+       i = 0;
+       while (DPAA_PER_LCORE_DQRR_SIZE) {
+               if (DPAA_PER_LCORE_DQRR_HELD & (1 << i)) {
+                       qman_dca_index(i, 0);
+                       mbuf = DPAA_PER_LCORE_DQRR_MBUF(i);
+                       mbuf->seqn = DPAA_INVALID_MBUF_SEQN;
+                       DPAA_PER_LCORE_DQRR_HELD &= ~(1 << i);
+                       DPAA_PER_LCORE_DQRR_SIZE--;
+               }
+               i++;
+       }
+       DPAA_PER_LCORE_DQRR_HELD = 0;
+
+       if (timeout_ticks)
+               wait_time_ticks = timeout_ticks;
+       else
+               wait_time_ticks = portal->timeout_us;
+
+       do {
+               /* Lets dequeue the frames */
+               num_frames = qman_portal_dequeue(ev, nb_events, buffers);
+               if (irq)
+                       irq = 0;
+               if (num_frames)
+                       break;
+               if (wait_time_ticks) { /* wait for time */
+                       if (dpaa_event_dequeue_wait(wait_time_ticks) > 0) {
+                               irq = 1;
+                               continue;
+                       }
+                       break; /* no event after waiting */
+               }
+               cur_ticks = rte_get_timer_cycles();
+       } while (cur_ticks < wait_time_ticks);
+
+       return num_frames;
+}
+
+static uint16_t
+dpaa_event_dequeue_intr(void *port,
+                       struct rte_event *ev,
+                       uint64_t timeout_ticks)
+{
+       return dpaa_event_dequeue_burst_intr(port, ev, 1, timeout_ticks);
+}
+
 static void
 dpaa_event_dev_info_get(struct rte_eventdev *dev,
                        struct rte_event_dev_info *dev_info)
 {
-       EVENTDEV_DRV_FUNC_TRACE();
+       EVENTDEV_INIT_FUNC_TRACE();
 
        RTE_SET_USED(dev);
-       dev_info->driver_name = "event_dpaa";
+       dev_info->driver_name = "event_dpaa1";
        dev_info->min_dequeue_timeout_ns =
                DPAA_EVENT_MIN_DEQUEUE_TIMEOUT;
        dev_info->max_dequeue_timeout_ns =
                DPAA_EVENT_MAX_DEQUEUE_TIMEOUT;
        dev_info->dequeue_timeout_ns =
-               DPAA_EVENT_MIN_DEQUEUE_TIMEOUT;
+               DPAA_EVENT_PORT_DEQUEUE_TIMEOUT_NS;
        dev_info->max_event_queues =
                DPAA_EVENT_MAX_QUEUES;
        dev_info->max_event_queue_flows =
@@ -108,8 +368,7 @@ dpaa_event_dev_configure(const struct rte_eventdev *dev)
        int ret, i;
        uint32_t *ch_id;
 
-       EVENTDEV_DRV_FUNC_TRACE();
-
+       EVENTDEV_INIT_FUNC_TRACE();
        priv->dequeue_timeout_ns = conf->dequeue_timeout_ns;
        priv->nb_events_limit = conf->nb_events_limit;
        priv->nb_event_queues = conf->nb_event_queues;
@@ -119,26 +378,18 @@ dpaa_event_dev_configure(const struct rte_eventdev *dev)
        priv->nb_event_port_enqueue_depth = conf->nb_event_port_enqueue_depth;
        priv->event_dev_cfg = conf->event_dev_cfg;
 
