dpdk.git
7 years agoconfig: enable vhost NUMA awareness by default
Ilya Maximets [Thu, 29 Jun 2017 05:59:20 +0000 (08:59 +0300)]
config: enable vhost NUMA awareness by default

It is safe to enable LIBRTE_VHOST_NUMA by default for all
configurations where libnuma is already a default dependency.

Signed-off-by: Ilya Maximets <i.maximets@samsung.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>
7 years agomem: balanced allocation of hugepages
Ilya Maximets [Thu, 29 Jun 2017 05:59:19 +0000 (08:59 +0300)]
mem: balanced allocation of hugepages

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>
7 years agomaintainers: update for i40e
Beilei Xing [Thu, 29 Jun 2017 01:11:38 +0000 (09:11 +0800)]
maintainers: update for i40e

Signed-off-by: Beilei Xing <beilei.xing@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
7 years agocrypto/scheduler: remove deprecated functions
Fan Zhang [Thu, 25 May 2017 07:29:47 +0000 (08:29 +0100)]
crypto/scheduler: remove deprecated functions

This patch remove the deprecated functions as well as notice for
scheduler mode set/get API changes.

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agotest/crypto: fix overflow
Pablo de Lara [Wed, 28 Jun 2017 09:31:03 +0000 (10:31 +0100)]
test/crypto: fix overflow

In one of the SNOW3G unit tests, insufficient memory
was allocated, leading to a buffer overflow.

Fixes: 2b52e1e95aab ("app/test: add SNOW 3G UEA2 with offset")
Cc: stable@dpdk.org
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocryptodev: fix device stop function
Kirill Rybalchenko [Wed, 14 Jun 2017 08:42:46 +0000 (09:42 +0100)]
cryptodev: fix device stop function

Flag dev_started should be cleared after dev_stop() function call
because the flag is checked inside the dev_stop() function.

Fixes: d11b0f30df88 ("cryptodev: introduce API and framework for crypto devices")
Cc: stable@dpdk.org
Signed-off-by: Kirill Rybalchenko <kirill.rybalchenko@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agodoc: fix typo in l2fwd-crypto usage
Pablo de Lara [Thu, 22 Jun 2017 12:02:35 +0000 (13:02 +0100)]
doc: fix typo in l2fwd-crypto usage

Fixes: ba7b86b1419b ("doc: add l2fwd-crypto sample app guide")
Cc: stable@dpdk.org
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
7 years agoapp/crypto-perf: fix error message
Pablo de Lara [Thu, 22 Jun 2017 12:02:34 +0000 (13:02 +0100)]
app/crypto-perf: fix error message

Fixes: f6cefe253cc8 ("app/crypto-perf: add range/list of sizes")
Cc: stable@dpdk.org
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
7 years agoexamples/l2fwd-crypto: fix auth info display
Pablo de Lara [Thu, 22 Jun 2017 12:02:33 +0000 (13:02 +0100)]
examples/l2fwd-crypto: fix auth info display

Fixes: 4790f99d2d31 ("examples/l2fwd-crypto: use cryptodev algorithm parser")
Cc: stable@dpdk.org
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
7 years agoexamples/l2fwd-crypto: fix application help
Pablo de Lara [Thu, 22 Jun 2017 12:02:32 +0000 (13:02 +0100)]
examples/l2fwd-crypto: fix application help

Chain parameter can be CIPHER_HASH, HASH_CIPHER,
CIPHER_ONLY or HASH_ONLY, but only the first two
were shown in the application help.

Fixes: 1a75e9f3fadb ("examples/l2fwd-crypto: add cipher/hash only cases")
Cc: stable@dpdk.org
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
7 years agotest/crypto: use existing algorithm strings
Pablo de Lara [Thu, 22 Jun 2017 12:02:31 +0000 (13:02 +0100)]
test/crypto: use existing algorithm strings

Instead of using duplicated functions to get the algorithm
strings, use the functions from the cryptodev library.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
7 years agotest/crypto: remove offsets in wireless algo vectors
Pablo de Lara [Thu, 22 Jun 2017 12:02:30 +0000 (13:02 +0100)]
test/crypto: remove offsets in wireless algo vectors

Instead of passing the authentication and cipher offset
from the test vectors, just get them from the IV and AAD lengths.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
7 years agotest/crypto: remove unnecessary offset
Pablo de Lara [Thu, 22 Jun 2017 12:02:29 +0000 (13:02 +0100)]
test/crypto: remove unnecessary offset

For AES/DES tests, there is no need to leave out
the first block from the crypto operation.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
7 years agotest/crypto: fix wrong AAD setting
Pablo de Lara [Thu, 22 Jun 2017 12:02:28 +0000 (13:02 +0100)]
test/crypto: fix wrong AAD setting

AAD should not point at IV for AES algorithms.
For AES-GCM, AAD will point at additional data in the mbuf.
For the other algorithms (such as 3DES CBC), AAD is not used.

Fixes: ffbe3be0d4b5 ("app/test: add libcrypto")
Cc: stable@dpdk.org
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
7 years agocrypto/aesni_mb: remove assert checks
Pablo de Lara [Thu, 22 Jun 2017 12:02:27 +0000 (13:02 +0100)]
crypto/aesni_mb: remove assert checks

Some assert checks in the driver were
incorrect, but they are not necessary anyway,
as application will panic in any case.

Fixes: 0f548b50a160 ("crypto/aesni_mb: process crypto op on dequeue")
Cc: stable@dpdk.org
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agocryptodev: remove unused driver struct
Pablo de Lara [Wed, 21 Jun 2017 06:30:33 +0000 (07:30 +0100)]
cryptodev: remove unused driver struct

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agocryptodev: remove unused PCI probe/remove
Pablo de Lara [Wed, 21 Jun 2017 06:30:20 +0000 (07:30 +0100)]
cryptodev: remove unused PCI probe/remove

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agocrypto/dpaa2_sec: do not use cryptodev driver
Pablo de Lara [Wed, 21 Jun 2017 06:30:06 +0000 (07:30 +0100)]
crypto/dpaa2_sec: do not use cryptodev driver

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocrypto/qat: do not use cryptodev driver
Pablo de Lara [Wed, 21 Jun 2017 06:28:26 +0000 (07:28 +0100)]
crypto/qat: do not use cryptodev driver

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agocryptodev: add PCI driver helpers
Pablo de Lara [Wed, 21 Jun 2017 06:28:25 +0000 (07:28 +0100)]
cryptodev: add PCI driver helpers

