event/dpaa: introduce PMD

Signed-off-by: Sunil Kumar Kori <sunil.kori@nxp.com>
Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

net/dpaa: support Rx queue configurations with eventdev

Given ethernet Rx queues can be attached with event queue in
parallel or atomic mode. Patch imlmplements Rx queue
configuration, attachment/detachment with given event queue and their
corresponding callbacks to handle events from respective queues.

Signed-off-by: Sunil Kumar Kori <sunil.kori@nxp.com>
Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

bus/dpaa: support dynamic logging

Signed-off-by: Sunil Kumar Kori <sunil.kori@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

bus/dpaa: support event dequeue and consumption

To receive events from given event port, corresponding
function needs to be added which receives events
from portal. Also added function to consume received
events based on entry index.

Signed-off-by: Sunil Kumar Kori <sunil.kori@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

config: enable compilation of DPAA eventdev PMD

Signed-off-by: Sunil Kumar Kori <sunil.kori@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

net/dpaa2: support atomic queues

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

bus/fslmc: add flag to configure DCA in QBMAN multi Tx

With the current QBMAN multi-tx API, we need to create separate
enqueue descriptors for each of the packet which is required to
be enqueued to the hardware, once we support Atomic Queues
(with DCA) in dpaa2 drivers. Creating enqueue descriptor for
each packet is costly and have significant performance impact.
This patch introduces a flag parameter in the QBMAN multi-tx API,
so that DCA configuration (and later on ORP/ODP for ordered queues)
can be passed using flags and be updated in the EQCR using this flag.

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

event/dpaa2: have separate structure to hold dqrr entries

This patch provides cleaner approach to store the DQRR entries,
which are yet to be consumed in case of atomic queues.

Also, this patch changes the storage of the DQRR entry index
into the mbuf->seqn instead of ev->opaque

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

event/dpaa2: use dqrr index to consume the DQRR entry

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

bus/fslmc: introduce API to consume dqrr using index

A new API qbman_swp_dqrr_idx_consume is defined which takes
input as DQRR index to consume corresponding DQRR entry.

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

event/dpaa2: replace static with dynamic logging

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

doc: add eventdev apps changes in release notes

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: John McNamara <john.mcnamara@intel.com>

doc: add pipeline atq test in testeventdev guide

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: John McNamara <john.mcnamara@intel.com>

doc: add pipeline queue test in testeventdev guide

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: John McNamara <john.mcnamara@intel.com>

app/eventdev: add pipeline atq worker functions

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

app/eventdev: add pipeline atq test

This is a pipeline test case that aims at testing the following with
``all types queue`` eventdev scheme.
1. Measure the end-to-end performance of an event dev with a ethernet dev.
2. Maintain packet ordering from Rx to Tx.

The atq queue test functions as same as ``pipeline_queue`` test.
The difference is, It uses, ``all type queue scheme`` instead of separate
queues for each stage and thus reduces the number of queues required to
realize the use case.

Note: The --prod_type_ethdev is mandatory for running the application.

Example command to run pipeline atq test:
sudo build/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
--test=pipeline_atq --wlcore=1 --prod_type_ethdev --stlist=ao

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

app/eventdev: add pipeline queue worker functions

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

app/eventdev: add pipeline queue test

This is a pipeline queue test case that aims at testing the following:
1. Measure the end-to-end performance of an event dev with a ethernet dev.
2. Maintain packet ordering from Rx to Tx.

The pipeline queue test configures the eventdev with Q queues and P ports,
where Q is (nb_ethdev * nb_stages) + nb_ethdev and P is nb_workers.

The user can choose the number of workers and number of stages through the
--wlcores and the --stlist application command line arguments respectively.
The probed ethernet devices act as producer(s) for this application.

The ethdevs are configured as event Rx adapters that enables them to
injects events to eventdev based the first stage schedule type list
requested by the user through --stlist the command line argument.

