> Very interesting. I am very curious, what kind of "offload" is possible
> with current stack are possible. Even if you know addresses. :-)
it's the same kind of offload we use in IPv4 -- we set the
net_device->flags to IFF_NOARP and can then talk IP to the card. That
means,
the card does ARP and the tasks of hard_header etc. itself. We register
ourselves to get informed as soon as an IP address has been added/removed
and can tell that information to the card. As soon as the card knows about
the IP addresses of the device in the stack, it can respond to ARP queries
of other nodes.
Now hardware header generation and ARP don't cost a lot of performance, but
this whole story gets in particular interesting, as an S/390 can share a
single card between partitions on the same physical box (and that's very
common; obvious by looking at the price tag of the card :-). The card
peeks into incoming packets and according to the destination address, it
passes the packets on to the right partition(s).
We have the problem, that we won't be informed about any address changes
in IPv6 other than somehow polling for them.
(un)register_inet6addr_notifier would help us here.
> What's about the patch... Do you understand that currently
> it is impossible to call notifiers for adding/deletion of each IPv6
address
> in an intelligible context? Not seeing uses of such notifiers,
> it is difficult to approve such feature because of danger of misuse
> now and even worse misuse in (near) future, when context will change.
Yes, we are fully aware, that we cannot call the notifier functions
ourselves
resp. cannot expect any useful result by doing so. We are only interested
in
being always up to date wrt. IP addresses.
Regards,
Utz
Linux for S/390 and zSeries
:wq
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/