Re: Re: [mdlabriola@xxxxxxxxx: [Xen-devel] Re: Re: [Xen-users] DomU routed traffic disappearing in vif.]

[Matej Zary <zary@xxxxxxxxx>]

On Mon, 2010-05-10 at 16:32 +0100, Ian Campbell wrote:
> On Mon, 2010-05-10 at 16:18 +0100, Konrad Rzeszutek Wilk wrote:
> > Ian had a patch that he backported from XCP that might fix this, not
> > sure thought. 
> 
> If possible can someone who is seeing this issue try this patch:
> 
> --- 
> 
> xen/netback: correctly setup skb->ip_summed on receive
> 
> In 2.6.18 CHECKSUM_PARTIAL and CHECKSUM_UNNECESSARY were both synonyms for
> CHECKSUM_HW. This is no longer the case and we need to select the correct one.
> 
>   data_validated csum_blank -> ip_summed
>   0              0             CHECKSUM_NONE
>   0              1             CHECKSUM_PARTIAL
>   1              0             CHECKSUM_UNNECESSARY
>   1              1             CHECKSUM_PARTIAL
> 
> Signed-off-by: Ian Campbell <ian.campbell@xxxxxxxxxx>
> 
> diff -r 3a1424756f50 drivers/xen/netback/netback.c
> --- a/drivers/xen/netback/netback.c   Fri May 07 14:38:37 2010 +0100
> +++ b/drivers/xen/netback/netback.c   Fri May 07 14:40:31 2010 +0100
> @@ -1384,14 +1384,10 @@
>                       netif_idx_release(pending_idx);
>               }
>  
> -             /*
> -              * Old frontends do not assert data_validated but we
> -              * can infer it from csum_blank so test both flags.
> -              */
> -             if (txp->flags & (NETTXF_data_validated|NETTXF_csum_blank))
> +             if (txp->flags & NETTXF_csum_blank)
>                       skb->ip_summed = CHECKSUM_PARTIAL;
> -             else
> -                     skb->ip_summed = CHECKSUM_NONE;
> +             else if (txp->flags & NETTXF_data_validated)
> +                     skb->ip_summed = CHECKSUM_UNNECESSARY;
>  
>               netbk_fill_frags(skb);
>  
> 
> 
Works for me!

The DomU router now works like charm. Thanks! :) Also the problem with
ICMP traffic disappeared (non UDP/TCP packet dropped in syslog).

Patched the 2.6.32.11 kernel, performance wise, the 2.6.18.8 had bit
better some numbers (~ 60 Mbit difference with Iperf), will have some
more numbers later. 

Regards

Matej

<<attachment: winmail.dat>>

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel