pci: store numa_node per device

Signed-off-by: Intel

pci: configure SR-IOV with max_vfs

Signed-off-by: Intel

pci: fix probing blacklisted device with RTE_PCI_DRV_MULTIPLE flag

When blacklisting network adapters that share a PCI bus address with multiple
ports such as Mellanox ConnectX-3 (librte_pmd_mlx4), RTE applications
enter infinite loops while probing PCI devices.

This commit removes this probe on subsequent instances for blacklisted devices
that have RTE_PCI_DRV_MULTIPLE set.

Signed-off-by: Julien Courtat <julien.courtat@6wind.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>

pci: don't unbind resources on exit

unbind operations must be done before starting eal application

Signed-off-by: Intel

pci: rework interrupt handling

Signed-off-by: Intel <intel.com>

pci: support multiple PCI regions per device

Need to change PCI code to support multiple I/O regions on a single device.
Some devices like VMXNET3 have multiple PCI memory regions, and some
have none.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Intel

mbuf: force inlining

Signed-off-by: Intel

mem: force inlining

Signed-off-by: Intel

mem: remove RLIMIT_NOFILE update

Memory init won't try to increase its RLIMIT_NOFILE limit, this must be done
prior to starting it.

Signed-off-by: Intel

mem: find most little element from heap

Allocations now return the most little element from heap matching size
constraint instead of the first one available.

Signed-off-by: Intel

ring: optimisations

Signed-off-by: Intel

timer: add a rte_pause in a tight loop

Signed-off-by: Intel

timer: prefer TSC to HPET

Signed-off-by: Intel

timer: optimize for empty case

In many application there are no timers queued, and the call to
rte_timer_managecan be optimized in that case avoid reading HPET and
lock overhead.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Reviewed-by: Vincent Jardin <vincent.jardin@6wind.com>

timer: option --vmware-tsc-map for VMware guest

The VMWare TSC mapping uses a hook to RDPMC to read the physical TSC
in the case of VMware ESXi.

Signed-off-by: Damien Millescamps <damien.millescamps@6wind.com>
Acked-by: Jean-Mickael Guerin <jmg@6wind.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Introduce new option --vmware-tsc-map, ignored if
CONFIG_RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is not set.

Default is CONFIG_RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT=y.

if CONFIG_RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT is set:
    rte_rdtsc() selects at runtime between Vmware mapping of
    TSC or native TSC
else
    rte_rdtsc() always uses native rdtsc.

When running DPDK on VMware guest, enable --vmware-tsc-map to
read the physical TSC.
Caution: ESXi should pass monitor_control.pseudo_perfctr = TRUE
othewise it results in general protection fault.

Signed-off-by: Jean-Mickael Guerin <jean-mickael.guerin@6wind.com>
Acked-by: Vincent Jardin <vincent.jardin@6wind.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>

hash: jhash optimisation

Signed-off-by: Intel

hash: fix for multi-process apps

Signed-off-by: Intel

sched: initial import

Signed-off-by: Intel

meter: initial import

Signed-off-by: Intel

power: initial import

Signed-off-by: Intel

lpm: add IPv6 support

Signed-off-by: Intel

log: optimize log/panic with attribute cold

Both logging and calls to panic are never in the critical path.
Use the GCC attribute cold to mark these functions as cold,
which generates more optimised code.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Reviewed-by: Vincent Jardin <vincent.jardin@6wind.com>

log: get full path as syslog id

It partially reverts commit 04210699eee.

Signed-off-by: Intel <intel.com>

log: add ability to override syslog parameters

By default, DPDK based applications would only allow logging
to syslog as "rte", DAEMON; but for any production application more
control is desired to allow using actual application name and
overriding the facility.

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Reviewed-by: Vincent Jardin <vincent.jardin@6wind.com>

eal: cosmetic changes

Signed-off-by: Intel

eal: add application usage hook

Signed-off-by: Intel

eal: memory barriers use intrinsic functions

Signed-off-by: Intel

eal: use intrinsic functions from compiler

RTE_FORCE_INTRINSICS makes it possible to force use of intrinsic
functions (defaults to n).