Based on the number of stages to process(selected through --stlist),
the application forwards the event to next upstream queue and when it
reaches last stage in the pipeline if the event type is ATOMIC it is
enqueued onto ethdev Tx queue else to maintain ordering the event type is
set to ATOMIC and enqueued onto the last stage queue.
On packet Tx, application increments the number events processed and print
periodically in one second to get the number of events processed in one
second.

Note: The --prod_type_ethdev is mandatory for running the application.

Example command to run pipeline queue test:
sudo build/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
--test=pipeline_queue --wlcore=1 --prod_type_ethdev --stlist=ao

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

app/eventdev: launch pipeline lcores

The event master lcore's test termination and the logic to print the mpps
are common for the queue and all types queue test.

Move them as the common function.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

app/eventdev: add Tx service setup

Setup one port event port for Tx and link the respective event queue.
Register the Tx function as a service to be called from a service core.
The Tx function dequeues the events from the event queue and transmits
the packet to its respective ethernet port.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

app/eventdev: add event port and Rx adapter setup

Setup one port per worker and link to all queues and setup producer port
based on Rx adapter capabilities.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

app/eventdev: add pipeline ethport setup and destroy

Add common ethdev port setup and destroy along with event dev destroy.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

app/eventdev: add pipeline opt dump and check functions

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

app/eventdev: add mempool setup and destroy

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

app/eventdev: add pipeline test setup and destroy

Pipeline test has the queue and all types queue variants.
Introduce test_pipeline_common* to share the common code between those
tests.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

app/eventdev: add packet distribution logs

Add logs for packet distribution across worker cores to be printed
along with the test results.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

examples/eventdev: rename example

Rename eventdev_pipeline_sw_pmd to eventdev_pipeline as it is no longer
specific underlying event device.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

examples/eventdev: add mempool size configuration

Add option to configure the mempool size at run time instead of
hardcoding it to 16384 * num_ports.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

examples/eventdev: add atq single stage pipeline worker

Add optimized eventdev pipeline when ethdev supports thread safe Tx,
number of configured stages is one and all type queue option is enabled.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

examples/eventdev: add single stage pipeline worker

Add optimized eventdev pipeline when ethdev supports thread safe Tx
and number of configured stages is one.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

examples/eventdev: add all type queue option

Added configurable option to make queue type as all type queues i.e.
RTE_EVENT_QUEUE_CFG_ALL_TYPES based on event dev capability
RTE_EVENT_DEV_CAP_QUEUE_ALL_TYPES.

This can be enabled by supplying '-a' as a cmdline argument.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

examples/eventdev: add burst for thread safe pipeline

Add burst mode worker pipeline when Tx is multi thread safe.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

examples/eventdev: add thread safe Tx worker pipeline

Add worker pipeline when Tx is multi thread safe.
Probe Ethernet dev capabilities and select it it is supported.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

examples/eventdev: modify work cycles

The current work cycles function exchanges source and destination mac
address and also pauses the core for the given cycles.
This patch splits the function into two parts i.e. exchange mac and
pause the cores. The pause cores function is invoked at every stage
where as exchange mac is invoked when packet is transmitted.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

examples/eventdev: add non burst mode generic worker

Currently, worker uses burst dequeue and burst enqueue to forward events.
Add a non burst mode based on the event dev capabilities.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

examples/eventdev: add ops to check cmdline args

Each eventdev pipeline needs to allow different cmdline args combination
based on pipeline type.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

examples/eventdev: add generic worker pipeline

Rename existing pipeline as generic worker pipeline.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

examples/eventdev: add framework for caps based pipeline

Add framework to support capability based pipeline.
Based on the capability of event device and probed ethernet devices the
optimal pipeline configuration can be chosen.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

examples/eventdev: move common data into pipeline common

Move common structures and functions into pipeline_common.h so that they
can be used by different kinds of pipelines.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

examples/eventdev: support Rx adapter

Use event Rx adapter for packets Rx instead of explicit producer logic.
Use service run iter function for granular control instead of using
dedicated service lcore.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

test/event: register selftests

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

event/sw: update selftest ops

Update software eventdev ops to invoke selftest when application invokes
`rte_event_dev_selftest`.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

event/sw: make test standalone

Modify test_eventdev_sw to be standalone selftest independent of test
framework.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

event/sw: move test to driver

Move software eventdev specific test (test_eventdev_sw) to
driver/event/sw/.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

event/octeontx: add selftest to device arguments

Add selftest as a device argument that can be enabled by suppling
'self_test=1' as a vdev parameter

