bus/fslmc: support debug log

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

bus/fslmc: scan for net and crypto device

This patch will add support in fslmc vfio process to
scan and parse the dpni and dpseci object for net and crypto
devices. It will add the scanned devices to the fslmc bus.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

bus/fslmc: support VFIO

Add support for using VFIO for dpaa2 based fsl-mc bus.

There are some differences in the way vfio used for fsl-mc bus
from the eal vfio.
 - The scanning of bus for individual objects on the basis of
   the DPRC container.
 - The use and mapping of MC portal for object access

With the evolution of bus model, they can be further aligned with
eal vfio code.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

vfio: export utility functions in map file

adding extra vfio utility functions to map file.
They will be used by other vfio supported buses like fslmc bus
for NXP DPAA2 devices

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

bus/fslmc: support MC dpbp object

DPBP object represent a hw based buffer pool instance
in the DPAA2 hardware.

Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

bus/fslmc: support MC dpio object

This patch adds the DPIO object support in MC driver.

DPIO - Data Path Input Output represent the processing
context to access the QBMAN HW for packet I/O.

Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

bus/fslmc: introduce MC object functions

This patch introduces the DPAA2 MC(Management complex Driver).

This is a minimal set of low level functions to send and
receive commands to the fsl-mc. It includes support for basic
management commands and commands to manipulate MC objects.

This is common to be used by various DPAA2 PMDs. e.g.net, crypto
and other drivers.

This is a low level library also used in kernel.

Signed-off-by: Cristian Sovaiala <cristian.sovaiala@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

bus/fslmc: add QBMAN driver to bus

QBMAN, is a hardware block which interfaces with the other
accelerating hardware blocks (For e.g., WRIOP) on NXP's DPAA2
SoC for queue, buffer and packet scheduling.

This patch introduces a userspace driver for interfacing with
the QBMAN hw block.

The qbman-portal component provides APIs to do the low level
hardware bit twiddling for operations such as:
  -initializing Qman software portals
  -building and sending portal commands
  -portal interrupt configuration and processing

This same/similar code is used in kernel and compat file is used
to make it working in user space.

Signed-off-by: Geoff Thorpe <geoff.thorpe@nxp.com>
Signed-off-by: Roy Pledge <roy.pledge@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

bus/fslmc: introduce fsl-mc bus driver

The fslmc bus driver is a rte_bus driver which scans the fsl-mc bus
for NXP DPAA2 SoCs.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

mk: add CRC support to DPAA2 machine type

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

net/ena: calculate partial checksum if DF bit is disabled

When TSO is disabled we still have to calculate partial checksum if DF bit
if turned off. This is caused by firmware bug.

First of all, we must make sure that we are dealing with IPV4 packet.
If not, we will just skip further checking of this packet and move to
the next one.

If application will not set m2_len field, we assume we that it was Ethernet
frame because we have to look inside the packet to check for the DF flag.
To make it work properly, PMD is assuming that before sending
packet application called function rte_eth_tx_prepare().

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Jakub Palider <jpalider@gmail.com>
Acked-by: Jan Medala <jan.medala@outlook.com>

net/ena: cleanup if refilling of Rx descriptors fails

If wrong number of descriptors for refilling was passed to the Rx
repopulate function, there was memory leak which caused memory pool to
run out of resources in longer go.

In case of fail when refilling Rx descriptors, all additional mbufs
have to be released.

Fixes: 1173fca25af9 ("ena: add polling-mode driver")
Cc: stable@dpdk.org
Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Jakub Palider <jpalider@gmail.com>
Acked-by: Jan Medala <jan.medala@outlook.com>

net/ena: fix delayed cleanup of Rx descriptors

On RX path, after receiving bunch of packets, variable tracking
available descriptors in HW queue was not updated.

To fix this issue, variable tracking used descriptors must be updated
after receiving packets - it must be reduced by the amount of received
descriptors in current batch.

Additionally, variable next_to_clean in rx_ring must be updated before
entering ena_populate_rx_queue() to keep it up to date with the current
ring state.