Add PCI probe/remove/init/uninit functions in a separate
file rte_cryptodev_pci.h, which do not use cryptodev driver,
in order to be removed in next commits.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agocryptodev: release device if PCI probing fails
Pablo de Lara [Wed, 21 Jun 2017 06:28:24 +0000 (07:28 +0100)]
cryptodev: release device if PCI probing fails

Call rte_cryptodev_pmd_release_device() if probing a
PCI crypto device, instead of accessing the variables
directly. This will be useful when rte_cryptodev_pci_probe()
gets moved to a separate file.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agocryptodev: move vdev functions to a separate file
Pablo de Lara [Tue, 27 Jun 2017 21:16:33 +0000 (22:16 +0100)]
cryptodev: move vdev functions to a separate file

Move all functions handling virtual devices to a separate
header file "rte_cryptodev_vdev.h", in order to leave only
generic functions for any device in the rest of the files.

It also creates the file "rte_cryptodev_pmd.c", with the
implementations of these functions.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agocryptodev: set PCI info only when device is PCI
Pablo de Lara [Wed, 21 Jun 2017 06:28:22 +0000 (07:28 +0100)]
cryptodev: set PCI info only when device is PCI

Do not set PCI information in the device information structure
for any crypto device, just for the ones that are PCI, so
this is set internally in the PCI crypto PMDs (only QAT now).

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agocryptodev: simplify device list retrieval logic
Pablo de Lara [Wed, 21 Jun 2017 06:28:21 +0000 (07:28 +0100)]
cryptodev: simplify device list retrieval logic

rte_cryptodev_devices_get() function returns an array of devices
sharing the same driver.

Instead of having two different paths depending on the device being
virtual or physical, retrieve the driver name from rte_device structure.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agocryptodev: rename device retrieval argument
Pablo de Lara [Wed, 21 Jun 2017 06:28:20 +0000 (07:28 +0100)]
cryptodev: rename device retrieval argument

rte_cryptodev_devices_get() function was parsing a crypto
device name as an argument, but the function actually
returns device identifiers of devices that share the
same crypto driver, so the argument should be driver name, instead.

Fixes: 38227c0e3ad2 ("cryptodev: retrieve device info")
Cc: stable@dpdk.org
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agocryptodev: set driver name for all devices
Pablo de Lara [Wed, 21 Jun 2017 06:28:19 +0000 (07:28 +0100)]
cryptodev: set driver name for all devices

When retrieving device information for a crypto driver,
driver name was only set when it was a PCI driver.
Getting the driver name from rte_device structure
allows rte_cryptodev_get_info() function to return it
regardless they are virtual or physical devices.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agocryptodev: store device pointer in virtual devices
Pablo de Lara [Wed, 21 Jun 2017 06:28:18 +0000 (07:28 +0100)]
cryptodev: store device pointer in virtual devices

Only non virtual devices were storing the pointer to
rte_device structure in rte_cryptodev, which will be needed
to retrieve the driver name for any device.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agoexamples/l2fwd-crypto: fix option parsing
Pablo de Lara [Wed, 24 May 2017 10:28:02 +0000 (11:28 +0100)]
examples/l2fwd-crypto: fix option parsing

Statistics period time option is parsed with -T argument,
but -t was accepted by mistake, instead.

Fixes: 387259bd6c67 ("examples/l2fwd-crypto: add sample application")
Cc: stable@dpdk.org
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocrypto/aesni_gcm: fix leading spaces
Pablo de Lara [Mon, 22 May 2017 10:52:34 +0000 (11:52 +0100)]
crypto/aesni_gcm: fix leading spaces

Fixes: 26c2e4ad5ad4 ("cryptodev: add capabilities discovery")

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agomem: fix malloc element resize with padding
Jamie Lavigne [Thu, 8 Jun 2017 19:12:17 +0000 (19:12 +0000)]
mem: fix malloc element resize with padding

Currently when a malloc_elem is split after resizing, any padding
present in the elem is ignored.  This causes the resized elem to be too
small when padding is present, and user data can overwrite the beginning
of the following malloc_elem.

Solve this by including the size of the padding when computing where to
split the malloc_elem.

Fixes: af75078fece3 ("first public release")

Signed-off-by: Jamie Lavigne <lavignen@amazon.com>
Acked-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
7 years agotest/mbuf: remove global mempool
Santosh Shukla [Mon, 26 Jun 2017 10:04:54 +0000 (10:04 +0000)]
test/mbuf: remove global mempool

Currently, pool resources are allocated statically
and are not freed. Results of that test can not run more than once.

Fix removes static dependency from test application and
now allocating and freeing resources dynamically.
Test runs for more than once.

Fixes: af75078fece3 ("first public release")

Signed-off-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
7 years agopci: reduce debug log verbosity when probing
Ferruh Yigit [Tue, 27 Jun 2017 14:23:34 +0000 (15:23 +0100)]
pci: reduce debug log verbosity when probing

When debug level logging enabled (--log-level=8) each driver failed to
probe the device printed, like:

EAL: Driver (net_ark) doesn't match the device
EAL: Driver (net_avp) doesn't match the device
EAL: Driver (net_bnxt) doesn't match the device
EAL: Driver (net_cxgbe) doesn't match the device
EAL: Driver (net_e1000_igb) doesn't match the device
EAL: Driver (net_e1000_igb_vf) doesn't match the device
EAL: Driver (net_e1000_em) doesn't match the device
EAL: Driver (net_ena) doesn't match the device
EAL: Driver (net_enic) doesn't match the device
EAL: Driver (net_fm10k) doesn't match the device
EAL: Driver (net_i40e) doesn't match the device
EAL: Driver (net_i40e_vf) doesn't match the device
....

Overall hundreds of similar lines printed, because all drivers printed
for all devices. This is too much noise and there is already a log
message printed when device matched.

Removing the debug log completely.

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agomk: fix excluding files when installing docs
Luca Boccassi [Fri, 23 Jun 2017 18:41:47 +0000 (19:41 +0100)]
mk: fix excluding files when installing docs

The --exclude parameter must be passed before the input directory to
tar, otherwise it's silently ignored and the .doctrees directory is
installed by make install-doc.

