dpdk.git
6 years agonet/mrvl: fix mbuf to bpool lookup
Natalie Samsonov [Fri, 26 Jan 2018 07:55:22 +0000 (08:55 +0100)]
net/mrvl: fix mbuf to bpool lookup

Since in DPDK 17.11 port type was changed from uint8_t to uint16_t
the MBUF_INVALID_PORT value became 0xffff but in mrvl_tx_pkt_burst()
when trying to lookup bpool using mbuf port, we check if the port
is invalid according to value 0xff. This causes segmentation fault.

Solution: since the valid port value cannot exceed RTE_MAX_ETHPORTS
(size of bpool lookup table) any other values consider as invalid so
the packet should be returned to DPDK pool.

Fixes: afb4d0d0bf91 ("net/mrvl: add Rx/Tx support")
Cc: stable@dpdk.org
Signed-off-by: Natalie Samsonov <nsamsono@marvell.com>
6 years agonet/bnxt: fix number of pools for RSS
Ajit Khaparde [Fri, 26 Jan 2018 17:32:00 +0000 (09:32 -0800)]
net/bnxt: fix number of pools for RSS

While using RSS, the pool count should be 1.

Fixes: 8103a57ab432a ("net/bnxt: handle Rx multi queue creation properly")
Cc: stable@dpdk.org
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: add 100G speed detection
Ajit Khaparde [Fri, 26 Jan 2018 17:31:59 +0000 (09:31 -0800)]
net/bnxt: add 100G speed detection

When the driver is loaded on a 100G NIC, the port speed is not
displayed correctly. Parse the 100G speed before displaying it.

Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: support Rx/Tx queue start/stop
Ajit Khaparde [Fri, 26 Jan 2018 17:31:58 +0000 (09:31 -0800)]
net/bnxt: support Rx/Tx queue start/stop

Currently this is implemented entirely in the PMD as there is no
explicit support in the HW. Re-program the RSS Table without this queue
on stop and add it back to the table on start.

Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: check if MAC address is all zeros
Ajit Khaparde [Fri, 26 Jan 2018 17:31:57 +0000 (09:31 -0800)]
net/bnxt: check if MAC address is all zeros

In certain cases the MAC address of a port could be all zeros.
Catch it early, log a message and fail the initialization.

Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: register for more async events
Ajit Khaparde [Fri, 26 Jan 2018 17:31:56 +0000 (09:31 -0800)]
net/bnxt: register for more async events

Register for async events from the FW.
New events we are registering for include Link speed config changes,
PF driver unload and VF config change. Also log a message when the
async event arrives on the completion ring.

Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: use dynamic log type
Ajit Khaparde [Fri, 26 Jan 2018 17:31:55 +0000 (09:31 -0800)]
net/bnxt: use dynamic log type

This patch implements driver specific log type doing away with
usage of RTE_LOG() for logging.

Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: fix size of Tx ring in HW
Ajit Khaparde [Fri, 26 Jan 2018 17:31:54 +0000 (09:31 -0800)]
net/bnxt: fix size of Tx ring in HW

During Tx ring allocation, the actual ring size configured in the HW
ends up being twice the number of txd parameter specified to the driver.
The power of 2 ring size wrongly adds a +1 while sending the ring
create command to the FW.

Fixes: 6eb3cc2294fd ("net/bnxt: add initial Tx code")
Cc: stable@dpdk.org
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/e1000: fix VF Rx interrupt enabling
Wenzhuo Lu [Wed, 24 Jan 2018 08:16:52 +0000 (16:16 +0800)]
net/e1000: fix VF Rx interrupt enabling

When using UIO, after enabling the interrupt to get the PF
message, VF RX queue interrupt is not working.
It's expected, as UIO doesn't support multiple interrupt.
So, PMD should not try to enable RX queue interrupt. Then
APP can know the RX queue interrupt is not enabled and only
choose the polling mode.

Fixes: 316f4f1adc2e ("net/igb: support VF mailbox interrupt for link up/down")
CC: stable@dpdk.org
Signed-off-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
6 years agonet/ixgbe: fix VF Rx interrupt enabling
Wenzhuo Lu [Wed, 24 Jan 2018 08:16:51 +0000 (16:16 +0800)]
net/ixgbe: fix VF Rx interrupt enabling

When using UIO, after enabling the interrupt to get the PF
message, VF RX queue interrupt is not working.
It's expected, as UIO doesn't support multiple interrupt.
So, PMD should not try to enable RX queue interrupt. Then
APP can know the RX queue interrupt is not enabled and only
choose the polling mode.

Fixes: 77234603fba0 ("net/ixgbe: support VF mailbox interrupt for link up/down")
CC: stable@dpdk.org
Signed-off-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
6 years agonet/i40e: fix VF Rx interrupt enabling
Wenzhuo Lu [Wed, 24 Jan 2018 08:16:50 +0000 (16:16 +0800)]
net/i40e: fix VF Rx interrupt enabling

When using UIO, after enabling the interrupt to get the PF
message, VF RX queue interrupt is not working.
It's expected, as UIO doesn't support multiple interrupt.
So, PMD should not try to enable RX queue interrupt. Then
APP can know the RX queue interrupt is not enabled and only
choose the polling mode.

Fixes: ae19955e7c86 ("i40evf: support reporting PF reset")
CC: stable@dpdk.org
Signed-off-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
6 years agonet/avf: fix VF Rx interrupt enabling
Wenzhuo Lu [Wed, 24 Jan 2018 08:16:53 +0000 (16:16 +0800)]
net/avf: fix VF Rx interrupt enabling

As UIO doesn't support multiple interrupt, and the interrupt
is occupied by the control plane. PMD should not try to enable
RX queue interrupt. Then APP can know the RX queue interrupt
is not enabled and only choose the polling mode.