-       /* Check dequeue timeout method is per dequeue or global */
-       if (priv->event_dev_cfg & RTE_EVENT_DEV_CFG_PER_DEQUEUE_TIMEOUT) {
-               /*
-                * Use timeout value as given in dequeue operation.
-                * So invalidating this timetout value.
-                */
-               priv->dequeue_timeout_ns = 0;
-       }
-
        ch_id = rte_malloc("dpaa-channels",
                          sizeof(uint32_t) * priv->nb_event_queues,
                          RTE_CACHE_LINE_SIZE);
        if (ch_id == NULL) {
-               EVENTDEV_DRV_ERR("Fail to allocate memory for dpaa channels\n");
+               DPAA_EVENTDEV_ERR("Fail to allocate memory for dpaa channels\n");
                return -ENOMEM;
        }
        /* Create requested event queues within the given event device */
        ret = qman_alloc_pool_range(ch_id, priv->nb_event_queues, 1, 0);
        if (ret < 0) {
-               EVENTDEV_DRV_ERR("Failed to create internal channel\n");
+               DPAA_EVENTDEV_ERR("qman_alloc_pool_range %u, err =%d\n",
+                                priv->nb_event_queues, ret);
                rte_free(ch_id);
                return ret;
        }
@@ -148,30 +399,41 @@ dpaa_event_dev_configure(const struct rte_eventdev *dev)
        /* Lets prepare event ports */
        memset(&priv->ports[0], 0,
              sizeof(struct dpaa_port) * priv->nb_event_ports);
+
+       /* Check dequeue timeout method is per dequeue or global */
        if (priv->event_dev_cfg & RTE_EVENT_DEV_CFG_PER_DEQUEUE_TIMEOUT) {
-               for (i = 0; i < priv->nb_event_ports; i++) {
-                       priv->ports[i].timeout =
-                               DPAA_EVENT_PORT_DEQUEUE_TIMEOUT_INVALID;
-               }
-       } else if (priv->dequeue_timeout_ns == 0) {
-               for (i = 0; i < priv->nb_event_ports; i++) {
-                       dpaa_event_dequeue_timeout_ticks(NULL,
-                               DPAA_EVENT_PORT_DEQUEUE_TIMEOUT_NS,
-                               &priv->ports[i].timeout);
-               }
+               /*
+                * Use timeout value as given in dequeue operation.
+                * So invalidating this timeout value.
+                */
+               priv->dequeue_timeout_ns = 0;
+
+       } else if (conf->dequeue_timeout_ns == 0) {
+               priv->dequeue_timeout_ns = DPAA_EVENT_PORT_DEQUEUE_TIMEOUT_NS;
        } else {
-               for (i = 0; i < priv->nb_event_ports; i++) {
-                       dpaa_event_dequeue_timeout_ticks(NULL,
-                               priv->dequeue_timeout_ns,
-                               &priv->ports[i].timeout);
+               priv->dequeue_timeout_ns = conf->dequeue_timeout_ns;
+       }
+
+       for (i = 0; i < priv->nb_event_ports; i++) {
+               if (priv->intr_mode) {
+                       priv->ports[i].timeout_us =
+                               priv->dequeue_timeout_ns/1000;
+               } else {
+                       uint64_t cycles_per_second;
+
+                       cycles_per_second = rte_get_timer_hz();
+                       priv->ports[i].timeout_us =
+                               (priv->dequeue_timeout_ns * cycles_per_second)
+                                       / NS_PER_S;
                }
        }
+
        /*
         * TODO: Currently portals are affined with threads. Maximum threads
         * can be created equals to number of lcore.
         */
        rte_free(ch_id);
-       EVENTDEV_DRV_LOG("Configured eventdev devid=%d", dev->data->dev_id);
+       DPAA_EVENTDEV_INFO("Configured eventdev devid=%d", dev->data->dev_id);
 
        return 0;
 }
@@ -179,7 +441,7 @@ dpaa_event_dev_configure(const struct rte_eventdev *dev)
 static int
 dpaa_event_dev_start(struct rte_eventdev *dev)
 {
-       EVENTDEV_DRV_FUNC_TRACE();
+       EVENTDEV_INIT_FUNC_TRACE();
        RTE_SET_USED(dev);
 
        return 0;
@@ -188,32 +450,549 @@ dpaa_event_dev_start(struct rte_eventdev *dev)
 static void
 dpaa_event_dev_stop(struct rte_eventdev *dev)
 {
-       EVENTDEV_DRV_FUNC_TRACE();
+       EVENTDEV_INIT_FUNC_TRACE();
        RTE_SET_USED(dev);
 }
 
