dpdk.git
7 years agonet/tap: fix invalid queue file descriptor
Keith Wiles [Sun, 29 Jan 2017 02:12:05 +0000 (20:12 -0600)]
net/tap: fix invalid queue file descriptor

Rx and Tx queues share the common tap file descriptor, but save this
value separately.

Setting up Rx/Tx queue sets up both queues, release_queue close the
tap file but update file descriptor only for that queue.

This makes other queue's file descriptor invalid.

As a workaround, prevent release_queue callback to be called by default.

This is done by separating Rx/Tx setup functions, so that each only
setup its own queue, this prevents rte_eth_rx/tx_queue_setup() calling
release_queue before setup_queue.

Fixes: 02f96a0a82d1 ("net/tap: add TUN/TAP device PMD")

Signed-off-by: Keith Wiles <keith.wiles@intel.com>
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agodrivers/net: remove redundant new line from logs
Ferruh Yigit [Fri, 27 Jan 2017 15:16:32 +0000 (15:16 +0000)]
drivers/net: remove redundant new line from logs

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
7 years agonet/mlx5: fix flow pattern validation
Nélio Laranjeiro [Fri, 27 Jan 2017 15:35:20 +0000 (16:35 +0100)]
net/mlx5: fix flow pattern validation

Size of the mask is wrongly computed and make the validation process only
verify the first 4 bytes of the layer.

Fixes: 2097d0d1e2cc ("net/mlx5: support basic flow items and actions")

Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
7 years agonet/mlx5: fix flow pattern VLAN validation
Nélio Laranjeiro [Fri, 27 Jan 2017 15:35:19 +0000 (16:35 +0100)]
net/mlx5: fix flow pattern VLAN validation

TCI field is read from the wrong place due to an invalid cast. Moreover
there is no need to limit matching to VID since PCP and DEI bits can be
matched as well.

Fixes: 12475fb203ad ("net/mlx5: support VLAN flow item")

Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
7 years agonet/ixgbe: fix API comments for doxygen
Ferruh Yigit [Fri, 27 Jan 2017 12:27:20 +0000 (12:27 +0000)]
net/ixgbe: fix API comments for doxygen

Fixes: 57aa1fd284d5 ("net/ixgbe: move set VF functions from the ethdev")

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agonet/i40e: fix API comments for doxygen
Ferruh Yigit [Fri, 27 Jan 2017 12:27:19 +0000 (12:27 +0000)]
net/i40e: fix API comments for doxygen

Fixes: a541407fe4bc ("net/i40e: set VF MAC anti-spoofing from PF")
Fixes: 4cbc41efcbb2 ("net/i40e: set VF VLAN anti-spoofing from PF")
Fixes: c0ec14757c5f ("net/i40e: set VF unicast promiscuous mode from PF")
Fixes: ae57070ca8b7 ("net/i40e: set VF multicast promiscuous mode from PF")
Fixes: 83bb95e3fefc ("net/i40e: set VF VLAN insertion from PF")
Fixes: 61fff9b4c68b ("net/i40e: set VF broadcast mode from PF")
Fixes: c33abbc1440e ("net/i40e: set VF VLAN tag from PF")

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agonet/i40e: fix crash in close
Bernard Iremonger [Fri, 27 Jan 2017 10:22:47 +0000 (10:22 +0000)]
net/i40e: fix crash in close

Change the order of releasing the VSI's.
Release the VMDq VSI's first, then release the main VSI.

Fixes: 3cb446b4aeb2 ("i40e: free vmdq vsi when closing")
Cc: stable@dpdk.org
Signed-off-by: Bernard Iremonger <bernard.iremonger@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>
7 years agonet/enic: fix MAC address add and remove
John Daley [Thu, 26 Jan 2017 20:12:12 +0000 (12:12 -0800)]
net/enic: fix MAC address add and remove

The mac_addr_add callback function was simply replacing the primary MAC
address instead of adding new ones and the mac_addr_remove callback would
only remove the primary MAC form the adapter. Fix the functions to add or
remove new address. Allow up to 64 MAC addresses per port.

Fixes: fefed3d1e62c ("enic: new driver")

Signed-off-by: John Daley <johndale@cisco.com>
Reviewed-by: Nelson Escobar <neescoba@cisco.com>
7 years agodrivers/net: make PCI device id struct const
Ferruh Yigit [Thu, 26 Jan 2017 14:07:44 +0000 (14:07 +0000)]
drivers/net: make PCI device id struct const

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>
7 years agonet/cxgbe: remove unused variables
Emmanuel Roullit [Tue, 24 Jan 2017 20:48:58 +0000 (21:48 +0100)]
net/cxgbe: remove unused variables

Found with clang static analysis:
drivers/net/cxgbe/sge.c:900:3: warning:
Value stored to 'in_use' is never read
        in_use += q->size;
        ^         ~~~~~~~

Removing "in_use" variable makes "hw_cidx" variable unused, which is
removed too.

Fixes: 4a01078b4fd1 ("cxgbe: add Tx support")
Cc: stable@dpdk.org
Signed-off-by: Emmanuel Roullit <emmanuel.roullit@gmail.com>
Acked-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
7 years agonet/ixgbevf: fix max packet length
Yi Zhang [Wed, 14 Dec 2016 18:50:19 +0000 (02:50 +0800)]
net/ixgbevf: fix max packet length

Current ixgbevf driver get max_rx_pktlen = 15872, but in fact PF
supports 15872-byte jumbo frame and VF only supports 9728-byte jumbo
frame. If VF is running DPDK driver and set frame_size > 9728 ,PF
running kernel ixgbe driver will report an error and set VF failed.
This patch fixs DPDK ixgbevf driver to get correct jumbo frame size
of VF.

More datasheet references from Wei Dai:

In 82599 datasheet, there is an annotation in the chapter 1.3 Features
Summary (page 29)
 The 82599 supports full-size 15.5 KB (15872-byte) jumbo packets while
 in a basic mode of operation. When DCB mode is enabled,
 or security engines enabled or virtualization is enabled, the 82599
 supports 9.5 KB (9728-byte) jumbo packets.

In x540 datasheet, there is also an annotation in the chapter 1.3
Features Summary (page 13)
 The X540 and 82599 support full-size 15.5 KB jumbo packets while in a
 basic mode of operation. When DCB mode is enabled,
 or security engines enabled, or virtualization is enabled, or OS2BMC is
 enabled, then the X540 supports 9.5 KB jumbo packets.
 Packets to/from MC longer than 2KB are filtered out.

