On 8/17/2018 5:25 PM, Pavan Nikhilesh wrote:
Hi Nikhil,
Few comments inline.
Nit: Please use --in-reply-to while sending next versions.
(https://core.dpdk.org/contribute/)
Thanks,
Pavan
On Fri, Aug 17, 2018 at 09:50:52AM +0530, Nikhil Rao wrote:
This patch adds tests for the eth Tx adapter APIs. It also
tests the data path for the rte_service function based
implementation of the APIs.
Signed-off-by: Nikhil Rao <nikhil....@intel.com>
---
test/test/test_event_eth_tx_adapter.c | 676 ++++++++++++++++++++++++++++++++++
MAINTAINERS | 1 +
test/test/Makefile | 1 +
test/test/meson.build | 2 +
4 files changed, 680 insertions(+)
create mode 100644 test/test/test_event_eth_tx_adapter.c
diff --git a/test/test/test_event_eth_tx_adapter.c
b/test/test/test_event_eth_tx_adapter.c
new file mode 100644
index 0000000..2dc487b
--- /dev/null
+++ b/test/test/test_event_eth_tx_adapter.c
<snip>
+static int
+testsuite_setup(void)
+{
+ int err;
+ uint8_t count;
+ struct rte_event_dev_info dev_info;
+ uint8_t priority;
+ uint8_t queue_id;
+
+ count = rte_event_dev_count();
+ if (!count) {
+ printf("Failed to find a valid event device,"
+ " testing with event_sw0 device\n");
+ rte_vdev_init("event_sw0", NULL);
+ event_dev_delete = 1;
+ }
+
+ struct rte_event_dev_config config = {
+ .nb_event_queues = 1,
+ .nb_event_ports = 1,
+ };
+
+ struct rte_event_queue_conf wkr_q_conf = {
+ .schedule_type = RTE_SCHED_TYPE_ORDERED,
+ .priority = RTE_EVENT_DEV_PRIORITY_NORMAL,
+ .nb_atomic_flows = 1024,
+ .nb_atomic_order_sequences = 1024,
+ };
+
+ err = rte_event_dev_info_get(TEST_DEV_ID, &dev_info);
+ config.nb_event_queue_flows = dev_info.max_event_queue_flows;
+ config.nb_event_port_dequeue_depth =
+ dev_info.max_event_port_dequeue_depth;
+ config.nb_event_port_enqueue_depth =
+ dev_info.max_event_port_enqueue_depth;
+ config.nb_events_limit =
+ dev_info.max_num_events;
+
+ rte_log_set_level(0, RTE_LOG_DEBUG);
+ err = rte_event_dev_configure(TEST_DEV_ID, &config);
+ TEST_ASSERT(err == 0, "Event device initialization failed err %d\n",
+ err);
+ if (rte_event_queue_setup(TEST_DEV_ID, 0, &wkr_q_conf) < 0) {
+ printf("%d: error creating qid %d\n", __LINE__, 0);
+ return -1;
+ }
+ if (rte_event_port_setup(TEST_DEV_ID, 0, NULL) < 0) {
+ printf("Error setting up port %d\n", 0);
+ return -1;
+ }
+
+ priority = RTE_EVENT_DEV_PRIORITY_LOWEST;
queue_id is uninitilized here, garbage value might be passed.
Ok.
+ if (rte_event_port_link(TEST_DEV_ID, 0, &queue_id, &priority, 1) != 1) {
+ printf("Error linking port\n");
+ return -1;
+ }
+
+ err = init_ports();
+ TEST_ASSERT(err == 0, "Port initialization failed err %d\n", err);
+
+ err = rte_event_eth_tx_adapter_caps_get(TEST_DEV_ID, TEST_ETHDEV_ID,
+ &default_params.caps);
+ TEST_ASSERT(err == 0, "Failed to get adapter cap err %d\n",
+ err);
+
+ return err;
+}
+
<snip>
+static uint32_t eid, tid;
+
+static int
+tx_adapter_single(uint16_t port, uint16_t tx_queue_id,
+ struct rte_mbuf *m, uint8_t qid,
+ uint8_t sched_type)
+{
+ struct rte_event event;
+ struct rte_mbuf *r;
+ int ret;
+ unsigned int l;
+
+ event.queue_id = qid;
Set event_type to RTE_EVENT_TYPE_CPU so that underlying drivers don't mess up
the
packet.