 static int
 dpaa_event_dev_close(struct rte_eventdev *dev)
 {
-       EVENTDEV_DRV_FUNC_TRACE();
+       EVENTDEV_INIT_FUNC_TRACE();
+       RTE_SET_USED(dev);
+
+       return 0;
+}
+
+static void
+dpaa_event_queue_def_conf(struct rte_eventdev *dev, uint8_t queue_id,
+                         struct rte_event_queue_conf *queue_conf)
+{
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       RTE_SET_USED(dev);
+       RTE_SET_USED(queue_id);
+
+       memset(queue_conf, 0, sizeof(struct rte_event_queue_conf));
+       queue_conf->nb_atomic_flows = DPAA_EVENT_QUEUE_ATOMIC_FLOWS;
+       queue_conf->schedule_type = RTE_SCHED_TYPE_PARALLEL;
+       queue_conf->priority = RTE_EVENT_DEV_PRIORITY_HIGHEST;
+}
+
+static int
+dpaa_event_queue_setup(struct rte_eventdev *dev, uint8_t queue_id,
+                      const struct rte_event_queue_conf *queue_conf)
+{
+       struct dpaa_eventdev *priv = dev->data->dev_private;
+       struct dpaa_eventq *evq_info = &priv->evq_info[queue_id];
+
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       switch (queue_conf->schedule_type) {
+       case RTE_SCHED_TYPE_PARALLEL:
+       case RTE_SCHED_TYPE_ATOMIC:
+               break;
+       case RTE_SCHED_TYPE_ORDERED:
+               DPAA_EVENTDEV_ERR("Schedule type is not supported.");
+               return -1;
+       }
+       evq_info->event_queue_cfg = queue_conf->event_queue_cfg;
+       evq_info->event_queue_id = queue_id;
+
+       return 0;
+}
+
+static void
+dpaa_event_queue_release(struct rte_eventdev *dev, uint8_t queue_id)
+{
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       RTE_SET_USED(dev);
+       RTE_SET_USED(queue_id);
+}
+
+static void
+dpaa_event_port_default_conf_get(struct rte_eventdev *dev, uint8_t port_id,
+                                struct rte_event_port_conf *port_conf)
+{
+       EVENTDEV_INIT_FUNC_TRACE();
+
        RTE_SET_USED(dev);
+       RTE_SET_USED(port_id);
+
+       port_conf->new_event_threshold = DPAA_EVENT_MAX_NUM_EVENTS;
+       port_conf->dequeue_depth = DPAA_EVENT_MAX_PORT_DEQUEUE_DEPTH;
+       port_conf->enqueue_depth = DPAA_EVENT_MAX_PORT_ENQUEUE_DEPTH;
+}
+
+static int
+dpaa_event_port_setup(struct rte_eventdev *dev, uint8_t port_id,
+                     const struct rte_event_port_conf *port_conf)
+{
+       struct dpaa_eventdev *eventdev = dev->data->dev_private;
+
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       RTE_SET_USED(port_conf);
+       dev->data->ports[port_id] = &eventdev->ports[port_id];
 
        return 0;
 }
 
+static void
+dpaa_event_port_release(void *port)
+{
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       RTE_SET_USED(port);
+}
+
+static int
+dpaa_event_port_link(struct rte_eventdev *dev, void *port,
+                    const uint8_t queues[], const uint8_t priorities[],
+                    uint16_t nb_links)
+{
+       struct dpaa_eventdev *priv = dev->data->dev_private;
+       struct dpaa_port *event_port = (struct dpaa_port *)port;
+       struct dpaa_eventq *event_queue;
+       uint8_t eventq_id;
+       int i;
+
+       RTE_SET_USED(dev);
+       RTE_SET_USED(priorities);
+
+       /* First check that input configuration are valid */
+       for (i = 0; i < nb_links; i++) {
+               eventq_id = queues[i];
+               event_queue = &priv->evq_info[eventq_id];
+               if ((event_queue->event_queue_cfg
+                       & RTE_EVENT_QUEUE_CFG_SINGLE_LINK)
+                       && (event_queue->event_port)) {
+                       return -EINVAL;
+               }
+       }
+
+       for (i = 0; i < nb_links; i++) {
+               eventq_id = queues[i];
+               event_queue = &priv->evq_info[eventq_id];
+               event_port->evq_info[i].event_queue_id = eventq_id;
+               event_port->evq_info[i].ch_id = event_queue->ch_id;
+               event_queue->event_port = port;
+       }
+
+       event_port->num_linked_evq = event_port->num_linked_evq + i;
+
+       return (int)i;
+}
+
+static int
+dpaa_event_port_unlink(struct rte_eventdev *dev, void *port,
+                      uint8_t queues[], uint16_t nb_links)
+{
+       int i;
+       uint8_t eventq_id;
+       struct dpaa_eventq *event_queue;
+       struct dpaa_eventdev *priv = dev->data->dev_private;
+       struct dpaa_port *event_port = (struct dpaa_port *)port;
+
+       if (!event_port->num_linked_evq)
+               return nb_links;
+
+       for (i = 0; i < nb_links; i++) {
+               eventq_id = queues[i];
+               event_port->evq_info[eventq_id].event_queue_id = -1;
+               event_port->evq_info[eventq_id].ch_id = 0;
+               event_queue = &priv->evq_info[eventq_id];
+               event_queue->event_port = NULL;
+       }
+
+       if (event_port->num_linked_evq)
+               event_port->num_linked_evq = event_port->num_linked_evq - i;
+
+       return (int)i;
+}
+
+static int
+dpaa_event_eth_rx_adapter_caps_get(const struct rte_eventdev *dev,
+                                  const struct rte_eth_dev *eth_dev,
+                                  uint32_t *caps)
+{
+       const char *ethdev_driver = eth_dev->device->driver->name;
 