Signed-off-by: Intel

lib: add rte_hexdump and remove duplicated code

Signed-off-by: Intel

lib: fix various compilation warnings

Signed-off-by: Intel

lib: 32/64-bit cleanups

Signed-off-by: Intel

config: change max lcore to 64

Signed-off-by: Intel

app/testpmd: cosmetic changes

Signed-off-by: Intel

doc: fix file format (dos to unix)

Signed-off-by: Intel

file mode changes

Signed-off-by: Intel

set version to 1.3.1

Signed-off-by: Intel

update copyright date to 2013

Signed-off-by: Intel

examples: various changes

Signed-off-by: Intel

app: various changes

Signed-off-by: Intel

app: fix unused values

Fix warnings of type "Value stored to 'xxx' is never read".

Signed-off-by: Zijie Pan <zijie.pan@6wind.com>
Acked-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>

app: fix volatile read for GCC >= 4.6

When a memory address must be read, for instance a [mapped] PCI register,
the read value is assigned to a local variable that is not used after,
as for instance:
    x = *((uint8_t *) mem_addr);

Such instructions do not compile with gcc 4.6.
The fix consists in adding the "volatile" attribute to the accessed data type
and to not assign the read value:
    *((volatile uint8_t *) mem_addr);

Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>

app/testpmd: fix compliance with __rte_mbuf_sanity_check()

Build mbuf and lists of mbufs in a way compliant with the checks performed
by the function __rte_mbuf_sanity_check() when CONFIG_RTE_LIBRTE_MBUF_DEBUG=y

Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>

app/testpmd: fix config crash

A chained topology must always have an increment of 1.
Here, it was 2 if ports number is even.

Signed-off-by: Damien Millescamps <damien.millescamps@6wind.com>
Acked-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>

lib: minor changes

Signed-off-by: Intel

lib: fix non-C99 macros definitions in exported headers

The original definitions prevent external programs/libraries from compiling
without warnings when using these headers and -std=gnu99 (relaxed C99 mode).

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Acked-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Damien Millescamps <damien.millescamps@6wind.com>

lib: fix uninitialized value

Fix warning "The left expression of the compound assignment
is an uninitialized value".

Signed-off-by: Zijie Pan <zijie.pan@6wind.com>
Acked-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>

lib: fix uninitialized variables

Fix compilation errors caused by uninitialized variables.

Signed-off-by: Krzysztof Witek <krzysztof.witek@6wind.com>
Signed-off-by: Zijie Pan <zijie.pan@6wind.com>
Acked-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>

lib: fix unused values

Fix warnings of type "Value stored to 'xxx' is never read".

Signed-off-by: Zijie Pan <zijie.pan@6wind.com>
Acked-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>

kni: various changes

Signed-off-by: Intel

kni: disable IGB_DCA

Signed-off-by: Intel

kni: fix disabling

Signed-off-by: Intel

kni: fix build with kernel < 2.6.34

Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>

igb_uio: fix build with kernel >= 3.8

__devinit has been removed in linux commit 54b956b903
("Remove __dev* markings from init.h", 3.8-rc4).

Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>

igb: fix RX for Springville (i210)

Signed-off-by: Intel

pci: allow drivers to be bound several times to the same PCI device

Drivers with the flag RTE_PCI_DRV_MULTIPLE enabled will be bound several
times to the same device until they return an error.

These drivers must internally keep track of each device state in order for
this to work reliably.

This flag currently required for Ethernet adapters with several physical
ports but a single address on the PCI bus.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Acked-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Damien Millescamps <damien.millescamps@6wind.com>

mem: 64-bit mempool sizes

Signed-off-by: Intel

mem: error if requesting more segments than MAX_MEMSEG

Signed-off-by: Intel

config: increase MAX_MEMSEG from 32 to 256

Signed-off-by: Intel

eal: fix access to huge page with high address

The returned value of lseek is a virtual address
which can be different from the offset.
Indeed, if the return address has a 64-bit canonical form,
the 16 higher bits are all 1 if bit 47 is 1.

So the check was wrong. It is better to test against an error value.

Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Acked-by: Damien Millescamps <damien.millescamps@6wind.com>

eal: add bsf32

Signed-off-by: Intel

pmac: integration without lib

This library could be used for pattern matching and ACL.
Code of librte_pmac is not released as Open Source.