Fixes: d6bde6b5eae9 ("net/avf: enable Rx interrupt")

Signed-off-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
6 years agonet/virtio: fix typo in function name
Olivier Matz [Tue, 23 Jan 2018 15:54:43 +0000 (16:54 +0100)]
net/virtio: fix typo in function name

Fixes: c1f86306a026 ("virtio: add new driver")

Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Yuanhan Liu <yliu@fridaylinux.org>
6 years agonet/virtio: rationalize queue flushing
Olivier Matz [Tue, 23 Jan 2018 15:54:42 +0000 (16:54 +0100)]
net/virtio: rationalize queue flushing

Use the same kind of loop than in virtio_free_queues() and factorize
common code.

Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Yuanhan Liu <yliu@fridaylinux.org>
6 years agonet/virtio: fix memory leak when reinitializing device
Olivier Matz [Tue, 23 Jan 2018 15:54:41 +0000 (16:54 +0100)]
net/virtio: fix memory leak when reinitializing device

Free the previous queues and the attached mbufs before initializing new
ones.

The function virtio_dev_free_mbufs() is now called when reconfiguring the
device, so we also need to add a check to ensure that it won't crash for
uninitialized queues.

Cc: stable@dpdk.org
Fixes: 60e6f4707ef2 ("net/virtio: reinitialize device when configuring")

Signed-off-by: Zijie Pan <zijie.pan@6wind.com>
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Yuanhan Liu <yliu@fridaylinux.org>
6 years agonet/virtio: fix queue flushing with vector Rx enabled
Olivier Matz [Tue, 23 Jan 2018 15:54:40 +0000 (16:54 +0100)]
net/virtio: fix queue flushing with vector Rx enabled

When using vector Rx mode (use_simple_rx = 1), vq->vq_descx[] is not
kept up to date. To properly detach the mbufs in this case, browse
sw_ring[] instead, as it's done in virtqueue_rxvq_flush().

Since we need virtio_get_queue_type(), also move this function in
virtqueue.h as a static inline.

Fixes: fc3d66212fed ("virtio: add vector Rx")
Cc: stable@dpdk.org
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Yuanhan Liu <yliu@fridaylinux.org>
6 years agonet/virtio: fix Rx and Tx handler selection for ARM32
Samuel Gauthier [Thu, 14 Dec 2017 14:32:13 +0000 (15:32 +0100)]
net/virtio: fix Rx and Tx handler selection for ARM32

On arm32, we were always selecting the simple handler, but it is only
available if neon is present.

This is due to a typo in the name of the config option.
CONFIG_RTE_ARCH_ARM is for Makefiles. One should use RTE_ARCH_ARM.

Fixes: 2d7c37194ee4 ("net/virtio: add NEON based Rx handler")
Cc: stable@dpdk.org
Signed-off-by: Samuel Gauthier <samuel.gauthier@6wind.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Acked-by: Yuanhan Liu <yliu@fridaylinux.org>
6 years agonet/virtio-user: fix crash as features change
Jianfeng Tan [Tue, 23 Jan 2018 09:52:43 +0000 (09:52 +0000)]
net/virtio-user: fix crash as features change

Since commit 59fe5e17d930 ("vhost: propagate set features handling error"),
vhost does not allow to set different features without reset.

The virtio-user driver fails to reset the device in below commit.

To fix, we send the reset message as stopping the device.

Fixes: c12a26ee209e ("net/virtio-user: fix not properly reset device")
Cc: stable@dpdk.org
Reported-by: Lei Yao <lei.a.yao@intel.com>
Reported-by: Tiwei Bie <tiwei.bie@intel.com>
Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Acked-by: Yuanhan Liu <yliu@fridaylinux.org>
6 years agovhost: claim to support any layout feature
Zhihong Wang [Fri, 19 Jan 2018 19:02:50 +0000 (14:02 -0500)]
vhost: claim to support any layout feature

The VIRTIO_F_ANY_LAYOUT feature indicates the device accepts arbitrary
descriptor layouts. The vhost-user lib already supports it, but the
feature declaration is missing. This patch fixes the mismatch.

Signed-off-by: Zhihong Wang <zhihong.wang@intel.com>
Acked-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Acked-by: Yuanhan Liu <yliu@fridaylinux.org>
6 years agonet/ena: do not set Tx L4 offloads in Rx path
Rafal Kozik [Thu, 25 Jan 2018 15:27:43 +0000 (16:27 +0100)]
net/ena: do not set Tx L4 offloads in Rx path

Information about received packet type detected by NIC should be
stored in packet_type field of rte_mbuf. TX L4 offload flags should
not be set in RX path. Only fields that could be set in of_flags
during packet receiving are information if L4 and L3 checksum is
correct.

Fixes: 1173fca25af9 ("ena: add polling-mode driver")
Cc: stable@dpdk.org
Reported-by: Matthew Smith <mgsmith@netgate.com>
Signed-off-by: Rafal Kozik <rk@semihalf.com>
Signed-off-by: Michal Krawczyk <mk@semihalf.com>
6 years agonet/mlx5: map UAR address around huge pages
Xueming Li [Thu, 25 Jan 2018 15:00:24 +0000 (23:00 +0800)]
net/mlx5: map UAR address around huge pages

Reserving the memory space for the UAR near huge pages helps to
**reduce** the cases where the secondary process cannot start. Those
pages being physical pages they must be mapped at the same virtual
address as in the primary process to have a
working secondary process.

As this remap is almost the latest being done by the processes
(libraries, heaps, stacks are already loaded), similar to huge pages,
there is **no guarantee** this mechanism will always work.

Signed-off-by: Xueming Li <xuemingl@mellanox.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: fix memory region boundary checks
Shahaf Shuler [Thu, 25 Jan 2018 16:18:03 +0000 (18:18 +0200)]
net/mlx5: fix memory region boundary checks

Since commit f81ec748434b ("net/mlx5: fix memory region lookup") the
Memory Region (MR) are no longer overlaps.

Comparing the end address of the MR should be exclusive, otherwise two
contiguous MRs may cause wrong matching.

Fixes: f81ec748434b ("net/mlx5: fix memory region lookup")
Cc: stable@dpdk.org
Signed-off-by: Xueming Li <xuemingl@mellanox.com>
Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: fix memory region cache last index
Shahaf Shuler [Thu, 25 Jan 2018 16:18:02 +0000 (18:18 +0200)]
net/mlx5: fix memory region cache last index

In case Memory Region cache is full, the new mempool will be
inserted in the last index of the array.

