dpdk.git
9 years agohash: replace libc memcmp with optimized functions for arm64
Jerin Jacob [Mon, 23 Nov 2015 18:45:35 +0000 (00:15 +0530)]
hash: replace libc memcmp with optimized functions for arm64

The following measurements shows improvement over the default
libc memcmp function

Length(B) by X% over libc memcmp
  16   149.57%
  32   122.7%
  48   104.96%
  64   98.21%
  80   93.75%
  96   90.55%
 112   110.48%
 128   137.24%

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
9 years agonet: fix build with gcc 4.4.7 and strict aliasing
Daniel Mrzyglod [Tue, 24 Nov 2015 16:31:18 +0000 (17:31 +0100)]
net: fix build with gcc 4.4.7 and strict aliasing

This fix is for IPv6 checksum offload error on RHEL65.
Any optimalisation above -O0 provide error in IPv6 checksum
flag "-fstrict-aliasing" is default for optimalisation above -O0.

Step 1: testpmd -c 0x6 -n 4  -- -i --portmask=0x3 --disable-hw-vlan
                --enable-rx-cksum --crc-strip --txqflags=0
Step 2: settings and start
           set verbose 1
           set fwd csum
           start
Step 3: send scapy with bad checksum IPv6/TCP packet
           Ether(src="52:00:00:00:00:00",
  dst="90:e2:ba:4a:33:5d")/IPv6(src="::1")/TCP(chksum=0xf)/("X"*46)
Step 4: Received packets:
           RESULTS: IPv6/TCP': ['0xd41'] or other unexpected.
  EXPECTED RESULTS: IPv6/TCP': ['0x9f5e']

Fixes: 2b039d5f20a3 ("net: fix build with gcc 4.4.7 and strict aliasing")

Signed-off-by: Daniel Mrzyglod <danielx.t.mrzyglod@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
9 years agocryptodev: mark experimental state
Thomas Monjalon [Wed, 25 Nov 2015 17:27:21 +0000 (18:27 +0100)]
cryptodev: mark experimental state

The crypto API is in an early state.
It requires more discussions and experiments to declare it stable,
as discussed in http://dpdk.org/ml/archives/dev/2015-November/028634.html

A documentation section will be required in the guides.

Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
9 years agoexamples/l2fwd-crypto: add sample application
Declan Doherty [Wed, 25 Nov 2015 13:25:17 +0000 (13:25 +0000)]
examples/l2fwd-crypto: add sample application

This patch creates a new sample applicaiton based off the l2fwd
application which performs specified crypto operations on IP packet
payloads which are forwarding.

Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
9 years agoapp/test: add cryptodev unit and performance tests
Declan Doherty [Wed, 25 Nov 2015 13:25:16 +0000 (13:25 +0000)]
app/test: add cryptodev unit and performance tests

unit tests are run by using cryptodev_qat_autotest or
cryptodev_aesni_autotest from the test apps interactive console.

performance tests are run by using the cryptodev_qat_perftest or
cryptodev_aesni_mb_perftest command from the test apps interactive
console.

If you which to run the tests on a QAT device there must be one
bound to igb_uio kernel driver.

Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Signed-off-by: John Griffin <john.griffin@intel.com>
Signed-off-by: Des O Dea <des.j.o.dea@intel.com>
Signed-off-by: Fiona Trahe <fiona.trahe@intel.com>
Acked-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
9 years agoaesni_mb: add driver for multi buffer based crypto
Declan Doherty [Wed, 25 Nov 2015 13:25:15 +0000 (13:25 +0000)]
aesni_mb: add driver for multi buffer based crypto

This patch provides the initial implementation of the AES-NI multi-buffer
based crypto poll mode driver using DPDK's new cryptodev framework.

This PMD is dependent on Intel's multibuffer library, see the whitepaper
"Fast Multi-buffer IPsec Implementations on Intel® Architecture
Processors", see ref 1 for details on the library's design and ref 2 to
download the library itself. This initial implementation is limited to
supporting the chained operations of "hash then cipher" or "cipher then
hash" for the following cipher and hash algorithms:

Cipher algorithms:
  - RTE_CRYPTO_CIPHER_AES_CBC (with 128-bit, 192-bit and 256-bit keys supported)

Authentication algorithms:
  - RTE_CRYPTO_AUTH_SHA1_HMAC
  - RTE_CRYPTO_AUTH_SHA256_HMAC
  - RTE_CRYPTO_AUTH_SHA512_HMAC
  - RTE_CRYPTO_AUTH_AES_XCBC_MAC

Important Note:
Due to the fact that the multi-buffer library is designed for
accelerating IPsec crypto operation, the digest's generated for the HMAC
functions are truncated to lengths specified by IPsec RFC's, ie RFC2404
for using HMAC-SHA-1 with IPsec specifies that the digest is truncate
from 20 to 12 bytes.

Build instructions:
To build DPDK with the AESNI_MB_PMD the user is required to download
(ref 2) and compile the multi-buffer library on there system before
building DPDK. The environmental variable AESNI_MULTI_BUFFER_LIB_PATH
must be exported with the path where you extracted and built the multi
buffer library and finally set CONFIG_RTE_LIBRTE_PMD_AESNI_MB=y in
config/common_linuxapp.

Current status: It's doesn't support crypto operation
across chained mbufs, or cipher only or hash only operations.

ref 1:
https://www-ssl.intel.com/content/www/us/en/intelligent-systems/intel-technology/fast-multi-buffer-ipsec-implementations-ia-processors-p

ref 2: https://downloadcenter.intel.com/download/22972

Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
9 years agoqat: add driver for QuickAssist devices
Declan Doherty [Wed, 25 Nov 2015 13:25:14 +0000 (13:25 +0000)]
qat: add driver for QuickAssist devices

This patch adds a PMD for the Intel Quick Assist Technology DH895xxC
hardware accelerator.

This patch depends on a QAT PF driver for device initialization. See
the file docs/guides/cryptodevs/qat.rst for configuration details

This patch supports a limited subset of QAT device functionality,
currently supporting chaining of cipher and hash operations for the
following algorithmsd:

Cipher algorithms:
  - RTE_CRYPTO_CIPHER_AES_CBC (with 128-bit, 192-bit and 256-bit keys supported)

Hash algorithms:
  - RTE_CRYPTO_AUTH_SHA1_HMAC
  - RTE_CRYPTO_AUTH_SHA256_HMAC
  - RTE_CRYPTO_AUTH_SHA512_HMAC
  - RTE_CRYPTO_AUTH_AES_XCBC_MAC

Some limitation on this patchset which shall be contributed in a
subsequent release:
 - Chained mbufs are not supported.
 - Hash only is not supported.
 - Cipher only is not supported.
 - Only in-place is currently supported (destination address is
   the same as source address).
 - Only supports session-oriented API implementation (session-less
   APIs are not supported).

Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Signed-off-by: John Griffin <john.griffin@intel.com>
Signed-off-by: Des O Dea <des.j.o.dea@intel.com>
Signed-off-by: Fiona Trahe <fiona.trahe@intel.com>
Acked-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
9 years agombuf_offload: introduce library to attach offloads to mbuf
Declan Doherty [Wed, 25 Nov 2015 13:25:13 +0000 (13:25 +0000)]
mbuf_offload: introduce library to attach offloads to mbuf

This library add support for adding a chain of offload operations to a
mbuf. It contains the definition of the rte_mbuf_offload structure as
well as helper functions for attaching  offloads to mbufs and a mempool
management functions.

This initial implementation supports attaching multiple offload
operations to a single mbuf, but only a single offload operation of a
specific type can be attach to that mbuf.

Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
9 years agocryptodev: introduce API and framework for crypto devices
Declan Doherty [Wed, 25 Nov 2015 13:25:12 +0000 (13:25 +0000)]
cryptodev: introduce API and framework for crypto devices

This patch contains the initial proposed APIs and device framework for
integrating crypto packet processing into DPDK.

features include:
 - Crypto device configuration / management APIs
 - Definitions of supported cipher algorithms and operations.
 - Definitions of supported hash/authentication algorithms and
   operations.
 - Crypto session management APIs
 - Crypto operation data structures and APIs allocation of crypto
   operation structure used to specify the crypto operations to
   be performed  on a particular mbuf.
 - Extension of mbuf to contain crypto operation data pointer and
   extra flags.
 - Burst enqueue / dequeue APIs for processing of crypto operations.

Signed-off-by: Des O Dea <des.j.o.dea@intel.com>
Signed-off-by: John Griffin <john.griffin@intel.com>
Signed-off-by: Fiona Trahe <fiona.trahe@intel.com>
Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
9 years agombuf: get physical address of data
Declan Doherty [Wed, 25 Nov 2015 13:25:11 +0000 (13:25 +0000)]
mbuf: get physical address of data

Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
9 years agoeal: add packing and alignment macros
Declan Doherty [Wed, 25 Nov 2015 13:25:10 +0000 (13:25 +0000)]
eal: add packing and alignment macros

Adding a new macro for specifying __aligned__ attribute, and updating the
current __rte_cache_aligned macro to use it.

Also adding a new macro to specify the __packed__ attribute

Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
9 years agoethdev: add sanity checks to functions
Bruce Richardson [Tue, 24 Nov 2015 17:37:57 +0000 (17:37 +0000)]
ethdev: add sanity checks to functions

The functions rte_eth_rx_queue_count and rte_eth_descriptor_done are
supported by very few PMDs. Therefore, it is best to check for support
for the functions in the ethdev library, so as to avoid run-time crashes
at run-time if the application goes to use those APIs. Similarly, the
port parameter should also be checked for validity.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
9 years agoethdev: remove duplicated debug functions
Bruce Richardson [Tue, 24 Nov 2015 17:37:56 +0000 (17:37 +0000)]
ethdev: remove duplicated debug functions

The functions for rx/tx burst, for rx_queue_count and descriptor_done in
the ethdev library all had two copies of the code. One copy in
rte_ethdev.h was inlined for performance, while a second was in
rte_ethdev.c for debugging purposes only. We can eliminate the second
copy of the functions by moving the additional debug checks into the
copies of the functions in the header file. [Any compilation for
debugging at optimization level 0 will not inline the function so the
result should be same as when the function was in the .c file.]

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
9 years agoethdev: make error checking macros public
Declan Doherty [Wed, 25 Nov 2015 13:25:09 +0000 (13:25 +0000)]
ethdev: make error checking macros public

Move the function pointer and port id checking macros to rte_ethdev and
rte_dev header files, so that they can be used in the static inline
functions there. Also replace the RTE_LOG call within
RTE_PMD_DEBUG_TRACE so this macro can be built with the -pedantic flag

Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agoethdev: prefix internal error checking macros
Declan Doherty [Wed, 25 Nov 2015 13:25:08 +0000 (13:25 +0000)]
ethdev: prefix internal error checking macros

The macros to check that the function pointers and port ids are valid
for an ethdev are potentially useful to have in a common headers for
use with all PMDs. However, since they would then become externally
visible, we apply the RTE_ & RTE_ETH_ prefix to them as approtiate.

Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
9 years agoeal/linux: fix build with glibc < 2.12
Ferruh Yigit [Wed, 25 Nov 2015 11:13:55 +0000 (11:13 +0000)]
eal/linux: fix build with glibc < 2.12

pthread_setname_np() function added in glibc 2.12, using this function
in older glibc versions cause compile error:
error: implicit declaration of function "pthread_setname_np"

This patch adds "rte_thread_setname" macro and set it according
glibc >= 2.12 check, thread naming disabled for older glibc versions,
glibc versions that support "pthread_setname_np" will keep using this
function.

Fixes: 67b6d3039e9e ("eal: set name to threads")

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
9 years agosched: remove unused debug functions
Thomas Monjalon [Wed, 25 Nov 2015 00:45:41 +0000 (01:45 +0100)]
sched: remove unused debug functions

clang warns about unused functions when debug is enabled.

Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
9 years agosched: release enqueued mbufs when freeing port
Simon Kagstrom [Tue, 17 Nov 2015 08:14:48 +0000 (09:14 +0100)]
sched: release enqueued mbufs when freeing port

Otherwise mbufs will leak when the port is destroyed. The
rte_sched_port_qbase() and rte_sched_port_qsize() functions are used
in free now, so move them up.

Signed-off-by: Simon Kagstrom <simon.kagstrom@netinsight.net>
9 years agosched: allow more subports
Stephen Hemminger [Fri, 13 Nov 2015 17:58:36 +0000 (09:58 -0800)]
sched: allow more subports

Increase the number of possible subports per port to allow up to 16 bits.
It is still possible that this will require excessive RAM.