Fixes: 1daff5260ff8 ("net/ena: use unmasked head and tail")
Cc: stable@dpdk.org
Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Jakub Palider <jpalider@gmail.com>
Acked-by: Jan Medala <jan.medala@outlook.com>

net/ena: fix Rx descriptors allocation

When application tried to allocate 1024 descriptors, device was not
initializing properly.

This patch solves it by avoiding allocation of all descriptors in the
ring in one attempt. At least one descriptor must remain unused in the
HW ring.

Fixes: 1173fca25af9 ("ena: add polling-mode driver")
Cc: stable@dpdk.org
Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Jakub Palider <jpalider@gmail.com>
Acked-by: Jan Medala <jan.medala@outlook.com>

net/ark: report hardware status on init

Expose additional fpga status registers.
Report hardware status during PMD init.

Signed-off-by: Ed Czeck <ed.czeck@atomicrules.com>
Acked-by: John Miller <john.miller@atomicrules.com>

net/ark: set mbuf time stamp field on Rx

Time stamp was carried in the packet meta data, but not
place in the mbuf. The new time stamp field is the proper
destination.

Remove the setting of data offset since this is done by
rte_pktmbuf_free()

Signed-off-by: Ed Czeck <ed.czeck@atomicrules.com>
Acked-by: John Miller <john.miller@atomicrules.com>

net/mlx5: remove extra check on Rx

Removing this check improves performance as VLAN and CRC stripping are
enabled most of the time.

Convert MLX5_CQE_VLAN_STRIPPED to network order to speed up the check
instead of doing it on the completion queue entry field.

Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>

net/thunderx: fix stats access out of bounds

Trying to assign more queues to stats struct break only from one loop
when the maximum size is reached. Outside loop iteration is continued.
This leads to access an array out of bounds.

Fixes: 21e3fb0050b9 ("net/thunderx: add final bits for secondary queue support")
Cc: stable@dpdk.org
Signed-off-by: Marcin Wilk <marcin.wilk@caviumnetworks.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

net/ixgbe: move private APIs to a specific file

Create a new file rte_pmd_ixgbe.c for all the private
APIs. Move all the related code to the new file.

Signed-off-by: Wenzhuo Lu <wenzhuo.lu@intel.com>

net/i40e: move private APIs to a specific file

Create a new file rte_pmd_i40e.c for all the private
APIs. Move all the related code to the new file.

Signed-off-by: Wenzhuo Lu <wenzhuo.lu@intel.com>

net/i40e: fix default PFC setting

To keep consistent with previous release, Priority Flow Control (PFC)
need to be disabled by default. This patch fixes it.

This also fixes an issue where traffic was not forwarded by testpmd
occasionally. In those cases ~4770 pps seen on one of the ports rather
than the full rate (>20mpps).

Fixes: 6f0a707e5b55 ("net/i40e: enable DCB on SRIOV VFs")

Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Tested-by: David Hunt <david.hunt@intel.com>

net/i40e: fix hash input set on X722

There are some new PCTYPEs on X722, but they have not been announced
on the RTE lib, so if it can not set corresponding hash input set for
these packet type, the hash function won’t work.

So we need to handle them base on the translation of the new
PCTYPE and the original PCTYPE.

Fixes: b6a0ec418274 ("i40e: use AQ for Rx control register read/write")

Signed-off-by: Jeff Guo <jia.guo@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>

net/enic: remove initialization of some mbuf fields

Remove initialization of next and nb_segs mbuf fields in the Rx path
since they are now initialized in the mbuf pool.

See commit 8f094a9ac5d7 ("mbuf: set mbuf fields while in pool").

Signed-off-by: John Daley <johndale@cisco.com>

net/thunderx: use internal byte ordering macros

The base code was using native GCC macros
for endian detection. Change to portable
rte_byteorder.h based endian detection.

Fixes: 262d43fe20ee ("net/thunderx/base: add HW register definitions")

Suggested-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>

net/tap: update driver param string

Fixes: 2bc06869cd94 ("net/tap: add remote netdevice traffic capture")

Signed-off-by: Pascal Mazon <pascal.mazon@6wind.com>

net/ixgbe: remove option to disable offload flags

Having packets received without any offload flags given in the mbuf is not
very useful, and performance tests with testpmd indicates little
benefit is got with the current code by turning off the flags. This makes
the build-time option pointless, so we can remove it.

Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>

net/ixgbe: eliminate mbuf write on rearm

With the mbuf rework, we now have 8 contiguous bytes to be rearmed in the
mbuf just before the 8-bytes of olflags. If we don't do the rearm write
inside the descriptor ring replenishment function, and delay it to
receiving the packet, we can do a single 16B write inside the RX function
to set both the rearm data, and the flags together.

Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>

net/i40e: sync between x86 and arm64 vector PMD

Porting two changes from x86 SSE implematation.
    net/i40e: fix checksum flag in x86 vector Rx
    net/i40e: eliminate mbuf write on rearm

Signed-off-by: Jianbo Liu <jianbo.liu@linaro.org>

net/i40e: remove option to disable offload flags

Having packets received without any offload flags given in the mbuf is not
very useful, and performance tests with testpmd indicates little to no
benefit is got with the current code by turning off the flags. This makes
the build-time option pointless, so we can remove it.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Jianbo Liu <jianbo.liu@linaro.org>

net/i40e: eliminate mbuf write on rearm

With the mbuf rework, we now have 8 contiguous bytes to be rearmed in the
mbuf just before the 8-bytes of olflags. If we don't do the rearm write
inside the descriptor ring replenishment function, and delay it to
receiving the packet, we can do a single 16B write inside the RX function
to set both the rearm data, and the flags together.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Jianbo Liu <jianbo.liu@linaro.org>

net/liquidio: fix null pointer check

Fix null pointer check in release Rx/Tx queue APIs.

Coverity issue: 1423923
Coverity issue: 1423924
Fixes: 9a30013b9884 ("net/liquidio: add API to release Rx queue")
Fixes: cf6bfcbea178 ("net/liquidio: add API to release Tx queue")

Signed-off-by: Shijith Thotton <shijith.thotton@caviumnetworks.com>

net/liquidio/base: fix mbox command initialization

Initialize mail box command to request VF FLR. Data field was
uninitialized before as it was not required and caused the following
error during scan.

Coverity issue: 1384518
Fixes: cdb166963cae ("net/liquidio: add API for VF FLR")

Signed-off-by: Shijith Thotton <shijith.thotton@caviumnetworks.com>

net/i40e: add missing 25G link speed

Add missing 25G link speed decode in i40evf_dev_link_update.

Fixes: 75d133dd3296 ("net/i40e: enable 25G device")
Cc: stable@dpdk.org
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>

net/i40e: fix VF link speed

When PF notify link speed to VF, encode of link speed need to
be converted from ETH_SPEED_XXX to I40E_LINK_SPEED_XXX.

Fixes: bb6722fb5c0e ("net/i40e: fix VF bonded device link down")
Cc: stable@dpdk.org
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>

net/i40e: remove duplicate code

Make i40e_notify_all_vfs_link_status as a wrapper of
i40e_notify_vf_link_status so duplicate code can be removed.

Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>

net/i40e: fix QinQ eth pattern parsing

The eth keyword only should be accepted.

Fixes: d37705068ee8 ("net/i40e: parse QinQ pattern")

Signed-off-by: Bernard Iremonger <bernard.iremonger@intel.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>

net/sfc: remove unnecessary mbuf initialization on Rx

mbuf next is equal to NULL, nb_segs and reference counter are equal to 1
when mbuf is obtained from mempool.

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>

config: enable AVP driver only on x86 64-bit

The AVP devices are only supported on Intel 64-bit architectures so
adjusting the defconfig attributes accordingly.

Fixes: 908072e9d0e6 ("net/avp: support driver registration")

Signed-off-by: Allain Legacy <allain.legacy@windriver.com>

net/ark: integrate PMD

* Flesh out device configuration
* Add links dev_ops
* Allow dynamic extension loading
* Update release notes and feature listing

Signed-off-by: Shepard Siegel <shepard.siegel@atomicrules.com>
Signed-off-by: John Miller <john.miller@atomicrules.com>
Signed-off-by: Ed Czeck <ed.czeck@atomicrules.com>

net/ark: add Rx initial version

* Core RX packet moving functions

Signed-off-by: Ed Czeck <ed.czeck@atomicrules.com>

net/ark: add Tx initial version

* Core TX packet moving functions
* Flesh out ark_adapter struct to support TX code
(not all fields used at this patch)

