On Mon, Sep 19, 2016 at 08:47:47PM +0200, Paolo Bonzini wrote:
>
>
> On 19/09/2016 19:22, Michael S. Tsirkin wrote:
> > No but you do need to make sure that when get_buf returns
> > the first buffer, all buffers are available.
>
> Is that not the case already? But anyway I now agree it's better to put
> it in the descriptor, thanks for clarifying.
>
> > And from the achitecture POV, I feel this is a transport
> > thing, not a device specific thing.
>
> It depends, many devices (e.g. storage) are packet-based and bursts do
> not have any meaning. So I guess it's okay, but then I would prefer to
> make it a separate feature bit.
It just becomes an optimization feature then, and can be ignored
without a feature flag.
Maybe it's ok - will allow reusing some bits for something else - but
some people did complain about all the branches forced by the
feature bits.
> I'm mostly afraid that it would be hard for the spec to define it in
> general terms (while it's easy to define something specific to rxbuf
> merging). Having device-specific descriptor flags seems like a natural
> extension anyway...
>
> Paolo
Well - the batching of the index is kind of an undocumented property
in 1.0. I'll try to write it up, let's see what I can come up with.
--
MST
