dpdk.git
6 years agonet/bonding: fix slave activation simultaneously
Matan Azrad [Tue, 24 Apr 2018 11:29:30 +0000 (11:29 +0000)]
net/bonding: fix slave activation simultaneously

The bonding PMD decides to activate\deactivate its slaves according to
the slaves link statuses.
Thus, it registers to the LSC events of the slaves ports and
activates\deactivates them from its LSC callbacks called asynchronously
by the host thread when the slave link status is changed.

In addition, the bonding PMD uses the callback for slave activation
when it tries to start it, this operation is probably called by the
master thread.

Consequently, a slave may be activated in the same time by two
different threads and may cause a lot of optional errors, for example,
slave mempool recreation with the same name causes an error.

Synchronize the critical section in the LSC callback using a special
new spinlock.

Fixes: 414b202343ce ("bonding: fix initial link status of slave")
Fixes: a45b288ef21a ("bond: support link status polling")
Cc: stable@dpdk.org
Signed-off-by: Matan Azrad <matan@mellanox.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
6 years agonet/avf: fix Rx interrupt mapping
Jingjing Wu [Fri, 11 May 2018 15:09:48 +0000 (23:09 +0800)]
net/avf: fix Rx interrupt mapping

Vector used for rx mapping is different if WB_ON_ITR
is supported. The mapping table need to be updated.

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

Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Cc: stable@dpdk.org
6 years agonet/dpaa2: change VLAN strip value to offload flag
Shreyansh Jain [Mon, 14 May 2018 10:58:26 +0000 (16:28 +0530)]
net/dpaa2: change VLAN strip value to offload flag

Fixes: 0ebce6129bc6 ("net/dpaa2: support new ethdev offload APIs")

Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
6 years agonet/cxgbe: free resources during uninit
Rahul Lakkireddy [Sat, 12 May 2018 11:50:00 +0000 (17:20 +0530)]
net/cxgbe: free resources during uninit

Move freeing up resources from dev_close() to dev_uninit(). This fixes
NULL pointer de-reference when accessing adapter context needed by
other ports under same PF, but had been freed up by the first port.
This can happen if only the first port is started up and the check
to free up all resources is still satisfied. When dev_close is
called for other ports, adapter context is NULL since it was freed
up by the first port.

Thus, by moving to dev_uninit() all the ports can be teared down
safely without need for extra checks.

Fixes: 2195df6d11bd ("net/cxgbe: rework ethdev device allocation")

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
6 years agonet/mlx4: advertise supported RSS hash functions
Ophir Munk [Mon, 14 May 2018 10:07:32 +0000 (10:07 +0000)]
net/mlx4: advertise supported RSS hash functions

Advertise mlx4 supported RSS functions as part of dev_infos_get
callback.
Previous to this commit RSS support was reported as none. Since the
introduction of [1] it is required that all RSS configurations will be
verified.

[1] commit 8863a1fbfc66 ("ethdev: add supported hash function check")

Signed-off-by: Ophir Munk <ophirmu@mellanox.com>
6 years agonet/mlx4: avoid constant recreations in function
Ophir Munk [Mon, 14 May 2018 10:07:31 +0000 (10:07 +0000)]
net/mlx4: avoid constant recreations in function

Function mlx4_conv_rss_types() contains constant arrays variables
which are recreated with every call to the function. By changing the
arrays definitions from "const" to "static const" these recreations
can be saved.

Signed-off-by: Ophir Munk <ophirmu@mellanox.com>
6 years agonet/mlx5: use correct field in a union structure
Yongseok Koh [Sat, 12 May 2018 01:35:45 +0000 (18:35 -0700)]
net/mlx5: use correct field in a union structure

This is not a bug but it is better to use semantically correct field.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx5: use coherent I/O memory barrier
Yongseok Koh [Sat, 12 May 2018 01:35:44 +0000 (18:35 -0700)]
net/mlx5: use coherent I/O memory barrier

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx5: fix inlining segmented TSO packet
Yongseok Koh [Fri, 11 May 2018 17:39:13 +0000 (10:39 -0700)]
net/mlx5: fix inlining segmented TSO packet

When a multi-segmented packet is inlined, data can be further inlined even
after the first segment. In case of TSO packet, extra inline data after TSO
header should be carried by an inline DSEG which has 4B inline header
recording the length of the inline data. If more than one segment is
inlined, the length doesn't count from the second segment. This will cause
a fault in HW and CQE will have an error, which is ignored by PMD.

Fixes: f895536be4fa ("net/mlx5: enable inlining data from multiple segments")
Cc: stable@dpdk.org
Signed-off-by: Xueming Li <xuemingl@mellanox.com>
Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
6 years agoapp/testpmd: check if CRC strip offload supported
Ferruh Yigit [Wed, 9 May 2018 22:09:04 +0000 (23:09 +0100)]
app/testpmd: check if CRC strip offload supported

Testpmd set CRC_STRIP offload blindly, this is wrong according offload
API definition, and will cause error for the PMDs that doesn't support
CRC_STRIP like virtual PMDs.

Check if underlying device report this capability and don't set it if
not supported.

Fixes: 0074d02fca21 ("app/testpmd: convert to new Rx offloads API")

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Bernard Iremonger <bernard.iremonger@intel.com>
6 years agonet/failsafe: add an RSS hash update callback
Ophir Munk [Thu, 10 May 2018 14:38:10 +0000 (14:38 +0000)]
net/failsafe: add an RSS hash update callback

Add an RSS hash update callback to eth_dev_ops.

Signed-off-by: Ophir Munk <ophirmu@mellanox.com>
Acked-by: Gaetan Rivet <gaetan.rivet@6wind.com>
6 years agoethdev: improve doc for name by port ID API
Ivan Malov [Fri, 11 May 2018 14:05:51 +0000 (15:05 +0100)]
ethdev: improve doc for name by port ID API

Description of rte_eth_dev_get_name_by_port() calls
port ID argument a pointer, which is misleading.
Also, output buffer minimal size is not mentioned.
These points need to be improved.

Fixes: bde516d5a85a ("ethdev: get port by name")
Cc: stable@dpdk.org
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
6 years agonet/ixgbe: fix missing port representor data-path
Remy Horton [Fri, 11 May 2018 12:28:45 +0000 (13:28 +0100)]
net/ixgbe: fix missing port representor data-path

This patch adds Rx and Tx burst functions to the ixgbe
Port Representors, so that the implementation within
ixgbe PMD can be tested using applications such as
testpmd which require data-path functionality.

Fixes: cf80ba6e2038 ("net/ixgbe: add support for representor ports")

Signed-off-by: Remy Horton <remy.horton@intel.com>
Acked-by: Mohammad Abdul Awal <mohammad.abdul.awal@intel.com>
6 years agonet/i40e: fix missing port representor data-path
Remy Horton [Fri, 11 May 2018 12:28:44 +0000 (13:28 +0100)]
net/i40e: fix missing port representor data-path

This patch adds Rx and Tx burst functions to the i40e Port
Representors, so that the implementation within this PMD
can be tested using applications such as testpmd which
require data-path functionality.

Fixes: e0cb96204b71 ("net/i40e: add support for representor ports")

Signed-off-by: Remy Horton <remy.horton@intel.com>
Acked-by: Mohammad Abdul Awal <mohammad.abdul.awal@intel.com>
6 years agoethdev: fix checking Rx/Tx queue status
Yanglong Wu [Fri, 11 May 2018 08:22:28 +0000 (16:22 +0800)]
ethdev: fix checking Rx/Tx queue status

Relax the check for queue setup, since some device
may not update queue states during dev_stop.