Signed-off-by: Ed Czeck <ed.czeck@atomicrules.com>

net/ark: provide API for hardware modules pktchkr and pktgen

Provide C-level interface for Arkville's internal HW resources
pktchkr and pktgen

Signed-off-by: Ed Czeck <ed.czeck@atomicrules.com>

net/ark: provide API for hardware modules UDM and DDM

Provide C-level interface for Arkville's internal HW resources
DDM (Downstream Data Mover) and UDM (Upstream Data Mover) modules

Signed-off-by: Ed Czeck <ed.czeck@atomicrules.com>

net/ark: provide API for hardware modules MPU RQP and pktdir

Provide C-level interface for Arkville's internal HW resources
MPU, (Memory Prefetch Unit) pktdir (Packet director), and RQP
(Request Pacer) modules

Signed-off-by: Ed Czeck <ed.czeck@atomicrules.com>

net/ark: stub PMD for Atomic Rules Arkville

Enable Arkville on supported configurations
Add overview documentation
Minimum driver support for valid compile
Arkville PMD is not supported on ARM or PowerPC at this time

Signed-off-by: Ed Czeck <ed.czeck@atomicrules.com>
Signed-off-by: John Miller <john.miller@atomicrules.com>

net/i40e: fix allocation check

function i40evf_add_del_all_mac_addr without check return
value of rte_zmalloc

Fixes: 97ac72aa71a9 ("i40e: support setting VF MAC address")
Cc: stable@dpdk.org
Signed-off-by: Henry Cai <caihe@huawei.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>

net/cxgbe: fix possible null pointer dereference

Check return value of malloc.

Fixes: 3bd122eef2cc ("cxgbe/base: add hardware API for Chelsio T5 series adapters")
Cc: stable@dpdk.org
Signed-off-by: Henry Cai <caihe@huawei.com>
Acked-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>

test/mempool_perf: support default mempool

Mempool_perf autotest currently does perf regression for:
 * nochache
 * cache

Introducing default_pool, mainly targeted for ext-mempool regression
test. Ext-mempool don't need 'cache' modes so only adding test-case
support for 'nocache' mode.

So to run ext-mempool perf regression, user has to set
RTE_MBUF_DEFAULT_MEMPOOL_OPS="<>"

There is chance of duplication  ie.. if user sets

RTE_MBUF_DEFAULT_MEMPOOL_OPS="ring_mp_mc" then regression
will happen twice for 'ring_mp_mc'

Signed-off-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>

test/mempool_perf: free mempool on exit

Mempool_perf test not freeing pool memory.

Signed-off-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>

test/mempool_perf: remove mempool global vars

Signed-off-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>

test/mempool: support default mempool

Mempool test currently supports:
 * ring_mp_mc
 * stack

Adding a new default pool options. So, ring* + stack + default
(which can be 'stack' or 'ring')
 * This way, whatever the value of RTE_MBUF_DEFAULT_MEMPOOL_OPS is set,
   it would be verified.
 * even if that means duplicating some test (for example when "stack" is
   set as default and it already part of standard test)

Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Reviewed-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>

test/mempool: free mempool on exit

mempool autotest was not freeing the mempools.

Fixes: 8ef772ae ("app/test: rework mempool test")
Cc: stable@dpdk.org
Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Reviewed-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>

mk: link app with ring mempool by default

From the discussion in [1], it was observed that application should
have a default pool already linked even in case of shared builds.
Ring is especially important because packet mbuf creation API refer to
ring_mp_mc as default handler.

Documentation for this is pending.

[1] http://dpdk.org/ml/archives/dev/2017-April/063819.html

Fixes: 9a8e9b57f544 ("mempool: move ring handler as a driver")

Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>

doc: fix sphinx warning

Seen with recent version of sphinx:
WARNING: latex_preamble conflicts with latex_elements['preamble'], ignored.

Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Tested-by: Ferruh Yigit <ferruh.yigit@intel.com>

ethdev: fix build with strict flags

Compilation error seen while compiling mlx5 in debug mode
under RHEL 7.3:

rte_ethdev.h:1670:7: error: type of bit-field 'state' is a GCC extension
    [-Werror=pedantic]

Address it by removing the unnecessary bit-field width limitation.