Update the last entry being hit to reflect it.

Fixes: b0b093845793 ("net/mlx5: use buffer address for LKEY search")
Cc: stable@dpdk.org
Signed-off-by: Xueming Li <xuemingl@mellanox.com>
Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: warn for unsuccessful memory registration
Shahaf Shuler [Thu, 25 Jan 2018 16:18:01 +0000 (18:18 +0200)]
net/mlx5: warn for unsuccessful memory registration

Memory registration can fail, add the proper warning for such scenario
for it at least to be visible in debug mode.

Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Signed-off-by: Xueming Li <xuemingl@mellanox.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: remove assert un-accessible from secondary process
Shahaf Shuler [Thu, 25 Jan 2018 16:18:00 +0000 (18:18 +0200)]
net/mlx5: remove assert un-accessible from secondary process

Verbs structs such as ibv_mr are not accessible from the secondary
process.

Choose to remove the assert in favor of performing more checks on the
critical data path.

Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Signed-off-by: Xueming Li <xuemingl@mellanox.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: fix secondary process mempool registration
Shahaf Shuler [Thu, 25 Jan 2018 16:17:59 +0000 (18:17 +0200)]
net/mlx5: fix secondary process mempool registration

Secondary process is not allowed to register mempools on the flight.

The code will return invalid memory key for such case.

Fixes: 87ec44ce1651 ("net/mlx5: add operations for secondary process")
Cc: stable@dpdk.org
Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Signed-off-by: Xueming Li <xuemingl@mellanox.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: fix memory region cache lookup
Shahaf Shuler [Thu, 25 Jan 2018 16:17:58 +0000 (18:17 +0200)]
net/mlx5: fix memory region cache lookup

The Memory Region (MR) cache contains pointers to mlx5_mr.
The MR cache indexes are filled when a new MR is created. As it is
possible for MR to be created on the flight, an extra validation must be
added to avoid segmentation fault.

Fixes: b0b093845793 ("net/mlx5: use buffer address for LKEY search")
Cc: stable@dpdk.org
Signed-off-by: Xueming Li <xuemingl@mellanox.com>
Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: fix link state on device start
Shahaf Shuler [Thu, 25 Jan 2018 16:04:28 +0000 (18:04 +0200)]
net/mlx5: fix link state on device start

Following commit c7bf62255edf ("net/mlx5: fix handling link status event")
the link state must be up in order for the burst function to be set on
the device ops.

As the link may take time to move between down and up state it is
possible the rte_eth_dev_start call will return with wrong burst
function (either null or the empty burst function).

Fixing it by forcing the link to be up before returning from device
start. In case the link is still not up after 5 seconds fail the function.
In addition initialize the burst function on device probe to prevent
crashes before the link is up.

Fixes: c7bf62255edf ("net/mlx5: fix handling link status event")
Cc: stable@dpdk.org
Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: fix reception of multiple MAC addresses
Nélio Laranjeiro [Thu, 25 Jan 2018 08:22:19 +0000 (09:22 +0100)]
net/mlx5: fix reception of multiple MAC addresses

When promiscuous is disabled, adding/removing a mac address is ignored
causing the packet to not be received or still being received corresponding
to the add or remove request.

Fixes: 272733b5ebfd ("net/mlx5: use flow to enable unicast traffic")
Cc: stable@dpdk.org
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
6 years agonet/mlx5: use RSS table max size from config
Nélio Laranjeiro [Thu, 25 Jan 2018 08:19:46 +0000 (09:19 +0100)]
net/mlx5: use RSS table max size from config

Maximum indirection table size has not been updated accordingly.

Fixes: 7fe24446e946 ("net/mlx5: add device configuration structure")

Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
6 years agonet/mlx5: support IPv4 time-to-live filter
Nélio Laranjeiro [Tue, 23 Jan 2018 08:05:10 +0000 (09:05 +0100)]
net/mlx5: support IPv4 time-to-live filter

Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx5: fix flow item validation
Nélio Laranjeiro [Tue, 23 Jan 2018 08:05:09 +0000 (09:05 +0100)]
net/mlx5: fix flow item validation

Two masks were compared instead of verifying the spec was included in the
supported mask.

Fixes: 2097d0d1e2cc ("net/mlx5: support basic flow items and actions")
Cc: stable@dpdk.org
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx5: fix flow director conversion
Nélio Laranjeiro [Tue, 23 Jan 2018 08:05:08 +0000 (09:05 +0100)]
net/mlx5: fix flow director conversion

Flow director must provide the same spec and mask to be sure to be
validated.

Fixes: 4c3e9bcdd52e ("net/mlx5: support flow director")
Cc: stable@dpdk.org
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx5: fix missing RSS capability
Shahaf Shuler [Mon, 22 Jan 2018 20:52:14 +0000 (22:52 +0200)]
net/mlx5: fix missing RSS capability

The PMD was not reporting the supported RSS capabilities.

Fixes: 2f97422e7759 ("mlx5: support RSS hash update and get")
Cc: stable@dpdk.org
Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx5: use PCI address as port name
Yuanhan Liu [Mon, 22 Jan 2018 09:30:06 +0000 (17:30 +0800)]
net/mlx5: use PCI address as port name

It is suggested to use PCI BDF to identify a port for port addition
in OVS-DPDK. While mlx5 has its own naming style: name it by ib dev
name. This breaks the typical OVS DPDK use case and brings more puzzle
to the end users.

To fix it, this patch changes it to use PCI BDF as the name, too.
Also, a postfix " port %u" is added, just in case their might be more
than 1 port associated with a PCI device.

Signed-off-by: Yuanhan Liu <yliu@fridaylinux.org>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/i40e: fix RSS flow action parser
Qi Zhang [Thu, 25 Jan 2018 05:10:33 +0000 (13:10 +0800)]
net/i40e: fix RSS flow action parser

Parameter action_flag is not used correctly in i40e_flow_parse_rss_action.
Also change it from point type to value type since it is not an output
parameter.