In x550 datasheet, there is still also an annotation in the chapter 1.4
Feature Summary (page 23)
 All the products support full-size 15.5 KB jumbo packets while in a
 basic mode of operation. When DCB mode is enabled, or security
 engines enabled, or virtualization is enabled, or OS2BMC is enabled,
 then only 9.5 KB jumbo packets are supported. Packets to/
 from the MC longer than 2 KB are filtered out.

Fixes: 2144f6630fca ("ixgbe: add redirection table size in device info")
Cc: stable@dpdk.org
Signed-off-by: Yi Zhang <zhang.yi75@zte.com.cn>
Acked-by: Wei Dai <wei.dai@intel.com>
7 years agonet/i40e: fix checksum flag in x86 vector Rx
Qi Zhang [Tue, 24 Jan 2017 19:06:59 +0000 (14:06 -0500)]
net/i40e: fix checksum flag in x86 vector Rx

When no error reported in Rx descriptor, we should set
CKSUM_GOOD flag before return.

Fixes: 9966a00a0688 ("net/i40e: enable bad checksum flags in vector Rx")
Cc: stable@dpdk.org
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
7 years agonet/bonding: remove useless assignment
Emmanuel Roullit [Tue, 24 Jan 2017 21:15:55 +0000 (22:15 +0100)]
net/bonding: remove useless assignment

Found with clang static analysis:
drivers/net/bonding/rte_eth_bond_pmd.c:903:3:
warning: Value stored to 'num_not_send' is never read
        num_not_send += slave_bufs_pkts[RTE_MAX_ETHPORTS] - num_send;
        ^               ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Fixes: 06fe78b98ccd ("bond: add mode 6")
Cc: stable@dpdk.org
Signed-off-by: Emmanuel Roullit <emmanuel.roullit@gmail.com>
7 years agonet/cxgbe/base: initialize variable before reading EEPROM
Emmanuel Roullit [Tue, 24 Jan 2017 20:48:57 +0000 (21:48 +0100)]
net/cxgbe/base: initialize variable before reading EEPROM

data value could have been garbage if VPD access timed out for VPD read
request could not been issued.

Found with clang static analysis:
drivers/net/cxgbe/base/t4_hw.c:1577:22:
warning: The left operand of '&' is a garbage value
        } while ((stats_reg & 0x1) && --max_poll);
                  ~~~~~~~~~ ^

Fixes: fe0bd9ee5da3 ("net/cxgbe: support EEPROM access")
Cc: stable@dpdk.org
Signed-off-by: Emmanuel Roullit <emmanuel.roullit@gmail.com>
7 years agonet/ena: prepare TSO offload calculation
Jakub Palider [Tue, 24 Jan 2017 14:22:53 +0000 (15:22 +0100)]
net/ena: prepare TSO offload calculation

While ENA can handle checksum calculations in almost all cases,
it cannot do so when DF bit in IPv4 header is not set,
that is DF=0, and TSO is requested. For that situation pseudo
header must be prepared manually.

Signed-off-by: Jakub Palider <jpa@semihalf.com>
7 years agonet/ixgbe: fix API parameter checking
Tiwei Bie [Tue, 24 Jan 2017 09:00:49 +0000 (17:00 +0800)]
net/ixgbe: fix API parameter checking

Add checks to rte_pmd_ixgbe_macsec_* APIs to ensure that the
port is an ixgbe port.

Fixes: b35d309710fe ("net/ixgbe: add MACsec offload")
Cc: stable@dpdk.org
Signed-off-by: Tiwei Bie <tiwei.bie@intel.com>
7 years agonet/ixgbe: return success when SFP is not present
Wei Dai [Tue, 24 Jan 2017 08:14:20 +0000 (16:14 +0800)]
net/ixgbe: return success when SFP is not present

Ignore the error=IXGBE_ERR_SFP_NOT_PRESENT when SFP is not present.
If it is not ignored, testpmd will fail during the NIC initialization
process.
Ixgbe kernel driver ignores this error and works well. So DPDK
does same thing.

Signed-off-by: Wei Dai <wei.dai@intel.com>
Signed-off-by: Helin Zhang <helin.zhang@intel.com>
Tested-by: Yuan Peng <yuan.peng@intel.com>
Acked-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
7 years agonet/bnx2x: fix Rx mode configuration
Rasesh Mody [Tue, 24 Jan 2017 05:38:33 +0000 (21:38 -0800)]
net/bnx2x: fix Rx mode configuration

Check if promisc mode was set when setting allmulti mode and vice-versa.
Introduced BNX2X_RX_MODE_ALLMULTI_PROMISC for the same. If check is
absent the filter configuration gets over written.

Fixes: 540a211084a7 ("bnx2x: driver core")
Fixes: 5dbc53d7e5a2 ("net/bnx2x: restrict Rx mask flags sent to the PF")
Cc: stable@dpdk.org
Signed-off-by: Rasesh Mody <rasesh.mody@cavium.com>
7 years agonet/szedata2: fix device memory access
Matej Vido [Tue, 24 Jan 2017 10:49:52 +0000 (11:49 +0100)]
net/szedata2: fix device memory access

Fixes: 8acba705b119 ("net/szedata2: localize handling of PCI resources")

Signed-off-by: Matej Vido <vido@cesnet.cz>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agonet/i40e: fix deletion of all macvlan filters
Jingjing Wu [Mon, 23 Jan 2017 09:42:45 +0000 (17:42 +0800)]
net/i40e: fix deletion of all macvlan filters

filter_type is not set when removing all macvlan filters. It will
cause error when send AQ command to HW.
This patch fixes this issue.

Fixes: 4861cde46116 ("i40e: new poll mode driver")
Cc: stable@dpdk.org
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
7 years agonet/i40e: fix validation when deleting MAC address
Jingjing Wu [Mon, 23 Jan 2017 09:42:44 +0000 (17:42 +0800)]
net/i40e: fix validation when deleting MAC address

When VF sends request to remove MAC address, PF host will check
if it is a non-zero or unicast address. When VF remove a multicast
address, it will report error.
This patch fixes this issue.

Fixes: ec852c94af39 ("net/i40e: enhance sanity check of MAC")
Cc: stable@dpdk.org
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
7 years agonet/qede/base: fix FreeBSD build
Rasesh Mody [Mon, 23 Jan 2017 05:02:38 +0000 (21:02 -0800)]
net/qede/base: fix FreeBSD build

This patch addresses compilation errors on FreeBSD with clang 3.8.0.

drivers/net/qede/base/ecore_cxt.c:1257:2: error:
     shifting a negative signed value is undefined
          SET_FIELD(cdu_params, CDUC_NCIB, elems_per_page);
          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/net/qede/base/ecore.h:82:27: note: expanded from macro 'SET_FIELD'
          (value) &= ~(name##_MASK << name##_SHIFT);
                    ~~~~~~~~~~~ ^

Fixes: ec94dbc57362 ("qede: add base driver")
Cc: stable@dpdk.org
Signed-off-by: Rasesh Mody <rasesh.mody@cavium.com>
Tested-by: Andrew Rybchenko <arybchenko@solarflare.com>
7 years agonet/mlx5: fix memory leak when parsing device params
Shahaf Shuler [Sun, 22 Jan 2017 08:24:47 +0000 (10:24 +0200)]
net/mlx5: fix memory leak when parsing device params

in case of an error argument list is not freed.