--vdev="event_octeontx,self_test=1"

The selftest is run after vdev creation is successfully
complete.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

event/octeontx: update selftest ops

Update octeontx eventdev ops to invoke selftest when application
invokes `rte_event_dev_selftest`.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

event/octeontx: make test standalone

Modify test_eventdev_octeontx to be standalone selftest independent of
test framework.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>

event/octeontx: move test to driver

Move octeontx eventdev specific test (test_eventdev_octeontx.c) to
driver/event/octeontx.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

eventdev: add API to perform self test

Add API to perform self test on the underlying event device driver.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

eal: add test assert macros

Adding common test assertion macros for unit testing.
Replaced common macros in test/test.h with new RTE_TEST_ASSERT_* macros.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>

app/eventdev: fix event device queue count

Fix the event device queue count reported when producer type is Rx
adapter for perfomance queue and atq test.

Fixes: 20eb154e0435 ("app/testeventdev: add perf queue test")

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

event/sw: remove stale IQ references when reconfigured

This commit fixes a bug in which, when the sw PMD is reconfigured, it would
leave stale IQ chunk pointers in each queue's IQ structure. Now, the PMD
initializes all IQs at eventdev start time and releases all IQ chunk
pointers at eventdev stop time (which has the consequence that any events
in a queue when the eventdev is stopped will be lost). This approach should
be resilient to any reconfiguration done between the stop and start, such
as adding or removing queues.

This commit also fixes two potential issues in iq_chunk.h. iq_init()
now initializes the IQ's count field to 0, and iq_dequeue_burst() sets
iq->head to the appropriate next pointer.

Fixes: dca926ca9faa ("event/sw: use dynamically-sized IQs")

Reported-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Signed-off-by: Gage Eads <gage.eads@intel.com>
Reviewed-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>

doc: add eventdev OPDL PMD guide

Add the description about opdl pmd

Signed-off-by: Liang Ma <liang.j.ma@intel.com>
Signed-off-by: Peter Mccarthy <peter.mccarthy@intel.com>
Acked-by: Marko Kovacevic <marko.kovacevic@intel.com>

event/sw: apply new capability flags

Signed-off-by: Liang Ma <liang.j.ma@intel.com>
Signed-off-by: Peter Mccarthy <peter.mccarthy@intel.com>

event/octeontx: apply new capability flags

Signed-off-by: Liang Ma <liang.j.ma@intel.com>
Signed-off-by: Peter Mccarthy <peter.mccarthy@intel.com>

event/dpaa2: apply new capability flags

Signed-off-by: Liang Ma <liang.j.ma@intel.com>
Signed-off-by: Peter Mccarthy <peter.mccarthy@intel.com>

eventdev: extend capability flags

add new capability flags to express the opdl PMD limitations.

RTE_EVENT_DEV_CAP_NONSEQ_MODE

Event device is capable of operating in none sequential mode. The path
of the event is not necessary to be sequential. Application can change
the path of event at runtime. If the flag is not set, then event each event
will follow a path from queue 0 to queue 1 to queue 2 etc. If the flag is
set, events may be sent to queues in any order. If the flag is not set, the
eventdev will return an error when the application enqueues an event for a
qid which is not the next in the sequence.

RTE_EVENT_DEV_CAP_RUNTIME_PORT_LINK

Event device is capable of configuring the queue/port link at runtime.
If the flag is not set, the eventdev queue/port link is only can be
configured during  initialization.

