Age | Commit message (Collapse) | Author | Files | Lines |
|
Also fixes vcl client/server stats and closing procedure.
Change-Id: I7d5a274ea0a3c8ea13062bf61bf402248dfe1a19
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Otherwise the key destructor is not called on pthread_exit.
Change-Id: I11e6b9683a926eecd3f40a44aab41924ff9c3101
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I770607cb213d538a29d7bd5762682981b01e9d18
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I72d400401a308012b43513179870823f6f921e44
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I7fcfddc2bc7d9a64f8aa0d57ba5d11d325a15ce1
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I27532b3ab244dc95955e836a42b229a6e4e32818
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Allows app workers to listen on the same session endpoint. Incoming
connects are spread across the workers in a round-robin fashion
Change-Id: Ib5f5817230d9abc6127a85cdbdcad70d980c0f7f
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
-- Not drop session dgram hdr when to drop processed udp message from tx fifo
Change-Id: Id5ce5780817041bca05011f7b9b53e97a6d78da4
Signed-off-by: shubing guo <guo.shubing@zte.com.cn>
|
|
Add basic support for app registration of multiple workers. LDP does not
work with multi-worker apps.
Change-Id: I3fc421a2a591a077b275827463f874b261415a63
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: Iac6e1c32cf99c5392a29f7366401b7fc39e463e3
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I2a25bba675fc9c84e5d391533e92b9a041637405
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Refactor session layer to support multiple workers per application.
Change-Id: Ie67354688d396449d14bbbb8c56050206e307cd8
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I89e8052f2d2c36dd3de5255c4ee570722dc58227
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I0f5c1dc99610b65646c3b5cf8da7aea273371e4d
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
- replies should not need to contain client_index since
it is used to identify sender of requests to VPP
Change-Id: Iece3853b3f020054ee1652b149d0cf8d9580db4e
Signed-off-by: Ondrej Fabry <ofabry@cisco.com>
|
|
- support eventfd based mq signaling. Based on configuration, vcl
epoll/select can use either condvars or epoll on mq eventfds.
- add vcl support for memfd segments
- vpp explicitly registers cut-through segments with apps/vcl
- if using eventfd, make ldp allow one call to libc_epoll_create. Needed
for the message queue epfd
- update svm_queue_t to allow blocking calls with eventfd signaling.
Change-Id: I064151ac370bbe29bb16c968bf4e3659c8286bea
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I77a6e092a42290eed7201ad4a62e0d00ef997d2b
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: Ie6171c12055cde6915856de340839f5da1b1b1da
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I17ffec018f5d2fb06a7a3af6d8bc6128ffee3ae6
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Have vcl poll and wait on the event message queues as opposed to
constantly polling the session fifos. This also adds event signaling to
cut through sessions.
On the downside, because we can't wait on multiple condvars, i.e., when
we have multiple message queues because of cut-through registrations, we
do timed waits.
Change-Id: I29ade95dba449659fe46008bb1af502276a7c5fd
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I8d631121b104fb40c20701d4c1b428f2b71e5785
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I4505954c737e8c6a3d4177c6833f2b12099e6dd4
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I242056bc46ddb671064665916b2687860292dcb2
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I3c58367eec2243fe19b75be78a175c5261863e9e
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: Ibbba75d069ca1bbf9e5a1b8bd2f405d32021c656
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
VPP-1340
Change-Id: Ia3f2fc054dbfb2ba912a768c70466e7042024e32
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: Ic5daed1f6c23d9561a04e235dcbf257f190d066a
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I0573d0aff39581bba96e610228a10ae923a8ca06
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Also add single loop prefetch in session tx
Change-Id: Ib2725a2552fcc1a65050d7a2eb10491a7b34db62
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Does some session node cleanup as well
Change-Id: Ifd52b07b28ba4dec1f6f729476decc76eb963837
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: Ib88d703bb7d4b170059960b0688352c90c5fcc39
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: Idcba72fd84128547718dd32858e8c728925a6b1d
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
- cache and reuse tcp options and rcv_wnd for session layer tx bursts
- avoid reading/setting total_length_not_including_first_buffer. It's
part of a buffer's second cache line so it comes at a "cost".
