eal: whitespace change

Signed-off-by: Intel

eal: fix build with intrinsics and gcc < 4.8

Provide a fallback if the intrinsic __builtin_bswap16 is not available.

Signed-off-by: Intel

mk: always preprocess template config file

Signed-off-by: Intel

mk: avoid multiple inclusion of rte_config.h

Signed-off-by: Intel

version: 1.5.0-pre-release

Signed-off-by: Intel

examples: minor changes

Signed-off-by: Intel

examples/qat: upgrade to CRF 1.2

Signed-off-by: Intel

examples/l3fwd: no bare metal support

Signed-off-by: Intel

examples/l2fwd_fork: new app

Signed-off-by: Intel

examples/qos_sched: add stats

Signed-off-by: Intel

examples/qos_meter: add color policy

Signed-off-by: Intel

examples/kni: new parameters

Signed-off-by: Intel

examples/vmdq_dcb: new port mask parameter

Signed-off-by: Intel

examples/vmdq: new app

Signed-off-by: Intel

app/testpmd: VMDq and DCB updates

Signed-off-by: Intel

app/test: various updates

Signed-off-by: Intel

app/test: update kni

Signed-off-by: Intel

app/test: update interrupts test

Signed-off-by: Intel

app/test: fix memzone test

Signed-off-by: Intel

app/test: new timer test

Signed-off-by: Intel

app/test: add whitelist

Signed-off-by: Intel

app/test: add timer_perf

Signed-off-by: Intel

app/test: minor changes

Signed-off-by: Intel

app/test: remove pattern matching test

It needs librte_pmac which is not released.

Signed-off-by: Intel

kni: add vhost backend

Attach to vhost-net as raw socket backend.

Signed-off-by: Intel

kni: move FIFO functions

Move FIFO functions into kni_fifo.h in order to reuse it for vhost.

Signed-off-by: Intel

kni: allow multiple threads

In this new mode, each KNI device has its own kernel thread for Rx.
The core affinity is configurable.

Signed-off-by: Intel

kni: identify device by name

Some old API functions based on port_id are deprecated.

Signed-off-by: Intel

kni: generate random MAC address if needed

Replace the address based on "\0KNIxy" by a random MAC.

Signed-off-by: Intel

kni: fix build with kernel 3.10

- The flags NETIF_F_HW_VLAN_* have been renamed to NETIF_F_HW_VLAN_CTAG_*.
See Linux commit f646968f8f7c624587de729115d802372b9063dd.

- The VLAN protocol must be specified.
See Linux commits 86a9bad3ab6b6f858fd4443b48738cabbb6d094c
and 80d5c3689b886308247da295a228a54df49a44f6.

Signed-off-by: Intel

kni: clean logs

The debug is now disabled by default and can be enabled with
configuration option CONFIG_RTE_KNI_KO_DEBUG.

Signed-off-by: Intel

kni: minor changes

Signed-off-by: Intel

virtio: add new driver

This PMD can be used in a VM having virtio-net NIC.

Note: it is a different implementation than virtio-usermap extension.

Signed-off-by: Intel

ethdev: random MAC address

Factorize code by moving random_addr() function in only place.
It will be reused for virtio.

Signed-off-by: Intel

eal: increase I/O privilege

Set I/O privilege to the highest level (3).
It is needed for virtio.

Signed-off-by: Intel

pcap: add new driver

This PMD uses libpcap to send/receive packets to/from any NIC.
It can also read/write to/from a file.

Signed-off-by: Intel

ring: add new driver

This PMD is a set of FIFOs using rte_ring without any NIC.
It can be used as a loopback.

Signed-off-by: Intel

ethdev: allow device without registered driver

It is needed for non-pci devices (ring and pcap).

Signed-off-by: Intel

ethdev: export allocate function

The function rte_eth_dev_allocate() was called by rte_eth_dev_init().
In order to use it for non-pci devices, it is now in public API.

Signed-off-by: Intel

eal: introduce non-pci devices

This type of pseudo-device is needed for ring and pcap PMDs.
They are compatible with whitelist and are initialized in rte_eal_init().

Signed-off-by: Intel

ixgbe: add VMDq support

Signed-off-by: Intel

igb: add VMDq support

Signed-off-by: Intel

ethdev: add VMDq support

Signed-off-by: Intel

ixgbe: fix VF init without setup

In case of multi-process application, the secondary process can initialize
the driver without configuring queues. In this case the Rx/Tx functions
were not initialized because it was only done in queue setup.

Fix by reproducing the same behaviour as in eth_ixgbe_dev_init().

Signed-off-by: Intel

igb: fix VF init without setup

In case of multi-process application, the secondary process can initialize
the driver without configuring queues. In this case the Rx/Tx functions
were not initialized because it was only done in queue setup.

Fix by reproducing the same behaviour as in eth_igb_dev_init().

Signed-off-by: Intel

e1000: fix descriptor overflow

Allow rxq->rx_tail + offset > 65535
in eth_em_rx_descriptor_done().

Signed-off-by: Intel

e1000: minor changes

Signed-off-by: Intel

ethdev: reset unsupported stats

Initialize statistics structure to 0 before passing it to the PMD.
This way, the unsupported fields will be 0.

