[Qemu-devel] [PATCH] target-i386: upgrade family number of qemu64 CPU model to 15

[Qemu-devel] [PATCH] target-i386: upgrade family number of qemu64 CPU model to 15

[Ren, Yongjie]

qemu64 is the default CPU model for QEMU/KVM on x86-64, but its family number 
definition is not accurate. The vendor of 'qemu64' is defined as 'AMD'. For AMD 
processors, AMD K8 (i.e. Family 15) firstly introduced SSE3 instruction set. 
'qemu64' already has 'CPUID_EXT_SSE3' in its ext_features, but its family 
number is only 6 which is not equal to or greater than 15.
So, upgrade the CPU family number of qemu64 model to 15.
After this upgrade, 32bit Windows 8 can boot up with qemu64 CPU model.
This upgrade can be also a fix to the below bug about 32bit Windows 8 booting.
https://bugs.launchpad.net/qemu/+bug/1007269

Signed-off-by: Yongjie Ren <yongjie.ren@intel.com>
---
 target-i386/cpu.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/target-i386/cpu.c b/target-i386/cpu.c
index e2302d8..3f8e6c3 100644
--- a/target-i386/cpu.c
+++ b/target-i386/cpu.c
@@ -420,7 +420,7 @@ static x86_def_t builtin_x86_defs[] = {
         .name = "qemu64",
         .level = 4,
         .vendor = CPUID_VENDOR_AMD,
-        .family = 6,
+        .family = 15,
         .model = 2,
         .stepping = 3,
         .features = PPRO_FEATURES |
--
1.7.1

Re: [Qemu-devel] [PATCH] target-i386: upgrade family number of qemu64 CPU model to 15

[Gleb Natapov]

On Thu, Apr 18, 2013 at 10:45:38AM +0000, Ren, Yongjie wrote:
> qemu64 is the default CPU model for QEMU/KVM on x86-64, but its family number 
> definition is not accurate. The vendor of 'qemu64' is defined as 'AMD'. For AMD 
> processors, AMD K8 (i.e. Family 15) firstly introduced SSE3 instruction set. 
> 'qemu64' already has 'CPUID_EXT_SSE3' in its ext_features, but its family 
> number is only 6 which is not equal to or greater than 15.
> So, upgrade the CPU family number of qemu64 model to 15.
> After this upgrade, 32bit Windows 8 can boot up with qemu64 CPU model.
> This upgrade can be also a fix to the below bug about 32bit Windows 8 booting.
> https://bugs.launchpad.net/qemu/+bug/1007269

Changing cpuid info is not that simple since this is guest visible
information and can impact migration. Old machine models should still
expose the old family.

> 
> Signed-off-by: Yongjie Ren <yongjie.ren@intel.com>
> ---
>  target-i386/cpu.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
> 
> diff --git a/target-i386/cpu.c b/target-i386/cpu.c
> index e2302d8..3f8e6c3 100644
> --- a/target-i386/cpu.c
> +++ b/target-i386/cpu.c
> @@ -420,7 +420,7 @@ static x86_def_t builtin_x86_defs[] = {
>          .name = "qemu64",
>          .level = 4,
>          .vendor = CPUID_VENDOR_AMD,
> -        .family = 6,
> +        .family = 15,
>          .model = 2,
>          .stepping = 3,
>          .features = PPRO_FEATURES |
> --
> 1.7.1

--
			Gleb.

Re: [Qemu-devel] [PATCH] target-i386: upgrade family number of qemu64 CPU model to 15

[Ren, Yongjie]

