[arch-general] [arch-dev-public] [announcement] qemu/qemu-kvm announcement draft

[Alexander Duscheleit]

On Sun, 10 Jan 2010 11:07:19 +0100
Tobias Powalowski <t.powa at gmx.de> wrote:

> Am Sonntag 10 Januar 2010 schrieb Simon Boulay:
> > On 01/10/2010 09:48 AM, Tobias Powalowski wrote:
> > > Am Samstag 09 Januar 2010 schrieb Simon Boulay:
> > >> On 01/09/2010 09:09 PM, Tobias Powalowski wrote:
> > >>> Am Samstag 09 Januar 2010 schrieb Dan McGee:
> > >>>> On Sat, Jan 9, 2010 at 11:52 AM, Tobias
> > >>>> Powalowski<t.powa at gmx.de>   
> wrote:
> > >>>>>    >   Yes will change the install message.
> > >>>>>>
> > >>>>>> Yes there is no mention in the changelogs, really strange.
> > >>>>>> greetings
> > >>>>>> tpowa
> > >>>>>
> > >>>>> Ok like this?
> > >>>>>    echo ">>>   Since kernel 2.6.29:"
> > >>>>>    echo ">>>   Qemu package now provides standard qemu with
> > >>>>> kvm enabled." echo ""
> > >>>>>    echo ">>>   PLEASE READ FOR KVM USAGE!"
> > >>>>>    echo ">>>    Load the correct KVM module, you will need a
> > >>>>> KVM capable CPU!" echo ">>>    Add yourself to the group
> > >>>>> 'kvm'." echo ">>>    Use 'qemu --enable-kvm' to use KVM."
> > >>>>>    echo ""
> > >>>>>    echo "With the release of qemu and qemu-kvm 0.12.X, the
> > >>>>> kqemu kernel module" echo "is no longer supported and will be
> > >>>>> removed from the repositories. You" echo "can safely
> > >>>>> uninstall it from your system."
> > >>>>
> > >>>> Can we put some vercmp checks around messages like this? That
> > >>>> way people only have to see them once (when they upgrade the
> > >>>> first time to a 0.12.x version for the second message). The
> > >>>> first message should really be a post_install message.
> > >>>>
> > >>>> And with all that said, why are there two packages in extra if
> > >>>> "qemu package now provides standard qemu with kvm enabled"?
> > >>>>
> > >>>> -Dan
> > >>>
> > >>> Yes sure i can add those vercmp stuff.
> > >>> qemu and qemu-kvm is different.
> > >>> qemu-kvm is only for  kvm while qemu provides much more
> > >>> machines to emulate.
> > >>
> > >> I'm not sure about that. Both seems to share the same code for
> > >> machine emulation; only the kvm stuff is different. In fedora
> > >> 12, they build kvm and qemu-system-xxx from qemu-kvm 0.11. But I
> > >> don't know how this will evolve in the future.
> > >> If qemu and qemu-kvm are used for different purposes, one may
> > >> need to install both apps side by side but that's not possible
> > >> in archlinux.
> > >
> > > Why?
> > > qemu is for those who need more different emulation types.
> > > qemu-kvm is only for 86 emulation with kvm hardware support.
> > > Both differ in files you would need to hack bios file destination
> > > etc. I don't see any need to install both at the same time.
> > 
> > Because one may want to use x86 emulation with kvm hardware support
> > and qemu-system-arm for example on the same machine.
> > It is possible to build all targets with qemu-kvm but that's not the
> > default and I don't know if that'll be the case for future release.
> > For 0.11 release, qemu and qemu-kvm seems to converge, but with 0.12
> > that's not so clear (at least to me). As I understand it, the
> > development of platform emulation is done in qemu and kvm
> > virtualization is done in qemu-kvm (even if qemu has some kvm
> > support) but the qemu repository is regularly merged in qemu-kvm. I
> > don't find any official statement about that, so...
> normal qemu supports kvm too, just use --enable-kvm start parameter.
> So no need to install qemu-kvm.
> greetings
> tpowa
> 

In this light, what actually is qemu-kvm good for? We don't split
packges for -src, -devel, but for startup-parameters?

If there is no other difference then a few more binaries (which as far
as i know doesn't justify another package) why not kill qemu-kvm
alltogether and include something like /usr/bin/kvm:

---8<---
#!/bin/bash
qemu --enable-kvm $*
--->8---  

When I tested both packages here qemu with --enable-kvm *felt* a little
slower when running XP, but that's a) entirely subjective and b) I
dodn't test identical workloads.

So, again, what is the reason for there being a qemu-kvm package, when
it is apparently a subset of the qemu package?

Greetings,
	jinks

--