Fixes: e72dd09b614e ("net/mlx5: add support for configuration through kvargs")
Cc: stable@dpdk.org
Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
7 years agodoc: add known uio_pci_generic issue for i40e
Jeff Guo [Sun, 22 Jan 2017 02:03:41 +0000 (10:03 +0800)]
doc: add known uio_pci_generic issue for i40e

When bind the "uio_pci_generic" module in X710/XL710/XXV710,
the result is failed. The "uio_pci_generic" module is not
supported by X710/XL710/XXV710.

Signed-off-by: Jeff Guo <jia.guo@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-by: John McNamara <john.mcnamara@intel.com>
7 years agonet/sfc: fix TSO limits imposed to the number of Tx queues
Ivan Malov [Mon, 23 Jan 2017 11:06:15 +0000 (11:06 +0000)]
net/sfc: fix TSO limits imposed to the number of Tx queues

The number of Tx queues requested by the user must not be overridden;
instead, the limits imposed by TSO must be applied to the advertised
maximum

Fixes: fec33d5bb3eb ("net/sfc: support firmware-assisted TSO")
Cc: stable@dpdk.org
Signed-off-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Andrew Lee <alee@solarflare.com>
Reviewed-by: Andy Moreton <amoreton@solarflare.com>
7 years agonet/sfc: use eal I/O device memory barriers API
Andrew Rybchenko [Fri, 20 Jan 2017 13:53:51 +0000 (13:53 +0000)]
net/sfc: use eal I/O device memory barriers API

The previous version relied on the fact that DMA sync for device and
PIO write barrier in pair. Now each does its job.

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
7 years agonet/sfc: use eal I/O device memory read/write API
Andrew Rybchenko [Fri, 20 Jan 2017 13:53:50 +0000 (13:53 +0000)]
net/sfc: use eal I/O device memory read/write API

Use relaxed version of these functions to guarantee no changes on
the step.

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
7 years agonet/sfc: enable TSO by default
Andrew Rybchenko [Fri, 20 Jan 2017 15:22:56 +0000 (15:22 +0000)]
net/sfc: enable TSO by default

Remove RTE_LIBRTE_SFC_EFX_TSO config option since it is not
required any more:
 - unreasonable limit on number of Tx queues when TSO is not
   actually required should be solved using per-device parameter
 - performance difference with and without TSO compiled in is small

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
7 years agonet/sfc: do not panic if alarms are not supported
Andrew Rybchenko [Thu, 19 Jan 2017 11:12:20 +0000 (11:12 +0000)]
net/sfc: do not panic if alarms are not supported

Alarms are not supported on the FreeBSD.
Application must poll link status periodically itself using
rte_eth_link_get_nowait() to avoid management event queue overflow.

Fixes: 2de39f4e1310 ("net/sfc: periodic management EVQ polling using alarm")

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Andrew Lee <alee@solarflare.com>
Reviewed-by: Andy Moreton <amoreton@solarflare.com>
7 years agonet/mlx5: fix Rx packet validation and type
Nélio Laranjeiro [Fri, 20 Jan 2017 15:27:29 +0000 (16:27 +0100)]
net/mlx5: fix Rx packet validation and type

Rx checksum validation and packet type offloads are read from the wrong
location.

Fixes: 501505c5608a ("net/mlx: fix IPv4 and IPv6 packet type")
Fixes: 350f4c482e46 ("net/mlx5: fix Rx checksum macros")
Fixes: 6218063b39a6 ("net/mlx5: refactor Rx data path")
Cc: stable@dpdk.org
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
7 years agonet/mlx5: fix flow API mark combined with drop
Nélio Laranjeiro [Fri, 20 Jan 2017 15:27:28 +0000 (16:27 +0100)]
net/mlx5: fix flow API mark combined with drop

Fixes: ea3bc3b1df94 ("net/mlx5: support mark flow action")
Cc: stable@dpdk.org
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
7 years agonet/mlx5: fix queue wrap-around of multi-packet send
Yongseok Koh [Wed, 18 Jan 2017 00:51:55 +0000 (16:51 -0800)]
net/mlx5: fix queue wrap-around of multi-packet send

When the WQ is wrapped around, it wrongly checks the condition when
resetting the pointer. It should be compared against the end of the queue,
not the beginning of the queue. And this isn't even needed when the length
of the copying data crosses the boundary.

Fixes: fdcb0f53053b ("net/mlx5: use work queue buffer as a raw buffer")
Cc: stable@dpdk.org
Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
7 years agonet/mlx5: increase RSS indirection table size limit
Yongseok Koh [Wed, 18 Jan 2017 00:39:29 +0000 (16:39 -0800)]
net/mlx5: increase RSS indirection table size limit

The size of Rx RSS indirection table was limited by 256, but it is not
required anymore for all Mellanox NICs. However, the librte_ether still
limits the size by 512.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
7 years agonet/mlx5: add prefetching Rx completion queue
Yongseok Koh [Tue, 17 Jan 2017 02:09:40 +0000 (18:09 -0800)]
net/mlx5: add prefetching Rx completion queue

On receiving a compressed session of Rx completion, prefetch every entries
to be invalidated. Also, invalidate consumed completions per every 8
mini-completions, not to wait until the last entry is consumed. This helps
to reduce jitter in rx_burst.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
7 years agonet/i40e: fix Rx checksum flag
Xiao Wang [Thu, 19 Jan 2017 05:39:23 +0000 (21:39 -0800)]
net/i40e: fix Rx checksum flag

When no error reported in Rx descriptor, we should set CKSUM_GOOD flag
before return.

Fixes: b704f9071b09 ("net/i40e: implement new Rx checksum flag")
Cc: stable@dpdk.org
Signed-off-by: Xiao Wang <xiao.w.wang@intel.com>
7 years agonet/ixgbe: fix blocked interrupts
Qi Zhang [Mon, 16 Jan 2017 19:23:52 +0000 (14:23 -0500)]
net/ixgbe: fix blocked interrupts

While handling link status change (LSC) interrupt, all interrupts are
blocked until delayed interrupt handler finishes.

The wait duration is at least one second and this may cause timeouts in
VF to PF mailbox.

