linux_dsm_epyc7002/net/ipv4
Ido Schimmel df6afe2f7c nexthop: Fix performance regression in nexthop deletion
While insertion of 16k nexthops all using the same netdev ('dummy10')
takes less than a second, deletion takes about 130 seconds:

# time -p ip -b nexthop.batch
real 0.29
user 0.01
sys 0.15

# time -p ip link set dev dummy10 down
real 131.03
user 0.06
sys 0.52

This is because of repeated calls to synchronize_rcu() whenever a
nexthop is removed from a nexthop group:

# /usr/share/bcc/tools/offcputime -p `pgrep -nx ip` -K
...
    b'finish_task_switch'
    b'schedule'
    b'schedule_timeout'
    b'wait_for_completion'
    b'__wait_rcu_gp'
    b'synchronize_rcu.part.0'
    b'synchronize_rcu'
    b'__remove_nexthop'
    b'remove_nexthop'
    b'nexthop_flush_dev'
    b'nh_netdev_event'
    b'raw_notifier_call_chain'
    b'call_netdevice_notifiers_info'
    b'__dev_notify_flags'
    b'dev_change_flags'
    b'do_setlink'
    b'__rtnl_newlink'
    b'rtnl_newlink'
    b'rtnetlink_rcv_msg'
    b'netlink_rcv_skb'
    b'rtnetlink_rcv'
    b'netlink_unicast'
    b'netlink_sendmsg'
    b'____sys_sendmsg'
    b'___sys_sendmsg'
    b'__sys_sendmsg'
    b'__x64_sys_sendmsg'
    b'do_syscall_64'
    b'entry_SYSCALL_64_after_hwframe'
    -                ip (277)
        126554955

Since nexthops are always deleted under RTNL, synchronize_net() can be
used instead. It will call synchronize_rcu_expedited() which only blocks
for several microseconds as opposed to multiple milliseconds like
synchronize_rcu().

With this patch deletion of 16k nexthops takes less than a second:

# time -p ip link set dev dummy10 down
real 0.12
user 0.00
sys 0.04

Tested with fib_nexthops.sh which includes torture tests that prompted
the initial change:

# ./fib_nexthops.sh
...
Tests passed: 134
Tests failed:   0

Fixes: 90f33bffa3 ("nexthops: don't modify published nexthop groups")
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Acked-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Link: https://lore.kernel.org/r/20201016172914.643282-1-idosch@idosch.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2020-10-19 20:07:15 -07:00
..
bpfilter net: Revert "net: optimize the sockptr_t for unified kernel/user address spaces" 2020-08-10 12:06:44 -07:00
netfilter netfilter: nf_log: missing vlan offload tag and proto 2020-10-14 01:25:14 +02:00
af_inet.c io_uring: allow tcp ancillary data for __sys_recvmsg_sock() 2020-08-24 16:16:06 -07:00
ah4.c inet: Use fallthrough; 2020-03-12 15:55:00 -07:00
arp.c inet: Use fallthrough; 2020-03-12 15:55:00 -07:00
bpf_tcp_ca.c bpf: Change bpf_sk_storage_*() to accept ARG_PTR_TO_BTF_ID_SOCK_COMMON 2020-09-25 13:58:01 -07:00
cipso_ipv4.c cipso: fix 'audit_secid' kernel-doc warning in cipso_ipv4.c 2020-09-08 20:03:36 -07:00
datagram.c inet: stop leaking jiffies on the wire 2019-11-01 14:57:52 -07:00
devinet.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net 2020-05-31 17:48:46 -07:00
esp4_offload.c net: Add MODULE_DESCRIPTION entries to network modules 2020-06-20 21:33:57 -07:00
esp4.c ESP: Export esp_output_fill_trailer function 2020-02-19 13:52:32 +01:00
fib_frontend.c ipv4: Initialize flowi4_multipath_hash in data path 2020-09-14 14:54:56 -07:00
fib_lookup.h net: add net available in build_state 2020-03-29 22:30:57 -07:00
fib_notifier.c net: fib_notifier: propagate extack down to the notifier block callback 2019-10-04 11:10:56 -07:00
fib_rules.c fib: use indirect call wrappers in the most common fib_rules_ops 2020-07-28 17:42:31 -07:00
fib_semantics.c net: Fix the arp error in some cases 2020-06-18 20:21:51 -07:00
fib_trie.c ipv4: Silence suspicious RCU usage warning 2020-08-26 15:58:48 -07:00
fou.c genetlink: move to smaller ops wherever possible 2020-10-02 19:11:11 -07:00
gre_demux.c gre: fix uninit-value in __iptunnel_pull_header 2020-03-08 21:25:37 -07:00
gre_offload.c net: gre: recompute gre csum for sctp over gre tunnels 2020-08-03 15:29:44 -07:00
icmp.c icmp: randomize the global rate limiter 2020-10-16 16:47:09 -07:00
igmp.c ip*_mc_gsfget(): lift copyout of struct group_filter into callers 2020-05-20 20:31:27 -04:00
inet_connection_sock.c inet: remove icsk_ack.blocked 2020-09-30 14:21:30 -07:00
inet_diag.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net 2020-09-22 16:45:34 -07:00
inet_fragment.c inet: frags: re-introduce skb coalescing for local delivery 2019-08-08 15:55:10 -07:00
inet_hashtables.c net: ipv4: remove unused arg exact_dif in compute_score 2020-08-31 13:08:29 -07:00
inet_timewait_sock.c
inetpeer.c inetpeer: fix data-race in inet_putpeer / inet_putpeer 2019-11-07 16:15:56 -08:00
ip_forward.c ipv4: Revert removal of rt_uses_gateway 2019-09-20 18:23:33 -07:00
ip_fragment.c inet: frags: re-introduce skb coalescing for local delivery 2019-08-08 15:55:10 -07:00
ip_gre.c ip_gre: set dev->hard_header_len and dev->needed_headroom properly 2020-10-13 18:35:29 -07:00
ip_input.c bpf: Add socket assign support 2020-03-30 13:45:04 -07:00
ip_options.c net: clean up codestyle for net/ipv4 2020-08-25 06:28:02 -07:00
ip_output.c networking changes for the 5.10 merge window 2020-10-15 18:42:13 -07:00
ip_sockglue.c net: Remove duplicated midx check against 0 2020-08-25 06:23:59 -07:00
ip_tunnel_core.c iptunnel: use new function dev_fetch_sw_netstats 2020-10-13 17:33:49 -07:00
ip_tunnel.c ipv4: use dev_sw_netstats_rx_add() 2020-10-06 06:23:21 -07:00
ip_vti.c ipv4: use dev_sw_netstats_rx_add() 2020-10-06 06:23:21 -07:00
ipcomp.c ipcomp: assign if_id to child tunnel from parent tunnel 2020-07-09 12:55:37 +02:00
ipconfig.c Documentation: nfsroot.rst: Fix references to nfsroot.rst 2020-03-02 13:11:46 -07:00
ipip.c net: ipip: implement header_ops->parse_protocol for AF_PACKET 2020-06-30 12:29:39 -07:00
ipmr_base.c net: fib_notifier: propagate extack down to the notifier block callback 2019-10-04 11:10:56 -07:00
ipmr.c ipmr: Use full VIF ID in netlink cache reports 2020-09-10 12:25:51 -07:00
Kconfig net: ipv4: remove duplicate "the the" phrase in Kconfig text 2020-08-18 16:02:16 -07:00