RTE_EVENT_DEV_CAP_MULTIPLE_QUEUE_PORT

Event device is capable of setting up the link between multiple queue
with single port. If the flag is not set, the eventdev can only map a
single queue to each port or map a single queue to many port.

Signed-off-by: Liang Ma <liang.j.ma@intel.com>
Signed-off-by: Peter Mccarthy <peter.mccarthy@intel.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

event/opdl: add unit tests

This commit adds unit test inside the OPDL PMD. There is a PMd parameter
"self_test" can be used to triger the test when vdev bus probe opdl device

  e.g.

  sudo ./app/test --vdev="event_opdl0,self_test=1"

Signed-off-by: Liang Ma <liang.j.ma@intel.com>
Signed-off-by: Peter Mccarthy <peter.mccarthy@intel.com>

event/opdl: add enqueue/dequeue

Signed-off-by: Liang Ma <liang.j.ma@intel.com>
Signed-off-by: Peter Mccarthy <peter.mccarthy@intel.com>

event/opdl: add event port config get/set

Signed-off-by: Liang Ma <liang.j.ma@intel.com>
Signed-off-by: Peter Mccarthy <peter.mccarthy@intel.com>

event/opdl: add event queue config get/set

Signed-off-by: Liang Ma <liang.j.ma@intel.com>
Signed-off-by: Peter Mccarthy <peter.mccarthy@intel.com>

event/opdl: add PMD main body and helper function

This commit adds a OPDL implementation of the eventdev API. The
implementation here is intended to enable the community to use
the OPDL infrastructure under eventdev API.

The main components of the implementation is three files:
  - opdl_evdev.c              Creation, configuration, etc
  - opdl_evdev_xstats.c       helper function to support stats collection
  - opdl_evdev.h              include the main data structure of opdl
                              device and all the function prototype
                              need to be exposed to support eventdev API.

  - opdl_evdev_init.c         implement all initailization helper function

This commit only adds the implementation, no existing DPDK files
are modified.

Signed-off-by: Liang Ma <liang.j.ma@intel.com>
Signed-off-by: Peter Mccarthy <peter.mccarthy@intel.com>

event/opdl: add OPDL ring infrastructure library

OPDL ring is the core infrastructure of OPDL PMD. OPDL ring library
provide the core data structure and core helper function set. The Ring
implements a single ring multi-port/stage pipelined packet distribution
mechanism. This mechanism has the following characteristics:

• No multiple queue cost, therefore, latency is significant reduced.
• Fixed dependencies between queue/ports is more suitable for complex.
  fixed pipelines of stateless packet processing (static pipeline).
• Has decentralized distribution (no scheduling core).
• Packets remain in order (no reorder core(s)).
* Update build system to enable compilation.

Signed-off-by: Liang Ma <liang.j.ma@intel.com>
Signed-off-by: Peter Mccarthy <peter.mccarthy@intel.com>
Reviewed-by: Seán Harte <seanbh@gmail.com>

app/eventdev: add service core configuration

Add service core configuration for Rx adapter. The configuration picks
the least used service core to run the service on.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>

app/eventdev: add event Rx adapter setup

Add functions to setup and configure Rx adapter based on the number of
ethdev ports setup.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

app/eventdev: add ethernet device tear down

Add ethernet device destroy functions to stop and close ethdev ports
if they are configured when prod_type_ethdev option is enabled.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

app/eventdev: add ethernet device setup helpers

Add ethernet device setup functions to configure ethdev ports incase
prod_type_ethdev option is enabled.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

app/eventdev: add pktmbuf pool for ethdev

Add pktmbuf pool creation used when configuring ethernet device as event
producer.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

app/eventdev: modify setup to support ethdev

Modify app setup to accommodate event port and queue setup based on the
number of ethernet ports.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

app/eventdev: add ethernet device producer option

Add command line option --prod_type_ethdev to specify that the events
are generated by ethernet device.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

eventdev: use links map to unlink queues

The octeontx event device doesn't store the queues to port mapping as a
result it cannot return the exact number of queues unlinked from a port
when application wants to unlink all the queues mapped (supplies queues
param as NULL).

Using links_map we can determine the exact queues mapped to a specific
port and unlink them.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Gage Eads <gage.eads@intel.com>

eventdev: fix doxygen comments

Fix doxygen return values and indentation.

Fixes: 64103dbcd673 ("eventdev: add dev attribute get function")
Cc: stable@dpdk.org
Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Kevin Laatz <kevin.laatz@intel.com>

event/sw: simplify credit scheme

This commit modifies the sw PMD credit scheme such that credits are
consumed when enqueueing a NEW event and released when an event is
released -- typically, the beginning and end of a pipeline. Workers that
simply forward events do not interact with the credit pool.

Signed-off-by: Gage Eads <gage.eads@intel.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

eventdev: add implicit release disable capability

This commit introduces a capability for disabling the "implicit" release
functionality for a port, which prevents the eventdev PMD from issuing
outstanding releases for previously dequeued events when dequeuing a new
batch of events.

If a PMD does not support this capability, the application will receive an
error if it attempts to setup a port with implicit releases disabled.
Otherwise, if the port is configured with implicit releases disabled, the
application must release each dequeued event by invoking
rte_event_enqueue_burst() with RTE_EVENT_OP_RELEASE or
RTE_EVENT_OP_FORWARD.

Signed-off-by: Gage Eads <gage.eads@intel.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

event/sw: use dynamically-sized IQs

This commit introduces dynamically-sized IQs, by switching the underlying
data structure from a fixed-size ring to a linked list of queue 'chunks.'
This has a number of benefits:
- Certain corner cases were observed in which all of a pipeline's flows
  could be pinned to one port for extended periods, effectively turning a
  multi-core pipeline into single-core one. This was caused by an event
  producer having a larger new_event_threshold than the IQ depth, and
  injecting large numbers of packets that are ultimately backpressured in a
  worker's rx_ring, causing those packets' flows to be scheduled to that
  port.
  The dynamically sized IQ does not have this problem because each IQ can
  grow large enough to store all the system's events, such that
  backpressure will not reach the worker_ring.
- Slight performance improvement (~1-2%) in high throughput scenarios,
  tested with eventdev_pipeline_sw_pmd.

This implementation has a small increase in the queue storage memory
footprint (~70KB). This commit also removes the iq_size xstat, which no
longer applies to this implementation.

Signed-off-by: Gage Eads <gage.eads@intel.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

event/sw: fix queue memory leak and multi-link bug

This commit reinitializes a queue before it is reconfigured, such that
reorder buffer memory is not leaked.

This bug masked a few other problems, which this commit corrects as well:
- sw_port_link() allowed a port to link to a queue twice, such that the
  port could then successfully unlink the queue twice. Now the link
  function checks whether a port is already linked to the queue, and if so
  returns success but doesn't assign the a port a second slot in the
  queue's cq map.
- test_eventdev.c's test_eventdev_unlink() was unlinking a queue twice
  from the same port, and expecting the second unlink to succeed. Now the
  test unlinks, links, then unlinks again.
- test_eventdev.c's test_eventdev_link_get() was linking a single queue but
  expecting the unlink function to return nb_queues (where nb_queues > 1).
  The test now checks for a return value of 1.

Fixes: 5ffb2f142d95 ("event/sw: support event queues")
Fixes: 371a688fc159 ("event/sw: support linking queues to ports")
Fixes: f8f9d233ea0e ("test/eventdev: add unit tests")
Cc: stable@dpdk.org
Signed-off-by: Gage Eads <gage.eads@intel.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>

test/eventdev: use CPU event type

octeontx test application was using non RTE_EVENT_TYPE_CPU
event type to generate the event from CPU.  Upon the introduction
of ethdev Rx adapter, RTE_EVENT_TYPE_ETHDEV has special
meaning. So avoid using non RTE_EVENT_TYPE_CPU event types
to inject events from CPU.