Fixes: cac923cfea47 ("ethdev: support runtime queue setup")

Signed-off-by: Yanglong Wu <yanglong.wu@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
6 years agonet/i40e: print original value for global register change
Beilei Xing [Fri, 11 May 2018 18:17:07 +0000 (02:17 +0800)]
net/i40e: print original value for global register change

Currently, only new value is printed during global
register change. Add original value to help debugging
facility.

Fixes: bc66b9717c50 ("net/i40e: add debug logs when writing global registers")
Cc: stable@dpdk.org
Signed-off-by: Beilei Xing <beilei.xing@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
6 years agonet/virtio-user: fix multiple queues fail in server mode
Zhiyong Yang [Fri, 11 May 2018 03:31:37 +0000 (11:31 +0800)]
net/virtio-user: fix multiple queues fail in server mode

This patch fixes multiple queues failure when virtio-user works in
server mode.

This patch adds feature negotiation in the processing of virtio-user
connection and enables multiple-queue pairs.

Fixes: bd8f50a45d0f ("net/virtio-user: support server mode")
Cc: stable@dpdk.org
Signed-off-by: Zhiyong Yang <zhiyong.yang@intel.com>
Reviewed-by: Tiwei Bie <tiwei.bie@intel.com>
6 years agonet/i40e: fix missing VLAN offload capability
Yanglong Wu [Thu, 10 May 2018 06:50:38 +0000 (14:50 +0800)]
net/i40e: fix missing VLAN offload capability

VLAN offload capability should be exposed in VF
since i40e does support it.

Fixes: c3ac7c5b0b8a ("net/i40e: convert to new Rx offloads API")

Signed-off-by: Yanglong Wu <yanglong.wu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
6 years agonet/i40e: fix missing mbuf fast free offload
Qi Zhang [Fri, 11 May 2018 01:25:42 +0000 (09:25 +0800)]
net/i40e: fix missing mbuf fast free offload

Expose the missing mbuf fast free capability since i40 does
support it.

Fixes: 7497d3e2f777 ("net/i40e: convert to new Tx offloads API")

Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
6 years agoethdev: fix port removal notification timing
Matan Azrad [Thu, 10 May 2018 23:58:36 +0000 (01:58 +0200)]
ethdev: fix port removal notification timing

When an ethdev port is released, a destroy event is triggered to notify
the users about the released port.

A bit before the destroy event is triggered, the port becomes invalid
by changing its state to UNUSED and cleaning its data. Therefore, the
port is invalid for the destroy event callback process and the users
may get a wrong information of the port.

Move the destroy event emitting to be called before the port
invalidation.

Fixes: 133b54779aa1 ("ethdev: fix port data reset timing")
Fixes: 29aa41e36de7 ("ethdev: add notifications for probing and removal")
Cc: stable@dpdk.org
Signed-off-by: Matan Azrad <matan@mellanox.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
6 years agonet/failsafe: fix sub-device ownership race
Matan Azrad [Thu, 10 May 2018 23:58:35 +0000 (01:58 +0200)]
net/failsafe: fix sub-device ownership race

There is time between the sub-device port probing by the sub-device PMD
to the sub-device port ownership taking by a fail-safe port.

In this time, the port is available for the application usage. For
example, the port will be exposed to the applications which use
RTE_ETH_FOREACH_DEV iterator.

Thus, ownership unaware applications may manage the port in this time
what may cause a lot of problematic behaviors in the fail-safe
sub-device initialization.

Register to the ethdev NEW event to take the sub-device port ownership
before it becomes exposed to the application.

Fixes: a46f8d584eb8 ("net/failsafe: add fail-safe PMD")
Cc: stable@dpdk.org
Signed-off-by: Matan Azrad <matan@mellanox.com>
Acked-by: Gaetan Rivet <gaetan.rivet@6wind.com>
Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
6 years agoethdev: fix port probing notification
Thomas Monjalon [Thu, 10 May 2018 23:58:34 +0000 (01:58 +0200)]
ethdev: fix port probing notification

The new device was notified as soon as it was allocated.
It leads to use a device which is not yet initialized.

The notification must be published after the initialization is done
by the PMD, but before the state is changed, in order to let
notified entities taking ownership before general availability.

Fixes: 29aa41e36de7 ("ethdev: add notifications for probing and removal")
Cc: stable@dpdk.org
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
6 years agoethdev: fix port visibility before initialization
Thomas Monjalon [Thu, 10 May 2018 23:58:33 +0000 (01:58 +0200)]
ethdev: fix port visibility before initialization

The port was set to the state ATTACHED during allocation.
The consequence was to iterate over ports which are not initialized.

The state ATTACHED is now set as the last step of probing.

The uniqueness of port name is now checked before the availability
of a port id for allocation (order reversed).

As the state is not set on allocation anymore, it is also not checked
in the function telling whether a port is allocated or not.
The name of the port is set on allocation, so it is enough as a check.

Fixes: 5588909af21b ("ethdev: add device iterator")
Cc: stable@dpdk.org
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Signed-off-by: Matan Azrad <matan@mellanox.com>
Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
6 years agoethdev: add lock to port allocation check
Matan Azrad [Thu, 10 May 2018 23:58:32 +0000 (01:58 +0200)]
ethdev: add lock to port allocation check

When comparing the port name, there can be a race condition with
a thread allocating a new port and writing the name at the same time.
It can lead to match with a partial name by error.

The check of the port is now considered as a critical section
protected with locks.

This fix will be even more required for multi-process when the
port availability will rely only on the name, in a following patch.

Fixes: 84934303a17c ("ethdev: synchronize port allocation")
Cc: stable@dpdk.org
Signed-off-by: Matan Azrad <matan@mellanox.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
6 years agoethdev: allow ownership operations on unused port
Matan Azrad [Thu, 10 May 2018 23:58:31 +0000 (01:58 +0200)]
ethdev: allow ownership operations on unused port

When the state will be updated later than in allocation,
we may need to update the ownership of a port which is
still in state unused.

It will be used to take ownership of a port before it is
declared as available for other entities.

Cc: stable@dpdk.org
Signed-off-by: Matan Azrad <matan@mellanox.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
6 years agoethdev: add probing finish function
Thomas Monjalon [Thu, 10 May 2018 23:58:30 +0000 (01:58 +0200)]
ethdev: add probing finish function

A new hook function is added and called inside the PMDs at the end
of the device probing:
- in primary process, after allocating, init and config
- in secondary process, after attaching and local init

This new function is almost empty for now.
It will be used later to add some post-initialization processing.

For the PMDs calling the helpers rte_eth_dev_create() or
rte_eth_dev_pci_generic_probe(), the hook rte_eth_dev_probing_finish()
is called from here, and not in the PMD itself.

Note that the helper rte_eth_dev_create() could be used more,
especially for vdevs, avoiding some code duplication in PMDs.

Cc: stable@dpdk.org
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
6 years agodrivers/net: use higher level of probing helper for PCI
Thomas Monjalon [Thu, 10 May 2018 23:58:29 +0000 (01:58 +0200)]
drivers/net: use higher level of probing helper for PCI

The drivers avp, bnx2x and liquidio were using the helper function
rte_eth_dev_pci_allocate() and can be replaced by
rte_eth_dev_pci_generic_probe() which calls the former.

