On Fri, Sep 14, 2018 at 11:47 AM Paolo Abeni <pab...@redhat.com> wrote:
>
> Currently, the UDP GRO callback is always invoked, regardless of
> the existence of any actual user (e.g. a UDP tunnel). With retpoline
> enabled, this causes measurable overhead.
>
> This changeset introduces explicit accounting of the sockets requiring
> UDP GRO and updates the UDP offloads at runtime accordingly, so that
> the GRO callback is present (and invoked) only when there is at least
> one socket requiring it.
I have a difference solution both to the UDP socket lookup avoidance
and configurable GRO in general.
The first can be achieved by exporting the udp_encap_needed_key static key:
"
diff --git a/include/net/udp.h b/include/net/udp.h
index 8482a990b0bb..9e82cb391dea 100644
--- a/include/net/udp.h
+++ b/include/net/udp.h
@@ -443,8 +443,10 @@ int udpv4_offload_init(void);
void udp_init(void);
+DECLARE_STATIC_KEY_FALSE(udp_encap_needed_key);
void udp_encap_enable(void);
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index f4e35b2ff8b8..bd873a5b8a86 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -1889,7 +1889,7 @@ static int __udp_queue_rcv_skb(struct sock *sk,
struct sk_buff *skb)
return 0;
}
-static DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
+DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
void udp_encap_enable(void)
{
static_branch_enable(&udp_encap_needed_key);
diff --git a/net/ipv4/udp_offload.c b/net/ipv4/udp_offload.c
index 4f6aa95a9b12..f44fe328aa0f 100644
--- a/net/ipv4/udp_offload.c
+++ b/net/ipv4/udp_offload.c
@@ -405,7 +405,7 @@ static struct sk_buff *udp4_gro_receive(struct
list_head *head,
{
struct udphdr *uh = udp_gro_udphdr(skb);
- if (unlikely(!uh))
+ if (unlikely(!uh) || !static_branch_unlikely(&udp_encap_needed_key))
goto flush;
"
.. and same for ipv6.
The second is a larger patchset that converts dev_offload to
net_offload, so that all offloads share the same infrastructure, and a
sysctl interface to be able to disable all gro_receive types, not just
udp.
I've been sitting on it for too long. Let me slightly clean it up and
send it out for discussion sake..
>
> Tested with pktgen vs udpgso_bench_rx
> Before:
> udp rx: 27 MB/s 1613271 calls/s
>
> After:
> udp rx: 30 MB/s 1771537 calls/s
>
> Signed-off-by: Paolo Abeni <pab...@redhat.com>
