This patch also adds configuration necessary for compilation of DPAA
Mempool driver into the DPAA specific config file.
CONFIG_RTE_MBUF_DEFAULT_MEMPOOL_OPS=dpaa is also configured to allow
applications to use DPAA mempool as default.
Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
This option both sets the maximum number of segments for Rx/Tx packets and
whether scattered mode is supported at all. This commit removes the latter
as well as configuration file exposure since the most appropriate value
should be decided at run-time.
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
The patch simplifies DPDK applications analysis for developers which use
Intel® VTune Amplifier.
The empty cycles are such iterations that yielded no RX packets. As far as
DPDK is running in poll mode, wasting cycles is equal to wasting CPU time.
Tracing such iterations can identify that device is underutilized. Tracing
empty cycles becomes even more critical if a system uses a lot of Ethernet
ports.
The patch gives possibility to analyze empty cycles without changing
application code. All needs to be done is just to reconfigure and rebuild
the DPDK itself with CONFIG_RTE_ETHDEV_PROFILE_ITT_WASTED_RX_ITERATIONS
enbled. The important thing here is that this does not affect DPDK code.
The profiling code is not being compiled if user does not specify config
flag.
The patch provides common way to inject RX queues profiling and VTune
specific implementation.
Signed-off-by: Ilia Kurakin <ilia.kurakin@intel.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Current mlx4 OFED version has bug which returns error to
ibv destroy functions when the device was plugged out, in
spite of the resources were destroyed correctly.
Hence, failsafe PMD was aborted, only in debug mode, when
it tries to remove the device in plug-out process.
The workaround added option to replace all claim_zero
assertions with debugging messages, by the way, this option
affects non ibv destroy assertions.
DPDK 18.02 release should work with Mellanox OFED-4.2 which will
include the verbs fix to this bug, then, this patch can
be removed.
Signed-off-by: Matan Azrad <matan@mellanox.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Introduce the fail-safe poll mode driver initialization and enable its
build infrastructure.
This PMD allows for applications to benefit from true hot-plugging
support without having to implement it.
It intercepts and manages Ethernet device removal events issued by
slave PMDs and re-initializes them transparently when brought back.
It also allows defining a contingency to the removal of a device, by
designating a fail-over device that will take on transmitting operations
if the preferred device is removed.
Applications only see a fail-safe instance, without caring for
underlying activity ensuring their continued operations.
Signed-off-by: Gaetan Rivet <gaetan.rivet@6wind.com>
Acked-by: Olga Shern <olgas@mellanox.com>
Generic Receive Offload (GRO) is a widely used SW-based offloading
technique to reduce per-packet processing overhead. It gains
performance by reassembling small packets into large ones. This
patchset is to support GRO in DPDK. To support GRO, this patch
implements a GRO API framework.
To enable more flexibility to applications, DPDK GRO is implemented as
a user library. Applications explicitly use the GRO library to merge
small packets into large ones. DPDK GRO provides two reassembly modes.
One is called lightweight mode, the other is called heavyweight mode.
If applications want to merge packets in a simple way and the number
of packets is relatively small, they can use the lightweight mode.
If applications need more fine-grained controls, they can choose the
heavyweight mode.
rte_gro_reassemble_burst is the main reassembly API which is used in
lightweight mode and processes N packets at a time. For applications,
performing GRO in lightweight mode is simple. They just need to invoke
rte_gro_reassemble_burst. Applications can get GROed packets as soon as
rte_gro_reassemble_burst returns.
rte_gro_reassemble is the main reassembly API which is used in
heavyweight mode and tries to merge N inputted packets with the packets
in GRO reassembly tables. For applications, performing GRO in heavyweight
mode is relatively complicated. Before performing GRO, applications need
to create a GRO context object, which keeps reassembly tables of
desired GRO types, by rte_gro_ctx_create. Then applications can use
rte_gro_reassemble to merge packets. The GROed packets are in the
reassembly tables of the GRO context object. If applications want to get
them, applications need to manually flush them by flush API.
Signed-off-by: Jiayu Hu <jiayu.hu@intel.com>
Reviewed-by: Jianfeng Tan <jianfeng.tan@intel.com>
The dpdk-test-eventdev tool is a Data Plane Development Kit (DPDK)
application that allows exercising various eventdev use cases. This
application has a generic framework to add new eventdev based test cases
to verify functionality and measure the performance parameters of DPDK
eventdev devices.
This patch adds the skeleton of the dpdk-test-eventdev application.
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
Currently EAL allocates hugepages one by one not paying attention
from which NUMA node allocation was done.
Such behaviour leads to allocation failure if number of available
hugepages for application limited by cgroups or hugetlbfs and
memory requested not only from the first socket.
Example:
# 90 x 1GB hugepages availavle in a system
cgcreate -g hugetlb:/test
# Limit to 32GB of hugepages
cgset -r hugetlb.1GB.limit_in_bytes=34359738368 test
# Request 4GB from each of 2 sockets
cgexec -g hugetlb:test testpmd --socket-mem=4096,4096 ...