Fixes: dcd5c8112bc3 ("ethdev: add PCI driver helpers")
Cc: stable@dpdk.org
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
6 years agoethdev: add doxygen comments for each state
Thomas Monjalon [Thu, 10 May 2018 23:58:28 +0000 (01:58 +0200)]
ethdev: add doxygen comments for each state

The enum rte_eth_dev_state was not properly documented.
Its values did not appear in the doxygen output,
and may be misunderstood.

The state RTE_ETH_DEV_DEFERRED has no interest anymore
since the ownership mechanism brings a more flexible categorization.
This state could be removed later.

Fixes: d52268a8b24b ("ethdev: expose device states")
Fixes: cb894d99eceb ("ethdev: add deferred intermediate device state")
Fixes: 5b7ba31148a8 ("ethdev: add port ownership")
Fixes: 7106edc12380 ("ethdev: add devop to check removal status")

Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Acked-by: Matan Azrad <matan@mellanox.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
6 years agonet/failsafe: fix sub-device visibility
Thomas Monjalon [Thu, 10 May 2018 23:58:27 +0000 (01:58 +0200)]
net/failsafe: fix sub-device visibility

The iterator function rte_eth_find_next_owned_by(), used by the
iterator macro RTE_ETH_FOREACH_DEV_OWNED_BY, are ignoring the devices
which are neither ATTACHED nor REMOVED. Thus sub-devices, having
the state DEFERRED, cannot be seen with the ethdev iterator.
The state RTE_ETH_DEV_DEFERRED can be replaced by
RTE_ETH_DEV_ATTACHED + owner.

Fixes: dcd0c9c32b8d ("net/failsafe: use ownership mechanism for slaves")
Cc: stable@dpdk.org
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Acked-by: Matan Azrad <matan@mellanox.com>
Acked-by: Gaetan Rivet <gaetan.rivet@6wind.com>
Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
6 years agoethdev: fix debug log of owner id
Thomas Monjalon [Thu, 10 May 2018 23:58:26 +0000 (01:58 +0200)]
ethdev: fix debug log of owner id

The owner id is 64-bit.
On 32-bit environment, it must be printed with PRIX64.

Fixes: 5b7ba31148a8 ("ethdev: add port ownership")
Cc: stable@dpdk.org
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
6 years agoapp/testpmd: add commands to test new offload API
Wei Dai [Wed, 9 May 2018 12:13:50 +0000 (20:13 +0800)]
app/testpmd: add commands to test new offload API

Add following testpmd run-time commands to support test of
new Rx offload API:
show port <port_id> rx_offload capabilities
show port <port_id> rx_offload configuration
port config <port_id> rx_offload <offload> on|off
port <port_id> rxq <queue_id> rx_offload <offload> on|off
Above last 2 commands should be run when the port is stopped.
And <offload> can be one of "vlan_strip", "ipv4_cksum", ...

Add following testpmd run-time commands to support test of
new Tx offload API:
show port <port_id> tx_offload capabilities
show port <port_id> tx_offload configuration
port config <port_id> tx_offload <offload> on|off
port <port_id> txq <queue_id> tx_offload <offload> on|off
Above last 2 commands should be run when the port is stopped.
And <offload> can be one of "vlan_insert", "udp_cksum", ...

Signed-off-by: Wei Dai <wei.dai@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>
6 years agonet/i40e: print global register change info
Beilei Xing [Thu, 10 May 2018 22:48:38 +0000 (06:48 +0800)]
net/i40e: print global register change info

Global register change info during enabling
flexible payload is not printed.
This patch changes macro to print the global
register change info.

Fixes: d2f9fe8ae309 ("net/i40e: turn off flexible payload on driver init")
Cc: stable@dpdk.org
Signed-off-by: Beilei Xing <beilei.xing@intel.com>
6 years agonet/sfc: fix inner TCP/UDP checksum offload control
Andrew Rybchenko [Thu, 10 May 2018 11:59:42 +0000 (12:59 +0100)]
net/sfc: fix inner TCP/UDP checksum offload control

If application uses Tx offload API and sets ETH_TXQ_FLAGS_IGNORE flag,
it still should have inner TCP/UDP checksum offload enabled if it is
supported and TCP/UDP checksum offload is requested.

Fixes: c78d280e88ef ("net/sfc: convert to new Tx offload API")
Cc: stable@dpdk.org
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
6 years agodoc: add XXV710 support in i40e guide
Qiming Yang [Thu, 10 May 2018 11:10:25 +0000 (19:10 +0800)]
doc: add XXV710 support in i40e guide

Signed-off-by: Qiming Yang <qiming.yang@intel.com>
6 years agonet/enic: fix flow drop action
Hyong Youb Kim [Thu, 10 May 2018 08:51:13 +0000 (01:51 -0700)]
net/enic: fix flow drop action

Drop is a fate-deciding action, so mark it as FATE. It was missing in
a previous commit.

Fixes: cc17feb90413 ("ethdev: alter behavior of flow API actions")

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
6 years agonet/e1000: report Tx multi segment offload
Wei Dai [Thu, 10 May 2018 03:56:50 +0000 (11:56 +0800)]
net/e1000: report Tx multi segment offload

This feature has been confirmed with testpmd:
testpmd> set fwd txonly
testpmd> port stop all
testpmd> port config all txd 1024
testpmd> set txsplit on
testpmd> set txpkts 70,80,90,100
testpmd> start
It can be observed at peer port that UDP packets
with UDP data length 298 bytes.

Signed-off-by: Wei Dai <wei.dai@intel.com>
6 years agonet/i40e: fix link status update
Beilei Xing [Thu, 10 May 2018 02:26:29 +0000 (10:26 +0800)]
net/i40e: fix link status update

Link status is not updated correctly, link speed is 0
when link is up and link speed is not 0 when link is
down. This patch fixes the issue.

Fixes: eef2daf2e199 ("net/i40e: fix link update no wait")
Cc: stable@dpdk.org
Signed-off-by: Keith Wiles <keith.wiles@intel.com>
Signed-off-by: Beilei Xing <beilei.xing@intel.com>
6 years agoapp/testpmd: fix device configure with zero queue
Qi Zhang [Thu, 10 May 2018 02:22:02 +0000 (10:22 +0800)]
app/testpmd: fix device configure with zero queue

Setup number of Rx & Tx queues to 0 at rte_eth_dev_configure means
take driver's default queue number, so if during a re-configuration
previous queue number will be overwrite, this is not expected when
we configure dcb. The patch fix it by re-configure device with the
original queue number.

Fixes: 3be82f5cc5e ("ethdev: support PMD-tuned Tx/Rx parameters")

Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
Acked-by: Remy Horton <remy.horton@intel.com>
6 years agonet/mlx5: add Multi-Packet Rx support
Yongseok Koh [Wed, 9 May 2018 11:13:50 +0000 (04:13 -0700)]
net/mlx5: add Multi-Packet Rx support

Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe
bandwidth by posting a single large buffer for multiple packets. Instead of
posting a buffer per a packet, one large buffer is posted in order to
receive multiple packets on the buffer. A MPRQ buffer consists of multiple
fixed-size strides and each stride receives one packet.

Rx packet is mem-copied to a user-provided mbuf if the size of Rx packet is
comparatively small, or PMD attaches the Rx packet to the mbuf by external
buffer attachment - rte_pktmbuf_attach_extbuf(). A mempool for external
buffers will be allocated and managed by PMD.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
6 years agonet/mlx5: add a function to rdma-core glue
Yongseok Koh [Wed, 9 May 2018 11:13:49 +0000 (04:13 -0700)]
net/mlx5: add a function to rdma-core glue

mlx5dv_create_wq() is added for the Multi-Packet RQ (a.k.a Striding RQ).

