[libvirt] [PATCH for-4.0 v2] virtio: Provide version-specific variants of virtio PCI devices

[Eduardo Habkost]

On Mon, Nov 19, 2018 at 10:08:42PM -0500, Michael S. Tsirkin wrote:
> On Mon, Nov 19, 2018 at 07:47:40PM -0200, Eduardo Habkost wrote:
> > On Mon, Nov 19, 2018 at 01:07:59PM -0500, Michael S. Tsirkin wrote:
> > n
> > > On Mon, Nov 19, 2018 at 11:41:05AM +0100, Cornelia Huck wrote:
> > > > On Fri, 16 Nov 2018 01:45:51 -0200
> > > > Eduardo Habkost <ehabkost at redhat.com> wrote:
> > > > 
> > > > > On Thu, Nov 15, 2018 at 05:29:24PM +0100, Andrea Bolognani wrote:
> > > > 
> > > > > > One thing that I'm very much not convinced about is the naming,
> > > > > > specifically leaving the virtio revision out: I get it that we
> > > > > > Should Never Need™ another major version of the spec, but I'm
> > > > > > afraid discounting the possibility outright might prove to be
> > > > > > shortsighted and come back to bite us later, so I'd much rather
> > > > > > keep it.
> > > 
> > > That's not the claim. In fact the reverse is true - a major revision can
> > > come at any point. The claim is that virtio compatibility is not based
> > > on version numbers.  And another claim is that you can trust the
> > > virtio TC not to overload terminology that it put in the spec. So use
> > > that and you should be fine.  Come up with your own and end up writing
> > > another spec just to document it.
> > > 
> > > > > > 
> > > > > > And once that's done, "non-transitional" (while matching the
> > > > > > language of the spec) starts to look a bit unnecessary when you
> > > > > > could simply have
> > > > > > 
> > > > > >   virtio-*-pci
> > > > > >   virtio-*-pci-1
> > > > > >   virtio-*-pci-1-transitional
> > > > > > 
> > > > > > instead. But I don't feel as strongly about this as I do about
> > > > > > keeping the virtio revision in the device name :)  
> > > > > 
> > > > > I like that suggestion.  Makes the device names more explicit
> > > > > _and_ shorter.  I'll do that in v3.
> > > > 
> > > > OTOH, that would mean we'd need to introduce new device types if we
> > > > ever start to support a virtio 2.x standard. My understanding was that
> > > > we'll want separate device types for transitional and non-transitional
> > > > for two reasons: the bus which a device can be plugged into, and
> > > > changing ids. Do we really expect huge changes in a possible 2.x
> > > > standard that affect virtio-pci only, and not other virtio transports
> > > > as well?
> > > 
> > > Yes I think adding a version there is a mistake.
> > > transitional/legacy/non-transitional are spec terms so
> > > they are unlikely to change abruptly. OTOH virtio TC can
> > > just decide next version is 2.0 on a drop of a hat.
> > > 
> > > And I strongly believe command line users really really do not want all
> > > this mess. Even adding "pci" is the name confuses people (what are the
> > > other options?). For command line model=virtio is pretty much perfect.
> > > So all these names are there primarily for libvirt's benefit.
> > > And the only input from libvirt devs so far
> > > has been that it's unclear how does cross version
> > > migration work. That needs to be addressed in some way.
> > 
> > What still needs to be addressed?
> 
> I don't belive you answered Daniel's question.
> 
> >  Just keep the existing device
> > types on migration.  We could make additional promises about
> > compatibility with the disable-modern and disable-legacy
> > properties, but I don't see why we need to do it unless we start
> > deprecating the old device types.
> 
> Then the answer seems to be in the negative?

If the question is about the current state, it's "yes".  If it's
about promises about the future, then we need to understand what
kind of promise is expected.


> > > 
> > > So can we maybe start with a libvirt domain xml patch, with an
> > > implementation for existing QEMU, get that acked, and then just map it
> > > back to qemu command line as directly as possible?
> > 
> > I don't know what you mean here by "libvirt domain XML patch".
> > 
> > Do you mean solving the problems only on the libvirt side,
> > keeping the existing device types?  Why would we do that?  It
> > would be a hack making the situation even messier.
> > 
> > If libvirt needs us to provide better interfaces, let's cooperate
> > with them.  I'd like us to avoid having yet another "the problem
> > must be solved in the other layer first" deadlock here.
> 
> I mean IIUC libvirt is the solve user that will benefit from this patch.
> Let's at least get an ack confirming it does make their lives easier.

I understand this as an ACK;
https://www.mail-archive.com/qemu-devel@nongnu.org/msg567161.html

Quoted below:

| I don't have a objection from libvirt side.
| 
| Last time, I suggested/discussed this I was not convinced that the benefit
| was compelling enough to justify the  work across all levels in the stack.
| 
| Apps using the new device model names would either make themselves
| incompatible with older libvirt/QEMU, or they would increase their
| code complexity & testing matrix by having to support both old & new
| names. The usage would also harm migration to older hosts.
| 
| This just to be able to switch from i440fx to q35 for OS which don't
| support virtio-1.0, but for such old OS, q35 isn't offering any
| compelling features, so they might as well stick with the thing that
| is known to work well.
| 
| If QEMU supports this, we'd support it in libvirt, but my recommendation
| to apps would still likely be to not use it and simply stick with i440fx
| for such older OS.

Re: making their lives easier: easier than what?

-- 
Eduardo