Signed-off-by: Luca Boccassi <luca.boccassi@gmail.com>
Acked-by: John McNamara <john.mcnamara@intel.com>
7 years agomk: use make silent flag to print HTML doc version
Luca Boccassi [Fri, 23 Jun 2017 18:41:46 +0000 (19:41 +0100)]
mk: use make silent flag to print HTML doc version

Depending on the environment, make might echo the command being ran.
In mk/rte.sdkdoc.mk make is used to print the DPDK version to be
piped to doxygen. This causes the following to be written:

<div id="projectname">DPDK
&#160;<span id="projectnumber">/usr/bin/make-f/build/dpdk-jYjqnr/
 dpdk-16.11.2/mk/rte.sdkconfig.mkshowversion</span>
</div>

Use -s (--silent) to prevent echoing.

Signed-off-by: Luca Boccassi <luca.boccassi@gmail.com>
Acked-by: John McNamara <john.mcnamara@intel.com>
7 years agoethdev: add missing symbol in map
Luca Boccassi [Thu, 22 Jun 2017 12:04:59 +0000 (13:04 +0100)]
ethdev: add missing symbol in map

The function rte_eth_tx_done_cleanup() was missing in the map file
so it cannot be used by applications linking to shared libraries.
pktgen uses it since version 3.2.0.

Fixes: 44a718c457b5 ("ethdev: add API to free consumed buffers in Tx ring")
Cc: stable@dpdk.org
Signed-off-by: Luca Boccassi <luca.boccassi@gmail.com>
7 years agopci: set default numa node for broken systems
Tonghao Zhang [Thu, 11 May 2017 01:56:33 +0000 (18:56 -0700)]
pci: set default numa node for broken systems

The NUMA node information for PCI devices provided through
sysfs is invalid for AMD Opteron(TM) Processor 62xx and 63xx
on Red Hat Enterprise Linux 6, and VMs on some hypervisors.
It is good to see more checking for valid values.

Typical wrong numa node in some VMs:
$ cat /sys/devices/pci0000:00/0000:00:18.6/numa_node
-1

Signed-off-by: Tonghao Zhang <nic@opencloud.tech>
Acked-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
7 years agoring: fix return value for dequeue
Anand B Jyoti [Fri, 2 Jun 2017 06:29:51 +0000 (11:59 +0530)]
ring: fix return value for dequeue

The error return code for rte_ring_sc_dequeue_bulk() and
rte_ring_mc_dequeue_bulk() function should be -ENOENT rather
than -ENOBUFS as described in the function description.

Fixes: cfa7c9e6fc1f ("ring: make bulk and burst return values consistent")
Cc: stable@dpdk.org
Signed-off-by: Anand B Jyoti <anand.b.jyoti@intel.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
7 years agoeventdev: define default value for dequeue timeout
Jerin Jacob [Thu, 18 May 2017 08:48:27 +0000 (14:18 +0530)]
eventdev: define default value for dequeue timeout

Defining the value 0 as default value for dequeue timeout
will help the application reduce the configuration setup
if the application is interested only in default
timeout value.

removed "min_dequeue_limit" negative testcase as
min_dequeue_limit value could be zero(which is
default timeout now) if driver has
dev_info->min_dequeue_timeout_ns  = 1.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
7 years agotest/eventdev: verify priority test prerequisite
Jerin Jacob [Thu, 1 Jun 2017 06:41:23 +0000 (12:11 +0530)]
test/eventdev: verify priority test prerequisite

octeontx specific priority test expects priority of each
event queue to be a unique value. Verify that condition
before it processes to test the priority.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
7 years agoevent/octeontx: fix missing enqueue SMP barrier
Jerin Jacob [Fri, 9 Jun 2017 13:16:03 +0000 (18:46 +0530)]
event/octeontx: fix missing enqueue SMP barrier

Typically RTE_EVENT_OP_NEW issued by the producer
lcore. To reflect the write changes issued by the
producer lcore on worker lcore, an SMP write barrier
is required on producer enqueue. Fixing the missing
rte_smp_wmb() on enqueue with RTE_EVENT_OP_NEW.

Fixes: f10d322eff76 ("event/octeontx: support worker enqueue")
Cc: stable@dpdk.org
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Gage Eads <gage.eads@intel.com>
7 years agoevent/octeontx: improve dequeue performance
Jerin Jacob [Fri, 9 Jun 2017 12:06:50 +0000 (17:36 +0530)]
event/octeontx: improve dequeue performance

switch tag wait is a costly operation as it may
translate to IOB read if core swtag cache is not updated.
Do tag switch wait only when there is a tag request on
the same hardware work slot.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Gage Eads <gage.eads@intel.com>
7 years agoevent/skeleton: advertise the burst mode capability
Jerin Jacob [Tue, 20 Jun 2017 14:26:48 +0000 (19:56 +0530)]
event/skeleton: advertise the burst mode capability

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
7 years agoevent/sw: advertise the burst mode capability
Jerin Jacob [Wed, 14 Jun 2017 04:57:33 +0000 (10:27 +0530)]
event/sw: advertise the burst mode capability

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
7 years agoeventdev: introduce burst mode capability
Jerin Jacob [Wed, 14 Jun 2017 04:57:32 +0000 (10:27 +0530)]
eventdev: introduce burst mode capability

Introducing the burst mode capability flag to express the event device
is capable of operating in burst mode for enqueue(forward, release) and
dequeue operation. If the device is not capable, then the application
still uses the rte_event_dequeue_burst() and rte_event_enqueue_burst()
but PMD accepts only one event at a time which is any way transparent
with the current rte_event_*_burst API semantics.

It solves two purposes:
1) Fix performance regression on the PMD which supports only nonburst
mode, and this issue is two-fold.