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
6 years agonet/mlx5: separate filling Rx flags
Yongseok Koh [Wed, 9 May 2018 11:13:48 +0000 (04:13 -0700)]
net/mlx5: separate filling Rx flags

Filling in fields of mbuf becomes a separate inline function so that this
can be reused.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
6 years agonet/mlx4: add new memory region support
Yongseok Koh [Wed, 9 May 2018 11:09:06 +0000 (04:09 -0700)]
net/mlx4: add new memory region support

This is the new design of Memory Region (MR) for mlx PMD, in order to:
- Accommodate the new memory hotplug model.
- Support non-contiguous Mempool.

There are multiple layers for MR search.

L0 is to look up the last-hit entry which is pointed by mr_ctrl->mru (Most
Recently Used). If L0 misses, L1 is to look up the address in a fixed-sized
array by linear search. L0/L1 is in an inline function -
mlx4_mr_lookup_cache().

If L1 misses, the bottom-half function is called to look up the address
from the bigger local cache of the queue. This is L2 - mlx4_mr_addr2mr_bh()
and it is not an inline function. Data structure for L2 is the Binary Tree.

If L2 misses, the search falls into the slowest path which takes locks in
order to access global device cache (priv->mr.cache) which is also a B-tree
and caches the original MR list (priv->mr.mr_list) of the device. Unless
the global cache is overflowed, it is all-inclusive of the MR list. This is
L3 - mlx4_mr_lookup_dev(). The size of the L3 cache table is limited and
can't be expanded on the fly due to deadlock. Refer to the comments in the
code for the details - mr_lookup_dev(). If L3 is overflowed, the list will
have to be searched directly bypassing the cache although it is slower.

If L3 misses, a new MR for the address should be created -
mlx4_mr_create(). When it creates a new MR, it tries to register adjacent
memsegs as much as possible which are virtually contiguous around the
address. This must take two locks - memory_hotplug_lock and
priv->mr.rwlock. Due to memory_hotplug_lock, there can't be any
allocation/free of memory inside.

In the free callback of the memory hotplug event, freed space is searched
from the MR list and corresponding bits are cleared from the bitmap of MRs.
This can fragment a MR and the MR will have multiple search entries in the
caches. Once there's a change by the event, the global cache must be
rebuilt and all the per-queue caches will be flushed as well. If memory is
frequently freed in run-time, that may cause jitter on dataplane processing
in the worst case by incurring MR cache flush and rebuild. But, it would be
the least probable scenario.

To guarantee the most optimal performance, it is highly recommended to use
an EAL option - '--socket-mem'. Then, the reserved memory will be pinned
and won't be freed dynamically. And it is also recommended to configure
per-lcore cache of Mempool. Even though there're many MRs for a device or
MRs are highly fragmented, the cache of Mempool will be much helpful to
reduce misses on per-queue caches anyway.

'--legacy-mem' is also supported.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx4: remove memory region support
Yongseok Koh [Wed, 9 May 2018 11:09:05 +0000 (04:09 -0700)]
net/mlx4: remove memory region support

This patch removes current support of Memory Region (MR) in order to
accommodate the dynamic memory hotplug patch. This patch can be compiled
but traffic can't flow and HW will raise faults. Subsequent patches will
add new MR support.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx5: add new memory region support
Yongseok Koh [Wed, 9 May 2018 11:09:04 +0000 (04:09 -0700)]
net/mlx5: add new memory region support

This is the new design of Memory Region (MR) for mlx PMD, in order to:
- Accommodate the new memory hotplug model.
- Support non-contiguous Mempool.

There are multiple layers for MR search.

L0 is to look up the last-hit entry which is pointed by mr_ctrl->mru (Most
Recently Used). If L0 misses, L1 is to look up the address in a fixed-sized
array by linear search. L0/L1 is in an inline function -
mlx5_mr_lookup_cache().

If L1 misses, the bottom-half function is called to look up the address
from the bigger local cache of the queue. This is L2 - mlx5_mr_addr2mr_bh()
and it is not an inline function. Data structure for L2 is the Binary Tree.

If L2 misses, the search falls into the slowest path which takes locks in
order to access global device cache (priv->mr.cache) which is also a B-tree
and caches the original MR list (priv->mr.mr_list) of the device. Unless
the global cache is overflowed, it is all-inclusive of the MR list. This is
L3 - mlx5_mr_lookup_dev(). The size of the L3 cache table is limited and
can't be expanded on the fly due to deadlock. Refer to the comments in the
code for the details - mr_lookup_dev(). If L3 is overflowed, the list will
have to be searched directly bypassing the cache although it is slower.

If L3 misses, a new MR for the address should be created -
mlx5_mr_create(). When it creates a new MR, it tries to register adjacent
memsegs as much as possible which are virtually contiguous around the
address. This must take two locks - memory_hotplug_lock and
priv->mr.rwlock. Due to memory_hotplug_lock, there can't be any
allocation/free of memory inside.

In the free callback of the memory hotplug event, freed space is searched
from the MR list and corresponding bits are cleared from the bitmap of MRs.
This can fragment a MR and the MR will have multiple search entries in the
caches. Once there's a change by the event, the global cache must be
rebuilt and all the per-queue caches will be flushed as well. If memory is
frequently freed in run-time, that may cause jitter on dataplane processing
in the worst case by incurring MR cache flush and rebuild. But, it would be
the least probable scenario.

To guarantee the most optimal performance, it is highly recommended to use
an EAL option - '--socket-mem'. Then, the reserved memory will be pinned
and won't be freed dynamically. And it is also recommended to configure
per-lcore cache of Mempool. Even though there're many MRs for a device or
MRs are highly fragmented, the cache of Mempool will be much helpful to
reduce misses on per-queue caches anyway.

'--legacy-mem' is also supported.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx5: remove memory region support
Yongseok Koh [Wed, 9 May 2018 11:09:03 +0000 (04:09 -0700)]
net/mlx5: remove memory region support

This patch removes current support of Memory Region (MR) in order to
accommodate the dynamic memory hotplug patch. This patch can be compiled
but traffic can't flow and HW will raise faults. Subsequent patches will
add new MR support.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
6 years agoethdev: new Rx/Tx offloads API
Wei Dai [Thu, 10 May 2018 11:56:55 +0000 (19:56 +0800)]
ethdev: new Rx/Tx offloads API

This patch check if a input requested offloading is valid or not.
Any reuqested offloading must be supported in the device capabilities.
Any offloading is disabled by default if it is not set in the parameter
dev_conf->[rt]xmode.offloads to rte_eth_dev_configure() and
[rt]x_conf->offloads to rte_eth_[rt]x_queue_setup().
If any offloading is enabled in rte_eth_dev_configure() by application,
it is enabled on all queues no matter whether it is per-queue or
per-port type and no matter whether it is set or cleared in
[rt]x_conf->offloads to rte_eth_[rt]x_queue_setup().
If a per-queue offloading hasn't be enabled in rte_eth_dev_configure(),
it can be enabled or disabled for individual queue in
ret_eth_[rt]x_queue_setup().
A new added offloading is the one which hasn't been enabled in
rte_eth_dev_configure() and is reuqested to be enabled in
rte_eth_[rt]x_queue_setup(), it must be per-queue type,
otherwise trigger an error log.
The underlying PMD must be aware that the requested offloadings
to PMD specific queue_setup() function only carries those
new added offloadings of per-queue type.

This patch can make above such checking in a common way in rte_ethdev
layer to avoid same checking in underlying PMD.

