event/octeontx: support event Rx adapter
authorPavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Tue, 24 Oct 2017 13:00:36 +0000 (18:30 +0530)
committerJerin Jacob <jerin.jacob@caviumnetworks.com>
Wed, 25 Oct 2017 12:03:43 +0000 (14:03 +0200)
Add Rx adapter queue add and delete API for both generic eth_devices as
well as HW backed eth_octeontx which supports direct event injection to
event device.
The HW injected event needs to be converted into mbuf, previously this
was done in eth_octeontx during rx_burst now it is moved to
event_octeontx as events from Rx adapter are dequeued directly from
event device.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Reviewed-by: Nikhil Rao <nikhil.rao@intel.com>
doc/guides/eventdevs/octeontx.rst
drivers/event/octeontx/ssovf_evdev.c

index b43d515..7e601a0 100644 (file)
@@ -149,3 +149,8 @@ Burst mode support
 Burst mode is not supported. Dequeue and Enqueue functions accepts only single
 event at a time.
 
+Rx adapter support
+~~~~~~~~~~~~~~~~~~
+
+When eth_octeontx is used as Rx adapter event schedule type
+``RTE_SCHED_TYPE_PARALLEL`` is not supported.
index d829b49..7bdc85d 100644 (file)
@@ -36,6 +36,8 @@
 #include <rte_debug.h>
 #include <rte_dev.h>
 #include <rte_eal.h>
+#include <rte_ethdev.h>
+#include <rte_event_eth_rx_adapter.h>
 #include <rte_lcore.h>
 #include <rte_log.h>
 #include <rte_malloc.h>
@@ -395,6 +397,123 @@ ssows_dump(struct ssows *ws, FILE *f)
        fprintf(f, "\tpwqp=0x%"PRIx64"\n", val);
 }
 
+static int
+ssovf_eth_rx_adapter_caps_get(const struct rte_eventdev *dev,
+               const struct rte_eth_dev *eth_dev, uint32_t *caps)
+{
+       int ret;
+       RTE_SET_USED(dev);
+
+       ret = strncmp(eth_dev->data->name, "eth_octeontx", 12);
+       if (ret)
+               *caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
+       else
+               *caps = RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT;
+
+       return 0;
+}
+
+static int
+ssovf_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)
+{
+       int ret = 0;
+       const struct octeontx_nic *nic = eth_dev->data->dev_private;
+       pki_mod_qos_t pki_qos;
+       RTE_SET_USED(dev);
+
+       ret = strncmp(eth_dev->data->name, "eth_octeontx", 12);
+       if (ret)
+               return -EINVAL;
+
+       if (rx_queue_id >= 0)
+               return -EINVAL;
+
+       if (queue_conf->ev.sched_type == RTE_SCHED_TYPE_PARALLEL)
+               return -ENOTSUP;
+
+       memset(&pki_qos, 0, sizeof(pki_mod_qos_t));
+
+       pki_qos.port_type = 0;
+       pki_qos.index = 0;
+       pki_qos.mmask.f_tag_type = 1;
+       pki_qos.mmask.f_port_add = 1;
+       pki_qos.mmask.f_grp_ok = 1;
+       pki_qos.mmask.f_grp_bad = 1;
+       pki_qos.mmask.f_grptag_ok = 1;
+       pki_qos.mmask.f_grptag_bad = 1;
+
+       pki_qos.tag_type = queue_conf->ev.sched_type;
+       pki_qos.qos_entry.port_add = 0;
+       pki_qos.qos_entry.ggrp_ok = queue_conf->ev.queue_id;
+       pki_qos.qos_entry.ggrp_bad = queue_conf->ev.queue_id;
+       pki_qos.qos_entry.grptag_bad = 0;
+       pki_qos.qos_entry.grptag_ok = 0;
+
+       ret = octeontx_pki_port_modify_qos(nic->port_id, &pki_qos);
+       if (ret < 0)
+               ssovf_log_err("failed to modify QOS, port=%d, q=%d",
+                               nic->port_id, queue_conf->ev.queue_id);
+
+       return ret;
+}
+
+static int
+ssovf_eth_rx_adapter_queue_del(const struct rte_eventdev *dev,
+               const struct rte_eth_dev *eth_dev, int32_t rx_queue_id)
+{
+       int ret = 0;
+       const struct octeontx_nic *nic = eth_dev->data->dev_private;
+       pki_del_qos_t pki_qos;
+       RTE_SET_USED(dev);
+       RTE_SET_USED(rx_queue_id);
+
+       ret = strncmp(eth_dev->data->name, "eth_octeontx", 12);
+       if (ret)
+               return -EINVAL;
+
+       pki_qos.port_type = 0;
+       pki_qos.index = 0;
+       memset(&pki_qos, 0, sizeof(pki_del_qos_t));
+       ret = octeontx_pki_port_delete_qos(nic->port_id, &pki_qos);
+       if (ret < 0)
+               ssovf_log_err("Failed to delete QOS port=%d, q=%d",
+                               nic->port_id, queue_conf->ev.queue_id);
+       return ret;
+}
+
+static int
+ssovf_eth_rx_adapter_start(const struct rte_eventdev *dev,
+                                       const struct rte_eth_dev *eth_dev)
+{
+       int ret;
+       const struct octeontx_nic *nic = eth_dev->data->dev_private;
+       RTE_SET_USED(dev);
+
+       ret = strncmp(eth_dev->data->name, "eth_octeontx", 12);
+       if (ret)
+               return 0;
+       octeontx_pki_port_start(nic->port_id);
+       return 0;
+}
+
+
+static int
+ssovf_eth_rx_adapter_stop(const struct rte_eventdev *dev,
+               const struct rte_eth_dev *eth_dev)
+{
+       int ret;
+       const struct octeontx_nic *nic = eth_dev->data->dev_private;
+       RTE_SET_USED(dev);
+
+       ret = strncmp(eth_dev->data->name, "eth_octeontx", 12);
+       if (ret)
+               return 0;
+       octeontx_pki_port_stop(nic->port_id);
+       return 0;
+}
+
 static void
 ssovf_dump(struct rte_eventdev *dev, FILE *f)
 {
@@ -488,6 +607,13 @@ static const struct rte_eventdev_ops ssovf_ops = {
        .port_link        = ssovf_port_link,
        .port_unlink      = ssovf_port_unlink,
        .timeout_ticks    = ssovf_timeout_ticks,
+
+       .eth_rx_adapter_caps_get  = ssovf_eth_rx_adapter_caps_get,
+       .eth_rx_adapter_queue_add = ssovf_eth_rx_adapter_queue_add,
+       .eth_rx_adapter_queue_del = ssovf_eth_rx_adapter_queue_del,
+       .eth_rx_adapter_start = ssovf_eth_rx_adapter_start,
+       .eth_rx_adapter_stop = ssovf_eth_rx_adapter_stop,
+
        .dump             = ssovf_dump,
        .dev_start        = ssovf_start,
        .dev_stop         = ssovf_stop,