> -----Original Message-----
> From: Gleb Natapov [mailto:gleb@redhat.com]
> Sent: Thursday, April 18, 2013 7:08 PM
> To: Ren, Yongjie
> Cc: Marcelo Tosatti; kvm@vger.kernel.org; qemu-devel@nongnu.org;
> ehabkost@redhat.com
> Subject: Re: [PATCH] target-i386: upgrade family number of qemu64 CPU
> model to 15
> 
> On Thu, Apr 18, 2013 at 10:45:38AM +0000, Ren, Yongjie wrote:
> > qemu64 is the default CPU model for QEMU/KVM on x86-64, but its
> family number
> > definition is not accurate. The vendor of 'qemu64' is defined as 'AMD'.
> For AMD
> > processors, AMD K8 (i.e. Family 15) firstly introduced SSE3 instruction
> set.
> > 'qemu64' already has 'CPUID_EXT_SSE3' in its ext_features, but its family
> > number is only 6 which is not equal to or greater than 15.
> > So, upgrade the CPU family number of qemu64 model to 15.
> > After this upgrade, 32bit Windows 8 can boot up with qemu64 CPU
> model.
> > This upgrade can be also a fix to the below bug about 32bit Windows 8
> booting.
> > https://bugs.launchpad.net/qemu/+bug/1007269
> 
> Changing cpuid info is not that simple since this is guest visible
> information and can impact migration. Old machine models should still
> expose the old family.
> 
Sorry, I didn't consider migration issue. I just think the cupid info for qemu64
is not compatible with real-world CPU. Family 6 AMD CPU should not have
SSE3 but it has SSE3 in qemu64 model.
If it's not so simple to modify cupid info, just ignore my patch.  :-)

> >
> > Signed-off-by: Yongjie Ren <yongjie.ren@intel.com>
> > ---
> >  target-i386/cpu.c |    2 +-
> >  1 files changed, 1 insertions(+), 1 deletions(-)
> >
> > diff --git a/target-i386/cpu.c b/target-i386/cpu.c
> > index e2302d8..3f8e6c3 100644
> > --- a/target-i386/cpu.c
> > +++ b/target-i386/cpu.c
> > @@ -420,7 +420,7 @@ static x86_def_t builtin_x86_defs[] = {
> >          .name = "qemu64",
> >          .level = 4,
> >          .vendor = CPUID_VENDOR_AMD,
> > -        .family = 6,
> > +        .family = 15,
> >          .model = 2,
> >          .stepping = 3,
> >          .features = PPRO_FEATURES |
> > --
> > 1.7.1
> 
> --
> 			Gleb.

Re: [Qemu-devel] [PATCH] target-i386: upgrade family number of qemu64 CPU model to 15

[Eduardo Habkost]

On Thu, Apr 18, 2013 at 02:47:21PM +0000, Ren, Yongjie wrote:
> > -----Original Message-----
> > From: Gleb Natapov [mailto:gleb@redhat.com]
> > Sent: Thursday, April 18, 2013 7:08 PM
> > To: Ren, Yongjie
> > Cc: Marcelo Tosatti; kvm@vger.kernel.org; qemu-devel@nongnu.org;
> > ehabkost@redhat.com
> > Subject: Re: [PATCH] target-i386: upgrade family number of qemu64 CPU
> > model to 15
> > 
> > On Thu, Apr 18, 2013 at 10:45:38AM +0000, Ren, Yongjie wrote:
> > > qemu64 is the default CPU model for QEMU/KVM on x86-64, but its
> > family number
> > > definition is not accurate. The vendor of 'qemu64' is defined as 'AMD'.
> > For AMD
> > > processors, AMD K8 (i.e. Family 15) firstly introduced SSE3 instruction
> > set.
> > > 'qemu64' already has 'CPUID_EXT_SSE3' in its ext_features, but its family
> > > number is only 6 which is not equal to or greater than 15.
> > > So, upgrade the CPU family number of qemu64 model to 15.
> > > After this upgrade, 32bit Windows 8 can boot up with qemu64 CPU
> > model.
> > > This upgrade can be also a fix to the below bug about 32bit Windows 8
> > booting.
> > > https://bugs.launchpad.net/qemu/+bug/1007269
> > 
> > Changing cpuid info is not that simple since this is guest visible
> > information and can impact migration. Old machine models should still
> > expose the old family.
> > 
> Sorry, I didn't consider migration issue. I just think the cupid info for qemu64
> is not compatible with real-world CPU. Family 6 AMD CPU should not have
> SSE3 but it has SSE3 in qemu64 model.
> If it's not so simple to modify cupid info, just ignore my patch.  :-)

We have other fixes for CPUID information sent by other people that will
require special handling to allow migration compatibility as well.

The ongoing static-properties and X86CPU subclasses work would help us
on this, but they probably won't be in 1.5. So I am planning to gather
those CPUID fixes and make them use a simple (and temporary)
compatibility mechanism and submit them next week.

-- 
Eduardo