Typically the burst_worker main loop consists of following pseudo code:

while(1)
{
uint16_t nb_rx = rte_event_dequeue_burst(ev,..);

for (i=0; i < nb_rx; i++) {
process(ev[i]);
if (is_release_required(ev[i]))
release_the_event(ev);
}

        uint16_t nb_tx = rte_event_enqueue_burst(dev_id, port_id,
                                events, nb_rx);
        while (nb_tx < nb_rx)
            nb_tx += rte_event_enqueue_burst(dev_id, port_id,
            events + nb_tx, nb_rx - nb_tx);
}

Typically the non_burst_worker main loop consists of following pseudo code:
while(1)
{
    uint16_t nb_rx = rte_event_dequeue_burst(&ev, , 1);
    if (!nb_rx)
        continue;
    process(ev);
    while (rte_event_enqueue_burst(dev, port, &ev, 1) != 1);
}

Following overhead has been seen on nonburst mode capable PMDs with
burst mode version
- Extra explicit release(PMD does release on implicitly on next
dequeue) and thus avoids the cost additional driver function overhead.
- Extra "for" loop for event processing which compiler cannot detect at
runtime

2) Simplify the application configuration by avoiding the application to
find the correct enqueue and dequeue depth across different PMD.
If burst mode is not supported then, PMD can ignore depth field.
This will enable to write portable applications and makes
RFC eventdev_pipeline application works on OCTEONTX PMD
http://dpdk.org/dev/patchwork/patch/23799/

If an application wishes to get the maximum performance on nonburst
capable PMD then the application can write the code in a way that by
keeping packet processing function as inline functions and launch the
workers based on the capability.
The generic burst based worker still work on those PMDs without
any code change but this scheme needed only when the application wants
to gets the maximum performance out of nonburst capable PMDs.

This patch is based the on the real world test cases
http://dpdk.org/dev/patchwork/patch/24832/, Where without this scheme
20.9% performance drop observed per core.

See worker_wrapper(), perf_queue_worker(), perf_queue_worker_burst()
functions to use this scheme in a portable way without losing performance
on both sets of PMDs and achieving the portability.
http://dpdk.org/dev/patchwork/patch/24832/

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
7 years agoeventdev: make vdev init and uninit functions optional
Jerin Jacob [Fri, 9 Jun 2017 08:37:29 +0000 (14:07 +0530)]
eventdev: make vdev init and uninit functions optional

Made libeventdev library independent of VDEV bus by moving vdev pmd
specific function to rte_eventdev_pmd_vdev.h header file. Eventdev VDEV
PMD can include that for generic eventdev VDEV init and uninit function
enablement.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
7 years agoeventdev: make PCI probe and remove functions optional
Jerin Jacob [Fri, 9 Jun 2017 08:37:28 +0000 (14:07 +0530)]
eventdev: make PCI probe and remove functions optional

Made libeventdev library independent of PCI bus by moving pci pmd
specific function to rte_eventdev_pmd_pci.h header file. Eventdev PCI
PMD can include that for generic eventdev PCI probe and remove function
enablement.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
7 years agoeventdev: restructure release function
Jerin Jacob [Fri, 9 Jun 2017 08:37:27 +0000 (14:07 +0530)]
eventdev: restructure release function

Remove rte_event_dev_close() from rte_event_pmd_release() function so
that rte_event_pmd_release() can be used in stateless way. This will
enable rte_event_pmd_vdev_uninit() function to avoid using
eventdev_globals global variable and the need for exposing the a
global variable to PMD.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
7 years agoeventdev: remove PCI dependency from generic structures
Jerin Jacob [Fri, 9 Jun 2017 08:37:26 +0000 (14:07 +0530)]
eventdev: remove PCI dependency from generic structures

Remove the PCI dependency from generic data structures
and moved the PCI specific code to rte_event_pmd_pci*

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
7 years agoevent/sw: fix mapped qid count with parallel queue
Harry van Haaren [Wed, 7 Jun 2017 10:04:44 +0000 (11:04 +0100)]
event/sw: fix mapped qid count with parallel queue

This commit fixes the counting of mapped queues to a port,
when the type of queue type is PARALLEL. Not incrementing
the count here could lead to an underflow of the count when
unlinking at a later date.

Fixes: 371a688fc159 ("event/sw: support linking queues to ports")

Reported-by: Jesse Bruni <jesse.bruni@intel.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
7 years agoevent/octeontx: fix error msg in mbox wait response
Santosh Shukla [Mon, 29 May 2017 07:29:40 +0000 (12:59 +0530)]
event/octeontx: fix error msg in mbox wait response

Fixes: 6da9d2457 ("event/octeontx: add mailbox support")

Signed-off-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
7 years agoevent/octeontx: add driver name in info get
Jerin Jacob [Fri, 2 Jun 2017 09:26:06 +0000 (14:56 +0530)]
event/octeontx: add driver name in info get

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
7 years agoevent/sw: fix credit tracking in port dequeue
Harry van Haaren [Thu, 1 Jun 2017 15:45:54 +0000 (16:45 +0100)]
event/sw: fix credit tracking in port dequeue

Single-link optimized ports previously did not correctly track
credits when dequeued, and re-enqueued as a FORWARD type. This
could "inflate" the number of credits in the system.

A unit test is added to reproduce and verify the issue, and the
fixed implementation counts FORWARD packets, and reduces the
number of credits the port has if it is of single-link type.

Fixes: 656af9180014 ("event/sw: add worker core functions")

Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Gage Eads <gage.eads@intel.com>
7 years agoeventdev: clarify the worker thread workflow
Jerin Jacob [Thu, 18 May 2017 11:10:41 +0000 (16:40 +0530)]
eventdev: clarify the worker thread workflow

If the RTE_EVENT_DEV_CAP_DISTRIBUTED_SCHED capability flag
is not set indicates the device is centralized and thus needs
a dedicated scheduling thread that repeatedly calls
rte_event_schedule().

Update the worker thread code snippet to match
the description.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
7 years agoeventdev: clarify atomic and ordered queue config
Gage Eads [Wed, 21 Jun 2017 13:32:51 +0000 (19:02 +0530)]
eventdev: clarify atomic and ordered queue config

The nb_atomic_flows and nb_atomic_order_sequences fields are only inspected
if the queue is configured for atomic or ordered scheduling, respectively.
This commit updates the documentation to reflect that.

Signed-off-by: Gage Eads <gage.eads@intel.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
7 years agoevent/sw: add queue-to-port stats
Harry van Haaren [Thu, 11 May 2017 09:56:26 +0000 (10:56 +0100)]
event/sw: add queue-to-port stats

This commit adds a new statistic to the SW eventdev PMD.
The statistic shows how many packets were sent from a
queue to a port. This provides information on how traffic
from a specific queue is being load-balanced to worker cores.

Note that these numbers should be compared across all queue
stages - the load-balancing does not try to perfectly share
each queue's traffic, rather it balances the overall traffic
from all queues to the ports.

The statistic is printed from the rte_eventdev_dump() function,
as well as being made available via the xstats API.

Unit tests have been updated to expect more per-queue statistics,
and the correctness of counts and counts after reset is verified.

Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
7 years agoip_frag: handle MTU sizes not aligned to 8 bytes
Allain Legacy [Tue, 14 Mar 2017 15:14:47 +0000 (11:14 -0400)]
ip_frag: handle MTU sizes not aligned to 8 bytes

The rte_ipv4_fragment_packet API expects that the link/interface MTU value
passed in be divisible by 8 bytes.  Given the name of the parameter is
"mtu" rather than "frag_size" it is not necessarily the case that it will
be divisible by 8.  An MTU of 1500 happens to produce a max fragment size
of 1480 (1500 - sizeof(ipv4_hdr)) which is divisible by 8 but other MTU
values such as 1600 or 9000 do not produce values that are divisible by 8.

Unfortunately, the API checks that the frag_size value produced is
divisible by 8 with a call to RTE_ASSERT which is only enabled when the
RTE_LOG_LEVEL >= RTE_LOG_DEBUG.  In cases where the log level is set
normally the code silently continues and produces IP fragments that have
invalid fragment offset values.

An application may not have control over what MTU a user selects and rather
than have each application adjust the MTU to pass a suitable value to the
fragmentation API this change modifies the fragmentation API to handle
cases where the "mtu" argument is not divisible by 8 and automatically
adjust the internal "frag_size".

Signed-off-by: Allain Legacy <allain.legacy@windriver.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
7 years agoip_frag: free mbufs on reassembly table destroy
Dahir Osman [Mon, 5 Jun 2017 15:49:01 +0000 (11:49 -0400)]
ip_frag: free mbufs on reassembly table destroy

The rte_ip_frag_table_destroy procedure simply releases the memory for the
table without freeing the packet buffers that may be referenced in the hash
table for in-flight or incomplete packet reassembly operations.  To prevent
leaked mbufs go through the list of fragments and free each one
individually.

Fixes: 416707812c03 ("ip_frag: refactor reassembly code into a proper library")
Cc: stable@dpdk.org
Reported-by: Matt Peters <matt.peters@windriver.com>
Signed-off-by: Allain Legacy <allain.legacy@windriver.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
7 years agotest/bonding: remove socket id check
Pablo de Lara [Wed, 21 Jun 2017 05:07:33 +0000 (06:07 +0100)]
test/bonding: remove socket id check

When creating a virtual pmd to test link bonding,
the socket id was checked, if it was in the range
of available sockets.
This check is unnecessary, as the socket specified
might not have memory anyway, so it will fail
at memory allocation.

Therefore, the best solution is to remove this check.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agonet/bonding: remove socket id check
Pablo de Lara [Wed, 21 Jun 2017 05:07:32 +0000 (06:07 +0100)]
net/bonding: remove socket id check

Socket id parsed from the user was checked
if it was in the range of available sockets.
This check is unnecessary, as the socket specified
might not have memory anyway, so it will fail
at memory allocation.

Therefore, the best solution is to remove this check.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agocrypto/scheduler: remove socket id check
Pablo de Lara [Wed, 21 Jun 2017 05:07:31 +0000 (06:07 +0100)]
crypto/scheduler: remove socket id check

Socket id parsed from the user was checked
if it was in the range of available sockets.
This check is unnecessary, as the socket specified
might not have memory anyway, so it will fail
at memory allocation.

Therefore, the best solution is to remove this check.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agocryptodev: remove socket id check
Pablo de Lara [Wed, 21 Jun 2017 05:07:30 +0000 (06:07 +0100)]
cryptodev: remove socket id check

Socket id parsed from the user was checked
if it was in the range of available sockets.
This check is unnecessary, as the socket specified
might not have memory anyway, so it will fail
at memory allocation.

Therefore, the best solution is to remove this check.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
7 years agoapp/testpmd: always build VF and MACsec functions
Thomas Monjalon [Thu, 15 Jun 2017 10:37:20 +0000 (12:37 +0200)]
app/testpmd: always build VF and MACsec functions

These functions are supported only on ixgbe.
However, they should appear in the help and returns an error
if the function is not supported or not enabled.

Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>
7 years agolpm: fix index of tbl8
Wei Dai [Mon, 19 Jun 2017 04:14:38 +0000 (12:14 +0800)]
lpm: fix index of tbl8

From v20 to v1604, number of tbl8 can be up to 1<<24,
(uint8_t) or (uint16_t) may truncate the number of
index of tlb8 in v1604 and cause wrong number.

Fixes: dc81ebbacaeb ("lpm: extend IPv4 next hop field")
Cc: stable@dpdk.org
Signed-off-by: Wei Dai <wei.dai@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
7 years agovhost: log error for badly negotiated features
Dariusz Stojaczyk [Fri, 16 Jun 2017 14:32:05 +0000 (16:32 +0200)]
vhost: log error for badly negotiated features

Since vhost_user_set_features failure is not handled in any way, a
single error log has been added to at least to let the user know that
something has gone wrong.

Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
7 years agonet/virtio: zero the whole memory zone
Tiwei Bie [Mon, 12 Jun 2017 04:34:30 +0000 (12:34 +0800)]
net/virtio: zero the whole memory zone

Zero the whole memory zone instead of the first few bytes.

Fixes: c1f86306a026 ("virtio: add new driver")
Cc: stable@dpdk.org
Signed-off-by: Tiwei Bie <tiwei.bie@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
7 years agovhost: fix crash on NUMA
Yuanhan Liu [Fri, 2 Jun 2017 00:14:46 +0000 (08:14 +0800)]
vhost: fix crash on NUMA

The queue allocation was changed, from allocating one queue-pair at a
time to one queue at a time. Most of the changes have been done, but
just with one being missed: the size of copying the old queue is still
based on queue-pair at numa_realloc(), which leads to overwritten issue.
As a result, crash may happen.

Fix it by specifying the right copy size. Also, the net queue macros
are not used any more. Remove them.

Fixes: ab4d7b9f1afc ("vhost: turn queue pair to vring")
Cc: stable@dpdk.org
Reported-by: Ciara Loftus <ciara.loftus@intel.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Jens Freimann <jfreiman@redhat.com>
Tested-by: Ciara Loftus <ciara.loftus@intel.com>
7 years agovhost: access VhostUsrMsg via packed struct
Daniel Verkamp [Fri, 26 May 2017 11:59:15 +0000 (13:59 +0200)]
vhost: access VhostUsrMsg via packed struct

Accessing fields of a packed struct through unaligned pointers is
undefined behavior. Instead of passing pointers to particular fields,
a pointer to the root struct should be used. This patch does exactly
that.

