Age | Commit message (Collapse) | Author | Files | Lines |
|
Type: refactor
Change-Id: I0bce385c7e391fa2b74646d001980610f80f7062
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Type: refactor
Change-Id: Iff96a79323d7d428a779e9736e07c1dc9dddb518
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Type: improvement
Change-Id: Idb0b079df49c12643c9a93ee0effe011d3489068
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Type: refactor
Change-Id: I2de762953faa5056d5fefa678e4faafbe7710dc6
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
This code was actually never working. It was just used as workaround for
Cisco ENIC issue with tagged default vlan frames.
Today Cisco ENIC provides solution to this problem with devargs flags...
Change-Id: Ia8284274117cb200bf6c7f7911d945d5a093d878
Type: refactor
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Type: improvement
Previously multiple sw crypto scheduler queues per core design
caused unaverage frame processing rate for each async op ID –
the lower the op ID is the highly likely they are processed first.
For example, when a RX core is feeding both encryption and
decryption jobs of the same crypto algorithm to the queues at a
high rate, in the mean time the crypto cores have no enough
cycles to process all: the jobs in the decryption queue are less
likely being processed, causing packet drop.
To improve the situation this patch makes every core only owning
a two queues, one for encrypt operations and one for decrypt.
The queue is changed either after checking each core
or after founding a frame to process.
All crypto jobs with different algorithm are pushed to
thoses queues and are treated evenly.
In addition, the crypto async infra now uses unified dequeue handler,
one per engine. Only the active engine will be registered its
dequeue handler in crypto main.
Signed-off-by: DariuszX Kazimierski <dariuszx.kazimierski@intel.com>
Signed-off-by: PiotrX Kleski <piotrx.kleski@intel.com>
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Signed-off-by: Jakub Wysocki <jakubx.wysocki@intel.com>
Change-Id: I517ee8e31633980de5e0dd4b05e1d5db5dea760e
|
|
It is actually longer and slower...
Type: refactor
Change-Id: I0f126d4cdb13ecc60a2d370409f23820d7f7eb72
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Now DPDK have API to register external threads so we can remove this
mess...
Type: improvement
Change-Id: I71a21f0cd94bd668aa406710c75a0bcc63fdc840
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
TCP csum offload fails although udp seems to work.
Type: fix
Fixes: fa1fb60
Signed-off-by: Florin Coras <fcoras@cisco.com>
Change-Id: Ie0651887b09920365806eaad776b0d13059faee8
|
|
Type: improvement
This patch adds AES-CTR-128/192/256 + SHA1 linked algo support to dpdk
cryptodev.
Signed-off-by: PiotrX Kleski <piotrx.kleski@intel.com>
Change-Id: Idc162b29f4075ef8be9577abd3daf6de05f84faa
|
|
Originally cryptodev allocates mempools for seesion and session private
data during its initialization. Moreover the size of these mempools are
fixed resulting in limited session count (up to value specified in
CRYPTODEV_NB_SESSION macro).
This patch allows for session count to scale up by allocating new
mempools as they are needed during session creation.
Type: improvement
Signed-off-by: Dastin Wilski <dastin.wilski@gmail.com>
Change-Id: I6ae240b474d3089d3ff50ca5bc7ff48f149983db
|
|
Changed dpdk-input prefetch of vlib_buffer_t to prefetchw.
vlib_buffer_t was being prefetched without 'ownership', which may
cause a stall when the buffer is subsequently written to. This saves
4 clocks a packet when the buffer is shared a cross cores, and has no
impact when not sharing.
Type: improvement
Signed-off-by: Ray Kinsella <mdr@ashroe.eu>
Change-Id: I317af2a38ef536022e68552351a8507861f62dad
|
|
Type: improvement
Signed-off-by: Florin Coras <fcoras@cisco.com>
Change-Id: Ia25e671084dd2c0010c0577649bf51ba6495b6ac
|
|
Add a hint so that it's obvious that fall through in switch statement is
intentional.
Type: fix
Fixes: 34c54dff5c
Signed-off-by: Klement Sekera <ksekera@cisco.com>
Change-Id: I12271227424761fb89b03a390f626c2ab466472c
|
|
Type: refactor
Change-Id: I76ccf8970ebb3f180ce745d8b515c5e0724784d6
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
- move existing ethernet hash functions to hash infra (no l4
awareness)
- refactor code to use hash infra and add apis to request l4 aware
hashing functions
- hashing functions per interface
- code cleanup
Type: improvement
Signed-off-by: Florin Coras <fcoras@cisco.com>
Change-Id: Ia9f44db98d83f0f027aeb37718585a2e10ffd2c6
|
|
unformat_init_vector() expects a vector, not a NULL-terminated C-string.
