Patchwork ipv4: percpu nh_rth_output cache

login
register
mail settings
Submitter Eric Dumazet
Date July 31, 2012, 3:22 p.m.
Message ID <1343748166.21269.304.camel@edumazet-glaptop>
Download mbox | patch
Permalink /patch/174267/
State Superseded
Delegated to: David Miller
Headers show

Comments

Eric Dumazet - July 31, 2012, 3:22 p.m.
From: Eric Dumazet <edumazet@google.com>

Input path is mostly run under RCU and doesnt touch dst refcnt

But output path on forwarding or UDP workloads hits
badly dst refcount, and we have lot of false sharing, for example
in ipv4_mtu() when reading rt->rt_pmtu

Using a percpu cache for nh_rth_output gives a nice performance
increase at a small cost.

24 udpflood test on my 24 cpu machine (dummy0 output device)
(each process sends 1.000.000 udp frames, 24 processes are started)

before : 5.24 s
after : 2.06 s
For reference, time on linux-3.5 : 6.60 s

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
On top on previous "ipv4: Restore old dst_free() behavior" patch

We probably can remove all paddings in struct dst_entry

 include/net/ip_fib.h     |    3 ++-
 net/ipv4/fib_semantics.c |   19 ++++++++++++++++++-
 net/ipv4/route.c         |   20 +++++++++++++++-----
 3 files changed, 35 insertions(+), 7 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
Eric Dumazet - July 31, 2012, 3:29 p.m.
On Tue, 2012-07-31 at 17:22 +0200, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 

> +		p = per_cpu_ptr(nh->nh_pcpu_rth_output,
> +				raw_smp_processor_id());
> +	}

Note : this can use :

	p = __this_cpu_ptr(nh->nh_pcpu_rth_output);

and same thing in __mkroute_output()

prth = __this_cpu_ptr(FIB_RES_NH(*res).nh_pcpu_rth_output);


--
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

Patch

diff --git a/include/net/ip_fib.h b/include/net/ip_fib.h
index e521a03..a0f003d 100644
--- a/include/net/ip_fib.h
+++ b/include/net/ip_fib.h
@@ -21,6 +21,7 @@ 
 #include <linux/rcupdate.h>
 #include <net/fib_rules.h>
 #include <net/inetpeer.h>
+#include <linux/percpu.h>
 
 struct fib_config {
 	u8			fc_dst_len;
@@ -81,7 +82,7 @@  struct fib_nh {
 	__be32			nh_gw;
 	__be32			nh_saddr;
 	int			nh_saddr_genid;
-	struct rtable __rcu	*nh_rth_output;
+	struct rtable * __percpu *nh_pcpu_rth_output;
 	struct rtable __rcu	*nh_rth_input;
 	struct fnhe_hash_bucket	*nh_exceptions;
 };
diff --git a/net/ipv4/fib_semantics.c b/net/ipv4/fib_semantics.c
index 625cf18..c8bf86e 100644
--- a/net/ipv4/fib_semantics.c
+++ b/net/ipv4/fib_semantics.c
@@ -176,6 +176,22 @@  static void rt_nexthop_free(struct rtable __rcu **rtp)
 	dst_free(&rt->dst);
 }
 
+static void rt_nexthop_free_cpus(struct rtable * __percpu *rtp)
+{
+	int cpu;
+
+	if (!rtp)
+		return;
+
+	for_each_possible_cpu(cpu) {
+		struct rtable *rt = *per_cpu_ptr(rtp, cpu);
+
+		if (rt)
+			dst_free(&rt->dst);
+	}
+	free_percpu(rtp);
+}
+
 /* Release a nexthop info record */
 static void free_fib_info_rcu(struct rcu_head *head)
 {
@@ -186,7 +202,7 @@  static void free_fib_info_rcu(struct rcu_head *head)
 			dev_put(nexthop_nh->nh_dev);
 		if (nexthop_nh->nh_exceptions)
 			free_nh_exceptions(nexthop_nh);
-		rt_nexthop_free(&nexthop_nh->nh_rth_output);
+		rt_nexthop_free_cpus(nexthop_nh->nh_pcpu_rth_output);
 		rt_nexthop_free(&nexthop_nh->nh_rth_input);
 	} endfor_nexthops(fi);
 
@@ -817,6 +833,7 @@  struct fib_info *fib_create_info(struct fib_config *cfg)
 	fi->fib_nhs = nhs;
 	change_nexthops(fi) {
 		nexthop_nh->nh_parent = fi;
+		nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable *);
 	} endfor_nexthops(fi)
 
 	if (cfg->fc_mx) {
diff --git a/net/ipv4/route.c b/net/ipv4/route.c
index 2bd1074..d5efcfe 100644
--- a/net/ipv4/route.c
+++ b/net/ipv4/route.c
@@ -1206,11 +1206,16 @@  static inline void rt_free(struct rtable *rt)
 
 static void rt_cache_route(struct fib_nh *nh, struct rtable *rt)
 {
-	struct rtable *orig, *prev, **p = (struct rtable **)&nh->nh_rth_output;
+	struct rtable *orig, *prev, **p;
 
-	if (rt_is_input_route(rt))
+	if (rt_is_input_route(rt)) {
 		p = (struct rtable **)&nh->nh_rth_input;
-
+	} else {
+		if (!nh->nh_pcpu_rth_output)
+			goto nocache;
+		p = per_cpu_ptr(nh->nh_pcpu_rth_output,
+				raw_smp_processor_id());
+	}
 	orig = *p;
 
 	prev = cmpxchg(p, orig, rt);
@@ -1223,6 +1228,7 @@  static void rt_cache_route(struct fib_nh *nh, struct rtable *rt)
 		 * unsuccessful at storing this route into the cache
 		 * we really need to set it.
 		 */
+nocache:
 		rt->dst.flags |= DST_NOCACHE;
 	}
 }
@@ -1749,8 +1755,12 @@  static struct rtable *__mkroute_output(const struct fib_result *res,
 	fnhe = NULL;
 	if (fi) {
 		fnhe = find_exception(&FIB_RES_NH(*res), fl4->daddr);
-		if (!fnhe) {
-			rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_output);
+		if (!fnhe && FIB_RES_NH(*res).nh_pcpu_rth_output) {
+			struct rtable **prth;
+
+			prth = per_cpu_ptr(FIB_RES_NH(*res).nh_pcpu_rth_output,
+					   raw_smp_processor_id()); 
+			rth = rcu_dereference(*(__force __rcu struct rtable **)prth);
 			if (rt_cache_valid(rth)) {
 				dst_hold(&rth->dst);
 				return rth;