Fixes: ecad87d22383 ("net/i40e: move RSS to flow API")

Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
Acked-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: Zhiyong Yang <zhiyong.yang@intel.com>
6 years agonet/i40e: fix memory leak
Yong Wang [Thu, 25 Jan 2018 09:01:04 +0000 (04:01 -0500)]
net/i40e: fix memory leak

There are several func calls to rte_zmalloc() which don't have null
pointer check on the return value. And before return, the memory
is not freed. It fixes by adding null pointer check and rte_free().

Fixes: 078259773da9 ("net/i40e: store ethertype filter")
Fixes: 425c3325f0b0 ("net/i40e: store tunnel filter")
Fixes: c50474f31efe ("net/i40e: support tunnel filter to VF")
Fixes: 5c53c82c8174 ("net/i40e: store flow director filter")
Cc: stable@dpdk.org
Signed-off-by: Yong Wang <wang.yong19@zte.com.cn>
Acked-by: Beilei Xing <beilei.xing@intel.com>
6 years agonet/e1000: fix null pointer check
Yong Wang [Thu, 25 Jan 2018 07:43:04 +0000 (02:43 -0500)]
net/e1000: fix null pointer check

There are several func calls to rte_zmalloc() which don't have null
pointer check for the return value. It fixes that by adding null
pointer check.

Fixes: 22bb13410cb2 ("net/igb: create consistent filter")
Cc: stable@dpdk.org
Signed-off-by: Yong Wang <wang.yong19@zte.com.cn>
Acked-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
6 years agonet/i40e: remove unnecessary explicit type casting
Zhiyong Yang [Thu, 25 Jan 2018 07:55:00 +0000 (15:55 +0800)]
net/i40e: remove unnecessary explicit type casting

Remove some unnecessary explicit type casting, to clean the code.

Signed-off-by: Zhiyong Yang <zhiyong.yang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
6 years agoexamples/flow_filtering: fix port id size
Zhiyong Yang [Wed, 24 Jan 2018 06:10:06 +0000 (14:10 +0800)]
examples/flow_filtering: fix port id size

Ethdev port id has been extended 16bits from 8bits in DPDK 17.11 release,
the patch fixes mismatch use.

Fixes: 4a3ef59a10c8 ("examples/flow_filtering: add simple demo of flow API")
Cc: stable@dpdk.org
Signed-off-by: Zhiyong Yang <zhiyong.yang@intel.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
6 years agonet/bonding: check dequeue result before proceeding
Radu Nicolau [Wed, 24 Jan 2018 13:11:13 +0000 (13:11 +0000)]
net/bonding: check dequeue result before proceeding

Coverity issue: 257015
Fixes: 09150784a776 ("net/bonding: burst mode hash calculation")
Cc: stable@dpdk.org
Signed-off-by: Radu Nicolau <radu.nicolau@intel.com>
6 years agonet/bonding: add MAC set operation
Radu Nicolau [Thu, 25 Jan 2018 11:31:44 +0000 (11:31 +0000)]
net/bonding: add MAC set operation

Fixes: aa7791ba8de0 ("net/bonding: fix setting slave MAC addresses")

Signed-off-by: Radu Nicolau <radu.nicolau@intel.com>
6 years agonet/bonding: do not early mark device as bonded
Chas Williams [Wed, 17 Jan 2018 16:44:44 +0000 (11:44 -0500)]
net/bonding: do not early mark device as bonded

bonding immediately marks the incoming eth device as bonded and doesn't
clear this in later error paths.  Delay marking the dev until we are
certain that we are going to add this eth device to the bond group.

Signed-off-by: Chas Williams <chas3@att.com>
Acked-by: Radu Nicolau <radu.nicolau@intel.com>
6 years agonet/failsafe: fix Rx burst infinite loop
Matan Azrad [Wed, 24 Jan 2018 10:19:17 +0000 (10:19 +0000)]
net/failsafe: fix Rx burst infinite loop

In case of plugged out device, the fail-safe PMD uses failsafe_rx_burst
function for packet receiving.

This function iterates over the present sub-devices until it
receives a traffic from one of them or they are all cannot receive
packets.

The corrupted code didn't advance the sub-device pointer when the
sub-device was not present and caused to infinite loop.

Advance the sub-device pointer also in plugged-out sub-device case.

Fixes: 8052bbd9d548 ("net/failsafe: improve Rx sub-devices iteration")

Signed-off-by: Matan Azrad <matan@mellanox.com>
Acked-by: Gaetan Rivet <gaetan.rivet@6wind.com>
6 years agonet/enic: set L4 checksum flags for IPv6 packets
Hyong Youb Kim [Tue, 23 Jan 2018 01:05:27 +0000 (17:05 -0800)]
net/enic: set L4 checksum flags for IPv6 packets

enic_cq_rx_to_pkt_flags() currently sets checksum good/bad flags only
for IPv4.  The hardware actually validates the TCP/UDP checksum of
IPv6 packets too. Set PKT_RX_L4_CKSUM_{GOOD,BAD} accordingly.

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
6 years agonet/enic: add Tx prepare handler
Hyong Youb Kim [Tue, 23 Jan 2018 01:05:29 +0000 (17:05 -0800)]
net/enic: add Tx prepare handler

Like most NICs, this hardware (Cisco VIC) also requires partial
checksum in the packet for checksum offload and TSO. So, add
the tx_pkt_prepare handler like other PMDs do.

Technically, VIC has an offload mode that does not require partial
checksum for non-TSO packets. But, it has no such mode for TSO
packets, making tx_pkt_prepare unavoidable.

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
6 years agonet/enic: fix crash due to static max number of queues
Hyong Youb Kim [Tue, 23 Jan 2018 01:05:28 +0000 (17:05 -0800)]
net/enic: fix crash due to static max number of queues

ENIC_CQ_MAX, ENIC_WQ_MAX and others are arbitrary values that
prevent the app from using more queues when they are available on
hardware. Remove them and dynamically allocate vnic_cq and such
arrays to accommodate all available hardware queues.

