> On 5/23/12 9:40 AM, "Dennis Musselwhite" <email@example.com...> wrote:
> >When the VSWITCH is defined with the IP option, ARP responsibility
> >is offloaded to the OSA hardware, so the Linux host should *not* be
> >sending ARP frames on that interface.
> But the OSA simply swallows the unneeded ARP automagically if sent, so
> make people choose, particularly as these days more than 8 times out of
> they WANT the extra ARP, and it's clearly needed to make the device
> function? It still seems that the proposed default of "no" is just wrong.
> Suggestion: have the device driver check the interface layer type (it
> clearly knows how, since the driver can issue an error if there is a type
> mismatch for L2 vs L3) and issue the ARP if needed. You shouldn't have
> set that kind of thing if the driver can figure it out, and clearly the
> current solution is broken because it used to work and now it doesn't.
> Since IBM opened the case, I'd suggest telling SuSE that that's not an
> acceptable solution. Their solution breaks a important network
> self-defence mechanism that is an accepted RFC, and that shouldn't
It turns out my concern was unnecessary. I know in the past we had to
block outgoing ARP frames to avoid interfering with the OSA hardware ARP
management (APAR VM64162) but I had forgotten how much time had passed.
That fix is certainly in every z/VM system by now.