Type: fix
Change-Id: I20a266243f63d94b0c6fe24e25ee8346c08c8ff2
Signed-off-by: Benoît Ganne <bganne@cisco.com>
|
|
Type: improvement
Change-Id: Ic7c2ac4237ecd192def7c3530ae5f788c62cf9ad
Signed-off-by: Benoît Ganne <bganne@cisco.com>
|
|
Type: fix
Signed-off-by: Florin Coras <fcoras@cisco.com>
Change-Id: I89fb738903ee74ffcb7c77a041391f0388df6991
|
|
Type: improvement
Signed-off-by: Mohsin Kazmi <sykazmi@cisco.com>
Change-Id: I10141033030342881298d70742fa5bdea402b4c9
|
|
CLIB_PREFETCH (cop[1], CLIB_CACHE_LINE_BYTES * 3, STORE);
Note on 64 bytes cache line size arm machines,
CLIB_CACHE_LINE_BYTES 128
CLIB_CACHE_PREFETCH_BYTES 6
above CLIB_PREFETCH () macro will be expand to
ASSERT ((size) <= 4 * CLIB_CACHE_PREFETCH_BYTES);
it will hit assert due to size (i.e. 3 * 128) > 4 * 64
Solution:
Change to CLIB_PREFETCH (cop[1], sizeof(*cop[1]), STORE);
Type: fix
Signed-off-by: Tianyu Li <tianyu.li@arm.com>
Reviewed-by: Lijian Zhang <lijian.zhang@arm.com>
Change-Id: Id0981fd5bd2b25ff71db4197b25578d0b7a9803e
|
|
Coverity does not seem happy about exotic control
flows in switch/case statements
Change-Id: Ie006190058f811aacf85125ebc58330d9c53510f
Signed-off-by: Mohammed Hawari <mohammed@hawari.fr>
Type: fix
Fixes: 0b42ac565b970c186a9ad734f980b440f56fb25b
|
|
Change-Id: Ifa09dcdfd567ac3c2afab4bd3bd468260f766e1d
Signed-off-by: Mohammed Hawari <mohammed@hawari.fr>
Type: improvement
|
|
Type: feature
This patch adds IPsec ESP SPI as RSS key for better IPsec
inbound processing scaling. With this feature enabled,
The NIC will use the packets' SPI index as the RSS key
to distribute them to different queues.
The dpdk-input is also updated to support this feature.
Sample CLI command to enable Ipsec ESP SPI as RSS key:
test flow add src-ip any dst-ip any rss types esp
test flow enable eth0 index 0
Signed-off-by: Piotr Bronowski <piotrx.bronowski@intel.com>
Signed-off-by: Gabriel Oginski <gabrielx.oginski@intel.com>
Change-Id: I8b7d01cfc065e9099fad33042ce76898e16ddbf0
|
|
- add format_vnet_buffer and format_vnet_buffer_no_chain to mirror
format_vlib_buffer and format_vlib_buffer_no_chain
- format_vnet_buffer used to be the "no chain" version, replace all of
its current use with the corresponding format_vnet_buffer_no_chain
- add a function to dump vnet buffer details from gdb
Type: improvement
Change-Id: I143ce845f80e7ef937ea33a557b6e3b5988c5b8f
Signed-off-by: Benoît Ganne <bganne@cisco.com>
|
|
Type: improvement
Change-Id: Ic07010f11ef303f5213a33b0faf24aaedb62f110
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Type: improvement
Change-Id: I06eaf39b1e441045c3402cbf40339054ad26ade9
Signed-off-by: Mohsin Kazmi <sykazmi@cisco.com>
|
|
Type: improvement
Request use of RX interrupts for virtio if the system will support it,
which is done by applying the same check as in the virtio driver,
namely whether multiple interrupts are supported. This allows the use
of RX adaptive/interrupt mode instead of just polling, which is useful
in virtualised environments where functionality may be more important
than performance and so using polling mode is wasteful.
Signed-off-by: Robert Shearman <robertshearman@gmail.com>
Change-Id: I29527b6f04b0b1d0c9f9424751b2bd252ed10505
|
|
Type: refactor
Change-Id: Id10cbf52e8f2dd809080a228d8fa282308be84ac
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Type: improvement
Allow the use of the base-virtaddr config option to be passed through
to DPDK. This is useful to allow use of devices with an IOMMU in
nested VMs without resorting to PA IOVA mode.
Signed-off-by: Robert Shearman <robertshearman@gmail.com>
Change-Id: I32b6513377e6d20bf155e12c45f902d51ea982c4
|
|
Type: improvement
Set port type and supported flow actions on 2.5G Intel NICs which are
managed by the igc PMD.
Change-Id: Id144eaa158e24a3f10effb6b517d1c62fba0d2e8
Signed-off-by: Matthew Smith <mgsmith@netgate.com>
|
|
Type: fix
Signed-off-by: Piotr Bronowski <piotrx.bronowski@intel.com>
Change-Id: I76923ad5035498aae821db4fd42a127617476fbb
|
|
Type: feature
This patch bumps DPDK version to 21.05 and updated VPP to
accomodate the changes in DPDK latest version.
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Change-Id: If217441f70c9ab531196dca7ec7a486ec9931cff
|
|
Type: fix
This patch fixes the missing symbol of dpdk_plugin.so when
creating symmetric key. The solution is to add dependency
of libssl to dpdk cryptodev and disable cryptodev engine
when libssl is not presented.