Fixes: d0d654986018 ("net/octeontx: support event Rx adapter")
Cc: stable@dpdk.org
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Reviewed-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>

eventdev: set error code in port link/unlink functions

The return value for rte_event_port_{link, unlink}() is defined as the
"number of {links, unlinks} actually established." However, the eventdev
layer's error checking returns negative error values. This commit aligns
the eventdev code with the API definition by having it set rte_errno and
return 0 if it detects an error.

Fixes: 4f0804bbdfb9 ("eventdev: implement the northbound APIs")
Cc: stable@dpdk.org
Signed-off-by: Gage Eads <gage.eads@intel.com>

examples/bbdev: add sample app

- sample application performing a loop-back over ethernet using
 a bbbdev device
- 'turbo_sw' PMD must be enabled for the app to be functional
- a packet is received on an ethdev port -> enqueued for baseband
 encode operation -> dequeued -> enqueued for baseband decode
 operation-> dequeued -> compared with original signal -> looped-back
 to the ethdev port

Signed-off-by: Amr Mokhtar <amr.mokhtar@intel.com>

app/testbbdev: add test application for bbdev

- full test suite for bbdev
- test App works seamlessly on all PMDs registered with bbdev
 framework
- a python script is provided to make our life easier
- supports execution of tests by parsing Test Vector files
- test Vectors can be added/deleted/modified with no need for
 re-compilation
- various tests can be executed:
 (a) Throughput test
 (b) Offload latency test
 (c) Operation latency test
 (d) Validation test
 (c) Sanity checks

Signed-off-by: Amr Mokhtar <amr.mokhtar@intel.com>

bb/turbo_sw: add software turbo driver

- bbdev 'turbo_sw' is the software accelerated version of 3GPP L1
 Turbo coding operation using the optimized Intel FlexRAN SDK libraries.
- 'turbo_sw' pmd is disabled by default

Signed-off-by: Amr Mokhtar <amr.mokhtar@intel.com>
Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>

bb/null: add null base band device driver

- 'bbdev_null' is a basic pmd that performs a minimalistic
 bbdev operation
- useful for bbdev smoke testing and in measuring the overhead
 introduced by the bbdev library
- 'bbdev_null' pmd is enabled by default

Signed-off-by: Amr Mokhtar <amr.mokhtar@intel.com>
Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>

bbdev: introduce wireless base band device lib

- wireless baseband device (bbdev) library files
- bbdev is tagged as EXPERIMENTAL
- Makefiles and configuration macros definition
- bbdev library is enabled by default
- release notes of the initial version

Signed-off-by: Amr Mokhtar <amr.mokhtar@intel.com>
Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>

member: fix memory leak on error

rte_member may have allocated a tailq entry or setum before failure,
so free them.

Fixes: 857ed6c68cf2 ("member: implement main API")
Cc: stable@dpdk.org
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Yipeng Wang <yipeng1.wang@intel.com>

mempool/octeontx: fix natural alignment being optimized out

The mbox messages are naturally aligned and in some cases compiler
optimization might disregard natural alignment.
Use volatile key word to force compiler to disable optimizing and
maintain alignment.

Fixes: aecb8e093b52 ("event/octeontx: introduce specialized mbox message copy")
Cc: stable@dpdk.org
Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Acked-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>

mempool/dpaa: optimize phy to virt conversion

If the allocation is from a single memzone, optimize
the phy-virt address conversions.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

mbuf: fix performance of freeing with non atomic refcnt

When RTE_MBUF_REFCNT_ATOMIC=n, the decrement of the mbuf reference
counter uses an atomic operation. This is not necessary and impacts
the performance (seen with TRex traffic generator).

We cannot replace rte_atomic16_add_return() by rte_mbuf_refcnt_update()
because it would add an additional check.

Solves this by introducing __rte_mbuf_refcnt_update(), which
updates the reference counter without doing anything else.