Although mbuf structure is changed, it is ABI compatiable since it
just expands existing sched part of structure to overlap pre-existing hole
in the hash element of structure.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
9 years agosched: fix coding style
Stephen Hemminger [Fri, 13 Nov 2015 17:58:35 +0000 (09:58 -0800)]
sched: fix coding style

Make rte_sched conform to kernel/DPDK coding style.
Fix missing whitespace and some of the excessively long lines.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
9 years agosched: enable SSE optimizations in config
Stephen Hemminger [Fri, 13 Nov 2015 17:58:34 +0000 (09:58 -0800)]
sched: enable SSE optimizations in config

Make the SSE optimizations visible as a normal config option.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
9 years agosched: cleanup defined constants
Stephen Hemminger [Fri, 13 Nov 2015 17:58:33 +0000 (09:58 -0800)]
sched: cleanup defined constants

Only use RTE_SCHED_PORT_N_GRINDERS from config.
Use RTE_BUILD_BUG_ON for errors.
The remaining implementation constants can be put together.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
9 years agosched: drop WRR conditional
Stephen Hemminger [Fri, 13 Nov 2015 17:58:32 +0000 (09:58 -0800)]
sched: drop WRR conditional

The RTE_SCHED_WRR value is always defined as 1.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
9 years agosched: remove enqueue debug
Stephen Hemminger [Fri, 13 Nov 2015 17:58:31 +0000 (09:58 -0800)]
sched: remove enqueue debug

Remove RTE_SCHED_ENQUEUE ifdef that was always being set.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
9 years agosched: drop credit check debug
Stephen Hemminger [Fri, 13 Nov 2015 17:58:30 +0000 (09:58 -0800)]
sched: drop credit check debug

Drop conditional code which was for debugging credit checks.
It is hard to maintain code with any additional #ifdef baggage.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
9 years agosched: make debugging configurable
Stephen Hemminger [Fri, 13 Nov 2015 17:58:29 +0000 (09:58 -0800)]
sched: make debugging configurable

All #ifdefs in code should be enabled/disabled via DPDK config
(or better yet removed all together).

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
9 years agosched: cleanup comments
Stephen Hemminger [Fri, 13 Nov 2015 17:58:28 +0000 (09:58 -0800)]
sched: cleanup comments

Break block comments that exceed common practice for line length.
Shorten wording for obvious things.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
9 years agosched: drop deprecated port hierarchy structure
Stephen Hemminger [Fri, 13 Nov 2015 17:58:27 +0000 (09:58 -0800)]
sched: drop deprecated port hierarchy structure

The actual port_hierarchy was deprecated and hidden in 2.1
so drop it from view in DPDK 2.2.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
9 years agoigb_uio: use existing PCI macros
Ferruh Yigit [Thu, 10 Sep 2015 16:01:48 +0000 (17:01 +0100)]
igb_uio: use existing PCI macros

To get pci_dev and vf number from dev, benefit from
existing macros in pci.h

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
[Thomas note: it breaks the old 2.6.33 support]

9 years agompipe: fix build as shared library
Zhigang Lu [Tue, 17 Nov 2015 07:19:01 +0000 (15:19 +0800)]
mpipe: fix build as shared library

Without .map file, compiling error occurs when
CONFIG_RTE_BUILD_SHARED_LIB is enabled.

Reported-by: Guo Xin <gxin@ezchip.com>
Signed-off-by: Zhigang Lu <zlu@ezchip.com>
9 years agokni: fix build on Ubuntu 12.04 with kernel 3.13.0-30
Pablo de Lara [Tue, 24 Nov 2015 16:59:41 +0000 (16:59 +0000)]
kni: fix build on Ubuntu 12.04 with kernel 3.13.0-30

Fixes following error when Ubuntu 12.04 uses kernel 3.13.0-30-generic,
since skb_set_hash() is implemented in the kernel from 3.13.0-30,
which is declared as UBUNTU_KERNEL_VERSION(3,13,0,30,0) and not
UBUNTU_KERNEL_VERSION(3,13,0,30,54)

In file included
from /usr/src/linux-headers-3.13.0-30-generic/include/linux/if_ether.h:23:0,
from /tmp/dpdk/lib/librte_eal/linuxapp/kni/ethtool/igb/e1000_osdep.h:39,
from /tmp/dpdk/lib/librte_eal/linuxapp/kni/ethtool/igb/e1000_hw.h:31,
from /tmp/dpdk/lib/librte_eal/linuxapp/kni/ethtool/igb/e1000_api.h:31,
from /tmp/dpdk/lib/librte_eal/linuxapp/kni/ethtool/igb/e1000_mbx.h:31,
from /tmp/dpdk/x86_64-native-linuxapp-gcc/build/lib/librte_eal/linuxapp/kni/e1000_mbx.c:28:
/usr/src/linux-headers-3.13.0-30-generic/include/linux/skbuff.h:740:1:
note: previous definition of ‘skb_set_hash’ was here
 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
 ^

Fixes: e88c3b0a ("kni: fix build on Ubuntu 12.04.5")

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
9 years agovhost: fix kernel module insertion
Xiaobo Chi [Wed, 11 Nov 2015 06:57:23 +0000 (14:57 +0800)]
vhost: fix kernel module insertion

Problem:if I firstly insert my kmod_test.ko, then insert eventfd_link.ko,
error will happen with hint "Device or resource busy". This is because
the default minor device number, 0, has been occupied by my kmod_test.ko .

root@distro:~/test$ lsmod
Module                  Size  Used by
kmod_test                927  0
vboxsf                 35930  4
vboxguest             222130  1 vboxsf
microcode              10315  0
autofs4                25051  0
root@distro:~/test$ insmod ./eventfd_link.ko
insmod: ERROR: could not insert module ./eventfd_link.ko: Device or
resource busy

Explanation: For miscdevices, the major device_no is same, so the minor
device_no should be set to ditinguish different misc devices;  if not set
the minor, it may fail while insmod due to the default minor value, 0, has
been used by other miscdevice. MISC_DYNAMIC_MINOR means to let Linux
kernel dynamically assign one minor devide number while loading.

Signed-off-by: Xiaobo Chi <xiaobo.chi@nokia.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
9 years agovhost: fix enabling vring per queue
Victor Kaplansky [Tue, 24 Nov 2015 07:25:35 +0000 (15:25 +0800)]
vhost: fix enabling vring per queue

The VHOST_USER_SET_VRING_ENABLE request was sent for each queue-pair.
However, it's changed to be sent per queue in the queue-pair by QEMU
commit dc3db6ad ("vhost-user: start/stop all rings"). The change
is reasonable, as we send all other request per queue, instead of
queue-pair.

