KVM: x86: Add a help to dedup loading guest/host XCR0 and XSS

[PATCH] KVM: x86: Add a help to dedup loading guest/host XCR0 and XSS

Posted by Binbin Wu 1 month, 1 week ago

Add and use a helper, kvm_load_xfeatures(), to dedup the code that loads
guest/host xfeatures by passing XCR0 and XSS values accordingly.

No functional change intended.

Signed-off-by: Binbin Wu <binbin.wu@linux.intel.com>
---
The patch is based on the patch series
"KVM: x86: Cleanup #MC and XCR0/XSS/PKRU handling" [1], which is applied on 
top of kvm-x86 next branch(commit a996dd2a5e1ec54dcf7d7b93915ea3f97e14e68a).

[1] https://lore.kernel.org/all/20251030224246.3456492-1-seanjc@google.com
---
 arch/x86/kvm/x86.c | 25 +++++--------------------
 1 file changed, 5 insertions(+), 20 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index b8df02a083b2..35fc1333e198 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -1203,33 +1203,18 @@ void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 }
 EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_lmsw);
 
-static void kvm_load_guest_xfeatures(struct kvm_vcpu *vcpu)
+static void kvm_load_xfeatures(struct kvm_vcpu *vcpu, u64 xcr0, u64 xss)
 {
 	if (vcpu->arch.guest_state_protected)
 		return;
 
 	if (kvm_is_cr4_bit_set(vcpu, X86_CR4_OSXSAVE)) {
 		if (vcpu->arch.xcr0 != kvm_host.xcr0)
-			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
+			xsetbv(XCR_XFEATURE_ENABLED_MASK, xcr0);
 
 		if (guest_cpu_cap_has(vcpu, X86_FEATURE_XSAVES) &&
 		    vcpu->arch.ia32_xss != kvm_host.xss)
-			wrmsrq(MSR_IA32_XSS, vcpu->arch.ia32_xss);
-	}
-}
-
-static void kvm_load_host_xfeatures(struct kvm_vcpu *vcpu)
-{
-	if (vcpu->arch.guest_state_protected)
-		return;
-
-	if (kvm_is_cr4_bit_set(vcpu, X86_CR4_OSXSAVE)) {
-		if (vcpu->arch.xcr0 != kvm_host.xcr0)
-			xsetbv(XCR_XFEATURE_ENABLED_MASK, kvm_host.xcr0);
-
-		if (guest_cpu_cap_has(vcpu, X86_FEATURE_XSAVES) &&
-		    vcpu->arch.ia32_xss != kvm_host.xss)
-			wrmsrq(MSR_IA32_XSS, kvm_host.xss);
+			wrmsrq(MSR_IA32_XSS, xss);
 	}
 }
 
@@ -11310,7 +11295,7 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
 	if (vcpu->arch.guest_fpu.xfd_err)
 		wrmsrq(MSR_IA32_XFD_ERR, vcpu->arch.guest_fpu.xfd_err);
 
-	kvm_load_guest_xfeatures(vcpu);
+	kvm_load_xfeatures(vcpu, vcpu->arch.xcr0, vcpu->arch.ia32_xss);
 
 	if (unlikely(vcpu->arch.switch_db_regs &&
 		     !(vcpu->arch.switch_db_regs & KVM_DEBUGREG_AUTO_SWITCH))) {
@@ -11406,7 +11391,7 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
 	vcpu->mode = OUTSIDE_GUEST_MODE;
 	smp_wmb();
 
-	kvm_load_host_xfeatures(vcpu);
+	kvm_load_xfeatures(vcpu, kvm_host.xcr0, kvm_host.xss);
 
 	/*
 	 * Sync xfd before calling handle_exit_irqoff() which may

base-commit: a996dd2a5e1ec54dcf7d7b93915ea3f97e14e68a
prerequisite-patch-id: 9aafd634f0ab2033d7b032e227d356777469e046
prerequisite-patch-id: 656ce1f5aa97c77a9cf6125713707a5007b2c7ba
prerequisite-patch-id: d6328b8c0fdb8593bb534ab7378821edcf9f639d
prerequisite-patch-id: c7f36d1cedc4ae6416223d2225460944629b3d4f
-- 
2.46.0

Re: [PATCH] KVM: x86: Add a help to dedup loading guest/host XCR0 and XSS

Posted by Chao Gao 1 month, 1 week ago

s/help/helper in the subject.

On Thu, Nov 06, 2025 at 06:11:38PM +0800, Binbin Wu wrote:
>Add and use a helper, kvm_load_xfeatures(), to dedup the code that loads
>guest/host xfeatures by passing XCR0 and XSS values accordingly.
>
>No functional change intended.
>
>Signed-off-by: Binbin Wu <binbin.wu@linux.intel.com>

Reviewed-by: Chao Gao <chao.gao@intel.com>

<snip>

>@@ -11406,7 +11391,7 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
> 	vcpu->mode = OUTSIDE_GUEST_MODE;
> 	smp_wmb();
> 
>-	kvm_load_host_xfeatures(vcpu);
>+	kvm_load_xfeatures(vcpu, kvm_host.xcr0, kvm_host.xss);

