[Qemu-devel] xen is not working with qemu0.7/kqemu

[Qemu-devel] xen is not working with qemu0.7/kqemu

[G L]

Hi,

I guess this is a question for Fabrice only, as I didn't test qvm86 yet.

When using qemu as a test environment for Xen I have a crash for linux
domain0 when kqemu is loaded but it runs fine when launched with
-no-kqemu.

Crash seems to happen on hotplug (I can provide a backtrace if this is useful).

host is a debian sarge i386 with vanilla kernel 2.6.11.6
qemu 0.7
kqemu-0.6.2
xen 2.05 and domain0 kernel is 2.6.10-xen0

Please cc: me as I did not subscribe to qemu-devel (I parse the
archives anyway).

-- 
Gildas

[Qemu-devel] Re: xen is not working with qemu0.7/kqemu

[G L]

Hi,

So far no news for my problem, so I tested with qvm86 to see if it works...

I compiled qemu 0.7.0 and qvm86 (cvs checkout fresh from today).

I don't use devfs or udev so I did a "mknod qvm86 c 10 62" before
"modprobe qvm86".

I used the exact same image I used for kqemu, and I have a
segmentation fault when booting the domO (can't tell if I reached the
"hotplug stage", it happened too fast.

The last informations from a "strace qemu -hda xen.img" are :

sigreturn()                             = ? (mask now [])
ioctl(6, SPIOCSTYPE or VTXIOCGETINFO, 0xbffff160) = 14
--- SIGALRM (Alarm clock) @ 0 (0) ---
times({tms_utime=406, tms_stime=53, tms_cutime=0, tms_cstime=0}) = 576022673
sigreturn()                             = ? (mask now [])
ioctl(6, SPIOCSTYPE or VTXIOCGETINFO <unfinished ...>
+++ killed by SIGSEGV +++

Don't know of this help, I can do some more testing and provide the
system image used for the tests.

Please cc: me
Gildas

On 5/10/05, G L <3ntr0p13@gmail.com> wrote:
> Hi,
> 
> I guess this is a question for Fabrice only, as I didn't test qvm86 yet.
> 
> When using qemu as a test environment for Xen I have a crash for linux
> domain0 when kqemu is loaded but it runs fine when launched with
> -no-kqemu.
> 
> Crash seems to happen on hotplug (I can provide a backtrace if this is useful).
> 
> host is a debian sarge i386 with vanilla kernel 2.6.11.6
> qemu 0.7
> kqemu-0.6.2
> xen 2.05 and domain0 kernel is 2.6.10-xen0
> 
> Please cc: me as I did not subscribe to qemu-devel (I parse the
> archives anyway).
> 
> --
> Gildas
>

Re: [Qemu-devel] Re: xen is not working with qemu0.7/kqemu

[Ishwar Rattan]

Why would anyone want run an emulator in another emulator..

-ishwar

Re: [Qemu-devel] Re: xen is not working with qemu0.7/kqemu

[G L]

Mainly for testing purposes.

Xen is not an emulator but a Virtual Machine Monitor using paravirtualization.

Being able to run Xen in a VM allow to build complex test environments
with only one or a few machines, for instance, testing live VM
migration between physical servers.

On 6/13/05, Ishwar Rattan <ishwar@pali.cps.cmich.edu> wrote:
> Why would anyone want run an emulator in another emulator..
> 
> -ishwar

Re: [Qemu-devel] Re: xen is not working with qemu0.7/kqemu

[John R. Hogerhuis]

On Mon, 2005-06-13 at 10:28 -0400, Ishwar Rattan wrote:
> Why would anyone want run an emulator in another emulator..
> 
> -ishwar

Well at least it's hard to argue with the fact that it's an excellent
'pathological case' to test QEMU.

-- John.

Re: [Qemu-devel] Re: xen is not working with qemu0.7/kqemu

[G L]

>> If qemu with the kqemu module can run under xen, this
>> alleviates part of the pressure of getting xen to run
>> windows under a linux host.
>
>but of course, it will also kill the good performance
>xen could have with a windows support (which already
>worked, but is very unlikely to work in the future, as
>it requires M$ to adapt their OS to the xen architecture)
>
>best,
>Herbert

Won't it be possible to adapt kqemu/qvm86/qemu to make it aware of the
Xen architecture (i-e, "porting Qemu to xen") Please note that this is
a purely rethorical question, as I don't know what it really involves.

On the other hand, Mark Williamson, one of the Xen dev pointed on Xen
devel ML on a very interesting document from Microsoft
(http://download.microsoft.com/download/9/8/f/98f3fe47-dfc3-4e74-92a3-088782200fe7/TWAR05013_WinHEC05.ppt)

It seems that MS will follow an approach similar to Xen for their
future virtualization technology (maybe because MS research was
involved in Xen's first steps)

We can dream and hope that it will be possible to adapt Xen to run MS
VM  (though I believe MS will do whatever is possible to make this
impossible).

MH2C,
Gildas

Re: [Qemu-devel] Re: xen is not working with qemu0.7/kqemu

[Mark Williamson]

> >but of course, it will also kill the good performance
> >xen could have with a windows support (which already
> >worked, but is very unlikely to work in the future, as
> >it requires M$ to adapt their OS to the xen architecture)
> >
> >best,
> >Herbert
>
> Won't it be possible to adapt kqemu/qvm86/qemu to make it aware of the
> Xen architecture (i-e, "porting Qemu to xen") Please note that this is
> a purely rethorical question, as I don't know what it really involves.

Porting kqemu / qvm86 to run under Linux/Xen would seem to be the best way to 
run Windows under Xen right now.  Win4Lin also runs under Linux/Xen.

> On the other hand, Mark Williamson, one of the Xen dev pointed on Xen
> devel ML on a very interesting document from Microsoft
> (http://download.microsoft.com/download/9/8/f/98f3fe47-dfc3-4e74-92a3-08878
>2200fe7/TWAR05013_WinHEC05.ppt)
>
> It seems that MS will follow an approach similar to Xen for their
> future virtualization technology (maybe because MS research was
> involved in Xen's first steps)
>
> We can dream and hope that it will be possible to adapt Xen to run MS
> VM  (though I believe MS will do whatever is possible to make this
> impossible).

I think we'd be pleased to see MS adopt a sufficiently permissive license for 
their VMM interface specs to allow that to happen.

With VT (Intel) or Pacifica (AMD) hardware support, it'll be possible to run 
unmodified Windows under Xen, as you can under QEmu on vanilla x86.  The 
device emulation for this is actually supplied by QEmu-derived code.  Equally 
well, kqemu / qvm86 could also be adapted to leverage VT / Pacifica where 
available to allow native execution of (almost) all guest OS code under QEmu.  
I'd like to see both ;-)

Interestingly, the plan is for the MS hypervisor to *require* VT / Pacifica, 
to run virtual machines even with their paravirtualisation APIs.  This gives 
QEmu and Xen a bit of an edge for some users, although by their projected 
release date (seems to be in the 2007 - 2009 period) such hardware will be 
rather more common.

Cheers,
Mark