Fixes: d52268a8b24b ("ethdev: expose device states")

Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Gaetan Rivet <gaetan.rivet@6wind.com>

eventdev: fix build for clang 4

build error:
.../lib/librte_eventdev/rte_eventdev.c:371:6:
error: logical not is only applied to the left hand side of this
bitwise operator [-Werror,-Wlogical-not-parentheses]
  if (!dev_conf->event_dev_cfg & RTE_EVENT_DEV_CFG_PER_DEQUEUE_TIMEOUT)
      ^
Added parentheses after the '!' to evaluate the bitwise operator first.

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>

mk: fix build for clang 4

clang 4 gives "taking address of packed member may result in an
unaligned pointer value" warnings in a few locations [1].

Disabled "-Waddress-of-packed-member" warning for clang >= 4

[1] build errors:
.../lib/librte_eal/common/eal_common_memzone.c:275:25:
error: taking address of packed member 'mlock' of class or structure
'rte_mem_config' may result in an unaligned pointer value
[-Werror,-Waddress-of-packed-member]
        rte_rwlock_write_lock(&mcfg->mlock);
                               ^~~~~~~~~~~

.../lib/librte_ip_frag/rte_ipv4_reassembly.c:139:31:
error: taking address of packed member 'src_addr' of class or structure
'ipv4_hdr' may result in an unaligned pointer value
[-Werror,-Waddress-of-packed-member]
        psd = (unaligned_uint64_t *)&ip_hdr->src_addr;
                                     ^~~~~~~~~~~~~~~~

.../lib/librte_vhost/vhost_user.c:1037:34:
error: taking address of packed member 'payload' of class or structure
'VhostUserMsg' may result in an unaligned pointer value
[-Werror,-Waddress-of-packed-member]
vhost_user_set_vring_num(dev, &msg.payload.state);
       ^~~~~~~~~~~~~~~~~

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>

net/xenvirt: fix build error

build error:
.../drivers/net/xenvirt/rte_eth_xenvirt.c:676:9:
error: ‘struct rte_eth_dev’ has no member named ‘driver’
  eth_dev->driver = NULL;
         ^~

ethdev driver removed, removing assignment to it to fix the build error

Fixes: 9dca21fb80b6 ("ethdev: remove ethdev driver")

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>

ethdev: do not include PCI header

Since the PCI functionality has been moved to the PCI specific ethdev
header we don't need to include rte_pci.h from here anymore.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

ethdev: remove PCI helper from generic ethdev header

This moves the rte_eth_copy_pci_info() into the PCI specific ethdev
header. As a side effect this also removes it from the list of symbols
exported by the rte_ethdev library.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

ethdev: remove ethdev driver

This removes the now unused struct eth_driver.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

ethdev: remove ethdev PCI probe/remove

This removes the now unused rte_eth_dev_pci_probe() and
rte_eth_dev_pci_remove() functions.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

drivers/net: do not use ethdev driver

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Signed-off-by: Gaetan Rivet <gaetan.rivet@6wind.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

ethdev: add PCI driver helpers

This adds the following helper intended to be used by rte_pci_driver
implementations working with ethdev:

- rte_eth_dev_pci_allocate
- rte_eth_dev_pci_release
- rte_eth_dev_pci_generic_probe
- rte_eth_dev_pci_generic_remove

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

drivers/net: use ethdev allocation helper for vdev

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Signed-off-by: Gaetan Rivet <gaetan.rivet@6wind.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

ethdev: add allocation helper for virtual drivers

This helper should be used by ethdev drivers supporting virtual devices
to help allocating a new ethdev and properly filling the default fields.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

net/null: internalize create function

There is no need to export this API. Remaining users should use the
rte_eal_vdev_init() function instead.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

test: do not short-circuit null device creation

A virtual device should get initialized through the rte_eal_vdev_init()
function to properly initialize the driver.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

net/nfp: use ethdev DMA helper

This driver can use the library function rte_eth_dma_zone_reserve()
instead of duplicating the code.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

net/bonding: call through EAL on create/free

To properly embed the generic rte_device into the rte_eth_dev this reworks
the bonding API to call through rte_eal_vdev_init().

