1. Patchew
  2. linux
  3. [RFC PATCH 00/22] KVM: combined patchset for MBEC/GMET support
  4. View patch

[PATCH 18/22] KVM: SVM: add GMET bit definitions

Paolo Bonzini posted 22 patches 2 weeks ago
There is a newer version of this series
[PATCH 18/22] KVM: SVM: add GMET bit definitions
Posted by Paolo Bonzini 2 weeks ago 
GMET (Guest Mode Execute Trap) is an AMD virtualization feature,
essentially the nested paging version of SMEP.  Hyper-V uses it;
add it in preparation for making it available to hypervisors
running under KVM.

Cc: Borislav Petkov (AMD) <bp@alien8.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 arch/x86/include/asm/cpufeatures.h | 1 +
 arch/x86/include/asm/svm.h         | 1 +
 2 files changed, 2 insertions(+)

diff --git a/arch/x86/include/asm/cpufeatures.h b/arch/x86/include/asm/cpufeatures.h
index 123c023fe42c..95469c7d357f 100644
--- a/arch/x86/include/asm/cpufeatures.h
+++ b/arch/x86/include/asm/cpufeatures.h
@@ -382,6 +382,7 @@
 #define X86_FEATURE_AVIC		(15*32+13) /* "avic" Virtual Interrupt Controller */
 #define X86_FEATURE_V_VMSAVE_VMLOAD	(15*32+15) /* "v_vmsave_vmload" Virtual VMSAVE VMLOAD */
 #define X86_FEATURE_VGIF		(15*32+16) /* "vgif" Virtual GIF */
+#define X86_FEATURE_GMET		(15*32+17) /* "gmet" Guest Mode Execution Trap */
 #define X86_FEATURE_X2AVIC		(15*32+18) /* "x2avic" Virtual x2apic */
 #define X86_FEATURE_V_SPEC_CTRL		(15*32+20) /* "v_spec_ctrl" Virtual SPEC_CTRL */
 #define X86_FEATURE_VNMI		(15*32+25) /* "vnmi" Virtual NMI */
diff --git a/arch/x86/include/asm/svm.h b/arch/x86/include/asm/svm.h
index e2a29e1144a7..47353bef947c 100644
--- a/arch/x86/include/asm/svm.h
+++ b/arch/x86/include/asm/svm.h
@@ -239,6 +239,7 @@ struct __attribute__ ((__packed__)) vmcb_control_area {
 #define SVM_NESTED_CTL_NP_ENABLE	BIT(0)
 #define SVM_NESTED_CTL_SEV_ENABLE	BIT(1)
 #define SVM_NESTED_CTL_SEV_ES_ENABLE	BIT(2)
+#define SVM_NESTED_CTL_GMET_ENABLE	BIT(3)
 
 
 #define SVM_TSC_RATIO_RSVD	0xffffff0000000000ULL
-- 
2.52.0
Re: [PATCH 18/22] KVM: SVM: add GMET bit definitions
Posted by Borislav Petkov 1 week, 4 days ago 
On Sat, Mar 21, 2026 at 01:09:27AM +0100, Paolo Bonzini wrote:
> GMET (Guest Mode Execute Trap) is an AMD virtualization feature,
> essentially the nested paging version of SMEP.  Hyper-V uses it;
> add it in preparation for making it available to hypervisors
> running under KVM.
> 
> Cc: Borislav Petkov (AMD) <bp@alien8.de>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>  arch/x86/include/asm/cpufeatures.h | 1 +
>  arch/x86/include/asm/svm.h         | 1 +
>  2 files changed, 2 insertions(+)
> 
> diff --git a/arch/x86/include/asm/cpufeatures.h b/arch/x86/include/asm/cpufeatures.h
> index 123c023fe42c..95469c7d357f 100644
> --- a/arch/x86/include/asm/cpufeatures.h
> +++ b/arch/x86/include/asm/cpufeatures.h
> @@ -382,6 +382,7 @@
>  #define X86_FEATURE_AVIC		(15*32+13) /* "avic" Virtual Interrupt Controller */
>  #define X86_FEATURE_V_VMSAVE_VMLOAD	(15*32+15) /* "v_vmsave_vmload" Virtual VMSAVE VMLOAD */
>  #define X86_FEATURE_VGIF		(15*32+16) /* "vgif" Virtual GIF */
> +#define X86_FEATURE_GMET		(15*32+17) /* "gmet" Guest Mode Execution Trap */

s!"gmet"!!