Hence we should do proper changes to adapt to the QEMU change here.
Otherwise, a segfault will be triggered when last TX queue was enabled.

Signed-off-by: Victor Kaplansky <victork@redhat.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
9 years agovhost: fix mmap not aligned with hugepage size
Jianfeng Tan [Thu, 12 Nov 2015 06:06:01 +0000 (14:06 +0800)]
vhost: fix mmap not aligned with hugepage size

This patch fixes a bug under lower version linux kernel, mmap()
fails when length is not aligned with hugepage size. mmap()
without flag of MAP_ANONYMOUS, should be called with length
argument aligned with hugepagesz at older longterm version
Linux, like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
This bug was fixed in Linux kernel by commit:
dab2d3dc45ae7343216635d981d43637e1cb7d45
To avoid failure, make sure in caller to keep length aligned.

Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Acked-by: Huawei Xie <huawei.xie@intel.com>
9 years agovhost: fix guest descriptor closed on reset owner message
Tetsuya Mukawa [Tue, 24 Nov 2015 06:45:35 +0000 (15:45 +0900)]
vhost: fix guest descriptor closed on reset owner message

The patch fixes reset_owner message handling not to clear callfd,
because callfd will be valid while connection is established.

Signed-off-by: Tetsuya Mukawa <mukawa@igel.co.jp>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
9 years agomlx5: add environment variables section to documentation
Olga Shern [Mon, 23 Nov 2015 14:44:49 +0000 (15:44 +0100)]
mlx5: add environment variables section to documentation

Describe how applications can benefit from CQE compression.

Signed-off-by: Olga Shern <olgas@mellanox.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agomlx4: allow operation in secondary processes
Or Ami [Mon, 23 Nov 2015 14:44:42 +0000 (15:44 +0100)]
mlx4: allow operation in secondary processes

Secondary processes are expected to use queues and other resources
allocated by the primary, however Verbs resources can only be shared
between processes when inherited through fork().

This limitation can be worked around for TX by configuring separate queues
from secondary processes.

Signed-off-by: Or Ami <ora@mellanox.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agomlx5: fix available entries in Tx rings
Nelio Laranjeiro [Mon, 23 Nov 2015 14:44:48 +0000 (15:44 +0100)]
mlx5: fix available entries in Tx rings

The number of available entries in TX rings is taken before performing
completion, effectively making rings smaller than they are and causing
TX performance issues under load.

Fixes: 2e22920b85d9 ("mlx5: support non-scattered Tx and Rx")

Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agomlx4: fix available entries in Tx rings
Nelio Laranjeiro [Mon, 23 Nov 2015 14:44:41 +0000 (15:44 +0100)]
mlx4: fix available entries in Tx rings

The number of available entries in TX rings is taken before performing
completion, effectively making rings smaller than they are and causing
TX performance issues under load.

Fixes: 7fae69eeff13 ("mlx4: new poll mode driver")

Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agomlx5: fix local protection error when Tx MP to MR cache is full
Olga Shern [Mon, 23 Nov 2015 14:44:47 +0000 (15:44 +0100)]
mlx5: fix local protection error when Tx MP to MR cache is full

When MP to MR cache is full, the last (newest) MR is freed instead of the
first (oldest) one, causing local protection errors during TX.

Fixes: 2e22920b85d9 ("mlx5: support non-scattered Tx and Rx")

Signed-off-by: Olga Shern <olgas@mellanox.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agomlx4: fix local protection error when Tx MP to MR cache is full
Olga Shern [Mon, 23 Nov 2015 14:44:40 +0000 (15:44 +0100)]
mlx4: fix local protection error when Tx MP to MR cache is full

When MP to MR cache is full, the last (newest) MR is freed instead of the
first (oldest) one, causing local protection errors during TX.

Fixes: 7fae69eeff13 ("mlx4: new poll mode driver")

Signed-off-by: Olga Shern <olgas@mellanox.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agomlx5: fix Tx loss after initialization
Olga Shern [Mon, 23 Nov 2015 14:44:46 +0000 (15:44 +0100)]
mlx5: fix Tx loss after initialization

Pre-registering mbuf memory pools when creating TX queues avoids costly
registrations later in the data path.

Fixes: 2e22920b85d9 ("mlx5: support non-scattered Tx and Rx")

Signed-off-by: Olga Shern <olgas@mellanox.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agomlx4: fix Tx loss after initialization
Olga Shern [Mon, 23 Nov 2015 14:44:39 +0000 (15:44 +0100)]
mlx4: fix Tx loss after initialization

Pre-registering mbuf memory pools when creating TX queues avoids costly
registrations later in the data path.

Fixes: 7fae69eeff13 ("mlx4: new poll mode driver")

Signed-off-by: Olga Shern <olgas@mellanox.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agomlx5: fix scattered Tx with too many segments
Adrien Mazarguil [Mon, 23 Nov 2015 14:44:45 +0000 (15:44 +0100)]
mlx5: fix scattered Tx with too many segments

Buffers with too many segments are linearized to overcome
MLX5_PMD_SGE_WR_N, unfortunately the last segment is never sent.

Fixes: 3ee8444608a9 ("mlx5: support scattered Rx and Tx")

Signed-off-by: Jesper Wramberg <jesper.wramberg@gmail.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agomlx4: fix scattered Tx with too many segments
Adrien Mazarguil [Mon, 23 Nov 2015 14:44:38 +0000 (15:44 +0100)]
mlx4: fix scattered Tx with too many segments

Buffers with too many segments are linearized to overcome
MLX4_PMD_SGE_WR_N, unfortunately the last segment is never sent.

Fixes: be11b35817e0 ("mlx4: move scattered Tx processing to helper function")

Signed-off-by: Jesper Wramberg <jesper.wramberg@gmail.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agomlx5: fix memory registration for indirect mbuf data
Adrien Mazarguil [Mon, 23 Nov 2015 14:44:44 +0000 (15:44 +0100)]
mlx5: fix memory registration for indirect mbuf data

Indirect mbuf data may come from a different mempool which must be
registered separately as another memory region, otherwise such mbufs cannot
be sent.

Fixes: 2e22920b85d9 ("mlx5: support non-scattered Tx and Rx")

Signed-off-by: Jesper Wramberg <jesper.wramberg@gmail.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agomlx4: fix memory registration for indirect mbuf data
Adrien Mazarguil [Mon, 23 Nov 2015 14:44:37 +0000 (15:44 +0100)]
mlx4: fix memory registration for indirect mbuf data

Indirect mbuf data may come from a different mempool which must be
registered separately as another memory region, otherwise such mbufs cannot
be sent.

