Age | Commit message (Collapse) | Author | Files | Lines |
|
Type: fix
Change-Id: I5f309ca4db4ae4a3e475d87b8f0188c4ead5e562
Signed-off-by: lijinhui <lijh_7@chinatelecom.cn>
|
|
Currently sw_scheduler runs interchangeably over queues of one selected
type either ENCRYPT or DECRYPT, then switches the type for the next run.
This runs perfectly when we have elements in both ENCRYPT and DECRYPT
queues, however, this leads to performance degradation when only one
of the queues have elements i.e either all traffic is to be encrypted or
decrypted.
If all operations are encryption, then 50% of the time, the loop exits
without dequeueing.With this change, that dequeueing happens on every
loop. This increases the performance of single mode operation (ecryption
or decryption) by over 15%.
This change was also added in commit https://github.com/FDio/vpp/commit/61cdc0981084f049067626b0123db700035120df
to fix similar performance issue when the crypto-dispatch node is in interrupt
node, however was removed by https://github.com/FDio/vpp/commit/9a9604b09f15691d7c4ddf29afd99a31e7e31eed
which has its own limitations.
Type: improvement
Change-Id: I15c1375427e06187e9c4faf2461ab79935830802
Signed-off-by: Niyaz Murshed <niyaz.murshed@arm.com>
|
|
vl_api_lcp_itf_pair_get_v2_t_handler method was not be able to
handle invalid sw_if_index, it caused a seg fault.
With this fix, the method checks for an invalid sw_if_index and
sends error back to caller.
Type: fix
Change-Id: I4e89b0cab34f9109c4ce2689021ecfc2786055ab
Signed-off-by: Anton Nikolaev <anikolaev@netgate.com>
|
|
Type: fix
Signed-off-by: Florin Coras <fcoras@cisco.com>
Change-Id: Iadd6a4dd928002e764fc52343c14403620279aa7
|
|
Type: fix
Change-Id: I9d7e6dd099cf9f4b7f6bb06d9e8a17fac7d7e772
Signed-off-by: Benoît Ganne <bganne@cisco.com>
|
|
Type: fix
Change-Id: I0a43a37971d03a700926d59e848f0b6e6dbeb19a
Signed-off-by: Benoît Ganne <bganne@cisco.com>
|
|
Type: improvement
Change-Id: Ib824d0ca9efc7d8967e043db69017655b2dcf6b5
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
'queue' is declared on the stack and must be big enough to handle
queue_size + max processed elements per loops.
When AVX-512 was added, this additional max was not increased from 4 to
8 accordingly.
Type: fix
Fixes: 767a9ab4b1
Change-Id: I15f2f930fc316f92da2e4642f41852d856c06a48
Signed-off-by: Benoît Ganne <bganne@cisco.com>
|
|
Some installation destinations were already converted to the
GNUInstallDirs standard in e3cf4d0cf. This patch converts the share and
include directories (typically /usr/share and /usr/include respectively)
as well.
Type: improvement
Change-Id: Id481b16604f3f52203166a5c0647f43d6eb05239
Signed-off-by: Wim de With <wf@dewith.io>
|
|
Change-Id: Ic1fa3bd164e80c2ca1146be001870da0238a5f2e
Type: improvement
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Type: improvement
Change-Id: I51345fb75843c67c6bf6a4c56380e1f0899c45b1
Signed-off-by: Tom Jones <thj@freebsd.org>
|
|
The rewrite string can be up to max_size, and max_size can be up to
VNET_REWRITE_TOTAL_BYTES. Don't waste the last byte.
Type: fix
Change-Id: I2fb7e9873b6b4c1e6a55b172c7f753f3c5910802
Signed-off-by: Benoît Ganne <bganne@cisco.com>
|
|
In case of multiple path within tunnel, mpls lookup node
computes lb hash with mpls_compute_flow_hash config value 0,
so only mpls label and l4 ports gets accounted, not 5-tuple.
This leads to flow traffic polarization and disbalance over
mpls paths.
Use mpls hash config from lb instead, usually it'll be
MPLS_FLOw_HASH_DEFAULT with 5-tuple plus flowlabel.
As optimization, fix flow hash reuse from the previous lookup
node if present, like ip_lookup does. Previously mpls lookup
always calcs the hash.
Test lb distribution for both cases.
Also, use the same flow hash hex format in ip4/ip6 and mpls
traces for easier reading, most code changes is due fixstyle
formatting.
Type: fix
Signed-off-by: Vladislav Grishenko <themiron@yandex-team.ru>
Change-Id: Ib89e1ab3edec14269866fe825a3e887d6c817b7c
|
|
On unnumbered interfaces, ARP fails because there is no attached route.
