[Qemu-devel] [RFC] Hardware acceleration for Windows and OS X?

[Qemu-devel] [RFC] Hardware acceleration for Windows and OS X?

[Stefan Weil]

Hello,

is anybody working on KVM or something equivalent for Windows and Mac OS X?

I recently noticed that the QEMU based Android emulator uses KVM for Linux
and HAXM, an Intel hypervisor, for Windows and Mac OS X.

Would it be a good plan to merge HAXM support in latest QEMU
(the Android QEMU is based on a rather old version)? It only
supports Intel based hosts, is restricted to 32 bit guests (?)
and free, but not open source.

Regards,
Stefan Weil

Re: [Qemu-devel] [RFC] Hardware acceleration for Windows and OS X?

[Paolo Bonzini]

Il 19/08/2013 22:17, Stefan Weil ha scritto:
> Hello,
> 
> is anybody working on KVM or something equivalent for Windows and Mac OS X?
> 
> I recently noticed that the QEMU based Android emulator uses KVM for Linux
> and HAXM, an Intel hypervisor, for Windows and Mac OS X.
> 
> Would it be a good plan to merge HAXM support in latest QEMU
> (the Android QEMU is based on a rather old version)? It only
> supports Intel based hosts, is restricted to 32 bit guests (?)
> and free, but not open source.

Considering the level of openness in both Windows and Mac OS X, a
proprietary hypervisor would be the smallest of the problems for someone
using those OSes...

I guess "patches are welcome"---if they are of good quality, why not.
The community doesn't have that many kernel hackers for Windows and Mac
OS X.

Paolo

Re: [Qemu-devel] [RFC] Hardware acceleration for Windows and OS X?

[Peter Maydell]

On 20 August 2013 11:56, Paolo Bonzini <pbonzini@redhat.com> wrote:
> I guess "patches are welcome"---if they are of good quality, why not.

Well, sort of. Given the fairly minimal level of support we currently
have for OSX (ie Andreas and I test it a bit and fix egregious breakage),
I'd prefer it if patches for significant new features came attached to a
developer who was going to stay around and help maintain the platform :-)

-- PMM

Re: [Qemu-devel] [RFC] Hardware acceleration for Windows and OS X?

[Stefan Hajnoczi]

On Mon, Aug 19, 2013 at 10:17:15PM +0200, Stefan Weil wrote:
> Would it be a good plan to merge HAXM support in latest QEMU
> (the Android QEMU is based on a rather old version)? It only
> supports Intel based hosts, is restricted to 32 bit guests (?)
> and free, but not open source.

I think the answer will be "Yes, if there is someone around to actively
maintain it".

Stefan

Re: [Qemu-devel] [RFC] Hardware acceleration for Windows and OS X?

[Stefan Hajnoczi]

On Tue, Aug 20, 2013 at 12:12:21PM +0100, Peter Maydell wrote:
> On 20 August 2013 11:56, Paolo Bonzini <pbonzini@redhat.com> wrote:
> > I guess "patches are welcome"---if they are of good quality, why not.
> 
> Well, sort of. Given the fairly minimal level of support we currently
> have for OSX (ie Andreas and I test it a bit and fix egregious breakage),
> I'd prefer it if patches for significant new features came attached to a
> developer who was going to stay around and help maintain the platform :-)

Yes, the maintenance and testing is the hard part.

Whoever steps up to the maintainer: be prepared to run builds and test
at least release candidates.  And then get ready to git-bisect(1) when
broken commits were merged and write fixes.  The more often you do this,
the shorter the bisect.

It's a fair bit of work.

Stefan

Re: [Qemu-devel] [RFC] Hardware acceleration for Windows and OS X?

