Re: [PATCH RFC net-next] virtio_net: Relax queue requirement for using XDP

["Toke Høiland-Jørgensen" <toke@redhat.com>]

"Michael S. Tsirkin" <mst@redhat.com> writes:

> On Wed, Feb 26, 2020 at 08:58:47AM -0700, David Ahern wrote:
>> On 2/26/20 1:34 AM, Toke Høiland-Jørgensen wrote:
>> >>
>> >> OK so basically there would be commands to configure which TX queue is
>> >> used by XDP. With enough resources default is to use dedicated queues.
>> >> With not enough resources default is to fail binding xdp program
>> >> unless queues are specified. Does this sound reasonable?
>> > 
>> > Yeah, that was the idea. See this talk from LPC last year for more
>> > details: https://linuxplumbersconf.org/event/4/contributions/462/
>> 
>>  Hopefully such a design is only required for a program doing a Tx path
>> (XDP_TX or XDP_REDIRECT). i.e., a program just doing basic ACL, NAT, or
>> even encap, decap, should not have to do anything with Tx queues to load
>> and run the program.
>
> Well when XDP was starting up it wasn't too late to require
> meta data about which codes can be returned (e.g. whether program
> can do tx). But by now there's a body of binary programs out there,
> it's probably too late ...

Well, right now things just fail silently if the system is configured
without support for a feature the XDP program is using (e.g., redirect
to an unsupported iface will just drop the packet). So arguably,
rejecting a program is an improvement :) There's ongoing work to define
a notion of XDP features (see [0]). Whether we can turn it on by
default, or if it has to be opt-in on program load/attach remains to be
seem. But it is definitely something we should improve upon :)

-Toke

[0] See https://github.com/xdp-project/xdp-project/blob/master/xdp-project.org#xdp-feature-flags