From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Fri, 3 Jan 2003 05:15:33 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Fri, 3 Jan 2003 05:15:33 -0500 Received: from e1.ny.us.ibm.com ([32.97.182.101]:45799 "EHLO e1.ny.us.ibm.com") by vger.kernel.org with ESMTP id ; Fri, 3 Jan 2003 05:15:30 -0500 Date: Fri, 3 Jan 2003 15:56:00 +0530 From: Dipankar Sarma To: Dave Miller Cc: linux-kernel@vger.kernel.org Subject: [PATCH] barriers in lockfree rtcache Message-ID: <20030103102600.GC8582@in.ibm.com> Reply-To: dipankar@in.ibm.com Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.4i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Dave, This patch fixes several things in ipv4 route cache that I missed in the last patch - 1. All the memory barriers are SMP-only avoiding unnecessary overhead on UP. 2. My forward porting merge of the rt_rcu patch dropped two changes in rt_intern_hash() in around 2.5.43 that ordered the writes while inserting a dst entry at the start of a hash chain. The dst entry updates must be visible to other (weakly ordered) CPUs before it is inserted. The necessary smp_wmb()s are added. 3. Comments to go with the write ordering. Please apply. Thanks Dipankar diff -urN linux-2.5.54-base/net/ipv4/route.c linux-2.5.54-rt_barriers/net/ipv4/route.c --- linux-2.5.54-base/net/ipv4/route.c 2003-01-02 08:53:15.000000000 +0530 +++ linux-2.5.54-rt_barriers/net/ipv4/route.c 2003-01-03 14:44:31.000000000 +0530 @@ -234,7 +234,7 @@ { struct rt_cache_iter_state *st = seq->private; - read_barrier_depends(); + smp_read_barrier_depends(); r = r->u.rt_next; while (!r) { rcu_read_unlock(); @@ -718,7 +718,18 @@ if (compare_keys(&rth->fl, &rt->fl)) { /* Put it first */ *rthp = rth->u.rt_next; + /* + * Since lookup is lockfree, the deletion + * must be visible to another weakly ordered CPU before + * the insertion at the start of the hash chain. + */ + smp_wmb(); rth->u.rt_next = rt_hash_table[hash].chain; + /* + * Since lookup is lockfree, the update writes + * must be ordered for consistency on SMP. + */ + smp_wmb(); rt_hash_table[hash].chain = rth; rth->u.dst.__use++; @@ -900,7 +911,7 @@ while ((rth = *rthp) != NULL) { struct rtable *rt; - read_barrier_depends(); + smp_read_barrier_depends(); if (rth->fl.fl4_dst != daddr || rth->fl.fl4_src != skeys[i] || rth->fl.fl4_tos != tos || @@ -1148,7 +1159,7 @@ rcu_read_lock(); for (rth = rt_hash_table[hash].chain; rth; rth = rth->u.rt_next) { - read_barrier_depends(); + smp_read_barrier_depends(); if (rth->fl.fl4_dst == daddr && rth->fl.fl4_src == skeys[i] && rth->rt_dst == daddr && @@ -1740,7 +1751,7 @@ rcu_read_lock(); for (rth = rt_hash_table[hash].chain; rth; rth = rth->u.rt_next) { - read_barrier_depends(); + smp_read_barrier_depends(); if (rth->fl.fl4_dst == daddr && rth->fl.fl4_src == saddr && rth->fl.iif == iif && @@ -2105,7 +2116,7 @@ rcu_read_lock(); for (rth = rt_hash_table[hash].chain; rth; rth = rth->u.rt_next) { - read_barrier_depends(); + smp_read_barrier_depends(); if (rth->fl.fl4_dst == flp->fl4_dst && rth->fl.fl4_src == flp->fl4_src && rth->fl.iif == 0 && @@ -2335,7 +2346,7 @@ rcu_read_lock(); for (rt = rt_hash_table[h].chain, idx = 0; rt; rt = rt->u.rt_next, idx++) { - read_barrier_depends(); + smp_read_barrier_depends(); if (idx < s_idx) continue; skb->dst = dst_clone(&rt->u.dst);