+       EVENTDEV_INIT_FUNC_TRACE();
 
-static const struct rte_eventdev_ops dpaa_eventdev_ops = {
+       RTE_SET_USED(dev);
+
+       if (!strcmp(ethdev_driver, "net_dpaa"))
+               *caps = RTE_EVENT_ETH_RX_ADAPTER_DPAA_CAP;
+       else
+               *caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
+
+       return 0;
+}
+
+static int
+dpaa_event_eth_rx_adapter_queue_add(
+               const struct rte_eventdev *dev,
+               const struct rte_eth_dev *eth_dev,
+               int32_t rx_queue_id,
+               const struct rte_event_eth_rx_adapter_queue_conf *queue_conf)
+{
+       struct dpaa_eventdev *eventdev = dev->data->dev_private;
+       uint8_t ev_qid = queue_conf->ev.queue_id;
+       u16 ch_id = eventdev->evq_info[ev_qid].ch_id;
+       struct dpaa_if *dpaa_intf = eth_dev->data->dev_private;
+       int ret, i;
+
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       if (rx_queue_id == -1) {
+               for (i = 0; i < dpaa_intf->nb_rx_queues; i++) {
+                       ret = dpaa_eth_eventq_attach(eth_dev, i, ch_id,
+                                                    queue_conf);
+                       if (ret) {
+                               DPAA_EVENTDEV_ERR(
+                                       "Event Queue attach failed:%d\n", ret);
+                               goto detach_configured_queues;
+                       }
+               }
+               return 0;
+       }
+
+       ret = dpaa_eth_eventq_attach(eth_dev, rx_queue_id, ch_id, queue_conf);
+       if (ret)
+               DPAA_EVENTDEV_ERR("dpaa_eth_eventq_attach failed:%d\n", ret);
+       return ret;
+
+detach_configured_queues:
+
+       for (i = (i - 1); i >= 0 ; i--)
+               dpaa_eth_eventq_detach(eth_dev, i);
+
+       return ret;
+}
+
+static int
+dpaa_event_eth_rx_adapter_queue_del(const struct rte_eventdev *dev,
+                                   const struct rte_eth_dev *eth_dev,
+                                   int32_t rx_queue_id)
+{
+       int ret, i;
+       struct dpaa_if *dpaa_intf = eth_dev->data->dev_private;
+
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       RTE_SET_USED(dev);
+       if (rx_queue_id == -1) {
+               for (i = 0; i < dpaa_intf->nb_rx_queues; i++) {
+                       ret = dpaa_eth_eventq_detach(eth_dev, i);
+                       if (ret)
+                               DPAA_EVENTDEV_ERR(
+                                       "Event Queue detach failed:%d\n", ret);
+               }
+
+               return 0;
+       }
+
+       ret = dpaa_eth_eventq_detach(eth_dev, rx_queue_id);
+       if (ret)
+               DPAA_EVENTDEV_ERR("dpaa_eth_eventq_detach failed:%d\n", ret);
+       return ret;
+}
+
+static int
+dpaa_event_eth_rx_adapter_start(const struct rte_eventdev *dev,
+                               const struct rte_eth_dev *eth_dev)
+{
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       RTE_SET_USED(dev);
+       RTE_SET_USED(eth_dev);
+
+       return 0;
+}
+
+static int
+dpaa_event_eth_rx_adapter_stop(const struct rte_eventdev *dev,
+                              const struct rte_eth_dev *eth_dev)
+{
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       RTE_SET_USED(dev);
+       RTE_SET_USED(eth_dev);
+
+       return 0;
+}
+
+static int
+dpaa_eventdev_crypto_caps_get(const struct rte_eventdev *dev,
+                           const struct rte_cryptodev *cdev,
+                           uint32_t *caps)
+{
+       const char *name = cdev->data->name;
+
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       RTE_SET_USED(dev);
+
+       if (!strncmp(name, "dpaa_sec-", 9))
+               *caps = RTE_EVENT_CRYPTO_ADAPTER_DPAA_CAP;
+       else
+               return -1;
+
+       return 0;
+}
+
+static int
+dpaa_eventdev_crypto_queue_add_all(const struct rte_eventdev *dev,
+               const struct rte_cryptodev *cryptodev,
+               const struct rte_event *ev)
+{
+       struct dpaa_eventdev *priv = dev->data->dev_private;
+       uint8_t ev_qid = ev->queue_id;
+       u16 ch_id = priv->evq_info[ev_qid].ch_id;
+       int i, ret;
+
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       for (i = 0; i < cryptodev->data->nb_queue_pairs; i++) {
+               ret = dpaa_sec_eventq_attach(cryptodev, i,
+                               ch_id, ev);
+               if (ret) {
+                       DPAA_EVENTDEV_ERR("dpaa_sec_eventq_attach failed: ret %d\n",
+                                   ret);
+                       goto fail;
+               }
+       }
+       return 0;
+fail:
+       for (i = (i - 1); i >= 0 ; i--)
+               dpaa_sec_eventq_detach(cryptodev, i);
+
+       return ret;
+}
+
+static int
+dpaa_eventdev_crypto_queue_add(const struct rte_eventdev *dev,
+               const struct rte_cryptodev *cryptodev,
+               int32_t rx_queue_id,
+               const struct rte_event *ev)
+{
+       struct dpaa_eventdev *priv = dev->data->dev_private;
+       uint8_t ev_qid = ev->queue_id;
+       u16 ch_id = priv->evq_info[ev_qid].ch_id;
+       int ret;
+
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       if (rx_queue_id == -1)
+               return dpaa_eventdev_crypto_queue_add_all(dev,
+                               cryptodev, ev);
+
+       ret = dpaa_sec_eventq_attach(cryptodev, rx_queue_id,