This patch assumes that all PMDs in 18.05-rc2 have already
converted to offload API defined in 17.11 . It also assumes
that all PMDs can return correct offloading capabilities
in rte_eth_dev_infos_get().

In the beginning of [rt]x_queue_setup() of underlying PMD,
add offloads = [rt]xconf->offloads |
dev->data->dev_conf.[rt]xmode.offloads; to keep same as offload API
defined in 17.11 to avoid upper application broken due to offload
API change.
PMD can use the info that input [rt]xconf->offloads only carry
the new added per-queue offloads to do some optimization or some
code change on base of this patch.

Signed-off-by: Wei Dai <wei.dai@intel.com>
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
6 years agonet/mlx5: change device reference for secondary process
Yongseok Koh [Wed, 9 May 2018 11:04:50 +0000 (04:04 -0700)]
net/mlx5: change device reference for secondary process

rte_eth_devices[] is not shared between primary and secondary process, but
a static array to each process. The reverse pointer of device (priv->dev)
is invalid. Instead, priv has the pointer to shared data of the device,
  struct rte_eth_dev_data *dev_data;

Two macros are added,
  #define PORT_ID(priv) ((priv)->dev_data->port_id)
  #define ETH_DEV(priv) (&rte_eth_devices[PORT_ID(priv)])

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx4: fix CRC stripping capability report
Ophir Munk [Tue, 8 May 2018 12:26:03 +0000 (12:26 +0000)]
net/mlx4: fix CRC stripping capability report

There are two capabilities related to CRC stripping:
1. mlx4 HW capability to perform CRC stripping on a received packet.
This capability is built in mlx4 HW. It should be returned by the API
call mlx4_get_rx_queue_offloads().
2. mlx4 driver capability to enable/disable HW CRC stripping. This
capability is dependent on the driver version.

Before this commit the second capability was falsely returned by
the mentioned API. This commit fixes it by returning the first
capability.
mlx4 HW performs CRC stripping by default and this capability is
always reported as "true".

The ability to enable/disable CRC stripping is supported since this
commit and requires OFED version 4.3-1.5.0.0 or rdma-core version v18.
CRC stripping will be done by default regardless of its configuration
when working with OFED or rdma-core versions earlier than those
previously specified or before this commit.

Fixes: de1df14e6e6ec ("net/mlx4: support CRC strip toggling")
Cc: stable@dpdk.org
Signed-off-by: Ophir Munk <ophirmu@mellanox.com>
6 years agoethdev: fix corrupted device info in configure
Ferruh Yigit [Wed, 9 May 2018 22:16:49 +0000 (23:16 +0100)]
ethdev: fix corrupted device info in configure

Calling dev_infos_get() devops directly in rte_eth_dev_configure cause
random values in uninitialized fields because devops doesn't reset the
dev_info structure.

Call rte_eth_dev_info_get() API instead which memset the struct.

Also remove duplicated dev_infos_get existence check.

Fixes: 3be82f5cc5e3 ("ethdev: support PMD-tuned Tx/Rx parameters")

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
6 years agonet/failsafe: fix probe cleanup
Raslan Darawsheh [Wed, 9 May 2018 15:57:39 +0000 (18:57 +0300)]
net/failsafe: fix probe cleanup

The hot-plug alarm mechanism is responsible to practically execute both
plug in and out operations. It periodically tries to detect missed
sub-devices to be reconfigured and clean the resources of the removed
sub-devices.

The hot-plug alarm is started by the failsafe probe function, and it's
wrongly not stopped if failsafe instance got an error. for example
when starting failsafe with a MAC option, and giving it an invalid MAC
address this will lead to a NULL pointer for the dev private field. Then
when the hotplug alarm is called it will try to access this pointer,
which will lead to a segmentation fault.

Uninstall the hot-plug alarm in case of error in probe function.

Fixes: ebea83f8 ("net/failsafe: add plug-in support")
Cc: stable@dpdk.org
Signed-off-by: Raslan Darawsheh <rasland@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
Acked-by: Gaetan Rivet <gaetan.rivet@6wind.com>
6 years agonet/failsafe: advertise supported RSS functions
Ophir Munk [Wed, 9 May 2018 13:46:41 +0000 (13:46 +0000)]
net/failsafe: advertise supported RSS functions

Advertise failsafe supported RSS functions as part of dev_infos_get
callback. Set failsafe default RSS hash functions to be:
ETH_RSS_IP, ETH_RSS_UDP, and ETH_RSS_TCP.
The result of failsafe RSS hash functions is the logical AND of the
RSS hash functions among all failsafe sub_devices and failsafe own
defaults.

Previous to this commit RSS support was reported as none. Since the
introduction of [1] it is required that all RSS configurations be
verified.

[1] commit 8863a1fbfc66 ("ethdev: add supported hash function check")

Signed-off-by: Ophir Munk <ophirmu@mellanox.com>
Acked-by: Gaetan Rivet <gaetan.rivet@6wind.com>
6 years agonet/dpaa: update optimal burst size in device info
Shreyansh Jain [Wed, 9 May 2018 09:49:44 +0000 (15:19 +0530)]
net/dpaa: update optimal burst size in device info

Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
6 years agonet/dpaa: fix max push mode queue
Shreyansh Jain [Wed, 9 May 2018 09:49:43 +0000 (15:19 +0530)]
net/dpaa: fix max push mode queue

Split default and max push mode queues to 4 and 8, respectively.

Fixes: 0c504f6950b6 ("net/dpaa: support push mode")
Cc: stable@dpdk.org
Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
6 years agonet/tap: report on supported RSS hash functions
Ophir Munk [Tue, 8 May 2018 17:07:35 +0000 (17:07 +0000)]
net/tap: report on supported RSS hash functions

Report on TAP supported RSS functions as part of dev_infos_get
callback: ETH_RSS_IP, ETH_RSS_UDP and ETH_RSS_TCP.
Known limitation: TAP supports all of the above hash functions together
and not in partial combinations.
Previous to this commit RSS support was reported as none. Since the
introduction of [1] it is required that all RSS configurations will be
verified.

[1] commit 8863a1fbfc66 ("ethdev: add supported hash function check")

Signed-off-by: Ophir Munk <ophirmu@mellanox.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
6 years agoapp/testpmd: fix weak RSS hash key for flow
Adrien Mazarguil [Thu, 3 May 2018 11:59:44 +0000 (13:59 +0200)]
app/testpmd: fix weak RSS hash key for flow

The default RSS hash key automatically provided by testpmd for RSS actions
specified without one is so weak that traffic can't spread properly on L4
with it (as seen with TCPv6).

It is only 30 bytes long, zero-padded to RSS_HASH_KEY_LENGTH (64 bytes),
later truncated to 40 bytes for most PMDs. The presence of padding is
really what kills balancing.

This patch provides a full 64-byte (non-zero-terminated) string to address
this issue.

Fixes: d0ad8648b1c5 ("app/testpmd: fix RSS flow action configuration")
Cc: stable@dpdk.org
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
6 years agoapp/testpmd: fix empty list of RSS queues for flow
Adrien Mazarguil [Thu, 3 May 2018 11:59:42 +0000 (13:59 +0200)]
app/testpmd: fix empty list of RSS queues for flow

Since the commit referenced below, specifying a RSS action without any
queues (e.g. "actions rss queues end / end") does not override the default
set automatically generated by testpmd.

