[Qemu-devel] Re: [PATCHv2] Support for booting from virtio disks

[Gleb Natapov <gleb@redhat.com>]

On Wed, May 12, 2010 at 10:22:59AM +0300, Avi Kivity wrote:
> On 05/11/2010 03:31 PM, Gleb Natapov wrote:
> >On Tue, May 11, 2010 at 11:19:07AM +0300, Avi Kivity wrote:
> >>On 05/10/2010 06:48 PM, Anthony Liguori wrote:
> >>>On 05/10/2010 03:11 AM, Gleb Natapov wrote:
> >>>>This patch adds native support for booting from virtio disks to Seabios.
> >>>>
> >>>>Signed-off-by: Gleb Natapov<gleb@redhat.com>
> >>>A related problem that I think we need to think about how we solve
> >>>is indicating to Seabios which device we want to boot from
> >>>
> >>>With your patch, a user can select a virtio device explicitly or
> >>>if they use only one virtio device, it will Just Work.
> >>>
> >>>However, if a user uses IDE and virtio, or a user has multiple
> >>>disks, they cannot select a device via -boot.
> >>>
> >>>Is this something we need to address?  I don't think we'd break
> >>>libvirt if we didn't.
> >>BIOSes traditionally address this by storing the boot order in RTC
> >>non-volatile memory, and allow the user to configure the order via a
> >>menu.  We could do the same (storing the RTC memory in a small disk
> >>image).
> >>
> >Real BIOS can do that because it enumerates all bootable devices,
> >attach name for each one of them and then asks user to configure
> >boot order using names it attached to devices. In our case we
> >want to provide boot order on qemu command line before BIOS
> >enumerated devices, so qemu should be able to pass enough information
> >about boot device so that BIOS can uniquely identify it after it will
> >discover all bootable devices. bus/device pair can be such thing.
> 
> Having a BIOS menu is also useful, you don't have to drop to the
> management tool, instead you do everything from the console.
> 
In Seabios we have functional boot menu. But it is management who
controls what disk plugged were.

> >>Alternatively we can seed the order from the command line (-boot
> >>id1,id2,id3 where id* are some qdev property attached to disks, this
> >>is more flexible than the current syntax I think).
> >>
> >The problem is how to communicate this order to Seabios.
> 
> Topology (bus/device/lun).
> 
Yeah, that what I proposed too actually.

--
			Gleb.