On Fri, 2014-03-21 at 06:32 -0700, Eric Dumazet wrote: > On Fri, 2014-03-21 at 05:50 -0700, Eric Dumazet wrote: > > > Or the fact that we mix 16 bit stores and 32bit loads ? > > > > iph->tot_len = newlen; > > iph->check = 0; > > iph->check = ip_fast_csum(iph, 5); > > Yep definitely. Using 16 bit loads in ip_fast_csum() totally removes the > stall. I no longer see inet_gro_complete() in perf top... > > + if (__builtin_constant_p(ihl) && ihl == 5) { > + asm(" movw (%[iph]), %w[sum]\n" /* ihl/version/tos */ > + " addw 2(%[iph]), %w[sum]\n" /* tot_len */ > + " adcw 8(%[iph]), %w[sum]\n" /* ttl/protocol */ > + " adcw 10(%[iph]), %w[sum]\n" /* check */ > + " adcl 4(%[iph]), %[sum]\n" /* id/frag_off */ > + " adcl 12(%[iph]), %[sum]\n" /* saddr */ > + " adcl 16(%[iph]), %[sum]\n" /* daddr */ > + " adcl $0, %[sum]\n" > + : [sum] "=r" (sum) > + : [iph] "r" (iph) > + ); > + return csum_fold(sum); >
Another idea would be to move the ip_fast_csum() call at the end of inet_gro_complete() I'll try this : diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c index 8c54870db792..0ca8f350a532 100644 --- a/net/ipv4/af_inet.c +++ b/net/ipv4/af_inet.c @@ -1434,8 +1434,8 @@ static int inet_gro_complete(struct sk_buff *skb, int nhoff) int proto = iph->protocol; int err = -ENOSYS; - csum_replace2(&iph->check, iph->tot_len, newlen); iph->tot_len = newlen; + iph->check = 0; rcu_read_lock(); ops = rcu_dereference(inet_offloads[proto]); @@ -1447,6 +1447,7 @@ static int inet_gro_complete(struct sk_buff *skb, int nhoff) * inet_gro_receive(). */ err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph)); + iph->check = ip_fast_csum((u8 *)iph, 5); out_unlock: rcu_read_unlock(); -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/