EAL: SIGBUS: Cannot mmap more hugepages of size 1024 MB
EAL: 32 not 90 hugepages of size 1024 MB allocated
EAL: Not enough memory available on socket 1!
Requested: 4096MB, available: 0MB
PANIC in rte_eal_init():
Cannot init memory
This happens beacause all allocated pages are
on socket 0.
Fix this issue by setting mempolicy MPOL_PREFERRED for each hugepage
to one of requested nodes using following schema:
1) Allocate essential hugepages:
1.1) Allocate as many hugepages from numa N to
only fit requested memory for this numa.
1.2) repeat 1.1 for all numa nodes.
2) Try to map all remaining free hugepages in a round-robin
fashion.
3) Sort pages and choose the most suitable.
In this case all essential memory will be allocated and all remaining
pages will be fairly distributed between all requested nodes.
New config option RTE_EAL_NUMA_AWARE_HUGEPAGES introduced and
enabled by default for linuxapp except armv7 and dpaa2.
Enabling of this option adds libnuma as a dependency for EAL.
Fixes: 77988fc08dc5 ("mem: fix allocating all free hugepages")
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
Acked-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Tested-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Move all bypass functions to ixgbe pmd and remove function
pointers from the eth_dev_ops struct.
Signed-off-by: Radu Nicolau <radu.nicolau@intel.com>
Acked-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
TX coalescing waits for ETH_COALESCE_PKT_NUM packets to be coalesced
across bursts before transmitting them. For slow traffic, such as
100 PPS, this approach increases latency since packets are received
one at a time and tx coalescing has to wait for ETH_COALESCE_PKT
number of packets to arrive before transmitting.
To fix this:
- Update rx path to use status page instead and only receive packets
when either the ingress interrupt timer threshold (5 us) or
the ingress interrupt packet count threshold (32 packets) fires.
(i.e. whichever happens first).
- If number of packets coalesced is <= number of packets sent
by tx burst function, stop coalescing and transmit these packets
immediately.
Also added compile time option to favor throughput over latency by
default.
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Stub callbacks for the generic flow API and a new FLOW debug define.
Signed-off-by: John Daley <johndale@cisco.com>
Reviewed-by: Nelson Escobar <neescoba@cisco.com>
When building DPDK with musl, there is need not to disable
backtrace to remove some references to execinfo.h which is
not supported by musl now.
This also applies to some other libc implementation which
doesn't support backtrace() and backtrace_symbols().
musl is an implementation of the userspace portion
of the standard library functionality described in
the ISO C and POSIX standards, plus common extensions.
Got more details about musl from http://www.musl-libc.org .
Signed-off-by: Wei Dai <wei.dai@intel.com>
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Making AVX and AVX512 configurable is useful for performance and power
testing.
The similar kernel patch at https://patchwork.kernel.org/patch/9618883/.
AVX512 support like in rte_memcpy has been in DPDK since 16.04, but it's
still unproven in rich use cases in hardware. Therefore it's marked as
experimental for now, will enable it after enough field test and possible
optimization.
Signed-off-by: Zhihong Wang <zhihong.wang@intel.com>
Reviewed-by: Zhiyong Yang <zhiyong.yang@intel.com>
Reviewed-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
DPAA2 Hardware Mempool handlers allow enqueue/dequeue from NXP's
QBMAN hardware block.
CONFIG_RTE_MBUF_DEFAULT_MEMPOOL_OPS is set to 'dpaa2', if the pool
is enabled.
This memory pool currently supports packet mbuf type blocks only.
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Having packets received without any offload flags given in the mbuf is not
very useful, and performance tests with testpmd indicates little
benefit is got with the current code by turning off the flags. This makes
the build-time option pointless, so we can remove it.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Having packets received without any offload flags given in the mbuf is not
very useful, and performance tests with testpmd indicates little to no
benefit is got with the current code by turning off the flags. This makes
the build-time option pointless, so we can remove it.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Jianbo Liu <jianbo.liu@linaro.org>
Enable Arkville on supported configurations
Add overview documentation
Minimum driver support for valid compile
Arkville PMD is not supported on ARM or PowerPC at this time
Signed-off-by: Ed Czeck <ed.czeck@atomicrules.com>
Signed-off-by: John Miller <john.miller@atomicrules.com>
The crypto scheduler PMD has no external dependencies to enable that by
default.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
Add a library designed to calculate latency statistics and report them
to the application when queried. The library measures minimum, average and
maximum latencies, and jitter in nano seconds. The current implementation
supports global latency stats, i.e. per application stats.
Signed-off-by: Reshma Pattan <reshma.pattan@intel.com>
Signed-off-by: Remy Horton <remy.horton@intel.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
This patch adds a library that calculates peak and average data-rate
statistics. For ethernet devices. These statistics are reported using
the metrics library.