[Paolo Bonzini]

Il 21/08/2013 11:08, Stefan Hajnoczi ha scritto:
> On Tue, Aug 20, 2013 at 12:12:21PM +0100, Peter Maydell wrote:
>> On 20 August 2013 11:56, Paolo Bonzini <pbonzini@redhat.com> wrote:
>>> I guess "patches are welcome"---if they are of good quality, why not.
>>
>> Well, sort of. Given the fairly minimal level of support we currently
>> have for OSX (ie Andreas and I test it a bit and fix egregious breakage),
>> I'd prefer it if patches for significant new features came attached to a
>> developer who was going to stay around and help maintain the platform :-)
> 
> Yes, the maintenance and testing is the hard part.
> 
> Whoever steps up to the maintainer: be prepared to run builds and test
> at least release candidates.  And then get ready to git-bisect(1) when
> broken commits were merged and write fixes.  The more often you do this,
> the shorter the bisect.
> 
> It's a fair bit of work.

Still, I must say merging Xen has given us zero headaches.  It's taken a
while, but (at least from the QEMU project's POV) the wait has been
worthwhile.

I would hope that HAXN support would be almost as easy as adding
haxn-all.c and haxn-stub.c files, and a few small changes to cpus.c and
vl.c.

Paolo

Re: [Qemu-devel] [RFC] Hardware acceleration for Windows and OS X?

[Stefan Weil]

Am 21.08.2013 11:11, schrieb Paolo Bonzini:
> Still, I must say merging Xen has given us zero headaches.  It's taken a
> while, but (at least from the QEMU project's POV) the wait has been
> worthwhile.
>
> I would hope that HAXN support would be almost as easy as adding
> haxn-all.c and haxn-stub.c files, and a few small changes to cpus.c and
> vl.c.
>
> Paolo

A first version which integrates HAXM in QEMU's build needs these changes:

 configure                   |   13 +
 include/sysemu/hax.h        |   50 ++
 target-i386/Makefile.objs   |    2 +
 target-i386/cpu.h           |    4 +
 target-i386/hax-all.c       | 1080
+++++++++++++++++++++++++++++++++++++++++++
 target-i386/hax-darwin.c    |  322 +++++++++++++
 target-i386/hax-darwin.h    |   86 ++++
 target-i386/hax-i386.h      |   93 ++++
 target-i386/hax-interface.h |  418 +++++++++++++++++
 target-i386/hax-windows.c   |  528 +++++++++++++++++++++
 target-i386/hax-windows.h   |   68 +++
 vl.c                        |    6 +-

(stub file(s) still missing, code completely untested,
maybe hax-all.c should be moved)

Regards,
Stefan

Re: [Qemu-devel] [RFC] Hardware acceleration for Windows and OS X?

[Paolo Bonzini]

Il 21/08/2013 11:33, Stefan Weil ha scritto:
> Am 21.08.2013 11:11, schrieb Paolo Bonzini:
>> Still, I must say merging Xen has given us zero headaches.  It's taken a
>> while, but (at least from the QEMU project's POV) the wait has been
>> worthwhile.
>>
>> I would hope that HAXN support would be almost as easy as adding
>> haxn-all.c and haxn-stub.c files, and a few small changes to cpus.c and
>> vl.c.
> 
> A first version which integrates HAXM in QEMU's build needs these changes:
> 
>  configure                   |   13 +
>  include/sysemu/hax.h        |   50 ++
>  target-i386/Makefile.objs   |    2 +
>  target-i386/cpu.h           |    4 +
>  target-i386/hax-all.c       | 1080
> +++++++++++++++++++++++++++++++++++++++++++
>  target-i386/hax-darwin.c    |  322 +++++++++++++
>  target-i386/hax-darwin.h    |   86 ++++
>  target-i386/hax-i386.h      |   93 ++++
>  target-i386/hax-interface.h |  418 +++++++++++++++++
>  target-i386/hax-windows.c   |  528 +++++++++++++++++++++
>  target-i386/hax-windows.h   |   68 +++
>  vl.c                        |    6 +-
> 
> (stub file(s) still missing, code completely untested,
> maybe hax-all.c should be moved)

I'm surprised there's no change in cpus.c, but it looks promising.  Show
us the code! :)

Paolo