+                       ch_id, ev);
+       if (ret) {
+               DPAA_EVENTDEV_ERR(
+                       "dpaa_sec_eventq_attach failed: ret: %d\n", ret);
+               return ret;
+       }
+       return 0;
+}
+
+static int
+dpaa_eventdev_crypto_queue_del_all(const struct rte_eventdev *dev,
+                            const struct rte_cryptodev *cdev)
+{
+       int i, ret;
+
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       RTE_SET_USED(dev);
+
+       for (i = 0; i < cdev->data->nb_queue_pairs; i++) {
+               ret = dpaa_sec_eventq_detach(cdev, i);
+               if (ret) {
+                       DPAA_EVENTDEV_ERR(
+                               "dpaa_sec_eventq_detach failed:ret %d\n", ret);
+                       return ret;
+               }
+       }
+
+       return 0;
+}
+
+static int
+dpaa_eventdev_crypto_queue_del(const struct rte_eventdev *dev,
+                            const struct rte_cryptodev *cryptodev,
+                            int32_t rx_queue_id)
+{
+       int ret;
+
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       if (rx_queue_id == -1)
+               return dpaa_eventdev_crypto_queue_del_all(dev, cryptodev);
+
+       ret = dpaa_sec_eventq_detach(cryptodev, rx_queue_id);
+       if (ret) {
+               DPAA_EVENTDEV_ERR(
+                       "dpaa_sec_eventq_detach failed: ret: %d\n", ret);
+               return ret;
+       }
+
+       return 0;
+}
+
+static int
+dpaa_eventdev_crypto_start(const struct rte_eventdev *dev,
+                          const struct rte_cryptodev *cryptodev)
+{
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       RTE_SET_USED(dev);
+       RTE_SET_USED(cryptodev);
+
+       return 0;
+}
+
+static int
+dpaa_eventdev_crypto_stop(const struct rte_eventdev *dev,
+                         const struct rte_cryptodev *cryptodev)
+{
+       EVENTDEV_INIT_FUNC_TRACE();
+
+       RTE_SET_USED(dev);
+       RTE_SET_USED(cryptodev);
+
+       return 0;
+}
+
+static int
+dpaa_eventdev_tx_adapter_create(uint8_t id,
+                                const struct rte_eventdev *dev)
+{
+       RTE_SET_USED(id);
+       RTE_SET_USED(dev);
+
+       /* Nothing to do. Simply return. */
+       return 0;
+}
+
+static int
+dpaa_eventdev_tx_adapter_caps(const struct rte_eventdev *dev,
+                              const struct rte_eth_dev *eth_dev,
+                              uint32_t *caps)
+{
+       RTE_SET_USED(dev);
+       RTE_SET_USED(eth_dev);
+
+       *caps = RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT;
+       return 0;
+}
+
+static uint16_t
+dpaa_eventdev_txa_enqueue_same_dest(void *port,
+                                    struct rte_event ev[],
+                                    uint16_t nb_events)
+{
+       struct rte_mbuf *m[DPAA_EVENT_MAX_PORT_ENQUEUE_DEPTH], *m0;
+       uint8_t qid, i;
+
+       RTE_SET_USED(port);
+
+       m0 = (struct rte_mbuf *)ev[0].mbuf;
+       qid = rte_event_eth_tx_adapter_txq_get(m0);
+
+       for (i = 0; i < nb_events; i++)
+               m[i] = (struct rte_mbuf *)ev[i].mbuf;
+
+       return rte_eth_tx_burst(m0->port, qid, m, nb_events);
+}
+
+static uint16_t
+dpaa_eventdev_txa_enqueue(void *port,
+                          struct rte_event ev[],
+                          uint16_t nb_events)
+{
+       struct rte_mbuf *m = (struct rte_mbuf *)ev[0].mbuf;
+       uint8_t qid, i;
+
+       RTE_SET_USED(port);
+
+       for (i = 0; i < nb_events; i++) {
+               qid = rte_event_eth_tx_adapter_txq_get(m);
+               rte_eth_tx_burst(m->port, qid, &m, 1);
+       }
+
+       return nb_events;
+}
+
+static struct rte_eventdev_ops dpaa_eventdev_ops = {
        .dev_infos_get    = dpaa_event_dev_info_get,
        .dev_configure    = dpaa_event_dev_configure,
        .dev_start        = dpaa_event_dev_start,
        .dev_stop         = dpaa_event_dev_stop,
        .dev_close        = dpaa_event_dev_close,
+       .queue_def_conf   = dpaa_event_queue_def_conf,
+       .queue_setup      = dpaa_event_queue_setup,
+       .queue_release    = dpaa_event_queue_release,
+       .port_def_conf    = dpaa_event_port_default_conf_get,
+       .port_setup       = dpaa_event_port_setup,
+       .port_release       = dpaa_event_port_release,
+       .port_link        = dpaa_event_port_link,
+       .port_unlink      = dpaa_event_port_unlink,
        .timeout_ticks    = dpaa_event_dequeue_timeout_ticks,
+       .eth_rx_adapter_caps_get        = dpaa_event_eth_rx_adapter_caps_get,
+       .eth_rx_adapter_queue_add       = dpaa_event_eth_rx_adapter_queue_add,
+       .eth_rx_adapter_queue_del       = dpaa_event_eth_rx_adapter_queue_del,
+       .eth_rx_adapter_start           = dpaa_event_eth_rx_adapter_start,
+       .eth_rx_adapter_stop            = dpaa_event_eth_rx_adapter_stop,
+       .eth_tx_adapter_caps_get        = dpaa_eventdev_tx_adapter_caps,
+       .eth_tx_adapter_create          = dpaa_eventdev_tx_adapter_create,
+       .crypto_adapter_caps_get        = dpaa_eventdev_crypto_caps_get,
+       .crypto_adapter_queue_pair_add  = dpaa_eventdev_crypto_queue_add,
+       .crypto_adapter_queue_pair_del  = dpaa_eventdev_crypto_queue_del,
+       .crypto_adapter_start           = dpaa_eventdev_crypto_start,
+       .crypto_adapter_stop            = dpaa_eventdev_crypto_stop,
 };
 