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

eal: parse driver argument before probing drivers

In some cases the virtual device name should be totally different than
the driver being used for the device. Therefore lets parse the devargs for
the "driver" argument before probing drivers in vdev_probe_all_drivers().

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

eal: add name field to generic device

This adds a name field to the generic struct rte_device. The EAL is
checking for the name being populated when registering a device but
doesn't enforce global unique names as this is left to the bus
implementations.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

vdev: use generic vdev struct for probe and remove

This is a preparation to embed the generic rte_device into the rte_eth_dev
also for virtual devices.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Signed-off-by: Gaetan Rivet <gaetan.rivet@6wind.com>

vdev: make virtual bus use its device struct

This allows the virtual bus to be rescanned and probed by tracking the
creation of rte_vdev_device.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Tested-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>

vdev: add virtual device arguments helper function

This adds the rte_vdev_device_args() helper function to prepare for
changing the virtual drivers probe() functions take a rte_vdev_device
pointer instead of the name+args strings.

Signed-off-by: Jan Blunck <jblunck@infradead.org>

vdev: add virtual device name helper function

This adds the rte_vdev_device_name() helper function to retrieve the
rte_vdev_device name which makes moving the name of the low-level
device into struct rte_device easier in the future.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>

vdev: add virtual device struct

This adds the rte_vdev_device structure which embeds a generic rte_device.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Tested-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>

vdev: refactor probe/remove

This is a preparation for the introduction of the struct rte_vdev_device.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Tested-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>

eal: remove unused device init function

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Tested-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>

vdev: move virtual device probing into a bus

This is a refactoring of the virtual device probing which moves into into
a proper bus structure.

Signed-off-by: Jan Blunck <jblunck@infradead.org>
Tested-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>

eal: probe new virtual bus after other bus devices

Also see commit f4ce209a8ce5 ("eal: postpone vdev initialization").

Signed-off-by: Jan Blunck <jblunck@infradead.org>

eal: bump ABI version for bus refactoring

Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

pci: use bus driver for scan/probe

Remove EAL initiated direct PCI scan/probe and enable PCI Bus linkage.

Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Reviewed-by: Gaetan Rivet <gaetan.rivet@6wind.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

pci: add bus driver

Based on EAL Bus APIs, PCI bus callbacks and support functions are
introduced in this patch.

EAL continues to have direct PCI init/scan calls as well. These would be
removed in subsequent patches to enable bus only PCI devices.

Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Reviewed-by: Gaetan Rivet <gaetan.rivet@6wind.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

pci: remove loop over drivers in device detach

rte_eal_pci_detach calls pci_detach_all_drivers which loops over all
PCI drivers for detaching the device. This is unnecessary as the device
already has the PCI driver reference which can be used directly.

Removing pci_detach_all_drivers and restructuring rte_eal_pci_detach
and rte_eal_pci_detach_dev to work without looping over driver list.

Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Reviewed-by: Gaetan Rivet <gaetan.rivet@6wind.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

pci: split match and probe

Matching of PCI device address and driver ID table is being done at two
discreet locations duplicating the code. (rte_eal_pci_probe_one_driver
and rte_eal_pci_detach_dev).

Refactor the match logic as a single function.

Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Signed-off-by: Gaetan Rivet <gaetan.rivet@6wind.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

test: remove PCI tests

These tests are not suited for the rte_bus PCI implementation anymore.

Signed-off-by: Gaetan Rivet <gaetan.rivet@6wind.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>

net/virtio-user: fix multi-process attach

Secondary process doesn't properly attach to the rte_eth_device
initialized by the primary process.

Accessing device from secondary process (e.g. via rte_eth_rx_burst),
causes process to crash. because rte_eth_dev_data is not properly set.

The issue was flood by
'commit 7f95f78a8aea ("ethdev: clear data when allocating device")'
which now clears rte_eth_dev_data entry.
For pci devices the struct is initialized by rte_eth_dev_pci_probe
->eth_dev_attach_secondary().
However, for virtio-user virtio_user_pmd_probe() is called instead of
rte_eth_dev_pci_probe().