Change-Id: Id18219c2f7e07cf4c63ee74f9cdd9e5918904036
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
It would seem that this avoids using the stack for computing the
subsequent crc32 by storing the 2 x u64 ip4 5-tuple key into two
registers. Probably the cast to the "convenience" data structure "masks"
the fact that the key is 16B. Probably ... :-)
"sh run" on an ip4 tcp sender reports that the number of clocks spent in
tcp_input, where a session lookup is performed, drops ~25% (from ~100
clocks/packet to ~75 clocks/packet)
Change-Id: I50d647115ac72f8407bff7e3b73328ba42f0ddb5
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I4a7e4f616ed4a4df82be3112ac4702f7e4e2025c
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: Idf17a0633a1618b12c22b1119e40c2e9d3192df9
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
- Cleanup session state after last ack and avoid using a cleanup timer.
- Change session cleanup to free the session as opposed to waiting for
delete notify.
- When in close-wait, postpone sending the fin on close until all
outstanding data has been sent.
- Don't flush rx fifo unless in closed state
Change-Id: Ic2a4f0d5568b65c83f4b55b6c469a7b24b947f39
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Since the main thread is not used for session polling anymore, when vpp
is started with multiple wokers, allocate connections on the first. Also
add a simple udp make test.
Change-Id: Id869f5d89e0fced51048f0384fa86a5022258b7c
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: Ib8e332174d96bf9cfa4bbaaa5b8d8bc9958424b1
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
If the caller is the session owning thread or the main thread with a
worker barrier sync (cli/api) add an event to the pending disconnects
vector in the session node and entirely avoid using the event queue.
Useful for bursts of disconnects (like an app detach).
If disconnects come from a processes, be willing to retry enqueueing the
disconnect to the event queue multiple times.
Change-Id: Ieece1f1091b713f94c41c703b6e805bc8498816a
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
This api only supports default tables but v6 and v4 default tables are
different.
Change-Id: I7fe39251af99082278e84d7411511bdeca34e609
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: If37d644a26132c42a04927f612eec5e891066451
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
- fix newreno cwnd computation
- reset snd_una_max on entering recovery
- accept acks beyond snd_nxt but less than snd_congestion when in
recovery
- avoid entering fast recovery multiple times when using sacks
- avoid as much as possible sending small segments when doing fast
retransmit
- more event logging
Change-Id: I19dd151d7704e39d4eae06de3a26f5e124875366
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I6615bb612bcc3f795b5f822ea55209bb30ef35b5
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Add a session process node that handles main thread tx and retransmit in
order to avoid having a polling input node.
Change-Id: I3357e987c023a84b533b32793e37ab4204420f64
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: Idae4ecb60351f2e74bad2f2a33c073de8412fcb0
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Change-Id: I1f7877af61f3726cfb7b93ce7893f6df23e866a6
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
- rework the function to declutter and avoid building more than one tx
frame
- add dual loop although benefits in my tests seem to be minimal
- improve tcp/udp echo external apps. They have slightly better
throughput than internal echo apps.
- udp bugfixes
Change-Id: Iea4a245b1b1bb407a7f403dedcce2664a49f774b
Signed-off-by: Florin Coras <fcoras@cisco.com>
|
|
Object sizes must evenly divide alignment requests, or vice
versa. Otherwise, only the first object will be aligned as
requested.
Three choices: add CLIB_CACHE_LINE_ALIGN_MARK(align_me) at
the end of structures, manually pad to an even divisor or multiple of
the alignment request, or use plain vectors/pools.
static assert for enforcement.
Change-Id: I41aa6ff1a58267301d32aaf4b9cd24678ac1c147
Signed-off-by: Dave Barach <dbarach@cisco.com>
|
|
Change-Id: I906e58b4f9827a79a6ab673f8fa2e03036c69820
Signed-off-by: Florin Coras <fcoras@cisco.com>
|