Signed-off-by: Intel

mem: retrieve mempool cache only when needed

It is an optimization for the single consumer case,
or when cache is too small,
or when cache is disabled.

Signed-off-by: Intel

mem: retry malloc with smaller block size when failure

rte_malloc try to allocate memzone blocks with a minimum size.
It it fails, it retries for a smaller size than the standard one.
It will really fail if it cannot allocate block of the requested size.

Signed-off-by: Intel

timer: use a skip list

The skip list algorithm allows to improve the scalability.

Signed-off-by: Intel

lpm: rework rules storage

Signed-off-by: Intel

sched: minor changes

Do not define grinder_credits_check() if it is not used.

Signed-off-by: Intel

eal: use pause only with SSE2

The pause instruction is part of SSE2 extensions.
Note that some compilers define _mm_pause as "rep; nop" instead of "pause".
For compatible processors, they are equivalent.

http://www.intel.com/Assets/PDF/manual/325383.pdf:
"
When executing a spin-wait loop, a Pentium 4 or Intel Xeon processor suffers
a severe performance penalty when exiting the loop because it detects a
possible memory order violation.
The PAUSE instruction provides a hint to the processor that the code sequence
is a spin-wait loop. The processor uses this hint to avoid the memory order
violation in most situations, which greatly improves processor performance.
"

Signed-off-by: Intel

eal: allow to whitelist devices

The new option --use-device is a PCI whitelist.
It is the opposite to -b option.

Signed-off-by: Intel

eal: allow to blacklist address without domain prefix

These 2 formats are now accepted:
    domain:bus:device.function
           bus:device.function

Signed-off-by: Intel

eal: rework CPU mask parsing

The CPU mask was limited to "unsigned long long".
The limit was removed and parsing/init is less loosy.

Signed-off-by: Intel

eal: minor changes

Signed-off-by: Intel

mk: combined library

Allow to merge all libraries (shared or static) into one.

Signed-off-by: Intel

mk: shared libraries

Allow to build shared libraries (.so) instead of static ones (.a).

Signed-off-by: Intel

mk: link with libm

Signed-off-by: Intel

doc: whitespace changes in licenses

Signed-off-by: Intel

update version to 1.4.1

Signed-off-by: Intel

examples/ip_reassembly: various updates

- postpone calls to rte_pktmbuf_free() when a mbuf is not used anymore
- add some tx statistics

Signed-off-by: Intel

examples/dpdk_qat: rework port discovery

Signed-off-by: Intel

examples/qos_sched: minor changes

Signed-off-by: Intel

examples/qos_sched: add --msz for mempool size

Signed-off-by: Intel

examples/l3fwd-power: use DD bit rather than RX queue count

Prefer use of rte_eth_rx_descriptor_done to rte_eth_rx_queue_count.

Signed-off-by: Intel

ixgbe: fix DCB setup

Signed-off-by: Intel

ixgbe: check DD bit for specific RX descriptor

Signed-off-by: Intel

igb: check DD bit of specific RX descriptor

Signed-off-by: Intel

ethdev: check DD bit of specific RX descriptor

Signed-off-by: Intel

ixgbe: RX queue count is not implemented for VF

It was introduced by mistake in version 1.4.0.

Signed-off-by: Intel

ixgbe: use DD bit to count RX available descriptors

Signed-off-by: Intel

igb: use DD bit to count RX available descriptors

Signed-off-by: Intel

ethdev: fix doxygen comment for rte_eth_rx_queue_count

Signed-off-by: Intel

pci: check driver probe return code

Signed-off-by: Intel

mem: rework huge page mapping for secondary process

Signed-off-by: Intel

sched: only support TC 3 oversubscription

Signed-off-by: Intel

sched: add mtu parameter

Signed-off-by: Intel

sched: bitmap is now dynamically allocated

Signed-off-by: Intel

eal: use intrinsic function

Signed-off-by: Intel

update version to 1.4.0

Signed-off-by: Intel

examples/quota_watermark: initial import

Signed-off-by: Intel

examples/ip_reassembly: initial import

Signed-off-by: Intel

examples/l3fwd-vf: release resources on SIGINT

Signed-off-by: Intel

examples/l3fwd: various updates

Signed-off-by: Intel

examples/l2fwd: if no port configured, exit

Signed-off-by: Intel

examples/vmdq_dcb: display port mac address

Signed-off-by: Intel

examples/dpdk_qat: minor change

Signed-off-by: Intel

examples/dpdk_qat: fix cross compile check

Signed-off-by: Intel

examples: work only on defined lcores

Signed-off-by: Intel

examples: time fixes

Signed-off-by: Intel

examples: use global RTE_MAX_ETHPORTS

Signed-off-by: Intel

examples: numa updates

Signed-off-by: Intel

app/testpmd: update csumonly

Signed-off-by: Intel

app: fix build with GCC 4.8

GCC 4.8 was producing this error:
argument to ‘sizeof’ in ‘strncmp’ call is the same expression as the
second source; did you mean to provide an explicit length?
[-Werror=sizeof-pointer-memaccess]

Signed-off-by: Intel

app: only use one thread to run autotests

Signed-off-by: Intel

app: fix build when some libs are disabled

Signed-off-by: Intel

app: rework ring tests

Signed-off-by: Intel