From: Soheil Hassas Yeganeh <soh...@google.com>
We should only record RPS on normal reads and writes.
In single threaded processes, all calls record the same state. In
multi-threaded processes where a separate thread processes
errors, the RFS table mispredicts.
Note that, when CONFIG_RPS is disabled, sock_rps_record_flow
is a noop and no branch is added as a result of this patch.
Signed-off-by: Soheil Hassas Yeganeh <soh...@google.com>
Signed-off-by: Willem de Bruijn <will...@google.com>
Signed-off-by: Eric Dumazet <eduma...@google.com>
Signed-off-by: Neal Cardwell <ncardw...@google.com>
---
net/ipv4/af_inet.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
index bab98a4fedad..54cccdd8b1e3 100644
--- a/net/ipv4/af_inet.c
+++ b/net/ipv4/af_inet.c
@@ -790,7 +790,8 @@ int inet_recvmsg(struct socket *sock, struct msghdr *msg,
size_t size,
int addr_len = 0;
int err;
- sock_rps_record_flow(sk);
+ if (likely(!(flags & MSG_ERRQUEUE)))
+ sock_rps_record_flow(sk);
err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
flags & ~MSG_DONTWAIT, &addr_len);
--
2.16.0.rc0.223.g4a4ac83678-goog