In short, one cannot instantiate a RSS action with 0 target queues anymore
in order to determine how PMDs react (hint: this is currently undocumented
so they may reject it, however ideally they should interpret it as a
default setting like for other fields where empty values stand for
"defaults".)

Fixes: d0ad8648b1c5 ("app/testpmd: fix RSS flow action configuration")
Cc: stable@dpdk.org
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
6 years agonet/bnxt: add NVM specific HWRM commands
Ajit Khaparde [Thu, 3 May 2018 17:23:58 +0000 (10:23 -0700)]
net/bnxt: add NVM specific HWRM commands

This patch adds new NVM specific HWRM commands.
Code using these commands will be added in future patches.

Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: add HWRM commands for more filtering support
Ajit Khaparde [Thu, 3 May 2018 17:23:57 +0000 (10:23 -0700)]
net/bnxt: add HWRM commands for more filtering support

Add HWRM structures to support more filtering features
like metering, tunnel filters and more.

Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: add async event HWRM commands
Ajit Khaparde [Thu, 3 May 2018 17:23:56 +0000 (10:23 -0700)]
net/bnxt: add async event HWRM commands

add structures to add async_event support.

Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: update HWRM to version 1.9.2
Ajit Khaparde [Thu, 3 May 2018 17:23:55 +0000 (10:23 -0700)]
net/bnxt: update HWRM to version 1.9.2

Update HWRM structures to version 1.9.2

Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agodoc: remove mention of unreleased nics from enic guide
John Daley [Sat, 5 May 2018 02:34:24 +0000 (19:34 -0700)]
doc: remove mention of unreleased nics from enic guide

Company policy discourages the mention of unreleased hardware in
guides and release notes.

Fixes: 08df773 ("doc: update enic guide and features")

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Signed-off-by: John Daley <johndale@cisco.com>
6 years agonet/enic: update UDP RSS controls
Hyong Youb Kim [Thu, 3 May 2018 19:37:13 +0000 (12:37 -0700)]
net/enic: update UDP RSS controls

Current adapters which support UDP RSS piggyback on TCP RSS. Change
the controls to be forward compatible with future adapters, which will
have independent control of UDP and TCP.

Fixes: 9bd04182bb01 ("net/enic: support UDP RSS on 1400 series adapters")

Signed-off-by: John Daley <johndale@cisco.com>
Reviewed-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: Aaron Conole <aconole@redhat.com>
6 years agonet/enic: fix RSS hash type advertisement
Hyong Youb Kim [Thu, 3 May 2018 19:37:12 +0000 (12:37 -0700)]
net/enic: fix RSS hash type advertisement

The NIC can hash these RSS packet types, but they are not advertised
via flow_type_rss_offloads. So add them.
- Part of the IPv4 hash:
  ETH_RSS_FRAG_IPV4
  ETH_RSS_NONFRAG_IPV4_OTHER
- Part of the IPv6 hash:
  ETH_RSS_FRAG_IPV6
  ETH_RSS_NONFRAG_IPV6_OTHER
- Part of the UDP hash:
  ETH_RSS_IPV6_UDP_EX

Also, do not use NIC_CFG_RSS_HASH_TYPE_IPV6_EX and
NIC_CFG_RSS_HASH_TYPE_TCP_IPV6_EX, as they are not needed to enable
RSS over IPv6 with extension headers.

Fixes: c2fec27b5cb0 ("net/enic: allow to change RSS settings")

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
Reviewed-by: Aaron Conole <aconole@redhat.com>
6 years agodoc: update the enic guide and features
Hyong Youb Kim [Thu, 3 May 2018 19:37:11 +0000 (12:37 -0700)]
doc: update the enic guide and features

Add more descriptions regarding SR-IOV and RSS settings.

Remove 'Multicast MAC filter' and add 'Allmulticast mode' to the
features.

Cc: stable@dpdk.org
Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
Reviewed-by: Aaron Conole <aconole@redhat.com>
6 years agonet/enic: set rte errno to positive value
John Daley [Thu, 3 May 2018 19:37:10 +0000 (12:37 -0700)]
net/enic: set rte errno to positive value

Related to d9fff8a31, where rte_errno should always have positive
errno values.

Technically this is an ABI change since it fixes an error code
introduced in 18.02, but is minor and inconsequential.

Fixes: 1e81dbb5321b ("net/enic: add Tx prepare handler")
Cc: stable@dpdk.org
Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
Reviewed-by: Aaron Conole <aconole@redhat.com>
6 years agonet/enic: fix the MTU handler to rely on max packet length
Hyong Youb Kim [Thu, 3 May 2018 19:37:09 +0000 (12:37 -0700)]
net/enic: fix the MTU handler to rely on max packet length

The RQ setup functions (enic_alloc_rq and enic_alloc_rx_queue_mbufs)
have changed to rely on max_rx_pkt_len to determine the use of scatter
and buffer size. But, the MTU handler only updates ethdev's MTU
value. So make it update max_rx_pkt_len as well. Other PMDs also
update both mtu and max_rx_pkt_len in their MTU handlers.

Also the condition for taking a short cut (scatter is disabled) in the
MTU handler is wrong. Even when scatter is disabled, a change in
max_rx_pkt_len may affect the buffer size posted to the NIC. So remove
that condition.

Finally, fix a comment and a warning message condition.

Fixes: 422ba91716a7 ("net/enic: heed the requested max Rx packet size")

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
Reviewed-by: Aaron Conole <aconole@redhat.com>
6 years agonet/enic: enable RQ first and then post Rx buffers
Hyong Youb Kim [Thu, 3 May 2018 19:37:08 +0000 (12:37 -0700)]
net/enic: enable RQ first and then post Rx buffers

Future VIC adapters may require that the driver enable RQ before
posting new buffers to the NIC. So split enic_alloc_rx_queue_mbufs()
into two functions, one that allocates buffers and fills RQ and the
other that posts them (i.e. PIO write to a doorbell). And, call the
post function only after enabling RQ.

Currently released models are not affected by this change, as they
work fine whether the driver posts buffers before or after enabling RQ.

Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Reviewed-by: John Daley <johndale@cisco.com>
Reviewed-by: Aaron Conole <aconole@redhat.com>
6 years agonet/virtio-user: support memory hotplug
Maxime Coquelin [Tue, 8 May 2018 07:10:45 +0000 (15:10 +0800)]
net/virtio-user: support memory hotplug

When memory is hot-added or hot-removed, the virtio-user driver has to
notify the vhost-user backend with sending a VHOST_USER_SET_MEM_TABLE
request with the new memory map as payload.

This patch implements and registers a mem_event callback, it pauses the
datapath and updates memory regions to vhost in case of hot-add or
hot-remove event. This memory region update has only to be done when the
device is already started, so a new status flag is added to the device to
keep track of the status.

As the device can now be managed by different threads, a mutex is
introduced to protect against concurrent device configuration.

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Signed-off-by: Xiao Wang <xiao.w.wang@intel.com>
6 years agovhost: retranslate vring addr when memory table changes
Maxime Coquelin [Tue, 8 May 2018 07:10:44 +0000 (15:10 +0800)]
vhost: retranslate vring addr when memory table changes

When the vhost-user master sends memory updates using
VHOST_USER_SET_MEM request, the user backends unmap and then
mmap again the memory regions in its address space.

If the ring addresses have already been translated, it needs to
be translated again as they point to unmapped memory.

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
6 years agonet/mlx5: fix calculation of Tx TSO inline room size
Yongseok Koh [Tue, 8 May 2018 18:48:19 +0000 (11:48 -0700)]
net/mlx5: fix calculation of Tx TSO inline room size

rdma-core doesn't add up max_tso_header size to max_inline_data size. The
library takes bigger value between the two.