Allow replies to peer-to-peer addresses on unnumbered interfaces:
eg. 192.0.2.1/32 <-> 192.0.2.2/32
Type: fix
Change-Id: Ibeb8d8ebc8d58d5bfb0724739a17694e0217356e
Signed-off-by: Pim van Pelt <pim@ipng.nl>
|
|
Only set state to ready if session is not already closing.
Type: fix
Change-Id: Ic95667f43ed09d693f1cf7c9f1c16f7f995ea2d8
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Use mask_load_zero to avoid out-of-buffer reads in vectorized
function versions.
Type: fix
Signed-off-by: Dmitry Valter <d-valter@yandex-team.com>
Change-Id: I12bcb817ccf2db210c1c99fdfa444dc3f540035b
|
|
Type: fix
Fixes: 01fe7ab
Change-Id: I4425e809f0977521ddecf91b58b26fe4519dd6e0
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Type: fix
Change-Id: I739b7129c7e5a3fccefcdeeaf7f4a298223dd8eb
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Type: fix
Fixes: 01fe7ab
Change-Id: I4419107c4bcb7f85b76addfc62178b6e75e10a52
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
vl_api_prefix_t.len is 1 byte only, but unformat %d writes 4 bytes
add helper functions unformat_u(8|16) which don't write more than
appropriate amount of bytes
fix other similar errors in vpp/api/types.c
Type: fix
Change-Id: I74a61a377147c373f8c25ed083052b2287763c39
Signed-off-by: Klement Sekera <klement.sekera@gmail.com>
|
|
Type: fix
Change-Id: I56c1a2717f197c889425449b37f51b0f2cc89ea5
Signed-off-by: Maxime Peim <mpeim@cisco.com>
|
|
Type: test
Change-Id: I01500466f3d15c79e38028677ce7e5c75d427fdc
Signed-off-by: Dave Wallace <dwallacelf@gmail.com>
|
|
Type: fix
Signed-off-by: Florin Coras <fcoras@cisco.com>
Change-Id: I1464e93cd3020eaa26068df558924e39dd255ccb
|
|
- also fix memory leak in adj_glean_walk_proto()
Type: fix
Change-Id: I3cd72b14506e6bfc9d8d77a65d7b9b2703992367
Signed-off-by: Dave Wallace <dwallacelf@gmail.com>
|
|
Properly handle small buffers in the AVX256 version of array_mask_u32.
Type: fix
Signed-off-by: Dmitry Valter <d-valter@yandex-team.com>
Fixes: f62ed3f9c1ec3e8db36f63d6a54f46b7bea43723
Change-Id: Idf1cb43ccf37bd7c439d11e4f68fe30064c6d09a
|
|
After memclnt api client disconnect, read timeout is set to zero
and stays the same even after same client reconnect. It causes
client process to spin in timeout loop up to 100% cpu.
Fix it by resetting timeout process state upon every (re)connect.
Type: fix
Signed-off-by: Vladislav Grishenko <themiron@yandex-team.ru>
Change-Id: I56812972a69c343f869eebbdfebdcbefd3d201e0
|
|
Update to the MAP rule lookup (in IPv6) based on the rule's source
prefix instead of DMR
Type: improvement
Per RFC, the DMR is allowed to serve multiple MAP Basic Mapping Rules, but this
capability was prevented by the above logic.
Updates to the code include populating a new hash table based on the MAP rule
ip6 prefix and length, changing several functions to reference this new table,
and slight alterations to a few functions regarding pre-lookup bitmasking.
All changes are commented with [dgeist] and are in need of peer review,
especially the bitmask alterations.
An attempt was made at generating an additonal MAP rule in the test_map_br test
harness, but the coding appears very much oriented towards testing just one
rule. I would appreciate suggestions on how to test multi-rule cases.
Issue: VPP-2111
Change-Id: Id1fea280eba625e23cd893575d9b63aac7f48405
Signed-off-by: Dan Geist <dan@polter.net>
|
|
Erase removed elements to avoid memory leaks.
Type: fix
Change-Id: I1605bea8df7e08455691364efb0a12eb0d9dae93
Signed-off-by: Klement Sekera <klement.sekera@gmail.com>
|
|
fix use of vip after it was deleted
Type: fix
Fixes: 041eacc81656d2ed5bc01b96b15a7d03a1700f13
Change-Id: I5723485c5da7507fbc6c86ff6eb9f77127439f67
Signed-off-by: Georgy Borodin <bor1-go@yandex-team.ru>
|
|
l2 tunnels like vxlan, gtpu, geneva use vnet_l2_compute_flow_hash() to
compute flow hash for udp src port entropy. In case of inner mpls tunnels
to the same lsr ethernet src and dst macs are the same, so l2 flow hash
is also the same leading to no src port entropy and the only rss queue
overflow on receiver side.
Fix it for all the possible vnet_l2_compute_flow_hash callers by making
mpls playload hash in additon to ip4/ip6. Visible performance impact is
not expected as it's only one check for mpls ethertype for common cases.
