Files
linux/include/net
Eric Dumazet 5b8e2f61b9 net: sock_rps_record_flow() is for connected sockets
Paolo noticed a cache line miss in UDP recvmsg() to access
sk_rxhash, sharing a cache line with sk_drops.

sk_drops might be heavily incremented by cpus handling a flood targeting
this socket.

We might place sk_drops on a separate cache line, but lets try
to avoid wasting 64 bytes per socket just for this, since we have
other bottlenecks to take care of.

sock_rps_record_flow() should only access sk_rxhash for connected
flows.

Testing sk_state for TCP_ESTABLISHED covers most of the cases for
connected sockets, for a zero cost, since system calls using
sock_rps_record_flow() also access sk->sk_prot which is on the
same cache line.

A follow up patch will provide a static_key (Jump Label) since most
hosts do not even use RFS.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-12-07 10:47:35 -05:00
..
2014-09-18 10:54:36 +02:00
2015-03-04 00:23:23 -05:00
2015-06-28 16:55:44 -07:00
2016-06-27 15:06:17 -04:00
2016-07-08 12:20:57 +02:00
2016-04-25 16:44:27 -04:00
2016-04-25 16:44:27 -04:00
2016-06-09 23:41:03 -07:00
2015-03-06 21:50:02 -05:00
2015-09-17 17:18:37 -07:00
2016-09-23 08:38:50 -04:00
2016-02-16 20:21:48 -05:00
2013-11-07 19:28:58 -05:00
2016-05-20 18:03:16 -04:00
2016-04-27 22:48:25 -04:00
2016-05-20 18:03:17 -04:00
2016-11-03 15:41:11 -04:00
2016-08-17 19:36:23 -04:00
2014-01-03 20:56:48 -05:00
2015-10-23 06:26:42 -07:00
2016-10-03 02:00:22 -04:00
2016-10-04 02:11:51 -04:00
2016-12-03 15:54:02 -05:00
2016-07-08 12:20:57 +02:00
2016-03-23 22:09:58 -04:00
2016-04-07 16:53:30 -04:00
2015-03-12 22:58:12 -04:00
2016-05-03 16:08:14 -04:00
2015-10-26 22:24:22 -07:00
2016-11-15 11:59:38 -05:00