Ok.
<snip>
+
+static int
+tx_adapter_service(void)
+{
+ struct rte_event_eth_tx_adapter_stats stats;
+ uint32_t i;
+ int err;
+ uint8_t ev_port, ev_qid;
+ struct rte_mbuf bufs[RING_SIZE];
+ struct rte_mbuf *pbufs[RING_SIZE];
+ struct rte_event_dev_info dev_info;
+ struct rte_event_dev_config dev_conf;
+ struct rte_event_queue_conf qconf;
+ uint32_t qcnt, pcnt;
+ uint16_t q;
+ int internal_port;
+
+ err = rte_event_eth_tx_adapter_caps_get(TEST_DEV_ID, TEST_ETHDEV_ID,
+ &default_params.caps);
+ TEST_ASSERT(err == 0, "Failed to get adapter cap err %d\n",
+ err);
+
+ internal_port = !!(default_params.caps &
+ RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT);
+
+ if (internal_port)
+ return TEST_SUCCESS;
+
+ err = rte_event_eth_tx_adapter_queue_add(TEST_INST_ID, TEST_ETHDEV_ID,
+ -1);
+ TEST_ASSERT(err == 0, "Expected 0 got %d", err);
+
+ err = rte_event_eth_tx_adapter_event_port_get(TEST_INST_ID,
+ &ev_port);
+ TEST_ASSERT_SUCCESS(err, "Failed to get event port %d", err);
+
+ err = rte_event_dev_attr_get(TEST_DEV_ID, RTE_EVENT_DEV_ATTR_PORT_COUNT,
+ &pcnt);
+ TEST_ASSERT_SUCCESS(err, "Port count get failed");
+
+ err = rte_event_dev_attr_get(TEST_DEV_ID,
+ RTE_EVENT_DEV_ATTR_QUEUE_COUNT, &qcnt);
+ TEST_ASSERT_SUCCESS(err, "Queue count get failed");
+
+ err = rte_event_dev_info_get(TEST_DEV_ID, &dev_info);
+ TEST_ASSERT_SUCCESS(err, "Dev info failed");
+
memset dev_config to avoid invalid values.
Ok.
+ dev_conf.nb_event_queue_flows = dev_info.max_event_queue_flows;
+ dev_conf.nb_event_port_dequeue_depth =
+ dev_info.max_event_port_dequeue_depth;
+ dev_conf.nb_event_port_enqueue_depth =
+ dev_info.max_event_port_enqueue_depth;
+ dev_conf.nb_events_limit =
+ dev_info.max_num_events;
+ dev_conf.nb_event_queues = qcnt + 1;
+ dev_conf.nb_event_ports = pcnt;
+ err = rte_event_dev_configure(TEST_DEV_ID, &dev_conf);
+ TEST_ASSERT(err == 0, "Event device initialization failed err %d\n",
+ err);
+
+ ev_qid = qcnt;
+ qconf.nb_atomic_flows = dev_info.max_event_queue_flows;
+ qconf.nb_atomic_order_sequences = 32;
+ qconf.schedule_type = RTE_SCHED_TYPE_ATOMIC;
+ qconf.priority = RTE_EVENT_DEV_PRIORITY_HIGHEST;
+ qconf.event_queue_cfg = RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
+ err = rte_event_queue_setup(TEST_DEV_ID, ev_qid, &qconf);
+ TEST_ASSERT_SUCCESS(err, "Failed to setup queue %u", ev_qid);
On reconfigure, setup all the ports and queues so that the newly configured
values are seen by them.
Is that required since dev_conf matches dev_info except for the number
of queues ?
+
+ err = rte_event_port_link(TEST_DEV_ID, ev_port, &ev_qid, NULL, 1);
+ TEST_ASSERT(err == 1, "Failed to link queue port %u",
+ ev_port);
+
+ err = rte_event_eth_tx_adapter_start(TEST_INST_ID);
+ TEST_ASSERT(err == 0, "Expected 0 got %d", err);
+
+ err = rte_event_dev_service_id_get(0, &eid);
+ TEST_ASSERT(err == 0, "Expected 0 got %d", err);
An event device might not be needing a service core, check the capabilities
before requesting service id above.
The internal_port capability is checked at the beginning of the
function, if its not set, I think its Ok to assume that we need a
service core ?
Thanks,
Nikhil