Signed-off-by: Remy Horton <remy.horton@intel.com>
This patch adds a new information metrics library. This Metrics
library implements a mechanism by which producers can publish
numeric information for later querying by consumers. Metrics
themselves are statistics that are not generated by PMDs, and
hence are not reported via ethdev extended statistics.
Metric information is populated using a push model, where
producers update the values contained within the metric
library by calling an update function on the relevant metrics.
Consumers receive metric information by querying the central
metric data, which is held in shared memory.
Signed-off-by: Remy Horton <remy.horton@intel.com>
This adds the minimal changes to allow a SW eventdev implementation to
be compiled, linked and created at run time. The eventdev does nothing,
but can be created via vdev on commandline, e.g.
sudo ./x86_64-native-linuxapp-gcc/app/test --vdev=event_sw0
...
PMD: Creating eventdev sw device event_sw0, numa_node=0, sched_quanta=128
RTE>>
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
The skeleton driver facilitates, bootstrapping the new
eventdev driver and creates a platform to verify
the northbound eventdev common code.
The driver supports both VDEV and PCI based eventdev
devices.
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
This patch implements northbound eventdev API interface using
southbond driver interface
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Adds a header file with log macros for the AVP PMD
Signed-off-by: Allain Legacy <allain.legacy@windriver.com>
Signed-off-by: Matt Peters <matt.peters@windriver.com>
Acked-by: Vincent Jardin <vincent.jardin@6wind.com>
This commit introduces the AVP PMD file structure without adding any actual
driver functionality. Functional blocks will be added in later patches.
Signed-off-by: Allain Legacy <allain.legacy@windriver.com>
Signed-off-by: Matt Peters <matt.peters@windriver.com>
Acked-by: Vincent Jardin <vincent.jardin@6wind.com>
Add debug options to config file. Define macros used for log and make
use of config file options to enable them.
Signed-off-by: Shijith Thotton <shijith.thotton@caviumnetworks.com>
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Venkat Koppula <venkat.koppula@caviumnetworks.com>
Signed-off-by: Srisivasubramanian S <ssrinivasan@caviumnetworks.com>
Signed-off-by: Mallesham Jatharakonda <mjatharakonda@oneconvergence.com>
Enable Thunderx nicvf PMD driver in the common
config as it does not have build dependency
with any external library and/or architecture.
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Add KNI PMD which wraps librte_kni for ease of use.
KNI PMD can be used as any regular PMD to send / receive packets to the
Linux networking stack.
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Reviewed-by: Yong Wang <yongwang@vmware.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Yong Wang <yongwang@vmware.com>
Moved from lib/librte_mempool, stack mempool handler is an independent
driver.
Shared builds would now require to link in librte_mempool_stack for
"stack" mempool handler.
Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Moved from lib/librte_mempool, ring mempool is now an independent
driver.
Shared builds would now need to add librte_mempool_ring for:
* ring_mp_mc
* ring_sp_sc
* ring_sp_mc
* ring_mp_sc
Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
There was a compile time setting to enable a ring to yield when
it entered a loop in mp or mc rings waiting for the tail pointer update.
Build time settings are not recommended for enabling/disabling features,
and since this was off by default, remove it completely. If needed, a
runtime enabled equivalent can be used.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Reviewed-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
The debug option only provided statistics to the user, most of
which could be tracked by the application itself. Remove this as a
compile time option, and feature, simplifying the code.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Reviewed-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Users compiling DPDK should not need to know or care about the arrangement
of cachelines in the rte_ring structure. Therefore just remove the build
option and set the structures to be always split. On platforms with 64B
cachelines, for improved performance use 128B rather than 64B alignment
since it stops the producer and consumer data being on adjacent cachelines.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Reviewed-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Downstreams might want to provide different DPDK releases at the same
time to support multiple consumers of DPDK linked against older and newer
sonames.
Also due to the interdependencies that DPDK libraries can have applications
might end up with an executable space in which multiple versions of a
library are mapped by ld.so.
Think of LibA that got an ABI bump and LibB that did not get an ABI bump
but is depending on LibA.
Application
\-> LibA.old
\-> LibB.new -> LibA.new
That is a conflict which can be avoided by setting CONFIG_RTE_MAJOR_ABI.
If set CONFIG_RTE_MAJOR_ABI overwrites any LIBABIVER value.
An example might be ``CONFIG_RTE_MAJOR_ABI=16.11`` which will make all
libraries librte<?>.so.16.11 instead of librte<?>.so.<LIBABIVER>.
We need to cut arbitrary long stings after the .so now and this would work
for any ABI version in LIBABIVER:
$(Q)ln -s -f $< $(patsubst %.$(LIBABIVER),%,$@)
But using the following instead additionally allows to simplify the Make
File for the CONFIG_RTE_NEXT_ABI case.
$(Q)ln -s -f $< $(shell echo $@ | sed 's/\.so.*/.so/')
Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
Reviewed-by: Jan Blunck <jblunck@infradead.org>
Tested-by: Jan Blunck <jblunck@infradead.org>
