linux/net/sctp
Marcelo Ricardo Leitner 4c2f245496 sctp: linearize early if it's not GSO
Because otherwise when crc computation is still needed it's way more
expensive than on a linear buffer to the point that it affects
performance.

It's so expensive that netperf test gives a perf output as below:

Overhead  Command         Shared Object       Symbol
  18,62%  netserver       [kernel.vmlinux]    [k] crc32_generic_shift
   2,57%  netserver       [kernel.vmlinux]    [k] __pskb_pull_tail
   1,94%  netserver       [kernel.vmlinux]    [k] fib_table_lookup
   1,90%  netserver       [kernel.vmlinux]    [k] copy_user_enhanced_fast_string
   1,66%  swapper         [kernel.vmlinux]    [k] intel_idle
   1,63%  netserver       [kernel.vmlinux]    [k] _raw_spin_lock
   1,59%  netserver       [sctp]              [k] sctp_packet_transmit
   1,55%  netserver       [kernel.vmlinux]    [k] memcpy_erms
   1,42%  netserver       [sctp]              [k] sctp_rcv

# netperf -H 192.168.10.1 -l 10 -t SCTP_STREAM -cC -- -m 12000
SCTP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.10.1 () port 0 AF_INET
Recv   Send    Send                          Utilization       Service Demand
Socket Socket  Message  Elapsed              Send     Recv     Send    Recv
Size   Size    Size     Time     Throughput  local    remote   local   remote
bytes  bytes   bytes    secs.    10^6bits/s  % S      % S      us/KB   us/KB

212992 212992  12000    10.00      3016.42   2.88     3.78     1.874   2.462

After patch:
Overhead  Command         Shared Object      Symbol
   2,75%  netserver       [kernel.vmlinux]   [k] memcpy_erms
   2,63%  netserver       [kernel.vmlinux]   [k] copy_user_enhanced_fast_string
   2,39%  netserver       [kernel.vmlinux]   [k] fib_table_lookup
   2,04%  netserver       [kernel.vmlinux]   [k] __pskb_pull_tail
   1,91%  netserver       [kernel.vmlinux]   [k] _raw_spin_lock
   1,91%  netserver       [sctp]             [k] sctp_packet_transmit
   1,72%  netserver       [mlx4_en]          [k] mlx4_en_process_rx_cq
   1,68%  netserver       [sctp]             [k] sctp_rcv

# netperf -H 192.168.10.1 -l 10 -t SCTP_STREAM -cC -- -m 12000
SCTP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.10.1 () port 0 AF_INET
Recv   Send    Send                          Utilization       Service Demand
Socket Socket  Message  Elapsed              Send     Recv     Send    Recv
Size   Size    Size     Time     Throughput  local    remote   local   remote
bytes  bytes   bytes    secs.    10^6bits/s  % S      % S      us/KB   us/KB

212992 212992  12000    10.00      3681.77   3.83     3.46     2.045   1.849

Fixes: 3acb50c18d ("sctp: delay as much as possible skb_linearize")
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-19 17:09:42 -07:00
..
associola.c sctp: add SCTP_PR_SUPPORTED on sctp sockopt 2016-07-11 13:25:38 -07:00
auth.c
bind_addr.c sctp: fix copying more bytes than expected in sctp_add_bind_addr 2016-03-08 15:04:08 -05:00
chunk.c sctp: implement prsctp PRIO policy 2016-07-11 13:25:39 -07:00
debug.c
endpointola.c sctp: add SCTP_PR_SUPPORTED on sctp sockopt 2016-07-11 13:25:38 -07:00
input.c sctp: linearize early if it's not GSO 2016-08-19 17:09:42 -07:00
inqueue.c sctp: linearize early if it's not GSO 2016-08-19 17:09:42 -07:00
ipv6.c sctp: use inet_recvmsg to support sctp RFS well 2016-07-25 10:56:28 -07:00
Kconfig sctp: add the sctp_diag.c file 2016-04-15 17:29:36 -04:00
Makefile sctp: Add GSO support 2016-06-03 19:37:21 -04:00
objcnt.c
offload.c sctp: fix GSO for IPv6 2016-07-16 22:02:09 -07:00
output.c sctp: fix the issue sctp requeue auth chunk incorrectly 2016-07-30 22:06:22 -07:00
outqueue.c sctp: implement prsctp PRIO policy 2016-07-11 13:25:39 -07:00
primitive.c
probe.c net: sctp: Convert log timestamps to be y2038 safe 2016-03-01 17:18:44 -05:00
proc.c net/sctp: always initialise sctp_ht_iter::start_fail 2016-08-13 15:10:16 -07:00
protocol.c sctp: use inet_recvmsg to support sctp RFS well 2016-07-25 10:56:28 -07:00
sctp_diag.c sctp_diag: Respect ss adding TCPF_CLOSE to idiag_states 2016-08-08 12:51:58 -07:00
sm_make_chunk.c sctp: avoid identifying address family many times for a chunk 2016-07-13 18:10:14 -07:00
sm_sideeffect.c sctp: sctp should change socket state when shutdown is received 2016-06-10 23:21:23 -07:00
sm_statefuns.c sctp: only check for ECN if peer is using it 2016-07-13 18:10:14 -07:00
sm_statetable.c
socket.c sctp: allow receiving msg when TCP-style sk is in CLOSED state 2016-07-30 22:06:22 -07:00
ssnmap.c
sysctl.c
transport.c sctp: avoid refreshing heartbeat timer too often 2016-04-10 22:22:34 -04:00
tsnmap.c
ulpevent.c sctp: use event->chunk when it's valid 2016-08-08 14:31:23 -07:00
ulpqueue.c sctp: allow delivering notifications after receiving SHUTDOWN 2016-07-30 22:06:22 -07:00