As a side effect of removing ENIC_CQ_MAX, this commit fixes a segfault
that would happen when the app requests more than 16 CQs, because
enic_set_vnic_res() does not consider ENIC_CQ_MAX. For example, the
following command causes a crash.

testpmd -- --rxq=16 --txq=16

Fixes: ce93d3c36db0 ("net/enic: fix resource check failures when bonding devices")
Cc: stable@dpdk.org
Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
6 years agonet/mrvl: switch to the new Tx offload API
Tomasz Duszynski [Tue, 23 Jan 2018 08:46:15 +0000 (09:46 +0100)]
net/mrvl: switch to the new Tx offload API

Since the old Tx offload API was depracated
update the driver to use the latest one.

Signed-off-by: Tomasz Duszynski <tdu@semihalf.com>
6 years agonet/mrvl: switch to the new Rx offload API
Tomasz Duszynski [Tue, 23 Jan 2018 08:46:14 +0000 (09:46 +0100)]
net/mrvl: switch to the new Rx offload API

Since the old Rx offload API is now depracated
update the driver to use the latest one.

Signed-off-by: Tomasz Duszynski <tdu@semihalf.com>
6 years agoethdev: move internal callback list definition
David Marchand [Mon, 22 Jan 2018 12:25:35 +0000 (13:25 +0100)]
ethdev: move internal callback list definition

This structure is not exposed through public apis, we should just move it
to the core header.

Fixes: 331c447ad913 ("ethdev: separate internal structures into own header")

Signed-off-by: David Marchand <david.marchand@6wind.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
6 years agonet/mlx5: fix allocation when no memory on device NUMA node
Olivier Matz [Mon, 22 Jan 2018 12:33:38 +0000 (13:33 +0100)]
net/mlx5: fix allocation when no memory on device NUMA node

When no memory is available on the same numa node than the device, the
initialization of the device fails. However, the use case where the
cores and memory are on a different socket than the device is valid,
even if not optimal.

To fix this issue, this commit introduces an infrastructure to select
the socket on which to allocate the verbs objects based on the ethdev
configuration and the object type, rather than the PCI numa node.

Fixes: 1e3a39f72d5d ("net/mlx5: allocate verbs object into shared memory")
Cc: stable@dpdk.org
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: fix return value of start operation
Olivier Matz [Mon, 22 Jan 2018 12:33:37 +0000 (13:33 +0100)]
net/mlx5: fix return value of start operation

On error, mlx5_dev_start() does not return a negative value
as it is supposed to do. The consequence is that the application
(ex: testpmd) does not notice that the port is not started
and begins the rxtx on an uninitialized port, which crashes.

Fixes: e1016cb73383 ("net/mlx5: fix Rx interrupts management")
Cc: stable@dpdk.org
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: fix all multi verification code position
Nélio Laranjeiro [Thu, 11 Jan 2018 09:25:22 +0000 (10:25 +0100)]
net/mlx5: fix all multi verification code position

All multi code should not be handled in exit part of the code but in the
mainline of the function.

Fixes: 0a40a1363a4d ("net/mlx5: fix flow type for allmulti rules")
Cc: stable@dpdk.org
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
6 years agodoc: add i40e queue region support to release notes
Wei Zhao [Mon, 22 Jan 2018 05:18:30 +0000 (13:18 +0800)]
doc: add i40e queue region support to release notes

This patch adds information about i40e queue region related to
the release notes.

Cc: stable@dpdk.org
Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: John McNamara <john.mcnamara@intel.com>
6 years agodoc: add i40e tunnel support in release notes
Beilei Xing [Thu, 18 Jan 2018 02:24:03 +0000 (10:24 +0800)]
doc: add i40e tunnel support in release notes

Update release notes to declare MPLSoUDP/MPLSoGRE/GTP-U/GTP-C/PPPoE/
PPPoL2TP steering support in i40e driver.

Cc: stable@dpdk.org
Signed-off-by: Beilei Xing <beilei.xing@intel.com>
Acked-by: John McNamara <john.mcnamara@intel.com>
6 years agonet/ixgbe: check security enable bits
Radu Nicolau [Thu, 18 Jan 2018 12:46:40 +0000 (12:46 +0000)]
net/ixgbe: check security enable bits

Check if the security enable bits are not fused before setting
offload capabilities for security.

Signed-off-by: Radu Nicolau <radu.nicolau@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
6 years agomaintainers: update for cryptodev
Pablo de Lara [Wed, 24 Jan 2018 17:24:50 +0000 (17:24 +0000)]
maintainers: update for cryptodev

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
6 years agocrypto/qat: fix truncated response ring value
Fiona Trahe [Mon, 29 Jan 2018 18:33:40 +0000 (18:33 +0000)]
crypto/qat: fix truncated response ring value

Issue detected by coverity. Could never actually cause a
problem as truncated value (0x7f7f7f7f->0x7f) is what's needed.
But fix in code for correctness.

Coverity issue: 194998
Fixes: 571365dd4c5e ("crypto/qat: enable Rx head writes coalescing")
Cc: stable@dpdk.org
Signed-off-by: Fiona Trahe <fiona.trahe@intel.com>
6 years agotest/crypto: improve NULL authentication validation
Fiona Trahe [Thu, 25 Jan 2018 17:19:15 +0000 (17:19 +0000)]
test/crypto: improve NULL authentication validation

Add comparison to make sure memory pointed to by
digest pointer is not overwritten in NULL auth case.

Signed-off-by: Fiona Trahe <fiona.trahe@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
6 years agocrypto/qat: fix null auth algo overwrite
Fiona Trahe [Thu, 25 Jan 2018 17:19:14 +0000 (17:19 +0000)]
crypto/qat: fix null auth algo overwrite

If auth algorithm is RTE_CRYPTO_AUTH_NULL and digest_length is 0
in the xform and digest pointer is set in the op, then
the PMD may overwrite memory at the digest pointer.
With this patch the memory is not overwritten.

Fixes: db0e952a5c01 ("crypto/qat: add NULL capability")
Cc: stable@dpdk.org
Signed-off-by: Fiona Trahe <fiona.trahe@intel.com>
Acked-by: Radu Nicolau <radu.nicolau@intel.com>
6 years agocryptodev: fix session pointer cast
Zhiyong Yang [Tue, 23 Jan 2018 09:48:13 +0000 (17:48 +0800)]
cryptodev: fix session pointer cast

The wrong casts don't cause actual error, but they should conform to C
standard.