Type: fix
Signed-off-by: Vladislav Grishenko <themiron@yandex-team.ru>
Change-Id: I69153d42fb3d7c094a670c674fac8d14039c626a
|
|
This change allows to limit lcp interface queues
to be used by explicit host interface workers.
Type: improvement
Signed-off-by: Vladislav Grishenko <themiron@yandex-team.ru>
Change-Id: I0626f66021e5a5f251470156231dc44ddaea5ee6
|
|
Type: improvement
Change-Id: I8c248d9e224bd069b641a174da57d448371470af
Signed-off-by: Jon Loeliger <jdl@netgate.com>
|
|
Type: improvement
Change-Id: I6cf5adbd609d911e15dcc6d976cda4ad21ce89ad
Signed-off-by: Tom Jones <thj@freebsd.org>
|
|
Ensure counter index is valid before using it to lookup
the counter.
Type: fix
Change-Id: I423c7a6aa6b65f6367b18d8e99cf40f52e06b416
Signed-off-by: lijinhui <lijh_7@chinatelecom.cn>
|
|
Type: fix
Fixes: 01fe7ab
Change-Id: I72fdaca250468d91a31efcce2fb447c97ba49dc7
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
It may happen that rx queue is empty of free buffers due to previous
alloc failure.
Type: fix
Fixes: 01fe7ab
Change-Id: I344dcda11525444bd1358b3d36ffdf8ab9aa2677
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Type: fix
Change-Id: Ic015b37e18a43c49c3fb3dbff284a17fa2c5fd99
Signed-off-by: Fan Zhang <fanzhang.oss@gmail.com>
Signed-off-by: Kai Ji <kai.ji@intel.com>
|
|
Don't force tx rescheduling of tls session if no forward progress is
made. The session will still be rescheduled by the session infra if
there's pending tx data.
Type: fix
Signed-off-by: Florin Coras <fcoras@cisco.com>
Change-Id: Ic57b6ee79969055cec782938668c054bcc39f206
|
|
Type: fix
Signed-off-by: Florin Coras <fcoras@cisco.com>
Change-Id: Ieb4e9d9e39b937ad4c7316b3955b3ca296f0a191
|
|
Type: improvement
Change-Id: Iefe5c2e610a26241a88ca783ac548fd8f2317bb0
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Type: improvement
Signed-off-by: Florin Coras <fcoras@cisco.com>
Change-Id: I04f836d09a1cbd5a5b55dc64359d2d761dfc4988
|
|
First step towards moving to an 8 byte struct instead of u64.
Type: improvement
Signed-off-by: Florin Coras <fcoras@cisco.com>
Change-Id: Idd0b95520ab7158e175b9af1702fc09c0613a4bb
|
|
Type: feature
Auto core pinning now fetches vpp cpu affinity list
using pthread api. This enables us to do core-pinning in
environments where the host cpu list does not necessarily align
with cpus available to vpp
Change-Id: Ife8c2a2351c08c5c6c4fdf7c729eeff2697bc39a
Signed-off-by: hsandid <halsandi@cisco.com>
|
|
Type: refactor
Change-Id: Ifd7f5b351401cdcaaaf57fefc5dbbfdaf235054e
Signed-off-by: hsandid <halsandi@cisco.com>
|
|
Change-Id: I1f8adf1f5650ab6c04e03c95d7a8d0bfa39b5f2d
Type: improvement
Signed-off-by: Mohammed Hawari <mohammed@hawari.fr>
|
|
Make sure session is marked as invalid.
Type: improvement
Change-Id: I1c861645de95ef15a24acd4fe6dd5364a55b4fb8
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Type: improvement
Change-Id: I958e059384db3c13a29f64be96877f57617bbae2
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
On FreeBSD a races seems to occur between the states socket connecting
and VPP accepting the connection and calling getpeername. The recv_fd
call here will fail with no fds before the VPP process has time to
handle the connection.
Introducing a socket timeout removes this failure on FreeBSD and allows
the test suite to run. Add a timeout on all platforms to try to avoid
this race everywhere.
Type: improvement
Change-Id: I96f77f924a9491222bc213b534cb17001d081f3f
Signed-off-by: Tom Jones <thj@freebsd.org>
|
|
CMSG_LEN provides the length required to store an ancillary item, but it
does not consider padding. RFC 3542 requires portable applications to
use CMSG_SPACE(), this is documented in the python API:
https://docs.python.org/3/library/socket.html#socket.CMSG_LEN
Type: improvement
Change-Id: I0eab470585f2f39bea38639db95d6b0e047d2cff
Signed-off-by: Tom Jones <thj@freebsd.org>
|
|
Type: improvement
Change-Id: I43f59037181ec6b76499e8ee1d82c04670119dc9
Signed-off-by: Tom Jones <thj@freebsd.org>
|