From patchwork Thu Sep 24 20:44:53 2009 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Eric Dumazet X-Patchwork-Id: 34235 X-Patchwork-Delegate: davem@davemloft.net Return-Path: X-Original-To: patchwork-incoming@ozlabs.org Delivered-To: patchwork-incoming@ozlabs.org Received: from vger.kernel.org (vger.kernel.org [209.132.176.167]) by ozlabs.org (Postfix) with ESMTP id EEA51B7B7C for ; Fri, 25 Sep 2009 06:45:04 +1000 (EST) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753405AbZIXUox (ORCPT ); Thu, 24 Sep 2009 16:44:53 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1752535AbZIXUow (ORCPT ); Thu, 24 Sep 2009 16:44:52 -0400 Received: from gw1.cosmosbay.com ([212.99.114.194]:43446 "EHLO gw1.cosmosbay.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753181AbZIXUow (ORCPT ); Thu, 24 Sep 2009 16:44:52 -0400 Received: from [127.0.0.1] (localhost [127.0.0.1]) by gw1.cosmosbay.com (8.13.7/8.13.7) with ESMTP id n8OKirIq011509; Thu, 24 Sep 2009 22:44:53 +0200 Message-ID: <4ABBDA45.2040109@gmail.com> Date: Thu, 24 Sep 2009 22:44:53 +0200 From: Eric Dumazet User-Agent: Thunderbird 2.0.0.23 (Windows/20090812) MIME-Version: 1.0 To: Stephen Hemminger CC: "David S. Miller" , Linux Netdev List Subject: Re: [PATCH] inet_peer: Optimize inet_getid() References: <4ABBC2D8.2040901@gmail.com> <20090924123020.59a2ee6e@s6510> In-Reply-To: <20090924123020.59a2ee6e@s6510> X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-1.6 (gw1.cosmosbay.com [0.0.0.0]); Thu, 24 Sep 2009 22:44:53 +0200 (CEST) Sender: netdev-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: netdev@vger.kernel.org Stephen Hemminger a écrit : > On Thu, 24 Sep 2009 21:04:56 +0200 > Eric Dumazet wrote: > >> While investigating for network latencies, I found inet_getid() was a contention point >> for some workloads. >> >> Fix is straightforward, using cmpxchg() instead of >> a spin_lock_bh()/spin_unlock_bh() pair on a central lock. >> >> Another possibility was to use an atomic_t and atomic_add_return() but >> the size of struct inet_peer object would had doubled on x86_64 because of >> SLAB_HWCACHE_ALIGN constraint. >> >> Signed-off-by: Eric Dumazet > > I thought cmpxchg was not available on all architectures. Here is the second version of the patch, thanks Stephen [PATCH] inet_peer: Optimize inet_getid() While investigating for network latencies, I found inet_getid() was a contention point for some workloads. If __HAVE_ARCH_CMPXCHG is defined, we can use cmpxchg() instead of a spin_lock_bh()/spin_unlock_bh() pair on a central lock. On other arches, we can use an atomic_t instead of a u16, and atomic_add_return(). This might grow memory usage a bit, unless someone invents atomic16_t :) Signed-off-by: Eric Dumazet --- include/net/inetpeer.h | 43 ++++++++++++++++++++++++++++++++------- net/ipv4/inetpeer.c | 5 ---- net/ipv4/route.c | 2 - 3 files changed, 38 insertions(+), 12 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html diff --git a/include/net/inetpeer.h b/include/net/inetpeer.h index 15e1f8f..f1cde51 100644 --- a/include/net/inetpeer.h +++ b/include/net/inetpeer.h @@ -19,7 +19,11 @@ struct inet_peer struct inet_peer *avl_left, *avl_right; __be32 v4daddr; /* peer's address */ __u16 avl_height; +#if defined(__HAVE_ARCH_CMPXCHG) __u16 ip_id_count; /* IP ID for the next packet */ +#else + atomic_t ip_id_count; +#endif struct list_head unused; __u32 dtime; /* the time of last use of not * referenced entries */ @@ -37,17 +41,42 @@ struct inet_peer *inet_getpeer(__be32 daddr, int create); /* can be called from BH context or outside */ extern void inet_putpeer(struct inet_peer *p); -extern spinlock_t inet_peer_idlock; /* can be called with or without local BH being disabled */ static inline __u16 inet_getid(struct inet_peer *p, int more) { - __u16 id; + __u16 old; - spin_lock_bh(&inet_peer_idlock); - id = p->ip_id_count; - p->ip_id_count += 1 + more; - spin_unlock_bh(&inet_peer_idlock); - return id; + more++; +#if defined(__HAVE_ARCH_CMPXCHG) + while (1) { + old = p->ip_id_count; + if (cmpxchg(&p->ip_id_count, old, old + more) == old) + break; + } +#else + old = atomic_add_return(more, &p->ip_id_count) - more; +#endif + return old; } +static inline void inet_id_set(struct inet_peer *p, int val) +{ +#if defined(__HAVE_ARCH_CMPXCHG) + p->ip_id_count = val; +#else + atomic_set(&p->ip_id_count, val); +#endif +} + +static inline __u16 inet_id_read(const struct inet_peer *p) +{ +#if defined(__HAVE_ARCH_CMPXCHG) + return p->ip_id_count; +#else + return atomic_read(&p->ip_id_count); +#endif +} + + + #endif /* _NET_INETPEER_H */ diff --git a/net/ipv4/inetpeer.c b/net/ipv4/inetpeer.c index b1fbe18..3641831 100644 --- a/net/ipv4/inetpeer.c +++ b/net/ipv4/inetpeer.c @@ -67,9 +67,6 @@ * ip_id_count: idlock */ -/* Exported for inet_getid inline function. */ -DEFINE_SPINLOCK(inet_peer_idlock); - static struct kmem_cache *peer_cachep __read_mostly; #define node_height(x) x->avl_height @@ -390,7 +387,7 @@ struct inet_peer *inet_getpeer(__be32 daddr, int create) n->v4daddr = daddr; atomic_set(&n->refcnt, 1); atomic_set(&n->rid, 0); - n->ip_id_count = secure_ip_id(daddr); + inet_id_set(n, secure_ip_id(daddr)); n->tcp_ts_stamp = 0; write_lock_bh(&peer_pool_lock); diff --git a/net/ipv4/route.c b/net/ipv4/route.c index df93473..a090fcf 100644 --- a/net/ipv4/route.c +++ b/net/ipv4/route.c @@ -2855,7 +2855,7 @@ static int rt_fill_info(struct net *net, error = rt->u.dst.error; expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0; if (rt->peer) { - id = rt->peer->ip_id_count; + id = inet_id_read(rt->peer); if (rt->peer->tcp_ts_stamp) { ts = rt->peer->tcp_ts; tsage = get_seconds() - rt->peer->tcp_ts_stamp;