Make sure only LSC interrupt is blocked while waiting for delayed
interrupt handler to finish.

Fixes: 0a45657a6794 ("pci: rework interrupt handling")
Cc: stable@dpdk.org
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
7 years agonet/sfc: fix flow control settings on port start
Andrew Rybchenko [Thu, 12 Jan 2017 09:03:23 +0000 (09:03 +0000)]
net/sfc: fix flow control settings on port start

efx_phy_adv_cap_set() sets all advertised phy capabilities including
pause capabilities which are also configured using efx_mac_fcntl_set().

If we set speed and autonegotiation capabilities only, we should
preserve already configured pause capabilities.

Fixes: d23f3a89ab54 ("net/sfc: support link speed and duplex settings")

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Andrew Lee <alee@solarflare.com>
7 years agonet/sfc: fix link status when port is not started
Andrew Rybchenko [Thu, 12 Jan 2017 09:03:22 +0000 (09:03 +0000)]
net/sfc: fix link status when port is not started

Fixes: 886f8d8a05bf ("net/sfc: retrieve link info")

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Andrew Lee <alee@solarflare.com>
7 years agonet/sfc: synchronize link status with HW on start
Andrew Rybchenko [Thu, 12 Jan 2017 09:03:21 +0000 (09:03 +0000)]
net/sfc: synchronize link status with HW on start

Fixes: 886f8d8a05bf ("net/sfc: retrieve link info")

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Andrew Lee <alee@solarflare.com>
7 years agonet/sfc: avoid usage of possibly uninitialized link mode
Andrew Rybchenko [Thu, 12 Jan 2017 09:03:20 +0000 (09:03 +0000)]
net/sfc: avoid usage of possibly uninitialized link mode

In fact efx_port_poll() always initializes it, but it isn't
explicitly documented feature of the API. Moreover, the API
annocation suggests that return code should be checked.

Fixes: 886f8d8a05bf ("net/sfc: retrieve link info")

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Andrew Lee <alee@solarflare.com>
7 years agonet/virtio: optimize header reset on any layout
Yuanhan Liu [Wed, 11 Jan 2017 04:27:12 +0000 (12:27 +0800)]
net/virtio: optimize header reset on any layout

When any layout is used, the header is stored in the head room of mbuf.
mbuf is allocated and filled by user, means there is no gurateen the
header is all zero for non TSO case. Therefore, we have to do the reset
by ourself:

    memest(hdr, 0, head_size);

The memset has two impacts on performance:

- memset could not be inlined, which is a bit costly.
- more importantly, it touches the mbuf, which could introduce severe
  cache issues as described by former patch.

Similiary, we could do the same trick: reset just when necessary, when
the corresponding field is already 0, which is likely true for a simple
l2 forward case. It could boost the performance up to 20+% in micro
benchmarking.

Cc: stable@dpdk.org
Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
7 years agonet/virtio: fix performance regression due to TSO
Yuanhan Liu [Wed, 11 Jan 2017 04:27:11 +0000 (12:27 +0800)]
net/virtio: fix performance regression due to TSO

TSO is now enabled, but it's not actually being used by default in a
simple L2 forward mode. In such case, we have to zero the virtio net
headers, to inform the vhost backend that no offload is being used:

    hdr->csum_start = 0;
    hdr->csum_offset = 0;
    hdr->flags = 0;

    hdr->gso_type = 0;
    hdr->gso_size = 0;
    hdr->hdr_len = 0;

Such writes could be very costly; it introduces severe cache issues:
The above operations introduce cache write for each packet, which
stalls the read operation from the vhost backend.

The fact that virtio net header is initiated to zero in PMD driver
init stage means that these costly writes are unnecessary and could
be avoided:

    if (hdr->csum_start != 0)
        hdr->csum_start = 0;

And that's what the macro ASSIGN_UNLESS_EQUAL does. With this, the
performance drop introduced by TSO enabling is recovered: it could
be up to 20% in micro benchmarking.

Fixes: 58169a9c8153 ("net/virtio: support Tx checksum offload")
Fixes: 696573046e9e ("net/virtio: support TSO")
Cc: stable@dpdk.org
Cc: Olivier Matz <olivier.matz@6wind.com>
Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Olivier Matz <olivier.matz@6wind.com>
7 years agonet/vhost: fix unix socket not removed as closing
Jianfeng Tan [Tue, 24 Jan 2017 08:37:38 +0000 (08:37 +0000)]
net/vhost: fix unix socket not removed as closing

The commit aed0b12930b3 ("net/vhost: fix socket file deleted on stop")
moves rte_vhost_driver_register and rte_vhost_driver_unregister from
dev_start() and dev_stop() into driver's probe() and remove().

Apps, like testpmd, using vhost pmd in server mode, usually calls
dev_stop() and dev_close() as quitting, instead of driver-specific
remove(). Then those unix socket files have no chance to get removed.

Semantically, device-specific things should be put into device-specific
APIs. Fix this issue by moving rte_vhost_driver_unregister, plus other
structure free into dev_close().

Fixes: aed0b12930b3 ("net/vhost: fix socket file deleted on stop")
Cc: stable@dpdk.org
Reported-by: Lei Yao <lei.a.yao@intel.com>
Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
7 years agovhost: remove unneeded variable assignment
Emmanuel Roullit [Tue, 24 Jan 2017 20:31:29 +0000 (21:31 +0100)]
vhost: remove unneeded variable assignment

Found with clang static analysis:
lib/librte_vhost/vhost_user.c:996:3: warning:
Value stored to 'ret' is never read
        ret = vhost_user_get_vring_base(dev, &msg.payload.state);
        ^     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Signed-off-by: Emmanuel Roullit <emmanuel.roullit@gmail.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
7 years agovhost: do not GSO when no header is present
Emmanuel Roullit [Tue, 24 Jan 2017 20:36:03 +0000 (21:36 +0100)]
vhost: do not GSO when no header is present

Found with clang static analysis:
lib/librte_vhost/virtio_net.c:723:17: warning:
Access to field 'data_off' results in a dereference of a null pointer
(loaded from variable 'tcp_hdr')
        m->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
                     ^~~~~~~~~~~~~~~~~

Fixes: d0cf91303d73 ("vhost: add Tx offload capabilities")
Cc: stable@dpdk.org
Signed-off-by: Emmanuel Roullit <emmanuel.roullit@gmail.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
7 years agonet/virtio-user: check value returned from malloc
Jianfeng Tan [Thu, 26 Jan 2017 03:05:42 +0000 (03:05 +0000)]
net/virtio-user: check value returned from malloc

Value returned from malloc is not checked for errors before being used.
This patch fixes following coverity issue.