Fixes: 8f094a9ac5d7 ("mbuf: set mbuf fields while in pool")
Cc: stable@dpdk.org
Suggested-by: Hanoch Haim <hhaim@cisco.com>
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>

test/flow_classify: fix build warning

Move all static variables defined in header file to c file.

error log: /usr/bin/ld: Warning: size of symbol 'count' changed
  from 8 in test_pmd_perf.o to 24 in test_flow_classify.o

Fixes: 50bdac5916d9 ("flow_classify: remove table id parameter from API")

Signed-off-by: Jasvinder Singh <jasvinder.singh@intel.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>

test/mempool_perf: improve default mempool ops selection

Use of rte_eal_mbuf_default_mempool_ops() API will
allow to override pool handler through "--mbuf-pool-ops-name"
eal command line argument.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>

test/mempool: improve default mempool ops selection

Use of rte_eal_mbuf_default_mempool_ops() API will
allow to override pool handler through "--mbuf-pool-ops-name"
eal command line argument.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>

test/flow_classify: remove port bound requirement

Remove the requirement to have a port bound to igb_uio in
order to run this test suite.

Fixes: 9c9befea4f57 ("test: add flow classify unit tests")
Cc: stable@dpdk.org
Reported-by: Anatoly Burakov <anatoly.burakov@intel.com>
Signed-off-by: Bernard Iremonger <bernard.iremonger@intel.com>

test/memzone: fix freeing test

When reserving memzone for mz[], it will out of mz[RTE_MAX_MEMZONE] memory
bound after the counter reached to RTE_MAX_MEMZONE. It will flush the
counter's memory and lead to mz[] memory cannot be freed.

Fixd by extend to mz[RTE_MAX_MEMZONE + 1].

Fixes: ff909fe21f0a ("mem: introduce memzone freeing")
Cc: stable@dpdk.org
Signed-off-by: Phil Yang <phil.yang@arm.com>
Acked-by: Jianbo Liu <jianbo.liu@arm.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>

test/memzone: fix NULL freeing

No need to free a NULL memzone. It will cause test
termination.

Fixes: 71330483a193 ("test/memzone: fix memory leak")
Cc: stable@dpdk.org
Signed-off-by: Phil Yang <phil.yang@arm.com>
Acked-by: Jianbo Liu <jianbo.liu@arm.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>

test/memzone: fix wrong test

When reserving memzones in autotest, it makes no sense to expect a
failed memzone reserve when we specify both size flags - instead,
we should expect a memzone reserved with one of the two sizes.

Fixes: af75078fece3 ("first public release")
Cc: stable@dpdk.org
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>

test/memzone: fix copy-paste typo

Fixes: b77b5639726e ("mem: add huge page sizes for IBM Power")
Cc: stable@dpdk.org
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>

test/memzone: fix typo

Fixes: 71330483a193 ("test/memzone: fix memory leak")
Cc: stable@dpdk.org
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Reviewed-by: Radoslaw Biernacki <radoslaw.biernacki@linaro.org>

test/table: fix uninitialized parameter

delete_bulk() copies metadata to pointers provided by the entries
parameter, but in the unit test, they are uninitialized, leading
to rte_table attempting to memcpy into random garbage pointers.

Memsetting pointer table to zero will prevent that from happening.

Fixes: 48f2543cf0a8 ("app/test: add bulk adding and deleting")
Cc: stable@dpdk.org
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>

test: add malloc stats dump command

This can be useful for checking if an autotest leaks memory after
its execution.

Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>

test: register test as failed if setup failed

If test set up couldn't be completed, the test was previously
shown as succeeding, even though setup failed. Fix this to report
test as failed, and count all tests that should've been executed,
as failed as well.

Fixes: ffac67b1f71b ("app/test: new assert macros and test suite runner")
Cc: stable@dpdk.org
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>

test/crypto: use skipped return result

The cryptodev tests are updated to return SKIPPED instead
of failing if the PMD is not enabled, allowing test
infrastructure to identify that the test was not able to run.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>