Re: MSI-X vector limits and reserving a virtio device ID

[Stefan Hajnoczi <stefanha@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 2369 bytes --]

On Thu, May 28, 2026 at 08:47:25AM +0300, Manos Pitsidianakis wrote:
> On Thu, May 28, 2026 at 8:22 AM Parav Pandit <parav@nvidia.com> wrote:
> >
> >
> > > From: Demi Marie Obenour <demiobenour@gmail.com>
> > > Sent: 28 May 2026 05:23 AM
> > > To: virtio-comment@lists.linux.dev
> > > Subject: MSI-X vector limits and reserving a virtio device ID
> > >
> > > I'd like to reserve a virtio device ID for virtio vhost-guest,
> > > formally virtio vhost-user.  Would this be possible?
> > >
> > Vhost user is an implementation of the device.
> > I believe it stays as implementation and not a new device type.
> 
> This exactly.
> 
> Furthermore, we already have a mechanism for "providing" an arbitrary
> virtio device; it's called a transport.
> 
> Demi, I suggest you look into virtio-msg transport, which would allow
> you to do what you want.

I think the "virtio vhost-guest" device is a continuation of the
virtio-vhost-user device work:
https://wiki.qemu.org/Features/VirtioVhostUser

It is a VIRTIO device type, not a transport.

The idea is that a driver VM runs a normal VIRTIO driver (e.g.
virtio-net-pci) and a second device VM runs the virtio-net device
implementation. It's similar to running a vhost-user-net device on the
host, except now the device implementation runs inside a VM instead of a
process on the host. The virtio-vhost-user device is what the device VM
sees and uses to provide the virtio-net device to the driver VM.

Stefan

> 
> >
> > > Also, I'd like to know which systems have a small limit on the number
> > > of MSI-X vectors.
> > I didn’t follow the question. Typically, Intel cpu has a limit of 256 vectors per core.
> >
> > > The vhost-guest device uses lots of MSI-X vectors.
> > > Typical implementations will use 256 * 3 + 7 = 775 of them!  That can,
> > > of course, be reduced by sharing vectors, but this is undesirable
> > > as it comes at a performance cost.
> > Did you get a chance to measure the performance, for which device type?
> > Interrupts are generally moderated.
> > In other devices (scalable functions of mlx5) I have measured perf of 8 to 16 queues sharing a vectors has < 4% perf drop.
> >
> > > --
> > > Sincerely,
> > > Demi Marie Obenour (she/her/hers)
> 
> -- 
> Manos Pitsidianakis
> Emulation and Virtualization Engineer at Linaro Ltd
> 

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]