    static struct vhost_memory_kernel *
    prepare_vhost_memory_kernel(void)
    {
        ...
        vm = malloc(sizeof(struct vhost_memory_kernel) +
                    max_regions *
                    sizeof(struct vhost_memory_region));
        ...
    >>>     CID 140744:    (NULL_RETURNS)
    >>>     Dereferencing a null pointer "vm".
                mr = &vm->regions[k++];

Coverity issue: 140744
Fixes: e3b434818bbb ("net/virtio-user: support kernel vhost")

Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
7 years agonet/virtio: fix crash when number of virtio devices > 1
Yuanhan Liu [Sun, 22 Jan 2017 08:47:00 +0000 (16:47 +0800)]
net/virtio: fix crash when number of virtio devices > 1

The vtpci_ops assignment needs the 'hw->port_id' as an input parameter.
That said, we should set 'hw->port_id' firstly, then do the vtpci_ops
assignment, while the code does reversely. That would result to a crash
when more than one virtio devices are used, because we keep assigning
proper vtpci_ops to virtio_hw_internal[0]->vtpci_ops, leaving the pointer
for other ports being NULL.

Reverse the order fixes this issue.

Fixes: 9470427c88e1 ("net/virtio: do not store PCI device pointer at shared memory")
Cc: stable@dpdk.org
Reported-by: Lei Yao <lei.a.yao@intel.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
7 years agovhost: fix long stall of negotiation
Yuanhan Liu [Sun, 22 Jan 2017 08:46:59 +0000 (16:46 +0800)]
vhost: fix long stall of negotiation

Setting up the mapping from GPA (guest physical address) to HPA (guest
physical address) could be very time consuming when the guest memory is
backened with small pages (4K). The bigger the guest memory, the longer
it takes. This could lead a very long vhost-user negotiation.

Since the mapping is only needed in zero copy mode so far, we could
avoid such time consuming settup when zero copy is turned off (which is
the default case).

It's actually a workaround, a right fix might be to start a new thread,
and hide the big latency there.

Fixes: e246896178e6 ("vhost: get guest/host physical address mappings")
Cc: stable@dpdk.org
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
7 years agovhost: fix dead loop in enqueue path
Yuanhan Liu [Sun, 22 Jan 2017 08:46:58 +0000 (16:46 +0800)]
vhost: fix dead loop in enqueue path

If a malicious guest forges a dead loop desc chain (let desc->next point
to itself) and desc->len is zero, this could lead to a dead loop in
copy_mbuf_to_desc(following is a simplified code to show this issue
clearly):

    while (mbuf_is_not_totally_consumed) {
        if (desc_avail == 0) {
            desc = &descs[desc->next];
            desc_avail = desc->len;
        }

        COPY(desc, mbuf, desc_avail);
    }

I have actually fixed a same issue before: commit a436f53ebfeb ("vhost:
avoid dead loop chain"); it fixes the dequeue path though, leaving the
enqueue path still vulnerable.

The fix is the same. Add a var nr_desc to avoid the dead loop.

Fixes: f1a519ad981c ("vhost: fix enqueue/dequeue to handle chained vring descriptors")
Cc: stable@dpdk.org
Reported-by: Xieming Katty <katty.xieming@huawei.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
7 years agocrypto/qat: make PCI device id struct const
Ferruh Yigit [Thu, 26 Jan 2017 14:07:42 +0000 (14:07 +0000)]
crypto/qat: make PCI device id struct const

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>
7 years agodoc: describe new performance test application
Slawomir Mrozowicz [Wed, 25 Jan 2017 16:27:35 +0000 (17:27 +0100)]
doc: describe new performance test application

Add documentation to describe using the new performance test application.

Signed-off-by: Slawomir Mrozowicz <slawomirx.mrozowicz@intel.com>
Signed-off-by: Piotr Azarewicz <piotrx.t.azarewicz@intel.com>
7 years agoapp/crypto-perf: add test vectors files
Slawomir Mrozowicz [Wed, 25 Jan 2017 16:27:34 +0000 (17:27 +0100)]
app/crypto-perf: add test vectors files

Add test vectors example files to the performance test application.

Signed-off-by: Marcin Kerlin <marcinx.kerlin@intel.com>
7 years agoapp/crypto-perf: introduce performance test application
Slawomir Mrozowicz [Wed, 25 Jan 2017 16:27:33 +0000 (17:27 +0100)]
app/crypto-perf: introduce performance test application

This patchset introduce new application which allows measuring
performance parameters of PMDs available in crypto tree. The goal of
this application is to replace existing performance tests in app/test.
Parameters available are: throughput (--ptest throughput) and latency
(--ptest latency). User can use multiply cores to run tests on but only
one type of crypto PMD can be measured during single application
execution. Cipher parameters, type of device, type of operation and
chain mode have to be specified in the command line as application
parameters. These parameters are checked using device capabilities
structure.
Couple of new library functions in librte_cryptodev are introduced for
application use.
To build the application a CONFIG_RTE_APP_CRYPTO_PERF flag has to be set
(it is set by default).
Example of usage: -c 0xc0 --vdev crypto_aesni_mb_pmd -w 0000:00:00.0 --
--ptest throughput --devtype crypto_aesni_mb --optype cipher-then-auth
--cipher-algo aes-cbc --cipher-op encrypt --cipher-key-sz 16 --auth-algo
sha1-hmac --auth-op generate --auth-key-sz 64 --auth-digest-sz 12
--total-ops 10000000 --burst-sz 32 --buffer-sz 64

Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Signed-off-by: Slawomir Mrozowicz <slawomirx.mrozowicz@intel.com>
Signed-off-by: Piotr Azarewicz <piotrx.t.azarewicz@intel.com>
Signed-off-by: Marcin Kerlin <marcinx.kerlin@intel.com>
Signed-off-by: Michal Kobylinski <michalx.kobylinski@intel.com>
7 years agocryptodev: retrieve device info
Slawomir Mrozowicz [Wed, 25 Jan 2017 16:27:32 +0000 (17:27 +0100)]
cryptodev: retrieve device info

This patch adds helper functions for new performance application which
provide identifiers and number of crypto device and
provide and check capabilities available for defined device and algorithm.
The performance application can be used to measure throughput and latency
of cryptography operation performed by crypto device.

Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Signed-off-by: Slawomir Mrozowicz <slawomirx.mrozowicz@intel.com>
Signed-off-by: Marcin Kerlin <marcinx.kerlin@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocrypto/scheduler: add documentation
Fan Zhang [Tue, 24 Jan 2017 16:24:01 +0000 (16:24 +0000)]
crypto/scheduler: add documentation

Adds the description of the cryptodev scheduler PMD overview,
limitations, build, instructions, modes, etc.

