[PATCH net-next] ipv6: let trace_fib6_table_lookup() dereference the fib table

[PATCH net-next] ipv6: let trace_fib6_table_lookup() dereference the fib table

[Paolo Abeni]

The perf traces for ipv6 routing code show a relevant cost around
trace_fib6_table_lookup(), even if no trace is enabled. This is
due to the fib6_table de-referencing currently performed by the
caller.

Let's the tracing code pay this overhead, passing to the trace
helper the table pointer. This gives small but measurable
performance improvement under UDP flood.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
---
 include/trace/events/fib6.h |  6 +++---
 net/ipv6/route.c            | 12 ++++++------
 2 files changed, 9 insertions(+), 9 deletions(-)

diff --git a/include/trace/events/fib6.h b/include/trace/events/fib6.h
index d60096cddb2a..b34bed17abc7 100644
--- a/include/trace/events/fib6.h
+++ b/include/trace/events/fib6.h
@@ -12,9 +12,9 @@
 TRACE_EVENT(fib6_table_lookup,
 
 	TP_PROTO(const struct net *net, const struct rt6_info *rt,
-		 u32 tb_id, const struct flowi6 *flp),
+		 struct fib6_table *table, const struct flowi6 *flp),
 
-	TP_ARGS(net, rt, tb_id, flp),
+	TP_ARGS(net, rt, table, flp),
 
 	TP_STRUCT__entry(
 		__field(	u32,	tb_id		)
@@ -34,7 +34,7 @@ TRACE_EVENT(fib6_table_lookup,
 	TP_fast_assign(
 		struct in6_addr *in6;
 
-		__entry->tb_id = tb_id;
+		__entry->tb_id = table->tb6_id;
 		__entry->oif = flp->flowi6_oif;
 		__entry->iif = flp->flowi6_iif;
 		__entry->tos = ip6_tclass(flp->flowlabel);
diff --git a/net/ipv6/route.c b/net/ipv6/route.c
index 2e8842fa6450..d638fd993287 100644
--- a/net/ipv6/route.c
+++ b/net/ipv6/route.c
@@ -945,7 +945,7 @@ static struct rt6_info *ip6_pol_route_lookup(struct net *net,
 
 	rcu_read_unlock();
 
-	trace_fib6_table_lookup(net, rt, table->tb6_id, fl6);
+	trace_fib6_table_lookup(net, rt, table, fl6);
 
 	return rt;
 
@@ -1670,7 +1670,7 @@ struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
 	if (rt == net->ipv6.ip6_null_entry) {
 		rcu_read_unlock();
 		dst_hold(&rt->dst);
-		trace_fib6_table_lookup(net, rt, table->tb6_id, fl6);
+		trace_fib6_table_lookup(net, rt, table, fl6);
 		return rt;
 	} else if (rt->rt6i_flags & RTF_CACHE) {
 		if (ip6_hold_safe(net, &rt, true)) {
@@ -1678,7 +1678,7 @@ struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
 			rt6_dst_from_metrics_check(rt);
 		}
 		rcu_read_unlock();
-		trace_fib6_table_lookup(net, rt, table->tb6_id, fl6);
+		trace_fib6_table_lookup(net, rt, table, fl6);
 		return rt;
 	} else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
 			    !(rt->rt6i_flags & RTF_GATEWAY))) {
@@ -1714,7 +1714,7 @@ struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
 		}
 
 uncached_rt_out:
-		trace_fib6_table_lookup(net, uncached_rt, table->tb6_id, fl6);
+		trace_fib6_table_lookup(net, uncached_rt, table, fl6);
 		return uncached_rt;
 
 	} else {
@@ -1742,7 +1742,7 @@ struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
 		}
 		local_bh_enable();
 		rcu_read_unlock();
-		trace_fib6_table_lookup(net, pcpu_rt, table->tb6_id, fl6);
+		trace_fib6_table_lookup(net, pcpu_rt, table, fl6);
 		return pcpu_rt;
 	}
 }
@@ -2183,7 +2183,7 @@ static struct rt6_info *__ip6_route_redirect(struct net *net,
 
 	rcu_read_unlock();
 
-	trace_fib6_table_lookup(net, rt, table->tb6_id, fl6);
+	trace_fib6_table_lookup(net, rt, table, fl6);
 	return rt;
 };
 
-- 
2.13.6

Re: [PATCH net-next] ipv6: let trace_fib6_table_lookup() dereference the fib table

[Steven Rostedt]

On Thu, 19 Oct 2017 09:31:43 +0200
Paolo Abeni <pabeni@redhat.com> wrote:

> The perf traces for ipv6 routing code show a relevant cost around
> trace_fib6_table_lookup(), even if no trace is enabled. This is
> due to the fib6_table de-referencing currently performed by the
> caller.
> 
> Let's the tracing code pay this overhead, passing to the trace
> helper the table pointer. This gives small but measurable
> performance improvement under UDP flood.
> 
> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
> ---

Acked-by: Steven Rostedt (VMware) <rostedt@goodmis.org>

-- Steve

Re: [PATCH net-next] ipv6: let trace_fib6_table_lookup() dereference the fib table

[David Ahern]

On 10/19/17 1:31 AM, Paolo Abeni wrote:
> The perf traces for ipv6 routing code show a relevant cost around
> trace_fib6_table_lookup(), even if no trace is enabled. This is
> due to the fib6_table de-referencing currently performed by the
> caller.
> 
> Let's the tracing code pay this overhead, passing to the trace
> helper the table pointer. This gives small but measurable
> performance improvement under UDP flood.
> 
> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
> ---
>  include/trace/events/fib6.h |  6 +++---
>  net/ipv6/route.c            | 12 ++++++------
>  2 files changed, 9 insertions(+), 9 deletions(-)

Makes sense.

Acked-by: David Ahern <dsa@cumulusnetworks.com>

Re: [PATCH net-next] ipv6: let trace_fib6_table_lookup() dereference the fib table

[Martin KaFai Lau]

On Thu, Oct 19, 2017 at 07:31:43AM +0000, Paolo Abeni wrote:
> The perf traces for ipv6 routing code show a relevant cost around
> trace_fib6_table_lookup(), even if no trace is enabled. This is
> due to the fib6_table de-referencing currently performed by the
> caller.
> 
> Let's the tracing code pay this overhead, passing to the trace
> helper the table pointer. This gives small but measurable
> performance improvement under UDP flood.
> 
> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>

Re: [PATCH net-next] ipv6: let trace_fib6_table_lookup() dereference the fib table

[David Miller]

From: Paolo Abeni <pabeni@redhat.com>
Date: Thu, 19 Oct 2017 09:31:43 +0200

> The perf traces for ipv6 routing code show a relevant cost around
> trace_fib6_table_lookup(), even if no trace is enabled. This is
> due to the fib6_table de-referencing currently performed by the
> caller.
> 
> Let's the tracing code pay this overhead, passing to the trace
> helper the table pointer. This gives small but measurable
> performance improvement under UDP flood.
> 
> Signed-off-by: Paolo Abeni <pabeni@redhat.com>

Looks good, applied.