linux/net/ipv4
Eric Dumazet 0b7d7f6b22 tcp: add tcp_tx_skb_cache sysctl
Feng Tang reported a performance regression after introduction
of per TCP socket tx/rx caches, for TCP over loopback (netperf)

There is high chance the regression is caused by a change on
how well the 32 KB per-thread page (current->task_frag) can
be recycled, and lack of pcp caches for order-3 pages.

I could not reproduce the regression myself, cpus all being
spinning on the mm spinlocks for page allocs/freeing, regardless
of enabling or disabling the per tcp socket caches.

It seems best to disable the feature by default, and let
admins enabling it.

MM layer either needs to provide scalable order-3 pages
allocations, or could attempt a trylock on zone->lock if
the caller only attempts to get a high-order page and is
able to fallback to order-0 ones in case of pressure.

Tests run on a 56 cores host (112 hyper threads)

-	35.49%	netperf 		 [kernel.vmlinux]	  [k] queued_spin_lock_slowpath
   - 35.49% queued_spin_lock_slowpath
	  - 18.18% get_page_from_freelist
		 - __alloc_pages_nodemask
			- 18.18% alloc_pages_current
				 skb_page_frag_refill
				 sk_page_frag_refill
				 tcp_sendmsg_locked
				 tcp_sendmsg
				 inet_sendmsg
				 sock_sendmsg
				 __sys_sendto
				 __x64_sys_sendto
				 do_syscall_64
				 entry_SYSCALL_64_after_hwframe
				 __libc_send
	  + 17.31% __free_pages_ok
+	31.43%	swapper 		 [kernel.vmlinux]	  [k] intel_idle
+	 9.12%	netperf 		 [kernel.vmlinux]	  [k] copy_user_enhanced_fast_string
+	 6.53%	netserver		 [kernel.vmlinux]	  [k] copy_user_enhanced_fast_string
+	 0.69%	netserver		 [kernel.vmlinux]	  [k] queued_spin_lock_slowpath
+	 0.68%	netperf 		 [kernel.vmlinux]	  [k] skb_release_data
+	 0.52%	netperf 		 [kernel.vmlinux]	  [k] tcp_sendmsg_locked
	 0.46%	netperf 		 [kernel.vmlinux]	  [k] _raw_spin_lock_irqsave