Makefile udp_tunnel: add central NIC RX port offload infrastructure 2020-07-10 13:54:00 -07:00
metrics.c
netfilter.c
netlink.c
nexthop.c nexthop: Fix performance regression in nexthop deletion 2020-10-19 20:07:15 -07:00
ping.c net: clean up codestyle 2020-08-31 12:33:34 -07:00
proc.c tcp: skip DSACKs with dubious sequence ranges 2020-09-24 20:15:45 -07:00
protocol.c
raw_diag.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net 2020-03-12 22:34:48 -07:00
raw.c networking changes for the 5.10 merge window 2020-10-15 18:42:13 -07:00
route.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net 2020-10-15 12:43:21 -07:00
syncookies.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net 2020-10-05 18:40:01 -07:00
sysctl_net_ipv4.c tcp: reflect tos value received in SYN to the socket 2020-09-10 13:15:40 -07:00
tcp_bbr.c tcp_bbr: improve arithmetic division in bbr_update_bw() 2020-01-21 10:45:49 +01:00
tcp_bic.c tcp: fix stretch ACK bugs in BIC 2020-03-16 18:26:54 -07:00
tcp_bpf.c net: sk_msg: Simplify sk_psock initialization 2020-08-21 15:16:11 -07:00
tcp_cdg.c
tcp_cong.c tcp: Simplify tcp_set_congestion_control() load=false case 2020-09-10 20:53:01 -07:00
tcp_cubic.c tcp_cubic: fix spurious HYSTART_DELAY exit upon drop in min RTT 2020-06-25 16:08:47 -07:00
tcp_dctcp.c
tcp_dctcp.h
tcp_diag.c inet_diag: Move the INET_DIAG_REQ_BYTECODE nlattr to cb->data 2020-02-27 18:50:19 -08:00
tcp_fastopen.c bpf: tcp: Add bpf_skops_established() 2020-08-24 14:35:00 -07:00
tcp_highspeed.c Replace HTTP links with HTTPS ones: IPv* 2020-07-06 13:23:03 -07:00
tcp_htcp.c Replace HTTP links with HTTPS ones: IPv* 2020-07-06 13:23:03 -07:00
tcp_hybla.c
tcp_illinois.c
tcp_input.c tcp: use semicolons rather than commas to separate statements 2020-10-13 17:11:52 -07:00
tcp_ipv4.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net 2020-10-08 15:44:50 -07:00
tcp_lp.c
tcp_metrics.c genetlink: move to smaller ops wherever possible 2020-10-02 19:11:11 -07:00
tcp_minisocks.c bpf: tcp: Do not limit cb_flags when creating child sk from listen sk 2020-10-02 11:34:48 -07:00
tcp_nv.c
tcp_offload.c
tcp_output.c tcp: add exponential backoff in __tcp_send_ack() 2020-09-30 14:21:30 -07:00
tcp_rate.c
tcp_recovery.c tcp: move tcp_mark_skb_lost 2020-09-25 17:17:14 -07:00
tcp_scalable.c net: ipv4: delete repeated words 2020-08-24 17:31:20 -07:00
tcp_timer.c inet: remove icsk_ack.blocked 2020-09-30 14:21:30 -07:00
tcp_ulp.c bpf: sockmap: Only check ULP for TCP sockets 2020-03-09 22:34:58 +01:00
tcp_vegas.c tcp: use semicolons rather than commas to separate statements 2020-10-13 17:11:52 -07:00
tcp_vegas.h
tcp_veno.c Replace HTTP links with HTTPS ones: IPv* 2020-07-06 13:23:03 -07:00
tcp_westwood.c
tcp_yeah.c tcp: fix stretch ACK bugs in Yeah 2020-03-16 18:26:55 -07:00
tcp.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net 2020-10-05 18:40:01 -07:00
tunnel4.c tunnel4: add cb_handler to struct xfrm_tunnel 2020-07-09 12:51:36 +02:00
udp_bpf.c net: sk_msg: Simplify sk_psock initialization 2020-08-21 15:16:11 -07:00
udp_diag.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net 2020-03-12 22:34:48 -07:00
udp_impl.h net: pass a sockptr_t into ->setsockopt 2020-07-24 15:41:54 -07:00
udp_offload.c udp: initialize is_flist with 0 in udp_gro_receive 2020-03-30 10:35:03 -07:00
udp_tunnel_core.c udp_tunnel: add central NIC RX port offload infrastructure 2020-07-10 13:54:00 -07:00
udp_tunnel_nic.c udp_tunnel: add the ability to share port tables 2020-09-28 12:50:12 -07:00
udp_tunnel_stub.c udp_tunnel: add central NIC RX port offload infrastructure 2020-07-10 13:54:00 -07:00
udp.c net: ipv4: delete repeated words 2020-08-24 17:31:20 -07:00
udplite.c net/ipv4: remove compat_ip_{get,set}sockopt 2020-07-19 18:16:41 -07:00
xfrm4_input.c xfrm: state: remove extract_input indirection from xfrm_state_afinfo 2020-05-06 09:40:08 +02:00
xfrm4_output.c xfrm: fix unused variable warning if CONFIG_NETFILTER=n 2020-05-11 15:12:27 +02:00
xfrm4_policy.c net: add bool confirm_neigh parameter for dst_ops.update_pmtu 2019-12-24 22:28:54 -08:00
xfrm4_protocol.c xfrm: add route lookup to xfrm4_rcv_encap 2019-12-09 09:59:07 +01:00
xfrm4_state.c xfrm: remove output_finish indirection from xfrm_state_afinfo 2020-05-06 09:40:08 +02:00
xfrm4_tunnel.c