Signed-off-by: Daniel Verkamp <daniel.verkamp@intel.com>
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
7 years agovhost: fix guest pages memory leak
Dariusz Stojaczyk [Fri, 26 May 2017 11:59:14 +0000 (13:59 +0200)]
vhost: fix guest pages memory leak

This patch fixes a memory leak.
virtio_net::guest_pages is allocated in vhost_setup_mem_table(),
reallocated in add_one_guest_page(), but never freed.

Fixes: e246896178e6 ("vhost: get guest/host physical address mappings")
Cc: stable@dpdk.org
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-by: Jens Freimann <jfreiman@redhat.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
7 years agovhost: fix malloc size too small
Dariusz Stojaczyk [Fri, 26 May 2017 11:59:13 +0000 (13:59 +0200)]
vhost: fix malloc size too small

Amount of allocated memory was too small, causing buffer overflow.

Fixes: eb32247457fe ("vhost: export guest memory regions")
Cc: stable@dpdk.org
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-by: Jens Freimann <jfreiman@redhat.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
7 years agovhost: support Rx queue count request
Zhihong Wang [Fri, 26 May 2017 17:18:02 +0000 (13:18 -0400)]
vhost: support Rx queue count request

This patch implements the ops rx_queue_count for vhost PMD by adding
a helper function rte_vhost_rx_queue_count in vhost lib.

The ops rx_queue_count gets vhost RX queue avail count and helps to
understand the queue fill level.

Signed-off-by: Zhihong Wang <zhihong.wang@intel.com>
Acked-by: Ciara Loftus <ciara.loftus@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
7 years agovhost: check allocation of guest pages
Jens Freimann [Thu, 11 May 2017 15:25:26 +0000 (17:25 +0200)]
vhost: check allocation of guest pages

When we try to allocate guest pages we need to check the return value of
malloc(). Print an error message and return when it fails.

Signed-off-by: Jens Freimann <jfreiman@redhat.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
7 years agomem: support page locking on FreeBSD
Thomas Monjalon [Thu, 15 Jun 2017 17:37:00 +0000 (19:37 +0200)]
mem: support page locking on FreeBSD

The function rte_mem_lock_page() was added for Linux only.
The file eal_common_memory.c is a better place to make it
available in FreeBSD also.

The issue is seen when trying to compile bnxt on FreeBSD:
bnxt_hwrm.c: undefined reference to `rte_mem_lock_page'

Fixes: 3097de6e6bfb ("mem: get physical address of any pointer")

Reported-by: Fangfang Wei <fangfangx.wei@intel.com>
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
7 years agoethdev: tidy up endianness handling in flow API
Adrien Mazarguil [Thu, 15 Jun 2017 15:48:59 +0000 (17:48 +0200)]
ethdev: tidy up endianness handling in flow API

The flow API defines several structures whose fields must be specified in
network order. This commit documents them using explicit type names and
related endianness conversion macros.

No ABI change.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
7 years agoeal: add static endianness conversion macros
Adrien Mazarguil [Thu, 15 Jun 2017 15:48:58 +0000 (17:48 +0200)]
eal: add static endianness conversion macros

These macros resolve to constant expressions that allow developers to
perform endianness conversion on static/const objects, even outside of
function scope as they do not translate to function calls.

This is most useful for static initializers and constant values (whenever
it has to be performed at compilation time). Run-time endianness conversion
of variable values should keep using rte_*_to_*() calls for best
performance.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
7 years agoeal: introduce big and little endian types
Nelio Laranjeiro [Thu, 15 Jun 2017 15:48:57 +0000 (17:48 +0200)]
eal: introduce big and little endian types

This commit introduces new rte_{le,be}{16,32,64}_t types and updates
rte_{le,be,cpu}_to_{le,be,cpu}_*() accordingly.

These types are added for documentation purposes, mainly to clarify the
byte ordering to use for storage when not CPU order. Doing so eliminates
uncertainty and conversion mistakes.

Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
7 years agoapp/testpmd: fix build with bypass without ixgbe
Thomas Monjalon [Thu, 15 Jun 2017 09:25:06 +0000 (11:25 +0200)]
app/testpmd: fix build with bypass without ixgbe

When ixgbe bypass is not explicitly disabled while ixgbe is disabled:
app/test-pmd/testpmd.c:304:27: error:
‘RTE_PMD_IXGBE_BYPASS_TMT_OFF’ undeclared here

The ixgbe bypass feature is meaningful only if ixgbe is enabled.
So we need to check both.

A best fix will be to enable bypass always and remove this option.

Fixes: e261265e42a1 ("ethdev: move bypass functions to ixgbe PMD")

Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
7 years agoapp/testpmd: fix build without ixgbe
Thomas Monjalon [Thu, 15 Jun 2017 09:34:16 +0000 (11:34 +0200)]
app/testpmd: fix build without ixgbe

cmd_set_vf_rxmode_parsed() was defined only in the build context
of RTE_LIBRTE_IXGBE_PMD:
app/test-pmd/cmdline.c:13817:27: error: ‘cmd_set_vf_rxmode’ undeclared here

Fixes: 4cfe399f6550 ("net/bnxt: support to set VF rxmode")

Reported-by: Yongseok Koh <yskoh@mellanox.com>
Reported-by: Jan Viktorin <viktorin@rehivetech.com>
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
7 years agofix typos using codespell utility
Jerin Jacob [Wed, 7 Jun 2017 05:05:06 +0000 (10:35 +0530)]
fix typos using codespell utility

Fixing typos across dpdk source code using codespell utility.
Skipped the ethdev driver's base code fixes to keep the base
code intact.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: John McNamara <john.mcnamara@intel.com>
7 years agopdump: remove unnecessary header include
Reshma Pattan [Mon, 12 Jun 2017 09:46:11 +0000 (10:46 +0100)]
pdump: remove unnecessary header include

Missed to remove unnecessary header file rte_pci.h.
Removed it now.

Fixes: bb900072ffaa ("pdump: revert PCI device name conversion")

Signed-off-by: Reshma Pattan <reshma.pattan@intel.com>
7 years agoethdev: add isolated mode to flow API
Adrien Mazarguil [Wed, 14 Jun 2017 14:48:51 +0000 (16:48 +0200)]
ethdev: add isolated mode to flow API

Isolated mode can be requested by applications on individual ports to avoid
ingress traffic outside of the flow rules they define.