The fix is to call rte_eth_dev_attach_secondary(), for secondary
process, from virtio_user_pmd_probe.

Fixes: 7f95f78a8aea ("ethdev: clear data when allocating device")
Cc: stable@dpdk.org
Signed-off-by: Ami Sabo <amis@radware.com>

ethdev: export secondary attach function

Today eth_dev_attach_secondary is defined as static and can only be
called by pci drivers. However, the functionality is also required for
non-pci drivers - so the patch export the function.

Signed-off-by: Ami Sabo <amis@radware.com>

ring: fix return value for single dequeue

The error return code for rte_ring_dequeue() function should be -ENOENT
rather than -ENOBUFS (which is the error value from the enqueue() fn).

Fixes: cfa7c9e6fc1f ("ring: make bulk and burst return values consistent")

Reported-by: Zhihong Wang <zhihong.wang@intel.com>
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>

test: remove unnecessary cast of void pointer

Remove unnecessary casts of void * pointers to a specific type.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>

lib: remove unnecessary void cast

Remove unnecessary casts of void * pointers to a specific type.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>

app/testpmd: fix IPv6 tunnel checksum

When ipv6 packet is tunnel packet, "PKT_TX_OUTER_IPV6" flag must
be set, to let prepare the correct mbuf meta data for tx forward.

Fixes: 2b76648872c9 ("net/e1000: add Tx preparation")
Cc: stable@dpdk.org
Signed-off-by: Jeff Guo <jia.guo@intel.com>

lib: fix IPv6 tunnel checksum

When packet is flag of "PKT_TX_OUTER_IPV6", it also need to be
considered to be tunnel case, in order to calculate the correct
csum value.

Fixes: 2b76648872c9 ("net/e1000: add Tx preparation")
Cc: stable@dpdk.org
Signed-off-by: Jeff Guo <jia.guo@intel.com>

examples: enable HW CRC strip by default

Since VF can not disable/enable HW CRC strip for non-DPDK PF drivers,
and kernel driver almost default enable that feature, if disable it in
example app's rxmode, VF driver will report the VF launch failure. So
this patch default to enable HW CRC strip to let VF launch successful.

Cc: stable@dpdk.org
Signed-off-by: Jeff Guo <jia.guo@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>

test: enable HW CRC strip by default

Since VF can not disable/enable HW CRC strip for non-DPDK PF drivers,
and kernel driver almost default enable that feature, if disable it in
test app's rxmode, VF driver will report the VF launch failure. So
this patch default to enable HW CRC strip to let VF launch successful.

Cc: stable@dpdk.org
Signed-off-by: Jeff Guo <jia.guo@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>

app: enable HW CRC strip by default

Since VF can not disable/enable HW CRC strip for non-DPDK PF drivers,
and kernel driver almost default enable that feature, if disable it in
app's rxmode, VF driver will report the VF launch failure. So this
patch default to enable HW CRC strip to let VF launch successful.

Cc: stable@dpdk.org
Signed-off-by: Jeff Guo <jia.guo@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>

mk: fix quoting for ARM mtune argument

The mtune argument needs an unquoted argument, otherwise the compiler
complains:

  arm-buildroot-linux-gnueabi-gcc.br_real: error:
    unrecognized argument in option ‘-mtune="cortex-a9"’

Fixes: 02a8686263de ("mk: introduce ARMv7 architecture")
Fixes: 4a7e4626975a ("mk: introduce NXP dpaa2 architecture based on armv8-a")
Cc: stable@dpdk.org
Reported-by: Jan Viktorin <viktorin@rehivetech.com>
Signed-off-by: Pascal Mazon <pascal.mazon@6wind.com>
Tested-by: Jan Viktorin <viktorin@rehivetech.com>

maintainers: rename section for szedata2 driver

The name of company is listed for other drivers.
Use the company name also for szedata2 driver.
Cards are available from Netcope rather than Cesnet.

Signed-off-by: Matej Vido <vido@cesnet.cz>

doc: update szedata2 guide

NICs are available from Netcope under the names NFB-*.
Change names and direct the links to Netcope web.

Signed-off-by: Matej Vido <vido@cesnet.cz>
Acked-by: John McNamara <john.mcnamara@intel.com>