Fixes: c261d1431bd8 ("security: introduce security API and framework")
Fixes: b3bbd9e5f265 ("cryptodev: support device independent sessions")
Cc: stable@dpdk.org
Signed-off-by: Zhiyong Yang <zhiyong.yang@intel.com>
Acked-by: Akhil Goyal <akhil.goyal@nxp.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
6 years agoapp/crypto-perf: fix out-of-bounds array access
Fan Zhang [Tue, 23 Jan 2018 14:22:55 +0000 (14:22 +0000)]
app/crypto-perf: fix out-of-bounds array access

Fixes: 27c2e7471961 ("app/crypto-perf: support IMIX")

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
6 years agoexamples/ipsec_secgw: fix security session
Fan Zhang [Tue, 23 Jan 2018 12:32:11 +0000 (12:32 +0000)]
examples/ipsec_secgw: fix security session

Fixes: 3da37f682173 ("examples/ipsec_secgw: create session mempools for ethdevs")

Some NICs do not have the rte_security context, this patch fixes the segment fault
caused by this.

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Radu Nicolau <radu.nicolau@intel.com>
6 years agocrypto/qat: fix parameter type
Tomasz Jozwiak [Mon, 22 Jan 2018 16:28:05 +0000 (17:28 +0100)]
crypto/qat: fix parameter type

This commit fixes right cast from qat_cipher_get_block_size
function. This function can return -EFAULT in case of
any error, and that value must be cast to int instead of uint8_t

Fixes: d18ab45f7654 ("crypto/qat: support DOCSIS BPI mode")
Cc: stable@dpdk.org
Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
6 years agocrypto/qat: fix typo in error message
Tomasz Jozwiak [Mon, 22 Jan 2018 16:28:04 +0000 (17:28 +0100)]
crypto/qat: fix typo in error message

This commit fixes typo in bpi_cipher_decrypt error message

Fixes: d18ab45f7654 ("crypto/qat: support DOCSIS BPI mode")
Cc: stable@dpdk.org
Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
6 years agocrypto/qat: fix out-of-bounds access
Tomasz Jozwiak [Mon, 22 Jan 2018 16:28:03 +0000 (17:28 +0100)]
crypto/qat: fix out-of-bounds access

This commit fixes
  - bpi_cipher_encrypt to prevent before 'array subscript is
    above array bounds' error
  - bpi_cipher_decrypt to prevent before 'array subscript is
    above array bounds' error

Fixes: d18ab45f7654 ("crypto/qat: support DOCSIS BPI mode")
Cc: stable@dpdk.org
Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
6 years agocrypto/dpaa_sec: support scatter gather
Akhil Goyal [Mon, 22 Jan 2018 08:46:38 +0000 (14:16 +0530)]
crypto/dpaa_sec: support scatter gather

Signed-off-by: Alok Makhariya <alok.makhariya@nxp.com>
Signed-off-by: Akhil Goyal <akhil.goyal@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
6 years agocrypto/dpaa2_sec: support scatter gather
Akhil Goyal [Mon, 22 Jan 2018 08:46:37 +0000 (14:16 +0530)]
crypto/dpaa2_sec: support scatter gather

Signed-off-by: Alok Makhariya <alok.makhariya@nxp.com>
Signed-off-by: Akhil Goyal <akhil.goyal@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
6 years agodoc: update feature list for cryptodevs
Akhil Goyal [Mon, 22 Jan 2018 08:46:36 +0000 (14:16 +0530)]
doc: update feature list for cryptodevs

Signed-off-by: Akhil Goyal <akhil.goyal@nxp.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
6 years agocrypto/dpaa2_sec: fix build with GCC 7
Thomas Monjalon [Mon, 29 Jan 2018 22:20:40 +0000 (23:20 +0100)]
crypto/dpaa2_sec: fix build with GCC 7

Seen with GCC 7.2.0, a switch fall through is detected and
cannot be fixed with a fall-through comment or attribute:

drivers/crypto/dpaa2_sec/hw/rta/operation_cmd.h:89:6: error:
this statement may fall through [-Werror=implicit-fallthrough=]
   if (rta_sec_era < RTA_SEC_ERA_2)
      ^

The check is disabled in dpaa2_sec Makefile but not in dpaa_sec Makefile
which uses source code shared by dpaa2_sec.

The workaround is to disable the check at the beginning of the file.

Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
6 years agocrypto/mrvl: fix export map file name
Thomas Monjalon [Mon, 29 Jan 2018 21:52:56 +0000 (22:52 +0100)]
crypto/mrvl: fix export map file name

Fixes: 8a61c83af2fa ("crypto/mrvl: add mrvl crypto driver")

Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
6 years agodoc: add ABI experimental tag in versioning guide
Neil Horman [Mon, 22 Jan 2018 01:48:07 +0000 (20:48 -0500)]
doc: add ABI experimental tag in versioning guide

Document the need to add the __experimental tag to appropriate functions

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: John McNamara <john.mcnamara@intel.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
6 years agomk: add experimental tag check
Neil Horman [Mon, 22 Jan 2018 01:48:05 +0000 (20:48 -0500)]
mk: add experimental tag check

Add checks during build to ensure that all symbols in the EXPERIMENTAL
version map section have __experimental tags on their definitions, and
enable the warnings needed to announce their use.  Also add an
ALLOW_EXPERIMENTAL_APIS define to allow individual libraries and files
to declare the acceptability of experimental api usage

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
6 years agoadd experimental tag to appropriate functions
Neil Horman [Mon, 22 Jan 2018 01:48:06 +0000 (20:48 -0500)]
add experimental tag to appropriate functions

Append the __rte_experimental tag to api calls appearing in the
EXPERIMENTAL section of their libraries version map

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
6 years agocompat: add experimental tag macro
Neil Horman [Mon, 22 Jan 2018 01:48:04 +0000 (20:48 -0500)]
compat: add experimental tag macro

