On 15/06/01 (月) 23:12, Eric Dumazet wrote:
On Mon, 2015-06-01 at 21:55 +0900, Toshiaki Makita wrote:
@@ -668,6 +753,9 @@ static int __init vlan_proto_init(void)
if (err < 0)
goto err5;
+ for (i = 0; i < ARRAY_SIZE(vlan_packet_offloads); i++)
+ dev_add_offload(&vlan_packet_offloads[i]);
+
vlan_ioctl_set(vlan_ioctl_handler);
return 0;
My concern about this is :
This might slow down GRO stack for other traffic, if dev_add_offload()
for vlan offloads is called after
dev_add_offload(&ip_packet_offload) /
dev_add_offload(&ipv6_packet_offload)
I didn't have that concern because there are already other similar
offloads (eth, mpls_uc, mpls_mc). But indeed, they and this could slow
down GRO stack.
This is because list_add_rcu is used and this inserts in front of the
offload_base list.
void dev_add_offload(struct packet_offload *po)
{
struct list_head *head = &offload_base;
spin_lock(&offload_lock);
list_add_rcu(&po->list, head);
spin_unlock(&offload_lock);
}
Can we ensure offload_base contains a sensible order of expected types ?
Add priority to packet_offload like nf_hook_ops?
Or have dev_add_offload() prioritize IP > IPV6 > others?
Toshiaki Makita
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