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocrypto/scheduler: add unit test
Fan Zhang [Tue, 24 Jan 2017 16:24:00 +0000 (16:24 +0000)]
crypto/scheduler: add unit test

Same as other cryptodev PMDs, it is necessary to carry out the unit
test for scheduler PMD. Currently the test is designed to attach 2
AESNI-MB cryptodev PMDs as slaves, sets the scheduling mode as round-
robin, and runs almost all AESNI-MB test items (except for sessionless
tests). In the end, the slaves are detached.

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocrypto/scheduler: enable compilation
Fan Zhang [Tue, 24 Jan 2017 16:23:58 +0000 (16:23 +0000)]
crypto/scheduler: enable compilation

Adds Makefile for scheduler cryptodev PMD, and updates existing
Makefiles. Different than other cryptodev PMDs, scheduler PMD
is required to be built as shared libraries.

Adds scheduler PMD enable and debug flags to config/common_base.

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocrypto/scheduler: register operation functions
Fan Zhang [Tue, 24 Jan 2017 16:23:57 +0000 (16:23 +0000)]
crypto/scheduler: register operation functions

Implements all standard operations required for cryptodev,
and register them to cryptodev operation function pointer table.

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocrypto/scheduler: register as vdev driver
Fan Zhang [Tue, 24 Jan 2017 16:23:56 +0000 (16:23 +0000)]
crypto/scheduler: register as vdev driver

Adds crypto scheduler's PMD's probe and remove function and the device's
enqueue and dequeue burst functions. A cryptodev scheduler PMD is
then registered in the end.

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocrypto/scheduler: add round-robin mode
Fan Zhang [Tue, 24 Jan 2017 16:23:55 +0000 (16:23 +0000)]
crypto/scheduler: add round-robin mode

Implements round-robin scheduling mode and register into cryptodev
scheduler ops structure. This mode enqueues a burst of operation
to one of its slaves, and iterates the next burst to the other
slave. Same procedure is done on dequeueing operations.

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocrypto/scheduler: add API implementations
Fan Zhang [Tue, 24 Jan 2017 16:23:54 +0000 (16:23 +0000)]
crypto/scheduler: add API implementations

Adds the implementations of the APIs for scheduler cryptodev PMD.

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocrypto/scheduler: add internal structures
Fan Zhang [Tue, 24 Jan 2017 16:23:53 +0000 (16:23 +0000)]
crypto/scheduler: add internal structures

Adds a number of internal structures for the cryptodev scheduler PMD. The
structures include the scheduler context, slave, queue pair context,
and session.

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocrypto/scheduler: add API
Fan Zhang [Tue, 24 Jan 2017 16:23:52 +0000 (16:23 +0000)]
crypto/scheduler: add API

Adds APIs and function prototypes for the scheduler PMD to perform extra
operations other than standard cryptodev APIs.

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocryptodev: add scheduler PMD name and type
Fan Zhang [Tue, 24 Jan 2017 16:23:51 +0000 (16:23 +0000)]
cryptodev: add scheduler PMD name and type

This patch adds the cryptodev scheduler PMD name and type identifier to
librte_cryptodev.

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agocryptodev: decouple from PCI device
Hemant Agrawal [Wed, 18 Jan 2017 09:16:10 +0000 (14:46 +0530)]
cryptodev: decouple from PCI device

This makes struct rte_cryptodev independent of struct rte_pci_device by
replacing it with a pointer to the generic struct rte_device.

This is inline with the recent changes in ethdev

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Acked-by: John Griffin <john.griffin@intel.com>
Reviewed-by: Shreyansh Jain <shreyansh.jain@nxp.com>
7 years agocryptodev: uninline some functions
Declan Doherty [Mon, 23 Jan 2017 12:18:35 +0000 (12:18 +0000)]
cryptodev: uninline some functions

rte_cryptodev_pmd_get_dev, rte_cryptodev_pmd_get_named_dev,
rte_cryptodev_pmd_is_valid_dev were incorrectly marked as inline and
therefore not useable from crypto PMDs when built as shared
libraries as they accessed the global rte_cryptodev_globals device
structure.

Fixes: d11b0f30 ("cryptodev: introduce API and framework for crypto devices")

Signed-off-by: Declan Doherty <declan.doherty@intel.com>
Acked-by: Fan Zhang <roy.fan.zhang@intel.com>
7 years agomk: parallelize make config
Ferruh Yigit [Mon, 30 Jan 2017 10:21:15 +0000 (10:21 +0000)]
mk: parallelize make config

make config dependency resolving was always running serial,
parallelize it for better performance.

$ time make T=x86_64-native-linuxapp-gcc config
real    0m12.633s

$ time make -j8 T=x86_64-native-linuxapp-gcc config
real    0m1.826s

When config creation done under a single make target, using a for loop,
make has no control on the action, and it needs to run as implemented in
the rule. But if for loop converted into multiple targets, make can
detect independent targets and run them parallel based on -j parameter.

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
7 years agodoc: add I/O device memory access API to release notes
Jerin Jacob [Thu, 19 Jan 2017 05:43:21 +0000 (11:13 +0530)]
doc: add I/O device memory access API to release notes

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: John McNamara <john.mcnamara@intel.com>
7 years agousertools: fix active interface detection when binding
Yoni Gilad [Tue, 10 Jan 2017 17:14:21 +0000 (17:14 +0000)]
usertools: fix active interface detection when binding

When adding crypto devices, the "Active" and "Ssh_if" attributes of
existing network devices were reset. This causes the following issues:

- Network interfaces aren't marked as "*Active*" in the --status output.
- Active network interfaces can be unbound without the --force option,
  causing loss of network connection.

The reset was caused by the call to devices[d].update in
get_crypto_details.

This patch prevents the update on non-crypto devices.

