Re: xen dom0 + dk(4)/gpt + hvm

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Thu, Sep 03, 2009 at 10:33:03PM -0400, Todd Kover wrote:
> 
> With both a 5.0 kernel and:
> 
> NetBSD tarvalon.omniscient.com 5.99.15 NetBSD 5.99.15 
> (OMNISCIENT_DOM0-$Revision: 1.5 $) #0: Sat Aug 29 05:03:43 UTC 2009  
> kovert%grolm.omniscient.com@localhost:/usr/obj/current/amd64/usr/src/os/NetBSD/src/sys/arch/amd64/compile/OMNISCIENT_DOM0
>  amd64
> 
> kernel, both of which are:
> 
> include "arch/amd64/conf/XEN3_DOM0"
> ident           "OMNISCIENT_DOM0-$Revision: 1.5 $"
> options         DKWEDGE_AUTODISCOVER    # Automatically add dk(4) instances
> options         DKWEDGE_METHOD_GPT      # Supports GPT partitions as wedges
> 
> if I'm doing a pvm domU (say, NetBSD 5.0/amd64) and I set the sole disk up as:
> 
>       disk = [ 'phy:/dev/dk9,hda,w' ]
> 
> everything works fine.  However, if I am trying to do an HVM (say, for
> windows) and want to use a wedge, as such:
> 
>       disk = [ 'phy:/dev/dk9,ioemu:hda,w']
> 
> Its seen as:
> 
> ata0 master: QEMU HARDDISK ATA-7 Hard Disk ( 0 MBytes)
> 
> despite the wedge being 20g.  Using a file works just fine.  (I didn't
> try a partition on a regular disklabel'd disk, not having any spares; I'm
> assuming this works, tho).
> 
> When it says its 0gb, muddling with fdisk and the like seems to support that
> its zero size.
> 
> Before I go and dig into this, does anyone have any inkling as to why this
> might be happening and what might be required?

This is probably related to qemu not trying DIOCGWEDGEINFO to get
the partition parameters, but only DIOCGPART. The best would maybe to
make it aware of DIOCGDISKINFO which is supposed to cover both
DIOCGWEDGEINFO and DIOCGPART, but I don't remember the details.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--