Fixes: 7fae69eeff13 ("mlx4: new poll mode driver")

Signed-off-by: Jesper Wramberg <jesper.wramberg@gmail.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agomlx5: fix possible crash when clearing device statistics
Adrien Mazarguil [Mon, 23 Nov 2015 14:44:43 +0000 (15:44 +0100)]
mlx5: fix possible crash when clearing device statistics

A typo causes TX stats indices to be retrieved from RX queues.

Fixes: 87011737b715 ("mlx5: add software counters")

Reported-by: Nicolas Harnois <nicolas.harnois@6wind.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agomlx4: fix possible crash when clearing device statistics
Adrien Mazarguil [Mon, 23 Nov 2015 14:44:36 +0000 (15:44 +0100)]
mlx4: fix possible crash when clearing device statistics

A typo causes TX stats indices to be retrieved from RX queues.

Fixes: 7fae69eeff13 ("mlx4: new poll mode driver")

Reported-by: Nicolas Harnois <nicolas.harnois@6wind.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
9 years agocxgbe: use trylock in Tx callback instead of spinning for a lock
Rahul Lakkireddy [Fri, 20 Nov 2015 13:13:38 +0000 (18:43 +0530)]
cxgbe: use trylock in Tx callback instead of spinning for a lock

CXGBE PMD depends on an alarm to periodically transmit any pending
coalesced packets and hence spins for a lock for each tx queue in the
alarm callback.

A better solution is to try to get a lock whenever possible, instead
of spinning for it.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
9 years agocxgbe: fix minimum descriptors requirement
Rahul Lakkireddy [Fri, 20 Nov 2015 13:13:37 +0000 (18:43 +0530)]
cxgbe: fix minimum descriptors requirement

Most dpdk examples and apps seem to initialize with a minimum of 128 rx
descriptors and 512 tx descriptors for queue setup.  However, CXGBE PMD
enforces a minimum of 1024 descriptors.  This causes the dpdk stack
to return -EINVAL because the apps do not request the minimum required.

The fix is to relax this enforcing in CXGBE PMD. A minimum of 128
descriptors seems to be a good number with the least impact on the
performance.

Fixes: 946c9ed95616 ("cxgbe: get descriptor limits")

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
9 years agoixgbe: fix VFIO interrupt vector
Cunming Liang [Fri, 13 Nov 2015 02:50:26 +0000 (10:50 +0800)]
ixgbe: fix VFIO interrupt vector

The vector number may change during 'dev_start'. Before enabling a new vector mapping,
it's necessary to disable/unmap the previous setting.

Fixes: 7ab8500037f6 ("ixgbe: fix VF start with PF stopped")

Reported-by: Yong Liu <yong.liu@intel.com>
Signed-off-by: Cunming Liang <cunming.liang@intel.com>
Tested-by: Yong Liu <yong.liu@intel.com>
Acked-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
9 years agoigb: fix VFIO interrupt vector
Cunming Liang [Fri, 13 Nov 2015 02:50:27 +0000 (10:50 +0800)]
igb: fix VFIO interrupt vector

The vector number may change during 'dev_start'.
Before enabling a new vector mapping,
it's necessary to disable/unmap the previous setting.

Fixes: fe685de2b1b6 ("igb: fix VF start with PF stopped")

Reported-by: Yong Liu <yong.liu@intel.com>
Signed-off-by: Cunming Liang <cunming.liang@intel.com>
Acked-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
9 years agofm10k: allow to disable vector driver
Chen Jing D(Mark) [Tue, 24 Nov 2015 06:15:53 +0000 (14:15 +0800)]
fm10k: allow to disable vector driver

fm10k driver will meet compile error on non-x86 platforms due to
SSE instructions. Original implementation didn't have switch to
turn off vPMD.
The improvement introduces a macro to turn on/off vPMD functions,
it's on by default. On non-x86 platforms, it can simply be turned
off to fix compile issue.

Signed-off-by: Chen Jing D(Mark) <jing.d.chen@intel.com>
9 years agofm10k: fix crash when closing
Chen Jing D(Mark) [Tue, 24 Nov 2015 03:10:16 +0000 (11:10 +0800)]
fm10k: fix crash when closing

When the fm10k port is closed, both func tx_queue_clean() and
fm10k_tx_queue_release_mbufs_vec() will try to release buffer in
SW ring. The latter func won't do sanity check on those pointers
and cause crash.

The fix removed Vector TX buffer release func since it can share
the release functions with regular TX.

Fixes: fb9066e479a6 ("fm10k: reset and release mbuf for vector Tx")

Signed-off-by: Chen Jing D(Mark) <jing.d.chen@intel.com>
Acked-by: Michael Qiu <michael.qiu@intel.com>
9 years agofm10k: fix VLAN in Rx mbuf
Shaopeng He [Fri, 20 Nov 2015 14:09:47 +0000 (22:09 +0800)]
fm10k: fix VLAN in Rx mbuf

vlan_tci should be copied from RX descriptor to mbuf,
and flag PKT_RX_VLAN_PKT should be set for every RX packet,
this patch fixes this issue.
Packets in fm10k device always carry at least one VLAN tag.
For those packets coming in without VLAN tag,
the port default VLAN tag will be used.
So in fm10k, always PKT_RX_VLAN_PKT flag is set and vlan_tci
is valid for each RX packet's mbuf.

Signed-off-by: Shaopeng He <shaopeng.he@intel.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
9 years agoi40evf: fix write flush
Jingjing Wu [Mon, 16 Nov 2015 07:08:30 +0000 (15:08 +0800)]
i40evf: fix write flush

For i40e vf driver, should use I40EVF_WRITE_FLUSH to flush
configuration but not I40E_WRITE_FLUSH. This patch fixed this issue.

Fixes: be6c228d4da3 (i40evf: support Rx interrupt)

Reported-by: Qian Xu <qian.q.xu@intel.com>
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
9 years agoi40e: fix BW info without DCB enabled
Jingjing Wu [Sun, 22 Nov 2015 12:35:20 +0000 (20:35 +0800)]
i40e: fix BW info without DCB enabled

If DCB is not enabled, the BW info is not stored for VSI. This
patch fixes this issue by merging functions i40e_vsi_dump_bw_config
and i40e_vsi_get_bw_info together.

