[v4] Enable FRED with KVM VMX

[PATCH v4 15/19] KVM: x86: Allow FRED/LKGS to be advertised to guests

Posted by Xin Li (Intel) 8 months, 3 weeks ago

From: Xin Li <xin3.li@intel.com>

Allow FRED/LKGS to be advertised to guests after changes required to
enable FRED in a KVM guest are in place.

LKGS is introduced with FRED to completely eliminate the need to swapgs
explicilty, because

1) FRED transitions ensure that an operating system can always operate
   with its own GS base address.

2) LKGS behaves like the MOV to GS instruction except that it loads
   the base address into the IA32_KERNEL_GS_BASE MSR instead of the
   GS segment’s descriptor cache, which is exactly what Linux kernel
   does to load a user level GS base.  Thus there is no need to SWAPGS
   away from the kernel GS base and an execution of SWAPGS causes #UD
   if FRED transitions are enabled.

A FRED CPU must enumerate LKGS.  When LKGS is not available, FRED must
not be enabled.

Signed-off-by: Xin Li <xin3.li@intel.com>
Signed-off-by: Xin Li (Intel) <xin@zytor.com>
Tested-by: Shan Kang <shan.kang@intel.com>
---
 arch/x86/kvm/cpuid.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
index 5e4d4934c0d3..8f290273aee1 100644
--- a/arch/x86/kvm/cpuid.c
+++ b/arch/x86/kvm/cpuid.c
@@ -992,6 +992,8 @@ void kvm_set_cpu_caps(void)
 		F(FZRM),
 		F(FSRS),
 		F(FSRC),
+		F(FRED),
+		F(LKGS),
 		F(AMX_FP16),
 		F(AVX_IFMA),
 		F(LAM),
-- 
2.48.1

Re: [PATCH v4 15/19] KVM: x86: Allow FRED/LKGS to be advertised to guests

Posted by Sean Christopherson 5 months, 3 weeks ago

The shortlog (and changelog intro) are wrong.  KVM isn't allowing FRED/LKGS to
be advertised to the guest.  Userspace can advertise whatever it wants.  The guest
will break badly without KVM support, but that doesn't stop userspace from
advertising a bogus vCPU model.

  KVM: x86: Advertise support for FRED/LKGS to userspace

On Fri, Mar 28, 2025, Xin Li (Intel) wrote:
> From: Xin Li <xin3.li@intel.com>
> 
> Allow FRED/LKGS to be advertised to guests after changes required to

Please explain what LKGS is early in the changelog.  I assumed it was a feature
of sorts; turns out it's a new instruction.

Actually, why wait this long to enumerate support for LKGS?  I.e. why not have a
patch at the head of the series to enumerate support for LKGS?  IIUC, LKGS doesn't
depend on FRED.

> enable FRED in a KVM guest are in place.
> 
> LKGS is introduced with FRED to completely eliminate the need to swapgs
> explicilty, because
> 
> 1) FRED transitions ensure that an operating system can always operate
>    with its own GS base address.
> 
> 2) LKGS behaves like the MOV to GS instruction except that it loads
>    the base address into the IA32_KERNEL_GS_BASE MSR instead of the
>    GS segment’s descriptor cache, which is exactly what Linux kernel
>    does to load a user level GS base.  Thus there is no need to SWAPGS
>    away from the kernel GS base and an execution of SWAPGS causes #UD
>    if FRED transitions are enabled.
> 
> A FRED CPU must enumerate LKGS.  When LKGS is not available, FRED must
> not be enabled.
> 
> Signed-off-by: Xin Li <xin3.li@intel.com>
> Signed-off-by: Xin Li (Intel) <xin@zytor.com>
> Tested-by: Shan Kang <shan.kang@intel.com>
> ---
>  arch/x86/kvm/cpuid.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
> index 5e4d4934c0d3..8f290273aee1 100644
> --- a/arch/x86/kvm/cpuid.c
> +++ b/arch/x86/kvm/cpuid.c
> @@ -992,6 +992,8 @@ void kvm_set_cpu_caps(void)
>  		F(FZRM),
>  		F(FSRS),
>  		F(FSRC),
> +		F(FRED),
> +		F(LKGS),

These need to be X86_64_F, no?