Signed-off-by: Intel

cmdline: fix for long input

Signed-off-by: Intel

mk: fix verbose display of install command

Correctly display executed install command in
  rte.install-post.mk install rules

Signed-off-by: Didier Pallard <didier.pallard@6wind.com>
Acked-by: Ivan Boule <ivan.boule@6wind.com>

mk: allow corei7-avx flag with gcc 4.7

In rte.toolchain-compat.mk, ensure MACHINE_CFLAGS is initialized when
using GCC 4.7. Update the GCC 4.6 test to also include GCC 4.7.

Signed-off-by: Krzysztof Witek <krzysztof.witek@6wind.com>
Signed-off-by: Zijie Pan <zijie.pan@6wind.com>
Acked-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>

mk: fix typo in LDFLAGS for 32-bit

Enable compilation for 32-bit architecture.

Signed-off-by: Thierry Herbelot <thierry.herbelot@6wind.com>
Acked-by: Ivan Boule <ivan.boule@6wind.com>

mk: use $CC to detect toolchain version

Using gcc is wrong, we should use the cross gcc to detect the
version of gcc.

Signed-off-by: Samuel Gauthier <samuel.gauthier@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>

set version to 1.3.0

Signed-off-by: Intel

examples/vmdq_dcb: update

Signed-off-by: Intel

examples/multi_process: update

Signed-off-by: Intel

examples/load_balancer: update

Signed-off-by: Intel

examples/link_status_interrupt: update

Signed-off-by: Intel

examples/l3fwd: update

Signed-off-by: Intel

examples/l3fwd-vf: update

Signed-off-by: Intel

examples/l2fwd: update

Signed-off-by: Intel

examples/ipv4_multicast: update

Signed-off-by: Intel

examples/ipv4_frag: update

Signed-off-by: Intel

examples/qat: update

Signed-off-by: Intel

examples/qat: rename config files

Signed-off-by: Intel

examples: update link status checks

Signed-off-by: Intel

examples: remove l2fwd-vf

Signed-off-by: Intel

app/testpmd: various updates

Signed-off-by: Intel

app/testpmd: add priority flow control

Signed-off-by: Intel

app/testpmd: add dcb support

Signed-off-by: Intel

app/testpmd: add port configuration settings

Signed-off-by: Intel

app/testpmd: add --total-num-mbufs option

Signed-off-by: Intel

app/testpmd: fix check on mac address

Signed-off-by: Intel

app/testpmd: fix fdir ipv4 filter

Signed-off-by: Intel

app/testpmd: add fdir ipv6 support

Signed-off-by: Intel

app/testpmd: add vlan offload support

Signed-off-by: Intel

app/testpmd: add stats per queue

Signed-off-by: Intel

app/testpmd: fix refcnt in mbuf allocation

test-pmd txonly leaks mbuf from the pool.
The function tx_mbuf_alloc() does not change the refcnt
and the refcnt is 0 when it is first allocated.
However, rte_pktmbuf_free_seg called by the driver's xmit code decrements
reference count to -1. So mbuf never goes back to the pool.
As a result, txonly can't send packets after it exhausts the mempool.

The function tx_mbuf_alloc() was getting mbuf directly from mempool and so
was bypassing mbuf API.
The dedicated function is rte_pktmbuf_alloc() but it is much slower because
it does unnecessary initializations in rte_pktmbuf_reset().
By using the internal API __rte_mbuf_raw_alloc(), refcnt is correctly handled
without adding too much overload.

Signed-off-by: Dongsu Han <dongsuh@cs.cmu.edu>
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>

app: various tests update

Signed-off-by: Intel

app: cosmetic

Signed-off-by: Intel

app: update test pci device ids

Signed-off-by: Intel

app: add new tests on eal flags

Signed-off-by: Intel

app: add reentrancy tests

Signed-off-by: Intel

app: change route table for lpm

Signed-off-by: Intel

app: split performance tests

Signed-off-by: Intel

app: rework autotest.py

Signed-off-by: Intel

app: refactor dump_cfg

Signed-off-by: Intel