Re: [RFC v4 3/5] virtio_ring: add packed ring support

[Jason Wang]

On 2018å05æ19æ 10:29, Tiwei Bie wrote:
> > I don't hope so.
> >
> > > I agreed driver should track the DMA addrs or some
> > > other necessary things from the very beginning. And
> > > I also repeated the spec to emphasize that it does
> > > make sense. And I'd like to do that.
> > >
> > > What I was saying is that, to support OOO, we may
> > > need to manage these context (which saves DMA addrs
> > > etc) via a list which is similar to the desc list
> > > maintained via `next` in split ring instead of an
> > > array whose elements always can be indexed directly.
> > My point is these context is a must (not only for OOO).
> Yeah, and I have the exactly same point after you
> pointed that I shouldn't get the addrs from descs.
> I do think it makes sense. I'll do it in the next
> version. I don't have any doubt about it. All my
> questions are about the OOO, instead of whether we
> should save context or not. It just seems that you
> thought I don't want to do it, and were trying to
> convince me that I should do it.

Right, but looks like I was wrong :)

> > > The desc ring in split ring is an array, but its
> > > free entries are managed as list via next. I was
> > > just wondering, do we want to manage such a list
> > > because of OOO. It's just a very simple question
> > > that I want to hear your opinion... (It doesn't
> > > means anything, e.g. It doesn't mean I don't want
> > > to support OOO. It's just a simple question...)
> > So the question is yes. But I admit I don't have better idea other than what
> > you propose here (something like split ring which is a little bit sad).
> > Maybe Michael had.
> Yeah, that's why I asked this question. It will
> make the packed ring a bit similar to split ring
> at least in the driver part. So I want to draw
> your attention on this to make sure that we're
> on the same page.

Yes. I think we are.

Thanks

> Best regards,
> Tiwei Bie