Re: [virtio-comment] Re: [PATCH 0/3] Rename queue index to queue number

["Michael S. Tsirkin" <mst@redhat.com>]

On Thu, Mar 02, 2023 at 04:12:48PM +0000, Parav Pandit wrote:
> 
> 
> > From: virtio-comment@lists.oasis-open.org <virtio-comment@lists.oasis-
> > open.org> On Behalf Of Halil Pasic
> > 
> > On Mon, 27 Feb 2023 12:39:39 -0500
> > "Michael S. Tsirkin" <mst@redhat.com> wrote:
> > 
> > > > This series is on top of [2].
> > > >
> > > > [1] https://github.com/oasis-tcs/virtio-spec/issues/163
> > > > [2]
> > > > https://lists.oasis-open.org/archives/virtio-dev/202302/msg00527.htm
> > > > l
> > >
> > > What about RSS description in net? That says:
> > >
> > > description.tex:Field \field{unclassified_queue} contains the 0-based
> > > index of
> > >
> > > is the index same as vq number? or something different?
> > 
> > It is something different. The full paragraph sound like this
> > 
> > """
> > Field \field{unclassified_queue} contains the 0-based index of the receive
> > virtqueue to place unclassified packets in. Index 0 corresponds to receiveq1.
> > """
> With vqn it can be as simply written as.
> Field \field{unclassified_queue} contains the receive virtqueue number to use for unclassified incoming packets.
> 
> Similarly rss description can be simplified.

Unfortunately this is not what it is. It's actually vqn / 2.

> > Here follows the explanation.
> > 
> > Disclaimer: For a brief math background see [1] and [2].
> > 
> > Here the index set the set of natural numbers (which includes 0, the neutral
> > element for the addition operation).
> > 
> > The indexed set is the set of receive virtqueues, each member of that set is
> > associated with an unique member of the index set.
> > 
> > """
> > \subsection{Virtqueues}\label{sec:Device Types / Network Device / Virtqueues}
> > 
> > \begin{description}
> > \item[0] receiveq1
> > \item[1] transmitq1
> > \item[\ldots]
> > \item[2(N-1)] receiveqN
> > \item[2(N-1)+1] transmitqN
> > \item[2N] controlq
> > \end{description}
> > """
> > 
> > But, as seen above, we happen to also use an other index set for the receive
> > queues of an virtio-net device. I guess the intention is to use these as names, in
> > any case the set is { "receiveq1", "receiveq2", ... } (yes, I'm a little sloppy here).
> > 
> > Thus we can say that:
> > * the virtqueue index 2*(N-1)
> > * the recieve virtqueue index N-1, and the
> > * virtqueue name receiveqN
> > refer to the very same virtqueue (for each N > 0)
> > 
> Receiveq1 is a name to virtuque number (currently index) 0.
> And naming this way doesn't seem to be a problem in the description.
> 
> Section 5.1.2 is liste them already.
> A short line tell about this mapping will suffice.

can't parse this.


> > And it depends on the context which scheme do we use. As far as I remember
> > the names are not a part of any virtio interface. I.e. they are only there for the
> > convenience of the spec and the reader. We could get rid of those. Although
> > the possibility of corresponding to some network device, Linux network device
> > or tooling naming convention is also to consider. But AFAIU Linux uses 0 based
> > indexing for the rx and tx queues. 
> Correct.
> 
> > If the 1 based naming is really just
> > constrained to the spec, we can change it without problem.
> Yes, most changes are trivial I looked at.
>