Re: VIRTIO_RING_F_INDIRECT_SIZE status

[Stefan Hajnoczi <stefanha@redhat.com>]

On Wed, Mar 01, 2023 at 01:55:14PM +0100, Christian Schoenebeck wrote:
> On Tuesday, February 28, 2023 1:05:21 PM CET Michael S. Tsirkin wrote:
> > On Tue, Feb 28, 2023 at 12:49:00PM +0100, Christian Schoenebeck wrote:
> > > On Monday, February 27, 2023 6:41:12 PM CET Michael S. Tsirkin wrote:
> > > > On Mon, Feb 27, 2023 at 05:13:12PM +0100, Christian Schoenebeck wrote:
> > > > > On Monday, February 27, 2023 4:45:45 PM CET Stefan Hajnoczi wrote:
> > > > > > On Mon, Feb 27, 2023 at 02:53:55PM +0000, Afsa, Baptiste wrote:
> > > > > > > The issue that we have now, is that this limitation does not seem to be enforced
> > > > > > > in Linux virtio drivers today. I came across:
> > > > > > > 
> > > > > > > https://github.com/oasis-tcs/virtio-spec/issues/122
> > > > > > > 
> > > > > > > which looks like a good base for us to build upon, but I'm not sure what is the
> > > > > > > status with this issue. Do you know whether there is any plan regarding this?
> > > > > > 
> > > > > > CCing Christian regarding extending queue size limits.
> > > > > 
> > > > > Status on this issue: it was suspended in May last year. AFAICR Michael
> > > > > expressed the need to give some more thought about it, and a new virtio spec
> > > > > release was just ahead at that time.
> > > > > 
> > > > > Michael, suggestions how to bring this forward?
> > > > > 
> > > > > Best regards,
> > > > > Christian Schoenebeck
> > > > > 
> > > > 
> > > > How about a summary letter listing various options, pros and cons?
> > > 
> > > Actually I had submitted a draft for this feature (latest version linked
> > > above), and AFAICR Michael was the only person who expressed concerns from
> > > design perspective. Comments by other people were already just aboout precise
> > > wording, but not about the design itself.
> > > 
> > > We stopped the discussion at the point where Michael expressed the need to
> > > think more about it, but as his expressed concerns were a bit vague, I still
> > > don't see how I could bring this issue forward.
> > > 
> > > Best regards,
> > > Christian Schoenebeck
> > > 
> > 
> > 
> > So the last time there were two things:
> > 
> > 
> > 1. bugs introduced during the packed ring work. For example:
> > 
> > 	> 
> > 	> I don't think so:
> > 	> 
> > 	>   2.6.5.3.1 Driver Requirements: Indirect Descriptors
> > 	>   ..
> > 	>   "A driver MUST NOT set both VIRTQ_DESC_F_INDIRECT and VIRTQ_DESC_F_NEXT in
> > 	>   flags."
> > 	> So as far as I can see it, the amount of direct descriptors is currently
> > 	> always exactly one if an indirect table is used.
> > 	> 
> > 	> Best regards,
> > 	> Christian Schoenebeck
> > 	> 
> > 
> > 	Oh. You are right.  Weirdly this text is not in packed-ring.tex - I
> > 	think this and a bunch of other cases like this are an oversight,
> > 	however we need to fix them first before adding features that assume
> > 	they are fixed ...
> > 
> > we need to get them fixed before we poke at core ring otherwise it's a
> > mess - maybe the TC will accept just fixing this
> > without a feature bit, or maybe people will feel a feature bit
> > is required.
> 
> It's been a while. Looking at v1.2 it says:
> 
>   2.8 Packed Virtqueues
>   ...
>   2.8.5 Scatter-Gather Support [1]
>   ...
>   While unusual (most implementations either create all lists solely using   
>   non-indirect descriptors, or always use a single indirect element), if both 
>   features have been negotiated, mixing indirect and non-indirect descriptors 
>   in a ring is valid, as long as each list only contains descriptors of a 
>   given type.
> 
>   [1] https://docs.oasis-open.org/virtio/virtio/v1.2/cs01/virtio-v1.2-cs01.html#x1-770005
> 
> To avoid misapprehensions: the way I understand it, same restrictions apply to
> packed queues as split queues, in the sense that you may neither chain several
> tables in a single message, nor multi-level nest tables, nor mix a list of
> direct descriptors and indirect descriptors on the same level within one
> message. So the explicit exception described here, only means you may use
> *one* indirect table in one message, while using chained direct descriptors in
> another message. But that's it, right?

Hi Christian,
Yes, for packed virtqueues it is forbidden to mix normal and indirect
descriptors in a single buffer. Further support for this interpretation
is in 2.8.19 Driver Requirements: Indirect Descriptors:

  A driver MUST NOT write direct descriptors with VIRTQ_DESC_F_INDIRECT
  set in a scatter-gather list linked by VIRTQ_DESC_F_NEXT. flags.

VIRTQ_DESC_F_INDIRECT cannot be set if the descriptor chain uses
VIRTQ_DESC_F_NEXT, therefore it's not possible to mix normal and
indirect descriptors in a single buffer.

However, the Split Virtqueues section documents an exception where
mixing normal and indirect descriptors in a single buffer is allowed,
but only in a specific order. 2.7.5.3.2 Device Requirements: Indirect
Descriptors says:

  The device MUST handle the case of zero or more normal chained
  descriptors followed by a single descriptor with
  flags&VIRTQ_DESC_F_INDIRECT. Note: While unusual (most implementations
  either create a chain solely using non-indirect descriptors, or use a
  single indirect element), such a layout is valid.

So what degree of mixing normal and indirect descriptors is allowed
depends on the virtqueue format.

Stefan

> > 2. Given this is a lot of work I am trying to find a way to
> > make the impact bigger. In particular to cover the use-case
> > of limiting s/g to 1k while making queues deeper (with
> > or without indirect). For this I proposed:
> > 
> > 	So I think that given this, we can limit the total number
> > 	of non-indirect descriptors, including non-indirect ones
> > 	in a chain + all the ones in indirect pointer table if any,
> > 	and excluding the indirect descriptor itself, and this
> > 	will address the issue you are describing here, right?
> > 
> > people seemed to be ok with this idea?
> 
> IIUIC it would not make a difference from design perspective from what I
> proposed, as virtio currently neither allows to mix, chain or mult-level nest
> indirect descriptor tables within a single message), and hence it would just
> boil down to adjusting the wording. So yes, it would therefore cover my
> intended use case.
> 
> Best regards,
> Christian Schoenebeck
> 
>

Attachment: signature.asc
Description: PGP signature