>  		F(AMX_FP16),
>  		F(AVX_IFMA),
>  		F(LAM),
> -- 
> 2.48.1
>

Re: [PATCH v4 15/19] KVM: x86: Allow FRED/LKGS to be advertised to guests

Posted by Xin Li 5 months, 3 weeks ago

On 6/24/2025 9:38 AM, Sean Christopherson wrote:
> The shortlog (and changelog intro) are wrong.  KVM isn't allowing FRED/LKGS to
> be advertised to the guest.  Userspace can advertise whatever it wants.  The guest
> will break badly without KVM support, but that doesn't stop userspace from
> advertising a bogus vCPU model.
> 
>    KVM: x86: Advertise support for FRED/LKGS to userspace
> 
> On Fri, Mar 28, 2025, Xin Li (Intel) wrote:
>> From: Xin Li <xin3.li@intel.com>
>>
>> Allow FRED/LKGS to be advertised to guests after changes required to
> 
> Please explain what LKGS is early in the changelog.  I assumed it was a feature
> of sorts; turns out it's a new instruction.
> 
> Actually, why wait this long to enumerate support for LKGS?  I.e. why not have a
> patch at the head of the series to enumerate support for LKGS?  IIUC, LKGS doesn't
> depend on FRED.

I will send LKGS as a separate patch, thus if you prefer you can take
it before the KVM FRED patch set.

> 
>> enable FRED in a KVM guest are in place.
>>
>> LKGS is introduced with FRED to completely eliminate the need to swapgs
>> explicilty, because
>>
>> 1) FRED transitions ensure that an operating system can always operate
>>     with its own GS base address.
>>
>> 2) LKGS behaves like the MOV to GS instruction except that it loads
>>     the base address into the IA32_KERNEL_GS_BASE MSR instead of the
>>     GS segment’s descriptor cache, which is exactly what Linux kernel
>>     does to load a user level GS base.  Thus there is no need to SWAPGS
>>     away from the kernel GS base and an execution of SWAPGS causes #UD
>>     if FRED transitions are enabled.
>>
>> A FRED CPU must enumerate LKGS.  When LKGS is not available, FRED must
>> not be enabled.
>>
>> Signed-off-by: Xin Li <xin3.li@intel.com>
>> Signed-off-by: Xin Li (Intel) <xin@zytor.com>
>> Tested-by: Shan Kang <shan.kang@intel.com>
>> ---
>>   arch/x86/kvm/cpuid.c | 2 ++
>>   1 file changed, 2 insertions(+)
>>
>> diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
>> index 5e4d4934c0d3..8f290273aee1 100644
>> --- a/arch/x86/kvm/cpuid.c
>> +++ b/arch/x86/kvm/cpuid.c
>> @@ -992,6 +992,8 @@ void kvm_set_cpu_caps(void)
>>   		F(FZRM),
>>   		F(FSRS),
>>   		F(FSRC),
>> +		F(FRED),
>> +		F(LKGS),
> 
> These need to be X86_64_F, no?

Yes.  Both LKGS and FRED are 64-bit only features.

However I assume KVM is 64-bit only now, so X86_64_F is essentially F,
right?

Thanks!
     Xin

Re: [PATCH v4 15/19] KVM: x86: Allow FRED/LKGS to be advertised to guests

Posted by Sean Christopherson 5 months, 3 weeks ago

On Wed, Jun 25, 2025, Xin Li wrote:
> On 6/24/2025 9:38 AM, Sean Christopherson wrote:
> > > ---
> > >   arch/x86/kvm/cpuid.c | 2 ++
> > >   1 file changed, 2 insertions(+)
> > > 
> > > diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
> > > index 5e4d4934c0d3..8f290273aee1 100644
> > > --- a/arch/x86/kvm/cpuid.c
> > > +++ b/arch/x86/kvm/cpuid.c
> > > @@ -992,6 +992,8 @@ void kvm_set_cpu_caps(void)
> > >   		F(FZRM),
> > >   		F(FSRS),
> > >   		F(FSRC),
> > > +		F(FRED),
> > > +		F(LKGS),
> > 
> > These need to be X86_64_F, no?
> 
> Yes.  Both LKGS and FRED are 64-bit only features.
> 
> However I assume KVM is 64-bit only now, so X86_64_F is essentially F,
> right?

Nope, KVM still supports 32-bit builds.  There are plans/efforts to kill off 32-bit
KVM x86, but we're not quite there yet.