1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright(c) 2017 Intel Corporation.
4 OPDL Eventdev Poll Mode Driver
5 ==================================
7 The OPDL (Ordered Packet Distribution Library) eventdev is a specific\
8 implementation of the eventdev API. It is particularly suited to packet\
9 processing workloads that have high throughput and low latency requirements.\
10 All packets follow the same path through the device. The order in which\
11 packets follow is determined by the order in which queues are set up.\
12 Events are left on the ring until they are transmitted. As a result packets\
13 do not go out of order
19 The OPDL eventdev implements a subset of features of the eventdev API;
23 * Ordered (Parallel is supported as parallel is a subset of Ordered)
27 * Load balanced (for Atomic, Ordered, Parallel queues)
28 * Single Link (for single-link queues)
31 Configuration and Options
32 -------------------------
34 The software eventdev is a vdev device, and as such can be created from the
35 application code, or from the EAL command line:
37 * Call ``rte_vdev_init("event_opdl0")`` from the application
39 * Use ``--vdev="event_opdl0"`` in the EAL options, which will call
40 rte_vdev_init() internally
44 .. code-block:: console
46 ./your_eventdev_application --vdev="event_opdl0"
52 It is possible to create a Single Port Queue ``RTE_EVENT_QUEUE_CFG_SINGLE_LINK``.
53 Packets dequeued from this queue do not need to be re-enqueued (as is the
54 case with an ordered queue). The purpose of this queue is to allow for
55 asynchronous handling of packets in the middle of a pipeline. Ordered
56 queues in the middle of a pipeline cannot delete packets.
62 As stated the order in which packets travel through queues is static in
63 nature. They go through the queues in the order the queues are setup at
64 initialisation ``rte_event_queue_setup()``. For example if an application
65 sets up 3 queues, Q0, Q1, Q2 and has 3 associated ports P0, P1, P2 and
66 P3 then packets must be
68 * Enqueued onto Q0 (typically through P0), then
70 * Dequeued from Q0 (typically through P1), then
72 * Enqueued onto Q1 (also through P1), then
74 * Dequeued from Q2 (typically through P2), then
76 * Enqueued onto Q3 (also through P2), then
78 * Dequeued from Q3 (typically through P3) and then transmitted on the relevant \
85 The opdl implementation has a number of limitations. These limitations are
86 due to the static nature of the underlying queues. It is because of this
87 that the implementation can achieve such high throughput and low latency
89 The following list is a comprehensive outline of the what is supported and
90 the limitations / restrictions imposed by the opdl PMD
92 - The order in which packets moved between queues is static and fixed \
93 (dynamic scheduling is not supported).
95 - NEW, RELEASE are not explicitly supported. RX (first enqueue) implicitly \
96 adds NEW event types, and TX (last dequeue) implicitly does RELEASE event types.
98 - All packets follow the same path through device queues.
100 - Flows within queues are NOT supported.
102 - Event priority is NOT supported.
104 - Once the device is stopped all inflight events are lost. Applications should \
105 clear all inflight events before stopping it.
107 - Each port can only be associated with one queue.
109 - Each queue can have multiple ports associated with it.
111 - Each worker core has to dequeue the maximum burst size for that port.
113 - For performance, the rte_event flow_id should not be updated once packet\
118 Validation & Statistics
119 ~~~~~~~~~~~~~~~~~~~~~~~
121 Validation can be turned on through a command line parameter
123 .. code-block:: console
125 --vdev="event_opdl0,do_validation=1,self_test=1"
127 If validation is turned on every packet (as opposed to just the first in
128 each burst), is validated to have come from the right queue. Statistics
129 are also produced in this mode. The statistics are available through the
130 eventdev xstats API. Statistics are per port as follows:
132 - claim_pkts_requested