Fixes: 43e9d9794cde ("net/mlx5: support upstream rdma-core")
Cc: stable@dpdk.org
Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
6 years agonet/mlx5: fix resource leak in case of error
Raslan Darawsheh [Tue, 8 May 2018 09:11:27 +0000 (12:11 +0300)]
net/mlx5: fix resource leak in case of error

If something went wrong in mlx5_pci_prob the allocated eth dev
will cause a memory leak.

This commit release the eth dev that was previously allocated.

Fixes: 771fa900b73a ("mlx5: introduce new driver for Mellanox ConnectX-4 adapters")
Cc: stable@dpdk.org
Signed-off-by: Raslan Darawsheh <rasland@mellanox.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx5: fix double free on error handling
Raslan Darawsheh [Tue, 8 May 2018 09:11:26 +0000 (12:11 +0300)]
net/mlx5: fix double free on error handling

When attr_ctx is NULL it will attempt to free the list of devices twice.
Avoid double freeing the list by directly going to error handling.

Fixes: 771fa900b73a ("mlx5: introduce new driver for Mellanox ConnectX-4 adapters")
Cc: stable@dpdk.org
Signed-off-by: Raslan Darawsheh <rasland@mellanox.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx5: fix SW parser enabling
Xueming Li [Wed, 9 May 2018 00:14:51 +0000 (08:14 +0800)]
net/mlx5: fix SW parser enabling

Fixes: 5f8ba81c4228 ("net/mlx5: support generic tunnel offloading")

Signed-off-by: Xueming Li <xuemingl@mellanox.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx5: fix SW parsing feature detection
Xueming Li [Wed, 9 May 2018 00:14:50 +0000 (08:14 +0800)]
net/mlx5: fix SW parsing feature detection

Fixes: 5f8ba81c4228 ("net/mlx5: support generic tunnel offloading")

Signed-off-by: Xueming Li <xuemingl@mellanox.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/mlx5: document update for Tx
Xueming Li [Wed, 9 May 2018 00:14:49 +0000 (08:14 +0800)]
net/mlx5: document update for Tx

Add document for hw header parsing and SWP.

Signed-off-by: Xueming Li <xuemingl@mellanox.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
6 years agonet/ixgbevf: save interrupt mask for performance
Tonghao Zhang [Sat, 28 Apr 2018 03:03:40 +0000 (20:03 -0700)]
net/ixgbevf: save interrupt mask for performance

If dpdk APPs call the rte_eth_dev_rx_intr_enable or
rte_eth_dev_rx_intr_disable frequently, and ixgbe vf will read
the IXGBE_VTEIMS register everytime. The patch saves the IXGBE_VTEIMS
to mask to avoid read frequently.

Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
Acked-by: Wei Dai <wei.dai@intel.com>
6 years agonet/ixgbe: write disable to EITR counter
Tonghao Zhang [Sat, 28 Apr 2018 03:03:39 +0000 (20:03 -0700)]
net/ixgbe: write disable to EITR counter

ixgbe doesn't write the EITR counter, disable it now.

Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
6 years agonet/ixgbe: set the default value for EITR
Tonghao Zhang [Sat, 28 Apr 2018 03:03:38 +0000 (20:03 -0700)]
net/ixgbe: set the default value for EITR

The ixgbe PF and VF use the same default value
for EITR defined in header file.

Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
6 years agonet/ixgbevf: set the interrupt interval for EITR
Tonghao Zhang [Sat, 28 Apr 2018 03:03:37 +0000 (20:03 -0700)]
net/ixgbevf: set the interrupt interval for EITR

Set EITR interval as default. This patch can improve the
performance when we enable the rx-interrupt to process the
packets because we hope rx-interrupt reduce CPU. For example,
the 200us value of EITR makes the performance better with
the low CPU. The default value of ITR is 500us, compatible
with RSC of ixgbe PF, and next patch will use the default value.

Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
6 years agonet/sfc/base: add Medford2 head-of-line blocking stats
Andrew Rybchenko [Thu, 3 May 2018 14:41:19 +0000 (15:41 +0100)]
net/sfc/base: add Medford2 head-of-line blocking stats

These stats are availble on Medford2 DPDK firmware variant
which support equal stride super-buffer Rx mode. RXDP_HLB_IDLE
capability bit is set when the stats are available.

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Andy Moreton <amoreton@solarflare.com>
6 years agonet/sfc/base: support RxDP scatter disabled truncate counter
Andrew Rybchenko [Thu, 3 May 2018 14:41:18 +0000 (15:41 +0100)]
net/sfc/base: support RxDP scatter disabled truncate counter

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Andy Moreton <amoreton@solarflare.com>
6 years agonet/sfc/base: generate Medford2 RxDP stats
Andrew Rybchenko [Thu, 3 May 2018 14:41:17 +0000 (15:41 +0100)]
net/sfc/base: generate Medford2 RxDP stats

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Andy Moreton <amoreton@solarflare.com>
6 years agonet/sfc/base: fix Medford2 FEC stats range
Andrew Rybchenko [Thu, 3 May 2018 14:41:16 +0000 (15:41 +0100)]
net/sfc/base: fix Medford2 FEC stats range

Fixes: 400ba3daeeb1 ("net/sfc/base: decode Medford2 FEC stats if available")

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Andy Moreton <amoreton@solarflare.com>
6 years agonet/bnxt: clear HWRM sniffer list for PFs
Randy Schacher [Tue, 1 May 2018 01:06:16 +0000 (18:06 -0700)]
net/bnxt: clear HWRM sniffer list for PFs

Clear HWRM sniffer list for DPDK PFs so that VFs on
DPDK PFs initialize successfully. DPDK PF driver does not
handle HWRM commands from VFs.

Signed-off-by: Randy Schacher <stuart.schacher@broadcom.com>
Signed-off-by: Scott Branden <scott.branden@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: fix usage of vnic id
Ajit Khaparde [Tue, 1 May 2018 01:06:15 +0000 (18:06 -0700)]
net/bnxt: fix usage of vnic id

VNIC ID returned by the FW is a 16-bit field.
We are incorrectly using it as a 32-bit value in few places.
This patch corrects that.

Fixes: daef48efe5e5 ("net/bnxt: support set MTU")
Cc: stable@dpdk.org
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Signed-off-by: Scott Branden <scott.branden@broadcom.com>
Reviewed-by: Michael Wildt <michael.wildt@broadcom.com>
Reviewed-by: Randy Schacher <stuart.schacher@broadcom.com>
Reviewed-by: Ray Jui <ray.jui@broadcom.com>
6 years agonet/bnxt: fix Rx mbuf and agg ring leak in dev stop
Xiaoxin Peng [Tue, 1 May 2018 01:06:14 +0000 (18:06 -0700)]
net/bnxt: fix Rx mbuf and agg ring leak in dev stop

In the start/stop_op operation, mbufs allocated for rings were not freed

1) add bnxt_free_tx_mbuf/bnxt_free_rx_mbuf in bnxt_dev_stop_op to free MBUF
   before freeing the rings.
2) MBUF allocation and free routines were not in sync. Allocation uses the
   ring->ring_size including any rounded up and multiple factors while the
   free routine uses the requested queue size.

