On 15/05/28 (木) 21:02, Eric Dumazet wrote:

On Thu, 2015-05-28 at 20:17 +0900, Toshiaki Makita wrote:

quoted

Currently packets with non-hardware-accelerated vlan cannot be handled
by GRO. This causes low performance for 802.1ad and stacked vlan, as their
vlan tags are currently not stripped by hardware.

This patch adds GRO support for non-hardware-accelerated vlan and
improves receive performance of them.

Very nice patch !

quoted

Signed-off-by: Toshiaki Makita <redacted>
---
  net/8021q/vlan.c | 94 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
  1 file changed, 94 insertions(+)

diff --git a/net/8021q/vlan.c b/net/8021q/vlan.c
index 59555f0..0a9e8e1 100644
--- a/net/8021q/vlan.c
+++ b/net/8021q/vlan.c

@@ -618,6 +618,90 @@ out:
  	return err;
  }

+		vhdr2 = (struct vlan_hdr *)(p->data + off_vlan);
+		if (memcmp(vhdr, vhdr2, VLAN_HLEN))
+			NAPI_GRO_CB(p)->same_flow = 0;
+	}

This memcmp() is quite expensive, you better use a helper like :

/* vlan header only guaranteed to be 16bit aligned */
static bool vlan_hdr_compare(const struct vlan_hdr *h1, const struct vlan_hdr *h2)
{
#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
	return *(u32 *)h1 != *(u32 *)h2;
#else
	return (((__force u32)h1->h_vlan_TCI ^ (__force u32)h2->h_vlan_TCI) |
		((__force u32)h1->h_vlan_encapsulated_proto ^
		 (__force u32)h2->h_vlan_encapsulated_proto)) != 0;
#endif
}

Hi Eric,

Thank you for your reviewing.
Indeed, memcmp() is not good for performance.
I'll include your feedback in v2.

Thanks,
Toshiaki Makita