Besides making ingress more deterministic, it allows PMDs to safely reuse
resources otherwise assigned to handle the remaining traffic, such as
global RSS configuration settings, VLAN filters, MAC address entries,
legacy filter API rules and so on in order to expand the set of possible
flow rule types.

To minimize code complexity, PMDs implementing this mode may provide
partial (or even no) support for flow rules when not enabled (e.g. no
priorities, no RSS action). Applications written to use the flow API are
therefore encouraged to enable it.

Once effective, leaving isolated mode may not be possible depending on PMD
implementation.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
7 years agohash: fix icc build
Ferruh Yigit [Tue, 13 Jun 2017 16:42:12 +0000 (17:42 +0100)]
hash: fix icc build

build error with icc version 17.0.4 (gcc version 7.0.0 compatibility):

In file included from .../dpdk/lib/librte_hash/rte_fbk_hash.h(59),
                 from .../dpdk/lib/librte_hash/rte_fbk_hash.c(54):
.../dpdk/x86_64-native-linuxapp-icc/include/rte_hash_crc.h(480):
 error #1292: unknown attribute "fallthrough"
                __attribute__ ((fallthrough));
                                ^

In file included from .../dpdk/lib/librte_hash/rte_fbk_hash.h(59),
                 from .../dpdk/lib/librte_hash/rte_fbk_hash.c(54):
.../dpdk/x86_64-native-linuxapp-icc/include/rte_hash_crc.h(486):
 error #1292: unknown attribute "fallthrough"
                __attribute__ ((fallthrough));
                                ^
This code patch hit when gcc > 7 installed and ICC doesn't recognize
fallthrough attribute.

Fixed by disabling code when compiled with ICC.

Fixes: 3dfb9facb055 ("lib: add switch fall-through comments")

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agonet/mlx5: fix build with gcc 7.1
Ferruh Yigit [Tue, 13 Jun 2017 16:42:11 +0000 (17:42 +0100)]
net/mlx5: fix build with gcc 7.1

build error:
.../dpdk/drivers/net/mlx5/mlx5_fdir.c:
  In function ‘fdir_filter_to_flow_desc’:
.../dpdk/drivers/net/mlx5/mlx5_fdir.c:146:18:
 error: this statement may fall through [-Werror=implicit-fallthrough=]
   desc->dst_port = fdir_filter->input.flow.udp4_flow.dst_port;
   ~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
.../dpdk/drivers/net/mlx5/mlx5_fdir.c:147:2: note: here
  case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
  ^~~~

Fixed by adding fallthrough comment to the code.

Fixes: 76f5c99e6840 ("mlx5: support flow director")
Cc: stable@dpdk.org
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
7 years agonet/enic: fix build with gcc 7.1
Ferruh Yigit [Tue, 13 Jun 2017 16:42:10 +0000 (17:42 +0100)]
net/enic: fix build with gcc 7.1

build error:

.../dpdk/drivers/net/enic/base/vnic_dev.c:
  In function ‘vnic_dev_get_mac_addr’:
.../dpdk/drivers/net/enic/base/vnic_dev.c:470:12:
  error: ‘a0’ is used uninitialized in this function
  [-Werror=uninitialized]
  args[0] = *a0;
            ^~~
...dpdk/drivers/net/enic/base/vnic_dev.c:
  In function ‘vnic_dev_classifier’:
...dpdk/drivers/net/enic/base/vnic_dev.c:471:12:
  error: ‘a1’ may be used uninitialized in this function
  [-Werror=maybe-uninitialized]
  args[1] = *a1;
            ^~~
Fixed by providing initial values.

Fixes: 9913fbb91df0 ("enic/base: common code")
Cc: stable@dpdk.org
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agonet/i40e: fix memset size
Ferruh Yigit [Tue, 13 Jun 2017 16:42:09 +0000 (17:42 +0100)]
net/i40e: fix memset size

This causes build error with gcc 7.1.1 :

...dpdk/drivers/net/i40e/i40e_flow.c:2357:2:
error: ‘memset’ used with length equal to number of elements without
       multiplication by element size [-Werror=memset-elt-size]
  memset(off_arr, 0, I40E_MAX_FLXPLD_FIED);
  ^~~~~~

...dpdk/drivers/net/i40e/i40e_flow.c:2358:2:
error: ‘memset’ used with length equal to number of elements without
       multiplication by element size [-Werror=memset-elt-size]
  memset(len_arr, 0, I40E_MAX_FLXPLD_FIED);
  ^~~~~~

Fixed by providing correct size to memset.

Fixes: 6ced3dd72f5f ("net/i40e: support flexible payload parsing for FDIR")

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
7 years agokni: fix build with gcc 7.1
Ferruh Yigit [Tue, 13 Jun 2017 16:42:08 +0000 (17:42 +0100)]
kni: fix build with gcc 7.1

build error:
.../dpdk/build/build/lib/librte_eal/linuxapp/kni/igb_main.c:
  In function ‘igb_kni_probe’:
.../dpdk/build/build/lib/librte_eal/linuxapp/kni/igb_main.c:2483:30:
  error: ‘%d’ directive output may be truncated writing between 1 and 5
  bytes into a region of size between 0 and 11
  [-Werror=format-truncation=]
        "%d.%d, 0x%08x, %d.%d.%d",
                              ^~
.../dpdk/build/build/lib/librte_eal/linuxapp/kni/igb_main.c:2483:8:
  note: directive argument in the range [0, 65535]
        "%d.%d, 0x%08x, %d.%d.%d",
        ^~~~~~~~~~~~~~~~~~~~~~~~~
.../dpdk/build/build/lib/librte_eal/linuxapp/kni/igb_main.c:2481:4:
  note: ‘snprintf’ output between 23 and 43 bytes into a destination of
  size 32
    snprintf(adapter->fw_version,
    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        sizeof(adapter->fw_version),
        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        "%d.%d, 0x%08x, %d.%d.%d",
        ~~~~~~~~~~~~~~~~~~~~~~~~~~
        fw.eep_major, fw.eep_minor, fw.etrack_id,
        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
        fw.or_major, fw.or_build, fw.or_patch);
        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Fixed by increasing buffer size to 43 as suggested in compiler log.