+static int flag_check_handler(__rte_unused const char *key,
+               const char *value, __rte_unused void *opaque)
+{
+       if (strcmp(value, "1"))
+               return -1;
+
+       return 0;
+}
+
 static int
-dpaa_event_dev_create(const char *name)
+dpaa_event_check_flags(const char *params)
+{
+       struct rte_kvargs *kvlist;
+
+       if (params == NULL || params[0] == '\0')
+               return 0;
+
+       kvlist = rte_kvargs_parse(params, NULL);
+       if (kvlist == NULL)
+               return 0;
+
+       if (!rte_kvargs_count(kvlist, DISABLE_INTR_MODE)) {
+               rte_kvargs_free(kvlist);
+               return 0;
+       }
+       /* INTR MODE is disabled when there's key-value pair: disable_intr = 1*/
+       if (rte_kvargs_process(kvlist, DISABLE_INTR_MODE,
+                               flag_check_handler, NULL) < 0) {
+               rte_kvargs_free(kvlist);
+               return 0;
+       }
+       rte_kvargs_free(kvlist);
+
+       return 1;
+}
+
+static int
+dpaa_event_dev_create(const char *name, const char *params)
 {
        struct rte_eventdev *eventdev;
        struct dpaa_eventdev *priv;
@@ -222,17 +1001,34 @@ dpaa_event_dev_create(const char *name)
                                           sizeof(struct dpaa_eventdev),
                                           rte_socket_id());
        if (eventdev == NULL) {
-               EVENTDEV_DRV_ERR("Failed to create eventdev vdev %s", name);
+               DPAA_EVENTDEV_ERR("Failed to create eventdev vdev %s", name);
                goto fail;
        }
+       priv = eventdev->data->dev_private;
 