Fixes: cb4a1d14bf3e ("tools: bind crypto devices")
Cc: stable@dpdk.org
Signed-off-by: Yoni Gilad <yonig@radcom.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agoexamples/server_node_efd: renamed from flow_distributor
Pablo de Lara [Tue, 24 Jan 2017 09:06:14 +0000 (09:06 +0000)]
examples/server_node_efd: renamed from flow_distributor

To avoid confusion with distributor app, this commit
renames the flow-distributor sample app to server_node_efd,
since it shows how to use the EFD library and it is based
on a server/nodes model.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agoapp/test: fix build with overflowed EFD value
Pablo de Lara [Wed, 25 Jan 2017 11:29:18 +0000 (11:29 +0000)]
app/test: fix build with overflowed EFD value

When RTE_EFD_VALUE_NUM_BITS is 32, there was a compilation issue
because of an overflow:

app/test/test_efd.c:157:55: error: overflow in expression;
result is 2147483647 with type 'int' [-Werror,-Winteger-overflow]
        data[0] = mrand48() & ((1 << RTE_EFD_VALUE_NUM_BITS) - 1);

This commit fixes the issue by using a setting a different
macro VALUE_BITMASK with a conditional

Fixes: 0e925aef2779 ("app/test: add EFD functional and perf tests")

Reported-by: Yong Liu <yong.liu@intel.com>
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agoapp/test: make PCI device id struct const
Ferruh Yigit [Fri, 27 Jan 2017 09:53:03 +0000 (09:53 +0000)]
app/test: make PCI device id struct const

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agopdump: fix coverage warning
Andrew Rybchenko [Wed, 18 Jan 2017 07:26:08 +0000 (07:26 +0000)]
pdump: fix coverage warning

Fix GCC 4.8.2 20140120 (Red Hat 4.8.2-16) (RHEL 7.0) false warning
when build with EXTRA_CFLAGS='--coverage'.

Fixes: 278f945402c5 ("pdump: add new library for packet capture")

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
7 years agoapp/testpmd: add offload capabilities query
Qiming Yang [Mon, 16 Jan 2017 02:31:26 +0000 (10:31 +0800)]
app/testpmd: add offload capabilities query

Add two new commands "show port cap <port>" and "show
port cap all" to display what offload capabilities supported
in ports. It will not only display all the capabilities of
the port, but also the enabling condition for each capability
in the running time.

Signed-off-by: Qiming Yang <qiming.yang@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-by: Beilei Xing <beilei.xing@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agoacl: fix flow data comments
Michał Mirosław [Tue, 13 Dec 2016 01:08:16 +0000 (02:08 +0100)]
acl: fix flow data comments

Signed-off-by: Michał Mirosław <michal.miroslaw@atendesoftware.pl>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
7 years agoacl: allow zero verdict
Michał Mirosław [Wed, 18 Jan 2017 19:32:51 +0000 (20:32 +0100)]
acl: allow zero verdict

This enables ACL matches to return 0 where the distinction
from no-match case is not needed.

Signed-off-by: Michał Mirosław <michal.miroslaw@atendesoftware.pl>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
7 years agoacl: remove invalid test
Michał Mirosław [Wed, 18 Jan 2017 19:32:51 +0000 (20:32 +0100)]
acl: remove invalid test

rte_acl_add_rules() has no way of checking rule size.

This was hidden because the test effectively checked that
adding a rule with userdata == 0 failed.

Signed-off-by: Michał Mirosław <michal.miroslaw@atendesoftware.pl>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
7 years agoefd: fix build by removing dependency to libmath
Olivier Matz [Fri, 27 Jan 2017 14:23:17 +0000 (15:23 +0100)]
efd: fix build by removing dependency to libmath

When we compile the dpdk with:
  CONFIG_RTE_LIBRTE_EFD=y
  CONFIG_RTE_LIBRTE_NFP_PMD=n
  CONFIG_RTE_LIBRTE_THUNDERX_NICVF_PMD=n
  CONFIG_RTE_LIBRTE_SCHED=n
  CONFIG_RTE_LIBRTE_METER=n

The linker gives the following error:
  lib/librte_efd.a(rte_efd.o): In function `rte_efd_create':
  lib/librte_efd/rte_efd.c:560: undefined reference to `log2'
  collect2: error: ld returned 1 exit status

This is because the '-lm' is missing in mk/rte.app.mk.

An alternative, which is proposed by this patch, is to use the compiler
builtin rte_bsf32() to process log2 instead of the libmath log2() that
requires to include math.h and link with -lm.

Fixes: 56b6ef874f80 ("efd: new Elastic Flow Distributor library")

Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agoethdev: remove useless pointer initialization
Emmanuel Roullit [Tue, 24 Jan 2017 20:28:35 +0000 (21:28 +0100)]
ethdev: remove useless pointer initialization

Found with clang static analysis:
lib/librte_ether/rte_ethdev.c:2467:22:
warning: Value stored to 'dev' during its initialization is never read
struct rte_eth_dev *dev = &rte_eth_devices[port_id];
                    ^~~   ~~~~~~~~~~~~~~~~~~~~~~~~~

Fixes: 88ac4396ad29 ("ethdev: add VMDq support")

Signed-off-by: Emmanuel Roullit <emmanuel.roullit@gmail.com>
7 years agoethdev: fix MAC address replay
Steve Shin [Fri, 27 Jan 2017 17:57:29 +0000 (09:57 -0800)]
ethdev: fix MAC address replay

This patch fixes a bug in replaying MAC address to the hardware
in rte_eth_dev_config_restore() routine. Added default MAC replay as well.

Fixes: 4bdefaade6d1 ("ethdev: VMDQ enhancements")

Signed-off-by: Steve Shin <jonshin@cisco.com>
Reviewed-by: Igor Ryzhov <iryzhov@nfware.com>
7 years agombuf: remove redundant assignment when attaching
Ilya V. Matveychikov [Fri, 20 Jan 2017 00:19:23 +0000 (04:19 +0400)]
mbuf: remove redundant assignment when attaching

mi->next will be assigned to NULL few lines later, trivial patch

Fixes: ea672a8b1655 ("mbuf: remove the rte_pktmbuf structure")

Signed-off-by: Ilya V. Matveychikov <matvejchikov@gmail.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
7 years agomempool: fix stack handler dequeue
Olivier Matz [Mon, 23 Jan 2017 17:11:03 +0000 (18:11 +0100)]
mempool: fix stack handler dequeue

The return value of the stack handler is wrong: it should be 0 on
success, not the number of objects dequeued.

This could lead to memory leaks depending on how the caller checks the
return value (ret < 0 or ret != 0). This was also breaking autotests
with debug enabled, because the debug cookies are only updated when the
function returns 0, so the cookies were not updated, leading to
an abort().

Fixes: 295a530b0844 ("mempool: add stack mempool handler")
Cc: stable@dpdk.org
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
7 years agodevargs: reset driver name pointer on parsing failure
Emmanuel Roullit [Tue, 24 Jan 2017 20:26:56 +0000 (21:26 +0100)]
devargs: reset driver name pointer on parsing failure

The pointer set by strdup() needs to be cleared on failure to avoid a
potential double-free from the caller.