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Change-Id: I30aa6e3e3af1faefa82883bad613e1d82235a2ec
|
|
Change-Id: I665d9d11fcdd9af33e38972a74e51ce3085f3e55
Signed-off-by: Mohammed Hawari <mohammed@hawari.fr>
Type: fix
|
|
If dpdk device starts failed for some reason, we should not try to get
stats and the interface must remain down.
Type: fix
Change-Id: I0079f89d3b43a9b47553991347da8ed4ce6ede8d
Signed-off-by: Benoît Ganne <bganne@cisco.com>
|
|
Type: fix
This patch fixes the possible segmentation fault in DPDK
cryptodev when deleting session.
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Change-Id: Ie81de09f7250eb5c1ed6ee471363137d8372fe27
|
|
Type: fix
Change-Id: I433fe3799975fe3ba00fa30226f6e8dae34e88fc
Signed-off-by: Mohsin Kazmi <sykazmi@cisco.com>
|
|
Type: improvement
Some tests i.e. ipsec see performance regression when offload flags
are moved to 2nd cacheline. This patch moves them back to 1st cacheline.
Change-Id: I6ead45ff6d2c467b0d248f409e27c2ba31758741
Signed-off-by: Mohsin Kazmi <sykazmi@cisco.com>
|
|
Enable configuration of DPDK's max-simd-bitwidth through the startup.conf
Type: improvement
Change-Id: I455148714ffc7caa257931526f310dbefd7ab01c
Signed-off-by: Ray Kinsella <mdr@ashroe.eu>
|
|
trajectory trace has been broken for a while because we used to save the
buffer trajectory in a vector pointed to in opaque2. This does not work
well when opaque2 is copied (eg. because of a clone) as 2 buffers end up
sharing the same vector.
This dedicates a full cacheline in the buffer metadata instead when
trajectory is compiled in. No dynamic allocation, no sharing, no tears.
Type: refactor
Change-Id: I6a028ca1b48d38f393a36979e5e452c2dd48ad3f
Signed-off-by: Benoît Ganne <bganne@cisco.com>
|
|
Type: make
Change-Id: I780c1c81a50cb92bc89c05856efd8ef88479c0ab
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
With this change, enable MD5 with AES-CBC support in dpdk cryptodev.
Type: improvement
Signed-off-by: Alexander Chernavin <achernavin@netgate.com>
Change-Id: Ic587aaa1fa0dc102e36eb34f329ef21a16156f26
|
|
Since the move to DPDK 21.02, compatibility build symbols were depreca-
ted, breaking the static linkage of the DPDK plugin to libibverbs.
Change-Id: I2fec0807e4c0eb00268618f1495af862eede9081
Type: fix
Fixes: 5f35a185d5afee2c364fe84f33922bac1a808113
Signed-off-by: Mohammed Hawari <mohammed@hawari.fr>
|
|
This patch fixes a vxlan flow type checking bug which is introduced in
commit 7a016e87e473.
It is the native avf instead of dpdk that doesn't support vxlan flow offloading.
So this patch removes Vxlan flow supporting in native avf_plugin and adds
the vxlan flow back to dpdk_plugin.
Type: fix
Signed-off-by: Chenmin Sun <chenmin.sun@intel.com>
Change-Id: I1a72629cee462b890327c589c846f6d532b2d61d
|
|
Type: fix
Signed-off-by: Chenmin Sun <chenmin.sun@intel.com>
Change-Id: I215b660a466e0579918e49de543163fd28f95c22
|
|
This patch fixes the below bugs in avf flow:
1. remove the VXLAN flow as it is not yet supported
2. fix a typo in IPv6 flow
3. fix the wrong type of the AH flows
Also did some cleanup in this patch
Type: fix
Signed-off-by: Chenmin Sun <chenmin.sun@intel.com>
Change-Id: I5eae131f1e89071364a41b7006a00e8425089a4f
|
|
Type: improvement
This patch combined cryptodev op and cryptodev raw API data
paths into one and makes the engine run-timely select which
data path is used: if all cryptodev devices support
RTE_CRYPTODEV_FF_SYM_RAW_DP feature flag, the raw data path
API is used, otherwise the traditional data path is used.
Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
Change-Id: Ibbd7c4405288bd9a48a34878954fd3040df7b4ad
|
|
The crypto op data offset passed into DPDK is relative to the mbuf
buffer address plus the mbuf data offset, therefore the mbuf data
offset needs to be set rather than left at whatever previous value it
was at, which is likely to be incorrect and result in the wrong
portion of the packet being encrypted/decrypted for GCM.
The fe->crypto_start_offset field is relative to the start of the vlib
buffer (as opposed to the current data pointer), so set the mbuf
data_off field to VLIB_BUFFER_PRE_DATA_SIZE when performing a GCM
crypto op enqueue to match the crypto_start_offset semantics. This
then matches the behaviour in the non-GCM case.
Type: fix
Change-Id: I0ac2a44139387158765a3e04cfcaa5ee6f11d395
Signed-off-by: Robert Shearman <robertshearman@gmail.com>
|
|
Type: improvement
Change-Id: I73383eb15186021cd6527d112da8443a0082f129
Signed-off-by: Damjan Marion <damarion@cisco.com>
|