The __rte_experimental macro tags a given exported function as being part of
the EXPERIMENTAL api.  Use of this tag will cause any caller of the
function (that isn't removed by dead code elimination) to emit a warning
that the user is making use of an API whos stabilty isn't guaranteed.
It also places the function in the .text.experimental section, which is
used to validate the tag against the corresponding library version map

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
6 years agobuildtools: add script to check experimental API exports
Neil Horman [Mon, 22 Jan 2018 01:48:03 +0000 (20:48 -0500)]
buildtools: add script to check experimental API exports

This tools reads the given version map for a directory, and checks to
ensure that, for each symbol listed in the export list, the corresponding
definition is tagged as __rte_experimental, erroring out if its not.  In this
way, we can ensure that the EXPERIMENTAL api is kept in sync with the tags

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
6 years agopmdinfogen: allow using stdin and stdout
Bruce Richardson [Thu, 25 Jan 2018 11:12:25 +0000 (11:12 +0000)]
pmdinfogen: allow using stdin and stdout

Rather than having to work off files all the time, allow stdin and stdout
to be used as the source and destination for pmdinfogen. This will allow
other possible usages from scripts, e.g. taking files from ar archive and
building a single .pmd.c file from all the .o files in it.

for f in `ar t librte_pmd_xyz.a` ; do
ar p librte_pmd_xyz.a $f | pmdinfogen - - >> xyz_info.c
done

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
6 years agoapp/procinfo: call EAL cleanup before exit
Harry van Haaren [Mon, 29 Jan 2018 16:37:32 +0000 (16:37 +0000)]
app/procinfo: call EAL cleanup before exit

This patch adds a call to the newly introduced cleanup()
function just before quitting the app.

Adding this function call before quitting from a secondary processes
is important, as otherwise it will leak hugepage memory. For a secondary
process that is run multiple times, this could cause hugepage memory
to become depleted and stop a secondary process from starting.

Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Vipin Varghese <vipin.varghese@intel.com>
6 years agoapp/pdump: call EAL cleanup before exit
Harry van Haaren [Mon, 29 Jan 2018 16:37:31 +0000 (16:37 +0000)]
app/pdump: call EAL cleanup before exit

This patch adds a call to the newly introduced cleanup()
function just before quitting the pdump app.

Adding this function call before quitting from a secondary processes
is important, as otherwise it will leak hugepage memory. For a secondary
process that is run multiple times, this could cause hugepage memory
to become depleted and stop a secondary process from starting.

Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Vipin Varghese <vipin.varghese@intel.com>
6 years agoeal: add function to release internal resources
Harry van Haaren [Mon, 29 Jan 2018 16:37:30 +0000 (16:37 +0000)]
eal: add function to release internal resources

This commit adds a new function rte_eal_cleanup().
The function serves as a hook to allow DPDK to release
internal resources (e.g.: hugepage allocations).

This function allows DPDK to become more like an ordinary
library, where the library context itself can be initialized
and cleaned up by the application.

The rte_exit() and rte_panic() functions must be considered,
particularly if they should call rte_eal_cleanup() to release any
resources or not. This patch adds the cleanup to rte_exit(),
but does not clean up on rte_panic(). The reason to not clean
up on panicing is that the developer may wish to inspect the
exact internal state of EAL and hugepages.

Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Vipin Varghese <vipin.varghese@intel.com>
6 years agoservice: restrict finalize to internal usage
Harry van Haaren [Mon, 29 Jan 2018 16:37:29 +0000 (16:37 +0000)]
service: restrict finalize to internal usage

This commit moves the rte_service_finalize() function
to be in the component header, and marks it as @internal.
The function is only called internally by rte_eal_finalize().

Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Vipin Varghese <vipin.varghese@intel.com>
6 years agobus/fslmc: register platform HW mempool on runtime
Hemant Agrawal [Mon, 29 Jan 2018 08:10:49 +0000 (13:40 +0530)]
bus/fslmc: register platform HW mempool on runtime

Detect if the DPAA2 mempool objects are present and register
it as platform default hw mempool

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
6 years agobus/dpaa: register platform HW mempool on runtime
Hemant Agrawal [Mon, 29 Jan 2018 08:10:48 +0000 (13:40 +0530)]
bus/dpaa: register platform HW mempool on runtime

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
6 years agoapp/testpmd: add log for preferred mempool ops
Pavan Nikhilesh [Mon, 29 Jan 2018 08:10:47 +0000 (13:40 +0530)]
app/testpmd: add log for preferred mempool ops

This patch adds the debug message to print the best selected
pktmbuf mempool ops name.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Reviewed-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
6 years agombuf: add pool create helper for specific mempool ops
Hemant Agrawal [Mon, 29 Jan 2018 08:10:46 +0000 (13:40 +0530)]
mbuf: add pool create helper for specific mempool ops

Introduce a new helper for pktmbuf pool, which will allow
the application to optionally specify the mempool ops name
as well.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
6 years agombuf: add pool ops selection functions
Hemant Agrawal [Mon, 29 Jan 2018 08:10:45 +0000 (13:40 +0530)]
mbuf: add pool ops selection functions

This patch add support for various mempool ops config helper APIs.

1.User defined mempool ops
2.Platform detected HW mempool ops (active).
3.Best selection of mempool ops by looking into user defined,
  platform registered and compile time configured.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
6 years agombuf: maintain user and compile time mempool ops name
Hemant Agrawal [Mon, 29 Jan 2018 08:10:44 +0000 (13:40 +0530)]
mbuf: maintain user and compile time mempool ops name

At present the userdefined mempool ops name overwrites
the default mempool ops name variable in internal_config.

This patch change the logic to maintain the value of
user defined only in the internal config.

The pktmbuf_create_pool is updated to reflect the same ie.
use user defined. If not present than use the default.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
6 years agoeal: prefix mbuf pool ops name with user defined
Hemant Agrawal [Mon, 29 Jan 2018 08:10:43 +0000 (13:40 +0530)]
eal: prefix mbuf pool ops name with user defined

This patch prefix the mbuf pool ops name with "user" to indicate
that it is user defined.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
6 years agombuf: fix VLAN flags documentation
Olivier Matz [Mon, 29 Jan 2018 09:36:06 +0000 (10:36 +0100)]
mbuf: fix VLAN flags documentation

Fix inconsistency between mbuf structure documentation and flags
documentation.

