Re: [PATCH 5/5] kvm: qemu: Improve virtio_net recv buffer allocation scheme

[Avi Kivity <avi@redhat.com>]

Mark McLoughlin wrote:
> Yeah, it's far from pretty. The current ABI basically says "you must
> supply >64k receive buffers" whereas this new ABI says "you must supply
> at least 18 >4k receive buffers".
>
> We could think about having the host expose the maximum rx packet size
> to the guest (and handle migrating to a host with a different max), but
> TBH I don't think it would be worth much until we have the prospect of
> running on a host with a larger maximum rx packet size.
>
> Requiring the guest to fill the ring with ~64k of buffers isn't onerous;
> the Linux guest impl currently re-fills the ring up to the max (e.g. 256
> x 4k)
>
>   

We should document the ABI somewhere.  All those hidden rules will make
implementing non-Linux drivers difficult, as well as maintaining
backwards compatibility.


-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.