Re: netback behaviour (mirage under xen)

[David Scott <scott.dj@xxxxxxxxx>]

Dave wondered:

>
> The only thing I can see that we might be doing wrong is we're updating the
> shared request pointer per fragment rather than per packet. This will allow
> the backend to see the initial fragment by itself if it has time to look
> (hence the sleep 1 triggering the problem).

 On Thu, Jan 10, 2013 at 4:43 PM, Steven Smith <sos22@xxxxxxxxxxxxx> wrote:

Yeah, don't do that, it doesn't work.  Netback assumes that you only
put whole packets on the ring before updating the indices, and will
get quite confused if you push in partial ones.  It's supposed to drop
the entire packet and return an error, rather than splitting it in
half, but that's not even slightly tested, so it really doesn't
surprise me that it's a bit broken.

> I suspect a workaround will probably be to update the request pointer per
> packet rather than per fragment.

That's the correct fix, yes.

Great, thanks for clearing that up!

Cheers,

Dave