"David S. Miller" wrote:
>
> Andrew Morton writes:
> > tells us that 54 packets have been sent with h/w checksums and this:
> >
> > tells us that they were all multi-fragment.
> >
> > I assume this is because the IP header is in a different frag?
>
> Right, the headers are all in the skb->data buffer and the
> application data sits in the SKB frags.
>
> > Is there ever a situation in which these numbers will differ?
>
> If we allowed SG without hw csumming, for example :-)
>
> It would let us identify such devices by just asking the
> user for a statistics dump which gathering debug information.
I see.
And just to clarify: it is currently the case that we
do support scatter/gather on devices which don't have
hardware checksums on transmit.
When tx checksums are disabled, with a 3c905C, SG is
still used, so /proc/net/dev says:
Inter-| Receive | Transmit
face |bytes packets errs drop fifo frame compressed multicast|bytes
packets errs drop fifo colls carrier compressed
lo: 2520 36 0 0 0 0 0 0 2520
36 0 0 0 0 0 0
eth0: 112364 549 0 0 0 0 0 0 69961
413 0 0 0 0 0 239
That is, 239 packets were sent using multiple fragments
and none of them used h/w checksums.
I think it's sunk in now :)
|