[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [Xen-devel] [PATCH v2] libxl: change default QEMU machi
From: |
Ian Campbell |
Subject: |
Re: [Qemu-devel] [Xen-devel] [PATCH v2] libxl: change default QEMU machine to pc-i440fx-1.6 |
Date: |
Fri, 11 Jul 2014 16:08:56 +0100 |
On Fri, 2014-07-11 at 15:54 +0100, Stefano Stabellini wrote:
> On Fri, 11 Jul 2014, Ian Campbell wrote:
> > On Fri, 2014-07-11 at 12:14 +0100, Stefano Stabellini wrote:
> > > On Thu, 10 Jul 2014, Ian Jackson wrote:
> > > > Stefano Stabellini writes ("Re: [PATCH v2] libxl: change default QEMU
> > > > machine to pc-i440fx-1.6"):
> > > > > ping?
> > > > >
> > > > > On Thu, 12 Jun 2014, Stefano Stabellini wrote:
> > > > ...
> > > > > > This patch does not change the emulated environment in the guest,
> > > > > > unless
> > > > > > soundhw='hda' is specified, in that case the xen-platform device is
> > > > > > moved to slot 3 (used to be always slot 2). This change might cause
> > > > > > problems to guests with soundhw='hda', migrating from 4.4 to 4.5.
> > > >
> > > > Is there a way for someone who experiences this problem to make it go
> > > > back to the way it was ?
> > >
> > > A guest reboot would solve the problem.
> >
> > That rather defeats the purpose of being able to migrate from N to N+1.
> >
> > What is the issue with hda which requires us to move slots around if it
> > is enabled where we didn't need to do so before?
>
> If we use -M pc or -M pc-i440fx-1.6, the xen-platform-pci device is
> created slightly later in the QEMU boot sequence. The Intel HDA device
> is able to "steal" its slot on the PCI bus.
>
> On the other hand it is worth pointing out that currently libxl switches
> from -M xenfv to -M pc silently depending on whether the
> xen_platform_pci option is enabled or disabled. If it is disabled we get
> pc, that is a moving target and not guaranteed to be compatible with
> anything.
Changing that option is a guest reboot scenario, which is completely
different to a migration. I think it is (just about) OK for the PCI id
to change on a reboot, but it certainly isn't OK on migrate.
If you can't get the platform device created earlier then perhaps you
can find a way to influence the slot used by the hda device?
Ian.