Fixes: 472c2e07ee ("tcp: add one skb cache for tx")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Feng Tang <feng.tang@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2019-06-14 20:18:28 -07:00
..
bpfilter SPDX update for 5.2-rc2, round 1 2019-05-21 12:33:38 -07:00
netfilter Merge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf 2019-05-23 14:45:36 -07:00
af_inet.c SPDX update for 5.2-rc3, round 1 2019-05-31 08:34:32 -07:00
ah4.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
arp.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
cipso_ipv4.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 13 2019-05-21 11:28:45 +02:00
datagram.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
devinet.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
esp4_offload.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net 2019-05-02 22:14:21 -04:00
esp4.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
fib_frontend.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
fib_lookup.h ipv4: Add fib_nh_common to fib_result 2019-04-03 21:50:20 -07:00
fib_notifier.c License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
fib_rules.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
fib_semantics.c net: ipv4: fib_semantics: fix uninitialized variable 2019-06-09 12:47:30 -07:00
fib_trie.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
fou.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
gre_demux.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
gre_offload.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
icmp.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
igmp.c SPDX update for 5.2-rc3, round 1 2019-05-31 08:34:32 -07:00
inet_connection_sock.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
inet_diag.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
inet_fragment.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
inet_hashtables.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
inet_timewait_sock.c treewide: Add SPDX license identifier for missed files 2019-05-21 10:50:45 +02:00
inetpeer.c net: ipv4: use a dedicated counter for icmp_v4 redirect packets 2019-02-08 21:50:15 -08:00
ip_forward.c ipv4: Prepare rtable for IPv6 gateway 2019-04-08 15:22:40 -07:00
ip_fragment.c net: remove unused struct inet_frag_queue.fragments field 2019-02-26 08:27:05 -08:00
ip_gre.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
ip_input.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
ip_options.c vrf: check accept_source_route on the original netdevice 2019-04-01 10:44:58 -07:00
ip_output.c net: correct udp zerocopy refcnt also when zerocopy only on append 2019-06-11 11:40:54 -07:00
ip_sockglue.c ip_sockglue: Fix missing-check bug in ip_ra_control() 2019-05-25 11:00:50 -07:00
ip_tunnel_core.c netlink: make validation more configurable for future strictness 2019-04-27 17:07:21 -04:00
ip_tunnel.c iptunnel: NULL pointer deref for ip_md_tunnel_xmit 2019-03-06 10:43:06 -08:00
ip_vti.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
ipcomp.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
ipconfig.c ipconfig: add carrier_timeout kernel parameter 2019-02-01 15:24:13 -08:00
ipip.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
ipmr_base.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net 2019-05-07 17:22:09 -07:00
ipmr.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
Kconfig treewide: Add SPDX license identifier - Makefile/Kconfig 2019-05-21 10:50:46 +02:00
Makefile xfrm: make xfrm modes builtin 2019-04-08 09:15:17 +02:00
metrics.c treewide: Add SPDX license identifier for missed files 2019-05-21 10:50:45 +02:00
netfilter.c netfilter: ipv4: remove useless export_symbol 2019-01-28 11:32:58 +01:00
netlink.c treewide: Add SPDX license identifier for missed files 2019-05-21 10:50:45 +02:00
ping.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
proc.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
protocol.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
raw_diag.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
raw.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
route.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net 2019-06-07 09:29:14 -07:00
syncookies.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
sysctl_net_ipv4.c tcp: add tcp_tx_skb_cache sysctl 2019-06-14 20:18:28 -07:00
tcp_bbr.c tcp_bbr: adapt cwnd based on ack aggregation estimation 2019-01-24 22:27:27 -08:00
tcp_bic.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp_bpf.c bpf, tcp: correctly handle DONT_WAIT flags and timeo == 0 2019-05-16 01:36:13 +02:00
tcp_cdg.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp_cong.c treewide: Add SPDX license identifier for missed files 2019-05-21 10:50:45 +02:00
tcp_cubic.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp_dctcp.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
tcp_dctcp.h tcp: refactor DCTCP ECN ACK handling 2018-10-10 22:26:00 -07:00
tcp_diag.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
tcp_fastopen.c tcp: pause Fast Open globally after third consecutive timeout 2017-12-13 15:51:12 -05:00
tcp_highspeed.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp_htcp.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp_hybla.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp_illinois.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp_input.c tcp: fix undo spurious SYNACK in passive Fast Open 2019-06-09 20:04:11 -07:00
tcp_ipv4.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
tcp_lp.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp_metrics.c tcp: refactor setting the initial congestion window 2019-05-01 11:47:54 -04:00
tcp_minisocks.c treewide: Add SPDX license identifier for missed files 2019-05-21 10:50:45 +02:00
tcp_nv.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp_offload.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
tcp_output.c treewide: Add SPDX license identifier for missed files 2019-05-21 10:50:45 +02:00
tcp_rate.c treewide: Add SPDX license identifier for missed files 2019-05-21 10:50:45 +02:00
tcp_recovery.c tcp: introduce tcp_skb_timestamp_us() helper 2018-09-21 19:37:59 -07:00
tcp_scalable.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp_timer.c treewide: Add SPDX license identifier for missed files 2019-05-21 10:50:45 +02:00
tcp_ulp.c treewide: Add SPDX license identifier for missed files 2019-05-21 10:50:45 +02:00
tcp_vegas.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp_vegas.h License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
tcp_veno.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp_westwood.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp_yeah.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
tcp.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
tunnel4.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
udp_diag.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
udp_impl.h udp: add missing rehash callback to udplite 2019-01-17 15:01:08 -08:00
udp_offload.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
udp_tunnel.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00
udp.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf 2019-06-07 14:46:47 -07:00
udplite.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
xfrm4_input.c xfrm: reset transport header back to network header after all input transforms ahave been applied 2018-09-04 10:26:30 +02:00
xfrm4_output.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
xfrm4_policy.c Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec-next 2019-04-30 09:26:13 -04:00
xfrm4_protocol.c treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 152 2019-05-30 11:26:32 -07:00
xfrm4_state.c License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
xfrm4_tunnel.c treewide: Add SPDX license identifier for more missed files 2019-05-21 10:50:45 +02:00