Fixes: c09f57b49c13 ("net/bnxt: add start/stop/link update operations")
Cc: stable@dpdk.org
Signed-off-by: Jay Ding <jay.ding@broadcom.com>
Signed-off-by: Scott Branden <scott.branden@broadcom.com>
Reviewed-by: Ray Jui <ray.jui@broadcom.com>
Reviewed-by: Randy Schacher <stuart.schacher@broadcom.com>
Signed-off-by: Xiaoxin Peng <xiaoxin.peng@broadcom.com>
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: fix to reset status of initialization
Ajit Khaparde [Tue, 1 May 2018 01:06:13 +0000 (18:06 -0700)]
net/bnxt: fix to reset status of initialization

clear flag on stop at proper location to avoid race conditions.

Fixes: ed2ced6fe927 ("net/bnxt: check initialization before accessing stats")
Cc: stable@dpdk.org
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: return error in stats if init is not complete
Jay Ding [Tue, 1 May 2018 01:06:12 +0000 (18:06 -0700)]
net/bnxt: return error in stats if init is not complete

return error if init is not complete before accessing stats.

Fixes: ed2ced6fe927 ("net/bnxt: check initialization before accessing stats")
Cc: stable@dpdk.org
Signed-off-by: Jay Ding <jay.ding@broadcom.com>
Signed-off-by: Scott Branden <scott.branden@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Randy Schacher <stuart.schacher@broadcom.com>
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: fix MTU calculation
Ajit Khaparde [Tue, 1 May 2018 01:06:11 +0000 (18:06 -0700)]
net/bnxt: fix MTU calculation

We were not considering the case of nested VLANs while
calculating MTU. This patch takes care of the same.

Fixes: daef48efe5e5 ("net/bnxt: support set MTU")
Cc: stable@dpdk.org
Signed-off-by: Qingmin Liu <qingmin.liu@broadcom.com>
Signed-off-by: Scott Branden <scott.branden@broadcom.com>
Reviewed-by: Jay Ding <jay.ding@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Randy Schacher <stuart.schacher@broadcom.com>
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: set MTU in dev config for jumbo packets
Qingmin Liu [Tue, 1 May 2018 01:06:10 +0000 (18:06 -0700)]
net/bnxt: set MTU in dev config for jumbo packets

MTU setting does not take effect after rte_eth_dev_configure
is called with jumbo enable unless it is configured using the
set_mtu dev_op.

Fixes: daef48efe5e5 ("net/bnxt: support set MTU")
Cc: stable@dpdk.org
Signed-off-by: Qingmin Liu <qingmin.liu@broadcom.com>
Signed-off-by: Scott Branden <scott.branden@broadcom.com>
Reviewed-by: Jay Ding <jay.ding@broadcom.com>
Reviewed-by: Randy Schacher <stuart.schacher@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: validate structs and pointers before use
Rahul Gupta [Tue, 1 May 2018 01:06:09 +0000 (18:06 -0700)]
net/bnxt: validate structs and pointers before use

Validate pointers aren't pointing to uninitialized areas
including txq and rxq before using them to avoid
bnxt driver from crashing.

Signed-off-by: Rahul Gupta <rahul.gupta@broadcom.com>
Signed-off-by: Jay Ding <jay.ding@broadcom.com>
Signed-off-by: Scott Branden <scott.branden@broadcom.com>
Reviewed-by: Ray Jui <ray.jui@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Randy Schacher <stuart.schacher@broadcom.com>
Tested-by: Randy Schacher <stuart.schacher@broadcom.com>
6 years agonet/bnxt: update returned error on invalid max ring
Jay Ding [Tue, 1 May 2018 01:06:08 +0000 (18:06 -0700)]
net/bnxt: update returned error on invalid max ring

Return EINVAL instead of ENOSPC when invalid queue_idx passed in
during rx and tx queue_setup_op routines.

Signed-off-by: Jay Ding <jay.ding@broadcom.com>
Signed-off-by: Scott Branden <scott.branden@broadcom.com>
Reviewed-by: Ray Jui <ray.jui@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: rename driver version from Cumulus to NetXtreme
Scott Branden [Tue, 1 May 2018 01:06:07 +0000 (18:06 -0700)]
net/bnxt: rename driver version from Cumulus to NetXtreme

Rename driver version from "Broadcom Cumulus driver" to
"Broadcom NetXtreme driver" to reflect this driver is applicable to
NetXtreme family beyond Cumulus.

Signed-off-by: Scott Branden <scott.branden@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: rename function checking MAC address
Scott Branden [Tue, 1 May 2018 01:06:06 +0000 (18:06 -0700)]
net/bnxt: rename function checking MAC address

rename check_zero_bytes to bnxt_check_zero_bytes to match proper prefix.

Signed-off-by: Scott Branden <scott.branden@broadcom.com>
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
6 years agonet/bnxt: add support for LSC interrupt event
Qingmin Liu [Tue, 1 May 2018 01:06:05 +0000 (18:06 -0700)]
net/bnxt: add support for LSC interrupt event

Add support to bnxt driver to register RTE_ETH_EVENT_INTR_LSC
event and monitor physical link status.

Signed-off-by: Qingmin Liu <qingmin.liu@broadcom.com>
Signed-off-by: Scott Branden <scott.branden@broadcom.com>
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Randy Schacher <stuart.schacher@broadcom.com>
6 years agonet/mlx4: fix UDP flow rule limitation enforcement
Adrien Mazarguil [Fri, 4 May 2018 15:17:44 +0000 (17:17 +0200)]
net/mlx4: fix UDP flow rule limitation enforcement

For some unknown reason, priorities do not have any effect on flow rules
that happen to match UDP destination ports. Those are always matched first
regardless.

This patch is a workaround that enforces this limitation at the PMD level;
such flow rules can only be created at the highest priority level for
correctness.

Fixes: a5171594fc3b ("net/mlx4: expose support for flow rule priorities")
Cc: stable@dpdk.org
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
6 years agonet/mlx5: fix flow validation
Nélio Laranjeiro [Thu, 3 May 2018 09:31:38 +0000 (11:31 +0200)]
net/mlx5: fix flow validation

Item spec and last are wrongly compared to the NIC capability causing a
validation failure when the mask is null.
This validation function should only verify the user is not configuring
unsupported matching fields.

Fixes: 2097d0d1e2cc ("net/mlx5: support basic flow items and actions")
Cc: stable@dpdk.org
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: fix probe return value polarity
Shahaf Shuler [Thu, 3 May 2018 07:59:37 +0000 (10:59 +0300)]
net/mlx5: fix probe return value polarity

mlx5 prefixed function returns a negative errno value.
the error handler on mlx5_pci_probe is doing the same.

Fixes: a6d83b6a9209 ("net/mlx5: standardize on negative errno values")

Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: fix socket connection return value
Shahaf Shuler [Thu, 3 May 2018 07:59:36 +0000 (10:59 +0300)]
net/mlx5: fix socket connection return value

Upon success, mlx5_socket_connect should return the fd descriptor of the
primary process

Fixes: a6d83b6a9209 ("net/mlx5: standardize on negative errno values")

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
6 years agonet/mlx5: add Rx and Tx tuning parameters
Shahaf Shuler [Tue, 1 May 2018 09:58:49 +0000 (12:58 +0300)]
net/mlx5: add Rx and Tx tuning parameters

A new ethdev API was exposed by
commit 3be82f5cc5e3 ("ethdev: support PMD-tuned Tx/Rx parameters")

Enabling the PMD to provide default parameters in case no strict request
from application in order to improve the out of the box experience.

While the current API lacks the means for the PMD to provide the best
possible value, providing the best default the PMD can guess.
The values are based on Mellanox performance report and depends on the
underlying NIC capabilities.

Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>