With that:

Acked-by: Borislav Petkov (AMD) <bp@alien8.de>

>  #define X86_FEATURE_X2AVIC		(15*32+18) /* "x2avic" Virtual x2apic */
>  #define X86_FEATURE_V_SPEC_CTRL		(15*32+20) /* "v_spec_ctrl" Virtual SPEC_CTRL */
>  #define X86_FEATURE_VNMI		(15*32+25) /* "vnmi" Virtual NMI */
> diff --git a/arch/x86/include/asm/svm.h b/arch/x86/include/asm/svm.h
> index e2a29e1144a7..47353bef947c 100644
> --- a/arch/x86/include/asm/svm.h
> +++ b/arch/x86/include/asm/svm.h
> @@ -239,6 +239,7 @@ struct __attribute__ ((__packed__)) vmcb_control_area {
>  #define SVM_NESTED_CTL_NP_ENABLE	BIT(0)
>  #define SVM_NESTED_CTL_SEV_ENABLE	BIT(1)
>  #define SVM_NESTED_CTL_SEV_ES_ENABLE	BIT(2)
> +#define SVM_NESTED_CTL_GMET_ENABLE	BIT(3)
>  
>  
>  #define SVM_TSC_RATIO_RSVD	0xffffff0000000000ULL
> -- 

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
Re: [PATCH 18/22] KVM: SVM: add GMET bit definitions
Posted by Borislav Petkov 1 week, 6 days ago 
On Sat, Mar 21, 2026 at 01:09:27AM +0100, Paolo Bonzini wrote:
> GMET (Guest Mode Execute Trap) is an AMD virtualization feature,
> essentially the nested paging version of SMEP.  Hyper-V uses it;
> add it in preparation for making it available to hypervisors
> running under KVM.
> 
> Cc: Borislav Petkov (AMD) <bp@alien8.de>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>  arch/x86/include/asm/cpufeatures.h | 1 +
>  arch/x86/include/asm/svm.h         | 1 +
>  2 files changed, 2 insertions(+)
> 
> diff --git a/arch/x86/include/asm/cpufeatures.h b/arch/x86/include/asm/cpufeatures.h
> index 123c023fe42c..95469c7d357f 100644
> --- a/arch/x86/include/asm/cpufeatures.h
> +++ b/arch/x86/include/asm/cpufeatures.h
> @@ -382,6 +382,7 @@
>  #define X86_FEATURE_AVIC		(15*32+13) /* "avic" Virtual Interrupt Controller */
>  #define X86_FEATURE_V_VMSAVE_VMLOAD	(15*32+15) /* "v_vmsave_vmload" Virtual VMSAVE VMLOAD */
>  #define X86_FEATURE_VGIF		(15*32+16) /* "vgif" Virtual GIF */
> +#define X86_FEATURE_GMET		(15*32+17) /* "gmet" Guest Mode Execution Trap */
						      ^^^^^^

Why do you need to show it in /proc/cpuinfo?

>  #define X86_FEATURE_X2AVIC		(15*32+18) /* "x2avic" Virtual x2apic */
>  #define X86_FEATURE_V_SPEC_CTRL		(15*32+20) /* "v_spec_ctrl" Virtual SPEC_CTRL */
>  #define X86_FEATURE_VNMI		(15*32+25) /* "vnmi" Virtual NMI */
> diff --git a/arch/x86/include/asm/svm.h b/arch/x86/include/asm/svm.h
> index e2a29e1144a7..47353bef947c 100644
> --- a/arch/x86/include/asm/svm.h
> +++ b/arch/x86/include/asm/svm.h
> @@ -239,6 +239,7 @@ struct __attribute__ ((__packed__)) vmcb_control_area {
>  #define SVM_NESTED_CTL_NP_ENABLE	BIT(0)
>  #define SVM_NESTED_CTL_SEV_ENABLE	BIT(1)
>  #define SVM_NESTED_CTL_SEV_ES_ENABLE	BIT(2)
> +#define SVM_NESTED_CTL_GMET_ENABLE	BIT(3)
>  
>  
>  #define SVM_TSC_RATIO_RSVD	0xffffff0000000000ULL
> -- 

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
Re: [PATCH 18/22] KVM: SVM: add GMET bit definitions
Posted by Paolo Bonzini 1 week, 6 days ago 
On Sat, Mar 21, 2026 at 12:58 PM Borislav Petkov <bp@alien8.de> wrote:
> > @@ -382,6 +382,7 @@
> >  #define X86_FEATURE_AVIC             (15*32+13) /* "avic" Virtual Interrupt Controller */
> >  #define X86_FEATURE_V_VMSAVE_VMLOAD  (15*32+15) /* "v_vmsave_vmload" Virtual VMSAVE VMLOAD */
> >  #define X86_FEATURE_VGIF             (15*32+16) /* "vgif" Virtual GIF */
> > +#define X86_FEATURE_GMET             (15*32+17) /* "gmet" Guest Mode Execution Trap */
>                                                       ^^^^^^
>
> Why do you need to show it in /proc/cpuinfo?

I prefer to be consistent with all the other ones in word 15.

Paolo

> >  #define X86_FEATURE_X2AVIC           (15*32+18) /* "x2avic" Virtual x2apic */
> >  #define X86_FEATURE_V_SPEC_CTRL              (15*32+20) /* "v_spec_ctrl" Virtual SPEC_CTRL */
> >  #define X86_FEATURE_VNMI             (15*32+25) /* "vnmi" Virtual NMI */
Re: [PATCH 18/22] KVM: SVM: add GMET bit definitions
Posted by Borislav Petkov 1 week, 6 days ago 
On Sat, Mar 21, 2026 at 02:51:47PM +0100, Paolo Bonzini wrote:
> I prefer to be consistent with all the other ones in word 15.