        eventdev->dev_ops       = &dpaa_eventdev_ops;
+       eventdev->enqueue       = dpaa_event_enqueue;
+       eventdev->enqueue_burst = dpaa_event_enqueue_burst;
+
+       if (dpaa_event_check_flags(params)) {
+               eventdev->dequeue       = dpaa_event_dequeue;
+               eventdev->dequeue_burst = dpaa_event_dequeue_burst;
+       } else {
+               priv->intr_mode = 1;
+               eventdev->dev_ops->timeout_ticks =
+                               dpaa_event_dequeue_timeout_ticks_intr;
+               eventdev->dequeue       = dpaa_event_dequeue_intr;
+               eventdev->dequeue_burst = dpaa_event_dequeue_burst_intr;
+       }
+       eventdev->txa_enqueue = dpaa_eventdev_txa_enqueue;
+       eventdev->txa_enqueue_same_dest = dpaa_eventdev_txa_enqueue_same_dest;
+
+       RTE_LOG(INFO, PMD, "%s eventdev added", name);
 
        /* For secondary processes, the primary has done all the work */
        if (rte_eal_process_type() != RTE_PROC_PRIMARY)
                return 0;
 
-       priv = eventdev->data->dev_private;
        priv->max_event_queues = DPAA_EVENT_MAX_QUEUES;
 
        return 0;
@@ -244,11 +1040,14 @@ static int
 dpaa_event_dev_probe(struct rte_vdev_device *vdev)
 {
        const char *name;
+       const char *params;
 
        name = rte_vdev_device_name(vdev);
-       EVENTDEV_DRV_LOG("Initializing %s", name);
+       DPAA_EVENTDEV_INFO("Initializing %s", name);
+
+       params = rte_vdev_device_args(vdev);
 
-       return dpaa_event_dev_create(name);
+       return dpaa_event_dev_create(name, params);
 }
 
 static int
@@ -257,7 +1056,7 @@ dpaa_event_dev_remove(struct rte_vdev_device *vdev)
        const char *name;
 
        name = rte_vdev_device_name(vdev);
-       EVENTDEV_DRV_LOG("Closing %s", name);
+       DPAA_EVENTDEV_INFO("Closing %s", name);
 
        return rte_event_pmd_vdev_uninit(name);
 }
@@ -268,3 +1067,5 @@ static struct rte_vdev_driver vdev_eventdev_dpaa_pmd = {
 };
 
 RTE_PMD_REGISTER_VDEV(EVENTDEV_NAME_DPAA_PMD, vdev_eventdev_dpaa_pmd);
+RTE_PMD_REGISTER_PARAM_STRING(EVENTDEV_NAME_DPAA_PMD,
+               DISABLE_INTR_MODE "=<int>");