Found with clang static analysis:
lib/librte_eal/common/eal_common_devargs.c:123:2:
warning: Attempt to free released memory
        free(buf);
        ^~~~~~~~~

Fixes: 0fe11ec592b2 ("eal: add vdev init and uninit")

Signed-off-by: Emmanuel Roullit <emmanuel.roullit@gmail.com>
7 years agoeal: fix warning about debug log at startup
Olivier Matz [Tue, 24 Jan 2017 16:21:12 +0000 (17:21 +0100)]
eal: fix warning about debug log at startup

The log "Debug logs available - lower performance" should
now only be displayed when dataplane debug logs are enabled.

The issue occurs only if the default log level (CONFIG_RTE_LOG_LEVEL) is
set to DEBUG in the configuration, which is not the case by default.

Fixes: 5d8f0baf69ea ("log: do not drop debug logs at compile time")

Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
7 years agokni: guard against unterminated name oops
Michał Mirosław [Tue, 13 Dec 2016 01:08:19 +0000 (02:08 +0100)]
kni: guard against unterminated name oops

If the name is too long, it triggers BUG in alloc_netdev().

Signed-off-by: Michał Mirosław <michal.miroslaw@atendesoftware.pl>
Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agokni: set interface name source as user-space
Michał Mirosław [Tue, 13 Dec 2016 01:08:18 +0000 (02:08 +0100)]
kni: set interface name source as user-space

Signed-off-by: Michał Mirosław <michal.miroslaw@atendesoftware.pl>
Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agoconfig: disable KNI ethtool by default
Ferruh Yigit [Tue, 17 Jan 2017 18:01:47 +0000 (18:01 +0000)]
config: disable KNI ethtool by default

KNI ethtool support (KNI control path) is not commonly used,
and it tends to break the build with new version of the Linux kernel.

KNI ethtool feature is disabled by default. KNI datapath is not effected
from this update.

It is possible to enable feature explicitly with config option:
"CONFIG_RTE_KNI_KMOD_ETHTOOL=y"

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
7 years agokni: add build option for ethtool support
Ferruh Yigit [Tue, 17 Jan 2017 18:01:46 +0000 (18:01 +0000)]
kni: add build option for ethtool support

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agobuildtools: allow symlinks within a source directory
Bruce Richardson [Mon, 23 Jan 2017 12:11:28 +0000 (12:11 +0000)]
buildtools: allow symlinks within a source directory

When creating the symlinks for header files to the include folder, the
relpath script dereferenced all symlinks. This made it impossible to
have file A.h renamed to B.h and then symlinked back to its original
name. This is useful to be able to do when refactoring or reworking
a library. Change this so that we just use the dirname of the path from
readlink, we can use the basename as it was originally, even if it was a
symlink.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
7 years agomk: remove default toolchain prefix for ThunderX
Thomas Monjalon [Mon, 23 Jan 2017 09:10:12 +0000 (10:10 +0100)]
mk: remove default toolchain prefix for ThunderX

The environment variable CROSS must be set when using a cross-toolchain.
However it is counter intuitive to set a default value, considering
the toolchain required to build this architecture is well known.
It is especially weird when using a native toolchain and requiring to
unset this variable on the command line.

Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Reviewed-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
7 years agodevtools: add more git headline checks for acronyms
Ferruh Yigit [Thu, 26 Jan 2017 14:16:47 +0000 (14:16 +0000)]
devtools: add more git headline checks for acronyms

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agodevtools: update git headline prefix check for drivers
Ferruh Yigit [Thu, 26 Jan 2017 14:07:43 +0000 (14:07 +0000)]
devtools: update git headline prefix check for drivers

For the patches that touch multiple drivers in same driver group,
script forces headline prefix start with drv group.

Like for net/a net/b net/c, patch title should be "net: x y z"

Update rule to let "driver" prefix in headline,
for above sample patch title becomes: "drivers/net: x y z"

This prevents patch confused with library with same name.

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agoethdev: fix data reset when allocating port
Yuanhan Liu [Fri, 20 Jan 2017 08:04:53 +0000 (16:04 +0800)]
ethdev: fix data reset when allocating port

Fix an silly error by auto-complete while managing the merge conflicts.
It's the eth_dev_data (but not eth_dev) entry should be memset.

Fixes: d948f596fee2 ("ethdev: fix port data mismatched in multiple process model")

Reported-by: Ferruh Yigit <ferruh.yigit@intel.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
7 years agonet/tap: fix build with old kernels
Keith Wiles [Fri, 20 Jan 2017 14:30:25 +0000 (08:30 -0600)]
net/tap: fix build with old kernels

IFF_MULTI_QUEUE does not exist in older kernels:
drivers/net/tap/rte_eth_tap.c:143:19: error: ‘IFF_MULTI_QUEUE’ undeclared

Signed-off-by: Keith Wiles <keith.wiles@intel.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
7 years agoeal: fix FreeBSD build
Pablo de Lara [Thu, 19 Jan 2017 14:21:35 +0000 (14:21 +0000)]
eal: fix FreeBSD build

rte_bus_scan() and rte_bus_probe() have been introduced
in eal.c, but it is missing the rte_bus.h header file,
for BSD systems.

Fixes: f44abbc12fa0 ("bus: add scanning")
Fixes: c3cec1d80708 ("bus: add probing")

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
7 years agoversion: 17.02-rc1
Thomas Monjalon [Thu, 19 Jan 2017 04:54:41 +0000 (05:54 +0100)]
version: 17.02-rc1

Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
7 years agobus: add probing
Shreyansh Jain [Wed, 18 Jan 2017 14:05:22 +0000 (19:35 +0530)]
bus: add probing

Bus implementations can implement a probe handler to match the devices
scanned against the drivers registered.

Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
7 years agobus: add scanning
Shreyansh Jain [Wed, 18 Jan 2017 14:05:21 +0000 (19:35 +0530)]
bus: add scanning

Scan for bus discovers the devices available on the bus and adds them
to a bus specific device list. Each bus mandatorily implements this
method.

Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
7 years agobus: introduce bus abstraction
Shreyansh Jain [Wed, 18 Jan 2017 14:05:17 +0000 (19:35 +0530)]
bus: introduce bus abstraction

This patch introduces the rte_bus abstraction for EAL.
The model is:
 - One or more devices are connected to a Bus
 - Drivers are running instances which manage one or more devices
 - Bus is responsible for identifying devices (and interrupt propogation)
 - Driver is responsible for initializing the device

This patch adds a 'rte_bus' base class which would be extended for
specific implementations. It also introduces Bus registration and
deregistration functions.

Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
7 years agocrypto/armv8: add documentation
Zbigniew Bodek [Wed, 18 Jan 2017 20:01:58 +0000 (21:01 +0100)]
crypto/armv8: add documentation

Add documentation about the driver and update
release notes.

Signed-off-by: Zbigniew Bodek <zbigniew.bodek@caviumnetworks.com>
Reviewed-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>