Sorry, no, /proc/cpuinfo is an ABI.

See Documentation/arch/x86/cpuinfo.rst

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
Re: [PATCH 18/22] KVM: SVM: add GMET bit definitions
Posted by Paolo Bonzini 1 week, 4 days ago 
Il sab 21 mar 2026, 16:42 Borislav Petkov <bp@alien8.de> ha scritto:
>
> On Sat, Mar 21, 2026 at 02:51:47PM +0100, Paolo Bonzini wrote:
> > I prefer to be consistent with all the other ones in word 15.
>
> Sorry, no, /proc/cpuinfo is an ABI.

I am not sure what you mean by ABI in this context, maybe "there are
rules for whether to add stuff to cpuinfo.rst"? If so, GMET matches
what cpuinfo.rst says:

* the kernel knows about the feature enough to have an X86_FEATURE bit
* the kernel supports it and is currently making it available to userspace
* the hardware supports it.

The bullet that applies here is the second from the above list: 1)
this series makes the feature available for userspace and guests to
use it, 2) that requires explicit support, it's not just a bunch of
new instructions.

I agree that not all word 15 features should be added to
/proc/cpuinfo, if that's what you meant by referencing cpuinfo.rst.
For example these days something like AVIC would not be added to
/proc/cpuinfo because it's transparent to userspace and guests.

Paolo
Re: [PATCH 18/22] KVM: SVM: add GMET bit definitions
Posted by Borislav Petkov 1 week, 4 days ago 
On Mon, Mar 23, 2026 at 08:53:32AM +0100, Paolo Bonzini wrote:
> The bullet that applies here is the second from the above list: 1)
> this series makes the feature available for userspace and guests to
> use it,

Also, if you're going to do this, then this has to be dependent on Kconfig and
be disabled when the kernel support for gmet is not build-enabled.

This explains the issue more:

https://lore.kernel.org/r/cover.1774008873.git.m.wieczorretman@pm.me

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
Re: [PATCH 18/22] KVM: SVM: add GMET bit definitions
Posted by Borislav Petkov 1 week, 4 days ago 
On Mon, Mar 23, 2026 at 08:53:32AM +0100, Paolo Bonzini wrote:
> I am not sure what you mean by ABI in this context, maybe "there are

It means that once we show it in /proc/cpuinfo, we're stuck with it forever
because something in userspace might depend on it.

So we try to avoid that if it can be solved in a different way.

> rules for whether to add stuff to cpuinfo.rst"? If so, GMET matches
> what cpuinfo.rst says:
> 
> * the kernel knows about the feature enough to have an X86_FEATURE bit
> * the kernel supports it and is currently making it available to userspace
> * the hardware supports it.
> 
> The bullet that applies here is the second from the above list: 1)
> this series makes the feature available for userspace and guests to
> use it,

This is the key question: what in userspace is going to parse /proc/cpuinfo
and look for "gmet"?

Because parsing /proc/cpuinfo is the silliest option available but glibc does
it and for some things like, "did the kernel *actually* enable this?" it is
the only way.

If the above applies, can this be solved in a different way? CPUID,
guest_cpu_cap_has(), whatever...

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
Re: [PATCH 18/22] KVM: SVM: add GMET bit definitions
Posted by Paolo Bonzini 1 week, 4 days ago 
On Mon, Mar 23, 2026 at 1:17 PM Borislav Petkov <bp@alien8.de> wrote:
> This is the key question: what in userspace is going to parse /proc/cpuinfo
> and look for "gmet"?

To be honest probably not. They can open /dev/kvm and call
KVM_GET_SUPPORTED_CPUID to get the same information.

The difference is that /proc/cpuinfo will always have gmet even if the
module parameter is zero, while /dev/kvm won't. Based on your other
message, it seems like you want this to be a bug rather than a
feature, so let's shelve this, can you ack it with "" instead?

Paolo
Re: [PATCH 18/22] KVM: SVM: add GMET bit definitions
Posted by Borislav Petkov 1 week, 4 days ago 
On Mon, Mar 23, 2026 at 01:22:10PM +0100, Paolo Bonzini wrote:
> On Mon, Mar 23, 2026 at 1:17 PM Borislav Petkov <bp@alien8.de> wrote:
> > This is the key question: what in userspace is going to parse /proc/cpuinfo
> > and look for "gmet"?
> 
> To be honest probably not. They can open /dev/kvm and call
> KVM_GET_SUPPORTED_CPUID to get the same information.
> 
> The difference is that /proc/cpuinfo will always have gmet even if the
> module parameter is zero, while /dev/kvm won't. Based on your other
> message, it seems like you want this to be a bug rather than a
> feature, so let's shelve this, can you ack it with "" instead?

Right, you don't need to add any "" anymore - just a normal comment and that's
it. Lemme reply to the patch.

And *if* we ever decide that we need to show it to userspace, we can always do
that later but then we'll actually have a proper justification for doing so.

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette

Patchew 2.1-devel-ea86937

© 2016 - 2026 Red Hat, Inc.