Re: [PATCH v2] machine: add missing doc for memory-backend option

[Peter Krempa <pkrempa@redhat.com>]

On Fri, Jan 15, 2021 at 10:02:04 +0000, Peter Maydell wrote:
> On Thu, 14 Jan 2021 at 23:48, Igor Mammedov <imammedo@redhat.com> wrote:
> >
> > Add documentation for '-machine memory-backend' CLI option and
> > how to use it.
> >
> > And document that x-use-canonical-path-for-ramblock-id,
> > is considered to be stable to make sure it won't go away by accident.
> 
> That's not what the x- prefix is supposed to mean.

This is the exact reason I was asking on behalf of the libvirt team for
adding such statement if we were to use it. We want guarantee that it's
considered stable since without that it will not be accepted into
libvirt.

> If we have an internal constraint that we mustn't delete
> the option in order to support some other must-be-stable
> interface (eg migration of some machines) we can document
> that in a comment, but that doesn't mean that we should
> document to users that direct use of an x-prefix option
> is supported as a stable interface.

AFAIK the issue is that with the new approach to configure system memory
(via a memory-backend=id, since the old one was deprecated) migration
fails from/to older qemus ...

> Alternatively, if the option is really stable for direct
> use by users then we should commit to making it so by
> removing the x-.

... thus the idea behind keeping this interface as is is it also fixes
the migration compatibility for qemu 5.0/5.1/5.2 which were already
released.

Removing the 'x-' will fix it only starting with qemu-6.0 and any
downstream which backports the removal of the prefix.

Obviously not using 'x-' prefixed options is strongly preferred in
libvirt.