Re: Processed: Re: [HVM} xen_platform_pci=0 doesn't prevent platform device creation and disk and nic take over by PV drivers.

From: Anthony PERARD <anthony.perard@citrix.com>
To: George Dunlap <George.Dunlap@eu.citrix.com>
Cc: Sander Eikelenboom <linux@eikelenboom.it>,
	xen-devel <xen-devel@lists.xen.org>,
	Ian Campbell <Ian.Campbell@citrix.com>,
	Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Subject: Re: Processed: Re: [HVM} xen_platform_pci=0 doesn't prevent platform device creation and disk and nic take over by PV drivers.
Date: Fri, 11 Oct 2013 18:31:36 +0100	[thread overview]
Message-ID: <20131011173136.GC3195@perard.uk.xensource.com> (raw)
In-Reply-To: <CAFLBxZbhrc67UaTqdPANY0qgAPmp3PDu0qaROjz2GUt-tX_vog@mail.gmail.com>

On Fri, Oct 11, 2013 at 06:10:32PM +0100, George Dunlap wrote:
> On Fri, Oct 11, 2013 at 5:53 PM, Anthony PERARD
> <anthony.perard@citrix.com> wrote:
> > On Fri, Oct 11, 2013 at 05:22:47PM +0200, Sander Eikelenboom wrote:
> >> > The original message was simply:
> >> >> While trying to get to the bottom of my passthrough problem with the rom bar,
> >> >> i noticed that specifiying "xen_platform_pci=0" in the config file does not prevent
> >> >> the platform device to appear in my HVM guest and consequently the disk and nic are taken over.
> >> >>
> >> >> Running latest xen-unstable, together with upstream qemu and upstream seabios.
> >
> > I've start looking at this, and the fix would be easy: removing the
> > creation of the xen-platform from QEMU, and let xl ask QEMU to create
> > it.
> >
> > But if one uses an older version of one project (qemu/xl) and a newer
> > version of the other, then we can end-up with no xen-platform or two
> > xen-platform ...
> 
> So what's the plan for compatibility here -- full forward and backward
> compatibily on both sides?  (i.e., newer qemu works with older Xen,
> newer Xen works with older qemu)?
> 
> > So should we try to be clever about this, or just backport patchs for
> > both qemu and xl?
> >
> > "Clever" option: There is a recent patch for xl that make use of
> > the -nodefault QEMU's command line option. So if it's present, we could
> > ask QEMU to not add xen-platform, and xl can decide.
> 
> I wouldn't mind doing something like this; but would it make more
> sense to have some kind of a "capability" or "version" interface, like
> we have between Xen and Linux, rather than relying on other quirks
> like the presence of "-nodefault"?

Well, we can always check if the xen-platform-pci is already there after
having started QEMU (through QMP) and react by adding it, display a
warning or do nothing more, depending the result and the config of the
vm.

-- 
Anthony PERARD