Fixes: 380a7aab1ae2 ("mbuf: rename deprecated VLAN flags")
Cc: stable@dpdk.org
Reported-by: Morten Brørup <mb@smartsharesystems.com>
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
6 years agombuf: rename Tx VLAN flags
Olivier Matz [Mon, 29 Jan 2018 09:37:07 +0000 (10:37 +0100)]
mbuf: rename Tx VLAN flags

For consistency with the Rx flags, the flags PKT_TX_VLAN_PKT and
PKT_TX_QINQ_PKT are respectively renamed as PKT_TX_VLAN and
PKT_TX_QINQ. The old defines are deprecated but will stay for some time
for compatibility.

Reported-by: Morten Brørup <mb@smartsharesystems.com>
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
6 years agombuf: fix NULL freeing when debug enabled
Olivier Matz [Mon, 29 Jan 2018 09:39:23 +0000 (10:39 +0100)]
mbuf: fix NULL freeing when debug enabled

Do not panic when calling rte_pktmbuf_free(NULL) with mbuf debug
enabled, it is a valid operation.

Fixes: af75078fece3 ("first public release")
Cc: stable@dpdk.org
Reported-by: Keith Wiles <keith.wiles@intel.com>
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
6 years agoeal/x86: use lock-prefixed instructions for SMP barrier
Konstantin Ananyev [Mon, 15 Jan 2018 15:09:31 +0000 (15:09 +0000)]
eal/x86: use lock-prefixed instructions for SMP barrier

On x86 it is possible to use lock-prefixed instructions to get
the similar effect as mfence.
As pointed by Java guys, on most modern HW that gives a better
performance than using mfence:
https://shipilev.net/blog/2014/on-the-fence-with-dependencies/
That patch adopts that technique for rte_smp_mb() implementation.
On BDW 2.2 mb_autotest on single lcore reports 2X cycle reduction,
i.e. from ~110 to ~55 cycles per operation.

Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
6 years agotest: introduce memory barrier test case
Konstantin Ananyev [Mon, 15 Jan 2018 15:04:39 +0000 (15:04 +0000)]
test: introduce memory barrier test case

Simple functional test for rte_smp_mb() implementations.
Also when executed on a single lcore could be used as rough
estimation how many cycles particular implementation of rte_smp_mb()
might take.

Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
6 years agoconfig/thunderx: disable C11 memory model ring
Jerin Jacob [Sun, 3 Dec 2017 12:37:30 +0000 (18:07 +0530)]
config/thunderx: disable C11 memory model ring

On thunderx and octeontx, ring_perf_autotest and
ring_pmd_perf_autotest test shows better performance
when disabling CONFIG_RTE_RING_USE_C11_MEM_MODEL.
On the other hand, Enabling CONFIG_RTE_RING_USE_C11_MEM_MODEL
shows better performance on thunderx2.
Since thunderx2 is using the default armv8 config,
no particular change is required.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
6 years agoring: introduce C11 memory model barrier option
Jia He [Mon, 22 Jan 2018 04:41:28 +0000 (20:41 -0800)]
ring: introduce C11 memory model barrier option

This patch is to support C11 memory model barrier in librte_ring.

There are 2 barrier implementation options in librte_ring (suggested
by Jerin).
1. use rte_smp_rmb
2. use load_acquire/store_release(refer to [1]).
The reason why providing 2 options is the performance benchmark
difference in different arm machines, refer to [2].

CONFIG_RTE_RING_USE_C11_MEM_MODEL is provided, and by default it is "n"
on any architectures and only "y" on arm64 so far.

[1] https://github.com/freebsd/freebsd/blob/master/sys/sys/buf_ring.h#L170
[2] http://dpdk.org/ml/archives/dev/2017-October/080861.html

Suggested-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Signed-off-by: Jia He <jia.he@hxt-semitech.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Jianbo Liu <jianbo.liu@arm.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
6 years agoring: move code in a new header file
Jia He [Mon, 22 Jan 2018 04:41:27 +0000 (20:41 -0800)]
ring: move code in a new header file

Move the common part of rte_ring.h into rte_ring_generic.h.
Move the memory barrier part into update_tail().

No functional changes here.

Suggested-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Suggested-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Signed-off-by: Jia He <jia.he@hxt-semitech.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
6 years agoeal/arm64: remove the braces in memory barrier macros
Jia He [Mon, 22 Jan 2018 04:41:26 +0000 (20:41 -0800)]
eal/arm64: remove the braces in memory barrier macros

for the code as follows:
if (condition)
rte_smp_rmb();
else
rte_smp_wmb();
Without this patch, compiler will report this error:
error: 'else' without a previous 'if'

Fixes: 84733fd0d75e ("eal/arm64: fix memory barrier definition")
Cc: stable@dpdk.org
Signed-off-by: Jia He <jia.he@hxt-semitech.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
6 years agonet/mlx5: fix synchronization on polling Rx completions
Yongseok Koh [Thu, 25 Jan 2018 21:02:50 +0000 (13:02 -0800)]
net/mlx5: fix synchronization on polling Rx completions

Polling a new packet is basically sensing the generation bit in a
completion entry. For some processors not having strongly-ordered memory
model, there has to be a memory barrier between reading the generation bit
and other fields of the entry in order to guarantee data is not stale.

Fixes: 570acdb1da8a ("net/mlx5: add vectorized Rx/Tx burst for ARM")
Cc: stable@dpdk.org
Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: replace I/O memory barrier with coherent version
Yongseok Koh [Thu, 25 Jan 2018 21:02:49 +0000 (13:02 -0800)]
net/mlx5: replace I/O memory barrier with coherent version

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx5: remove unnecessary memory barrier
Yongseok Koh [Thu, 25 Jan 2018 21:02:48 +0000 (13:02 -0800)]
net/mlx5: remove unnecessary memory barrier

As rte_write64() has an IO barrier, there's no need to have a barrier
before the call.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>