Fixes: c8b9a3e3fe1b (i40e: support DCB mode)

Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
9 years agoi40e: fix DCB configuration with firmware >= 5.x
Jingjing Wu [Mon, 16 Nov 2015 08:56:57 +0000 (16:56 +0800)]
i40e: fix DCB configuration with firmware >= 5.x

When NVM version is updated to 5.x, DCB can not be configured. This
issue is because of the FW version validation is not correct.
This patch fixed this issue.

Fixes: c8b9a3e3fe1b (i40e: support DCB mode)

Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Tested-by: Yulong Pei <yulong.pei@intel.com>
9 years agoi40e: skip link control as firmware workaround
Helin Zhang [Tue, 17 Nov 2015 07:09:47 +0000 (15:09 +0800)]
i40e: skip link control as firmware workaround

As firmware does not support any link control from software driver
side, any phy config should be ignored as a workaround. Otherwise
the link might not be up again after binding back to kernel driver.

Signed-off-by: Helin Zhang <helin.zhang@intel.com>
9 years agoi40e: fix reconfiguring hash enable flags
Helin Zhang [Mon, 23 Nov 2015 06:19:24 +0000 (14:19 +0800)]
i40e: fix reconfiguring hash enable flags

It fixes the issue of not re-configuring hash enable flags (HENA)
if there is no key.

Fixes: d0a349409bd7 ("i40e: support AQ based RSS config")

Signed-off-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>
9 years agoi40e: remove CRC size from byte counters
Harry van Haaren [Wed, 18 Nov 2015 10:48:09 +0000 (10:48 +0000)]
i40e: remove CRC size from byte counters

This patch removes the crc bytes from byte counter statistics.

Doing so fixes a bug that CRC bytes were included on TX but not
on RX, causing mismatch of bytes received / sent.

Fixes: 9aace75fc82e ("i40e: fix statistics")

Reported-by: Weichun Chen <weichunx.chen@intel.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
9 years agoixgbe: remove CRC size from byte counters
Harry van Haaren [Wed, 18 Nov 2015 10:48:08 +0000 (10:48 +0000)]
ixgbe: remove CRC size from byte counters

This patch removes the crc bytes from byte counter statistics.

Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
9 years agoe1000: remove CRC size from byte counters
Harry van Haaren [Wed, 18 Nov 2015 10:48:07 +0000 (10:48 +0000)]
e1000: remove CRC size from byte counters

This patch removes the crc bytes from byte counter statistics.

Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
9 years agoe1000: fix PCI device info with secondary process
Jon DeVree [Mon, 23 Nov 2015 15:13:53 +0000 (10:13 -0500)]
e1000: fix PCI device info with secondary process

This fixes a bug added to em and igb drivers which causes the pci info
seen by the primary process to become invalidated by secondary process
startup.
This call was added after the process type check in the other drivers.

Fixes: eeefe73f0af1 ("drivers: copy PCI device info to ethdev data")

Signed-off-by: Jon DeVree <nuxi@vault24.org>
Acked-by: Bernard Iremonger <bernard.iremonger@intel.com>
9 years agoring: fix memory leak when detaching
Mauricio Vasquez B [Fri, 20 Nov 2015 18:24:19 +0000 (19:24 +0100)]
ring: fix memory leak when detaching

When freeing the device it is also neccesary to free
rx_queues and tx_queues

Signed-off-by: Mauricio Vasquez B <mauricio.vasquezbernal@studenti.polito.it>
9 years agoconfig: disable i40e vector driver
Zhe Tao [Mon, 23 Nov 2015 18:47:11 +0000 (02:47 +0800)]
config: disable i40e vector driver

Issue: l3fwd app need the ptype in the mbuf to forward the packets properly.
But now some drivers like virtio driver and FVL vPMD will not set the ptype
in mbuf, so l3fwd cannot work properly on that kind of drivers.

Configure the vector PMD option as no for default as a work around for l3fwd.
After the l3fwd app can handle the undefined ptype or the i40e vPMD can
return the ptype, the option will be set as yes for default again.

Signed-off-by: Zhe Tao <zhe.tao@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
9 years agoethdev: add a missing sanity check for Tx queue setup
Rahul Lakkireddy [Fri, 20 Nov 2015 10:26:37 +0000 (15:56 +0530)]
ethdev: add a missing sanity check for Tx queue setup

Add a sanity check for number of tx descriptors requested during tx
queue setup.  Return -EINVAL if the number requested does not meet
the tx descriptor requirements of the device.

Fixes: 80a1deb4c77a ("ethdev: add API to retrieve queue information")

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
9 years agoethdev: fix missing symbol export
Panu Matilainen [Wed, 18 Nov 2015 07:33:52 +0000 (09:33 +0200)]
ethdev: fix missing symbol export

It prevents some drivers to load:
undefined symbol: rte_eth_dma_zone_reserve

Fixes: 719dbebceb81 ("xen: allow determining DOM0 at runtime")

Signed-off-by: Panu Matilainen <pmatilai@redhat.com>
Acked-by: John McNamara <john.mcnamara@intel.com>
9 years agoeal: fix plugin directory scan to be filesystem agnostic
Panu Matilainen [Wed, 18 Nov 2015 06:45:23 +0000 (08:45 +0200)]
eal: fix plugin directory scan to be filesystem agnostic

Not all filesystems supply struct dirent d_type field, in which case
everything in the specified directory would go ignored. One such
filesystem being XFS which RHEL 7 defaults to... stat() the entries
instead.

Fixes: 9f8eb1d9ca0f ("eal: support driver loading from directory")

Signed-off-by: Panu Matilainen <pmatilai@redhat.com>
9 years agoeal: fix plugin loading without requiring full path
Panu Matilainen [Mon, 23 Nov 2015 12:05:55 +0000 (14:05 +0200)]
eal: fix plugin loading without requiring full path

The added error checking on plugin initialization in
commit 9f8eb1d9ca0f56d6292db5858c52e6873d0abe51 broke the ability of
loading plugins by their basename from default linker locations.
Only use stat() for directory discovery and leave error handling
to dlopen() to restore former behavior.

Fixes: 9f8eb1d9ca0f ("eal: support driver loading from directory")

Signed-off-by: Panu Matilainen <pmatilai@redhat.com>
9 years agoeal: restrict empty declarations to doxygen
David Marchand [Fri, 13 Nov 2015 09:35:26 +0000 (10:35 +0100)]
eal: restrict empty declarations to doxygen

No need for those forward declarations (which breaks build when asking for
C++11 or adding pedantic flag).

