From mboxrd@z Thu Jan 1 00:00:00 1970 From: Andy Grover Subject: [PATCH 10/13] RDS: only put sockets that have seen congestion on the poll_waitq Date: Thu, 11 Mar 2010 15:50:04 -0800 Message-ID: <1268351407-7394-11-git-send-email-andy.grover@oracle.com> References: <1268351407-7394-1-git-send-email-andy.grover@oracle.com> Cc: rds-devel@oss.oracle.com To: netdev@vger.kernel.org Return-path: Received: from rcsinet11.oracle.com ([148.87.113.123]:45090 "EHLO rcsinet11.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754439Ab0CKXu6 (ORCPT ); Thu, 11 Mar 2010 18:50:58 -0500 Received: from acsinet15.oracle.com (acsinet15.oracle.com [141.146.126.227]) by rcsinet11.oracle.com (Switch-3.4.2/Switch-3.4.2) with ESMTP id o2BNouIp023385 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=OK) for ; Thu, 11 Mar 2010 23:50:57 GMT In-Reply-To: <1268351407-7394-1-git-send-email-andy.grover@oracle.com> Sender: netdev-owner@vger.kernel.org List-ID: rds_poll_waitq's listeners will be awoken if we receive a congestion notification. Bad performance may result because *all* polled sockets contend for this single lock. However, it should not be necessary to wake pollers when a congestion update arrives if they have never experienced congestion, and not putting these on the waitq will hopefully greatly reduce contention. Signed-off-by: Andy Grover --- net/rds/af_rds.c | 7 ++++++- net/rds/rds.h | 2 ++ net/rds/send.c | 4 +++- 3 files changed, 11 insertions(+), 2 deletions(-) diff --git a/net/rds/af_rds.c b/net/rds/af_rds.c index 853c52b..937ecda 100644 --- a/net/rds/af_rds.c +++ b/net/rds/af_rds.c @@ -159,7 +159,8 @@ static unsigned int rds_poll(struct file *file, struct socket *sock, poll_wait(file, sk->sk_sleep, wait); - poll_wait(file, &rds_poll_waitq, wait); + if (rs->rs_seen_congestion) + poll_wait(file, &rds_poll_waitq, wait); read_lock_irqsave(&rs->rs_recv_lock, flags); if (!rs->rs_cong_monitor) { @@ -181,6 +182,10 @@ static unsigned int rds_poll(struct file *file, struct socket *sock, mask |= (POLLOUT | POLLWRNORM); read_unlock_irqrestore(&rs->rs_recv_lock, flags); + /* clear state any time we wake a seen-congested socket */ + if (mask) + rs->rs_seen_congestion = 0; + return mask; } diff --git a/net/rds/rds.h b/net/rds/rds.h index 85d6f89..4bec6e2 100644 --- a/net/rds/rds.h +++ b/net/rds/rds.h @@ -388,6 +388,8 @@ struct rds_sock { /* flag indicating we were congested or not */ int rs_congested; + /* seen congestion (ENOBUFS) when sending? */ + int rs_seen_congestion; /* rs_lock protects all these adjacent members before the newline */ spinlock_t rs_lock; diff --git a/net/rds/send.c b/net/rds/send.c index 192a480..51e2def 100644 --- a/net/rds/send.c +++ b/net/rds/send.c @@ -894,8 +894,10 @@ int rds_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg, queue_delayed_work(rds_wq, &conn->c_conn_w, 0); ret = rds_cong_wait(conn->c_fcong, dport, nonblock, rs); - if (ret) + if (ret) { + rs->rs_seen_congestion = 1; goto out; + } while (!rds_send_queue_rm(rs, conn, rm, rs->rs_bound_port, dport, &queued)) { -- 1.6.3.3