Fixes: b9ee370557f1 ("kni: update kernel driver ethtool baseline")
Cc: stable@dpdk.org
Reported-by: Nirmoy Das <ndas@suse.de>
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Reviewed-by: Markos Chandras <mchandras@suse.de>
7 years agonet/thunderx: manage PCI device mapping for SQS VFs
Jerin Jacob [Fri, 9 Jun 2017 10:27:46 +0000 (15:57 +0530)]
net/thunderx: manage PCI device mapping for SQS VFs

Since the commit e84ad157b7bc ("pci: unmap resources if probe fails"),
EAL unmaps the PCI device if ethdev probe returns positive or
negative value.

nicvf thunderx PMD needs special treatment for Secondary queue set(SQS)
PCIe VF devices, where, it expects to not unmap or free the memory
without registering the ethdev subsystem.

Enable the same behavior by using RTE_PCI_DRV_KEEP_MAPPED_RES
PCI driver flag.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agoeal/pci: introduce a PCI driver flag
Jerin Jacob [Fri, 9 Jun 2017 10:27:45 +0000 (15:57 +0530)]
eal/pci: introduce a PCI driver flag

Some ethdev devices like nicvf thunderx PMD need special treatment for
Secondary queue set(SQS) PCIe VF devices, where, it expects to not unmap
or free the memory without registering the ethdev subsystem.

Introducing a new RTE_PCI_DRV_KEEP_MAPPED_RES
PCI driver flag to request PCI subsystem to not unmap the mapped PCI
resources(PCI BAR address) if unsupported device detected.

Suggested-by: Ferruh Yigit <ferruh.yigit@intel.com>
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agodrivers/net: document missing speed capabilities feature
Ferruh Yigit [Mon, 15 May 2017 12:30:46 +0000 (13:30 +0100)]
drivers/net: document missing speed capabilities feature

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: John McNamara <john.mcnamara@intel.com>
7 years agoethdev: remove driver name from device private data
Ferruh Yigit [Mon, 12 Jun 2017 15:25:12 +0000 (16:25 +0100)]
ethdev: remove driver name from device private data

rte_driver->name has the driver name and all physical and virtual
devices has access to it.

Previously it was not possible for virtual ethernet devices to access
rte_driver->name field (because eth_dev used to keep only pci_dev),
and it was required to save driver name in the device private struct.

After re-works on bus and vdev, it is possible for all bus types to
access rte_driver.

It is able to remove the driver name from ethdev device private data and
use eth_dev->device->driver->name.

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Acked-by: Jan Blunck <jblunck@infradead.org>
7 years agonet/ring: use EAL APIs in PMD specific API
Ferruh Yigit [Mon, 12 Jun 2017 15:25:11 +0000 (16:25 +0100)]
net/ring: use EAL APIs in PMD specific API

When ring PMD created via PMD specific API instead of EAL abstraction
it is missing the virtual device creation done by EAL vdev.

And this makes eth_dev unusable exact same as other PMDs used, because
of some missing fields, like rte_device->name.

Now API calls EAL APIs to create ring PMDs.

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
7 years agonet/ring: set ethernet device field
Ferruh Yigit [Mon, 12 Jun 2017 15:25:10 +0000 (16:25 +0100)]
net/ring: set ethernet device field

The eth_dev->device link was missing for ring PMD, adding it.

This is to generalize rte_device access from eth_dev.

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agonet/szedata2: add more supported firmwares
Matej Vido [Mon, 12 Jun 2017 12:03:22 +0000 (14:03 +0200)]
net/szedata2: add more supported firmwares

Add IBUF and OBUF offsets definitions for new firmwares.

Signed-off-by: Matej Vido <vido@cesnet.cz>
7 years agonet/szedata2: move ibuf and obuf to specific header
Matej Vido [Mon, 12 Jun 2017 12:03:21 +0000 (14:03 +0200)]
net/szedata2: move ibuf and obuf to specific header

Signed-off-by: Matej Vido <vido@cesnet.cz>
7 years agonet/szedata2: refactor ibuf and obuf address definition
Matej Vido [Mon, 12 Jun 2017 12:03:20 +0000 (14:03 +0200)]
net/szedata2: refactor ibuf and obuf address definition

This is to prepare for firmwares with multiple ibufs and obufs.
Ibufs and obufs are the modules in FPGA firmware implementing
the Ethernet port.
There is one ibuf+obuf per Ethernet port.
The cards and firmwares allow one physical port to be one Ethernet
port or split into more Ethernet ports, e.g. one 100GE physical
port can be one Ethernet port of 100GE or split into ten Ethernet
ports of 10GE.
All DMA queues in the device are shared between all Ethernet ports.
Offsets of ibufs and obufs are defined in array.
Functions which operate on ibufs and obufs iterate over this array.

Signed-off-by: Matej Vido <vido@cesnet.cz>
7 years agonet/szedata2: refactor ibuf and obuf read and write
Matej Vido [Mon, 12 Jun 2017 12:03:19 +0000 (14:03 +0200)]
net/szedata2: refactor ibuf and obuf read and write

Remove unused read and write functions.
Use rte_read*, rte_write* functions to access ibuf and obuf
address space.

Signed-off-by: Matej Vido <vido@cesnet.cz>
7 years agonet/szedata2: refactor ibuf and obuf names
Matej Vido [Mon, 12 Jun 2017 12:03:18 +0000 (14:03 +0200)]
net/szedata2: refactor ibuf and obuf names

Prefix "cgmii" is removed because it is too specific.
There are different ibuf/obuf modules in different firmwares
but the address space definition is the same.
This patch makes the name general.

Signed-off-by: Matej Vido <vido@cesnet.cz>
7 years agonet/igb: flush all the filter
Wei Zhao [Mon, 12 Jun 2017 06:48:28 +0000 (14:48 +0800)]
net/igb: flush all the filter

This patch adds a function to flush all the fliter list
and filter rule on a port.

Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
7 years agonet/igb: destroy consistent filter
Wei Zhao [Mon, 12 Jun 2017 06:48:27 +0000 (14:48 +0800)]
net/igb: destroy consistent filter

This patch adds a function to destroy the flow fliter.

Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
7 years agonet/igb: create consistent filter
Wei Zhao [Mon, 12 Jun 2017 06:48:26 +0000 (14:48 +0800)]
net/igb: create consistent filter

This patch adds a function to create the flow directory filter.

Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
7 years agonet/igb: parse flow API flex filter
Wei Zhao [Mon, 12 Jun 2017 06:48:25 +0000 (14:48 +0800)]
net/igb: parse flow API flex filter

check if the rule is a flex byte rule, and get the flex info.

Signed-off-by: Wei Zhao <wei.zhao1@intel.com>