Signed-off-by: David Marchand <david.marchand@6wind.com>
9 years agoeal: revert C++ app build fix
David Marchand [Fri, 13 Nov 2015 09:35:25 +0000 (10:35 +0100)]
eal: revert C++ app build fix

It does not build with every C++ compilers.
Reverts the _UNDERLYING_TYPE workarounds to prepare
for another fix in the next patch.

Fixes: 621389bbbe08 ("eal: fix C++ app build")

Signed-off-by: David Marchand <david.marchand@6wind.com>
9 years agoeal: fix alarm clock type for glibc < 2.12
Ferruh Yigit [Thu, 19 Nov 2015 11:23:42 +0000 (11:23 +0000)]
eal: fix alarm clock type for glibc < 2.12

CLOCK_MONOTONIC_RAW added in glibc 2.12, using this define in older
glibc versions cause compile error:
'error: identifier "CLOCK_MONOTONIC_RAW" is undefined'

This patch replaces "CLOCK_MONOTONIC_RAW" with "CLOCK_MONOTONIC" for
older glibc versions, versions that support "CLOCK_MONOTONIC_RAW"
will keep using this clock type.

Fixes: d08d304508a8 ("eal/linux: make alarm not affected by system time jump")

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
9 years agoszedata2: add scattered Rx
Matej Vido [Tue, 10 Nov 2015 14:18:16 +0000 (15:18 +0100)]
szedata2: add scattered Rx

Add new RX function for handling scattered packets.

Signed-off-by: Matej Vido <matejvido@gmail.com>
9 years agoszedata2: add Tx
Matej Vido [Tue, 10 Nov 2015 14:18:15 +0000 (15:18 +0100)]
szedata2: add Tx

TX function handles scattered and non-scattered packets.

Signed-off-by: Matej Vido <matejvido@gmail.com>
9 years agoszedata2: add non-scattered Rx
Matej Vido [Tue, 10 Nov 2015 14:18:14 +0000 (15:18 +0100)]
szedata2: add non-scattered Rx

Signed-off-by: Matej Vido <matejvido@gmail.com>
9 years agoszedata2: add new poll mode driver
Matej Vido [Tue, 10 Nov 2015 14:18:13 +0000 (15:18 +0100)]
szedata2: add new poll mode driver

Add virtual PMD which communicates with COMBO cards through sze2
layer using libsze2 library.

Since link_speed is uint16_t, there can not be used number for 100G
speed, therefore link_speed is set to ETH_LINK_SPEED_10G until the
type of link_speed is solved.

Signed-off-by: Matej Vido <matejvido@gmail.com>
9 years agomk: fix extra options when linking lib through compiler
Thomas Monjalon [Fri, 20 Nov 2015 13:56:56 +0000 (14:56 +0100)]
mk: fix extra options when linking lib through compiler

When using a linker option not known by the compiler like -rpath,
the library linkage was failing.
It is fixed by prefixing the option with -Wl, as it is done in other
makefiles.

Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
9 years agomk: fix build with icc 16
Declan Doherty [Fri, 20 Nov 2015 14:34:33 +0000 (14:34 +0000)]
mk: fix build with icc 16

This patch changes the ICC conditional check to disable
-no-inline-max-size and -no-inline-max-total-size to be
for all versions of icc greater than 14 and not just for version 15.

Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
9 years agoacl: fix native build on haswell with icc
Konstantin Ananyev [Fri, 20 Nov 2015 15:38:50 +0000 (15:38 +0000)]
acl: fix native build on haswell with icc

On HSW box with icc 16.0.0 build for x86_64-native-linuxapp-icc fails with:
icc: command line warning #10120: overriding '-march=native' with '-msse4.1'
...
dpdk.org/x86_64-native-linuxapp-icc/include/rte_memcpy.h(96): error: identifier "__m256i" is undefined

The reason is that icc treats "-march=native ... -msse4.1"
in a different way, then gcc and clang.
For icc it means override all flags enabled with
'-march=native' with '-msse4.1'.
Even when '-march=native' is a superset for '-msse4.1'.
To overcome the problem add a check is SSE4.1 compilation flag already enabled.
If yes, then no need to add '-msse4.1'
Similar change for avx2 compilation option.

Fixes: 074f54ad03ee ("acl: fix build and runtime for default target")

Reported-by: Declan Doherty <declan.doherty@intel.com>
Reported-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
9 years agoexamples/l2fwd-keepalive: add sample application
Remy Horton [Wed, 18 Nov 2015 14:05:16 +0000 (14:05 +0000)]
examples/l2fwd-keepalive: add sample application

Modified version of l2fwd to demonstrate keep-alive functionality.

Signed-off-by: Remy Horton <remy.horton@intel.com>
Signed-off-by: Maryam Tahhan <maryam.tahhan@intel.com>
Signed-off-by: John J Browne <john.j.browne@intel.com>
9 years agoeal: add keep alive monitoring
Remy Horton [Wed, 18 Nov 2015 14:05:14 +0000 (14:05 +0000)]
eal: add keep alive monitoring

Adds functions for detecting and reporting the live-ness of LCores,
the primary requirement of which is minimal overheads for the
core(s) being checked. Core failures are notified via an application
defined callback.

Signed-off-by: Remy Horton <remy.horton@intel.com>
9 years agokni: fix build with kernel 2.6.32
Helin Zhang [Mon, 9 Nov 2015 06:26:12 +0000 (14:26 +0800)]
kni: fix build with kernel 2.6.32

It fixes the compile issue on kernel version 2.6.32 or old ones.

Error logs:
lib/librte_eal/linuxapp/kni/kni_misc.c:121: error: unknown field id specified in initializer
lib/librte_eal/linuxapp/kni/kni_misc.c:121: error: excess elements in struct initializer
lib/librte_eal/linuxapp/kni/kni_misc.c:121: error: (near initialization for kni_net_ops)
lib/librte_eal/linuxapp/kni/kni_misc.c:122: error: unknown field size specified in initializer
lib/librte_eal/linuxapp/kni/kni_misc.c:122: error: excess elements in struct initializer
lib/librte_eal/linuxapp/kni/kni_misc.c:122: error: (near initialization for kni_net_ops)

Fixes: 72a7a2b2469e ("kni: allow per-net instances")

Signed-off-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
9 years agokni: use utsrelease to determine Ubuntu kernel version
Simon Kagstrom [Thu, 20 Aug 2015 06:51:06 +0000 (08:51 +0200)]
kni: use utsrelease to determine Ubuntu kernel version

/proc/version_signature is the version for the host machine, but in
e.g., chroots, this does not necessarily match that DPDK is built
for. DPDK will then build for the wrong kernel version - that of the
server, and not that installed in the (build) chroot.

The patch uses utsrelease.h from the kernel sources instead and fakes
the upload version.

Tested on a server with Ubuntu 12.04, building in a chroot for Ubuntu
14.04.

Signed-off-by: Simon Kagstrom <simon.kagstrom@netinsight.net>
Signed-off-by: Johan Faltstrom <johan.faltstrom@netinsight.net>
Acked-by: Helin Zhang <helin.zhang@intel.com>
9 years agoconfig: disable KNI module on tile
Zhigang Lu [Tue, 17 Nov 2015 06:23:54 +0000 (14:23 +0800)]
config: disable KNI module on tile

Commit 36080ff96b0e causes compiling error on tile, as tile
does not support KNI, so we disable the CONFIG_RTE_KNI_KMOD.

Fixes: 36080ff96b0e ("config: add KNI kmod option")

Reported-by: Guo Xin <gxin@ezchip.com>
Signed-off-by: Zhigang Lu <zlu@ezchip.com>
9 years agoeal: introduce SMP memory barriers
Jerin Jacob [Fri, 6 Nov 2015 09:40:32 +0000 (15:10 +0530)]
eal: introduce SMP memory barriers

This commit introduce rte_smp_mb(), rte_smp_wmb() and rte_smp_rmb(), in
order to enable memory barriers between lcores.
The patch does not provide any functional change for IA, the goal is to
have infrastructure for weakly ordered machines like ARM to work on DPDK.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
9 years agomk: add thunderx machine target based on armv8-a
Jerin Jacob [Fri, 6 Nov 2015 09:40:29 +0000 (15:10 +0530)]
mk: add thunderx machine target based on armv8-a

Created the new thunderx machine target to address difference
in "cache line size" and "-mcpu=thunderx" vs default armv8-a machine target

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
9 years agomk: introduce ARMv8 architecture
Jerin Jacob [Fri, 6 Nov 2015 09:40:28 +0000 (15:10 +0530)]
mk: introduce ARMv8 architecture

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
9 years agoacl: add NEON optimization for ARMv8
Jerin Jacob [Fri, 6 Nov 2015 09:40:27 +0000 (15:10 +0530)]
acl: add NEON optimization for ARMv8

The implementation uses NEON gcc intrinsic.
Verified with testacl and acl_autotest applications on arm64 architecture.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
9 years agoeal/arm: add CPU flags for ARMv8
Jerin Jacob [Fri, 6 Nov 2015 09:40:21 +0000 (15:10 +0530)]
eal/arm: add CPU flags for ARMv8

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
9 years agoeal/arm: add vector header for ARM NEON
Jerin Jacob [Mon, 2 Nov 2015 23:47:25 +0000 (00:47 +0100)]
eal/arm: add vector header for ARM NEON

Added the definition of rte_xmm and xmm_t for acl neon implementation.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
9 years agoeal/arm: add memcpy based on libc for ARMv8
Jerin Jacob [Fri, 6 Nov 2015 09:40:24 +0000 (15:10 +0530)]
eal/arm: add memcpy based on libc for ARMv8

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
9 years agoeal/arm: add atomic operations for ARMv8
Jerin Jacob [Fri, 6 Nov 2015 09:40:20 +0000 (15:10 +0530)]
eal/arm: add atomic operations for ARMv8

except rte_?wb() functions other functions are used from
RTE_FORCE_INTRINSICS=y scheme

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
9 years agoeal/arm: add prefetch operations for ARMv8
Jerin Jacob [Fri, 6 Nov 2015 09:40:22 +0000 (15:10 +0530)]
eal/arm: add prefetch operations for ARMv8

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
9 years agoeal/arm: add cpu cycle operations for ARMv8
Jerin Jacob [Fri, 6 Nov 2015 09:40:23 +0000 (15:10 +0530)]
eal/arm: add cpu cycle operations for ARMv8

cntcvt_el0 ticks are not based on cpu clk unlike rdtsc in x86.
Its a fixed clock running based at constant speed.
Though its a armv8-a implementer choice, typically it runs at 50 or 100 MHz

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
9 years agomk: introduce ARMv7 architecture
Vlastimil Kosar [Mon, 2 Nov 2015 23:47:27 +0000 (00:47 +0100)]
mk: introduce ARMv7 architecture

Make DPDK run on ARMv7-A architecture. This patch assumes
ARM Cortex-A9. However, it is known to be working on Cortex-A7
and Cortex-A15.

Signed-off-by: Vlastimil Kosar <kosar@rehivetech.com>
Signed-off-by: Jan Viktorin <viktorin@rehivetech.com>
Acked-by: David Marchand <david.marchand@6wind.com>
9 years agomk: ignore alignment errors for ARMv7
Jan Viktorin [Mon, 2 Nov 2015 23:47:26 +0000 (00:47 +0100)]
mk: ignore alignment errors for ARMv7

There several issues with alignment when compiling for ARMv7.
They are not considered to be fatal (ARMv7 supports unaligned
access of 32b words), so we just leave them as warnings. They
should be solved later, however.

Signed-off-by: Jan Viktorin <viktorin@rehivetech.com>
Signed-off-by: Vlastimil Kosar <kosar@rehivetech.com>
Acked-by: David Marchand <david.marchand@6wind.com>
9 years agoeal/arm: detect ARMv7 architecture in CPU flags
Jan Viktorin [Mon, 2 Nov 2015 23:47:23 +0000 (00:47 +0100)]
eal/arm: detect ARMv7 architecture in CPU flags

Based on the patch by David Hunt and Armuta Zende:

  lib: added support for armv7 architecture

Signed-off-by: Jan Viktorin <viktorin@rehivetech.com>
Signed-off-by: Amruta Zende <amruta.zende@intel.com>
Signed-off-by: David Hunt <david.hunt@intel.com>
Acked-by: David Marchand <david.marchand@6wind.com>
9 years agoeal/arm: add CPU flags for ARMv7
Vlastimil Kosar [Mon, 2 Nov 2015 23:47:22 +0000 (00:47 +0100)]
eal/arm: add CPU flags for ARMv7

This implementation is based on IBM POWER version of
rte_cpuflags. We use software emulation of HW capability
registers, because those are usually not directly accessible
from userspace on ARM.

Signed-off-by: Vlastimil Kosar <kosar@rehivetech.com>
Signed-off-by: Jan Viktorin <viktorin@rehivetech.com>
Acked-by: David Marchand <david.marchand@6wind.com>