Nit: given that xcr0/xss are either guest or host values, would it be slightly
better for this helper to accept a boolean (e.g., bool load_guest) to convey
that the API loads guest (or host) values rather than arbitrary xcr0/xss
values? like fpu_swap_kvm_fpstate().

static void kvm_load_xfeatures(struct kvm_vcpu *vcpu, bool load_guest)
{
	u64 xcr0 = load_guest ? vcpu->arch.xcr0 : kvm_host.xcr0;
	u64 xss  = load_guest ? vcpu->arch.ia32_xss : kvm_host.xss;

	if (vcpu->arch.guest_state_protected)
		return;

> 	/*
> 	 * Sync xfd before calling handle_exit_irqoff() which may
>
>base-commit: a996dd2a5e1ec54dcf7d7b93915ea3f97e14e68a
>prerequisite-patch-id: 9aafd634f0ab2033d7b032e227d356777469e046
>prerequisite-patch-id: 656ce1f5aa97c77a9cf6125713707a5007b2c7ba
>prerequisite-patch-id: d6328b8c0fdb8593bb534ab7378821edcf9f639d
>prerequisite-patch-id: c7f36d1cedc4ae6416223d2225460944629b3d4f
>-- 
>2.46.0
>
>

Re: [PATCH] KVM: x86: Add a help to dedup loading guest/host XCR0 and XSS

Posted by Xiaoyao Li 1 month, 1 week ago

On 11/7/2025 11:14 AM, Chao Gao wrote:
> s/help/helper in the subject.
> 
> On Thu, Nov 06, 2025 at 06:11:38PM +0800, Binbin Wu wrote:
>> Add and use a helper, kvm_load_xfeatures(), to dedup the code that loads
>> guest/host xfeatures by passing XCR0 and XSS values accordingly.
>>
>> No functional change intended.
>>
>> Signed-off-by: Binbin Wu <binbin.wu@linux.intel.com>
> 
> Reviewed-by: Chao Gao <chao.gao@intel.com>
> 
> <snip>
> 
>> @@ -11406,7 +11391,7 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
>> 	vcpu->mode = OUTSIDE_GUEST_MODE;
>> 	smp_wmb();
>>
>> -	kvm_load_host_xfeatures(vcpu);
>> +	kvm_load_xfeatures(vcpu, kvm_host.xcr0, kvm_host.xss);
> 
> Nit: given that xcr0/xss are either guest or host values, would it be slightly
> better for this helper to accept a boolean (e.g., bool load_guest) to convey
> that the API loads guest (or host) values rather than arbitrary xcr0/xss
> values? like fpu_swap_kvm_fpstate().

This really is a good idea!

Allow arbitrary input but skip the write when vcpu->arch.{xcr0,ia32_xss} 
== kvm_host.{xcr0,ia32_xss} does introduce confusion.

> static void kvm_load_xfeatures(struct kvm_vcpu *vcpu, bool load_guest)
> {
> 	u64 xcr0 = load_guest ? vcpu->arch.xcr0 : kvm_host.xcr0;
> 	u64 xss  = load_guest ? vcpu->arch.ia32_xss : kvm_host.xss;
> 
> 	if (vcpu->arch.guest_state_protected)
> 		return;
> 
>> 	/*
>> 	 * Sync xfd before calling handle_exit_irqoff() which may
>>
>> base-commit: a996dd2a5e1ec54dcf7d7b93915ea3f97e14e68a
>> prerequisite-patch-id: 9aafd634f0ab2033d7b032e227d356777469e046
>> prerequisite-patch-id: 656ce1f5aa97c77a9cf6125713707a5007b2c7ba
>> prerequisite-patch-id: d6328b8c0fdb8593bb534ab7378821edcf9f639d
>> prerequisite-patch-id: c7f36d1cedc4ae6416223d2225460944629b3d4f
>> -- 
>> 2.46.0
>>
>>
>

Re: [PATCH] KVM: x86: Add a help to dedup loading guest/host XCR0 and XSS

Posted by Binbin Wu 1 month, 1 week ago


On 11/7/2025 11:14 AM, Chao Gao wrote:
> s/help/helper in the subject.
>
> On Thu, Nov 06, 2025 at 06:11:38PM +0800, Binbin Wu wrote:
>> Add and use a helper, kvm_load_xfeatures(), to dedup the code that loads
>> guest/host xfeatures by passing XCR0 and XSS values accordingly.
>>
>> No functional change intended.
>>
>> Signed-off-by: Binbin Wu <binbin.wu@linux.intel.com>
> Reviewed-by: Chao Gao <chao.gao@intel.com>
>
> <snip>
>
>> @@ -11406,7 +11391,7 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
>> 	vcpu->mode = OUTSIDE_GUEST_MODE;
>> 	smp_wmb();
>>
>> -	kvm_load_host_xfeatures(vcpu);
>> +	kvm_load_xfeatures(vcpu, kvm_host.xcr0, kvm_host.xss);
> Nit: given that xcr0/xss are either guest or host values, would it be slightly
> better for this helper to accept a boolean (e.g., bool load_guest) to convey
> that the API loads guest (or host) values rather than arbitrary xcr0/xss
> values? like fpu_swap_kvm_fpstate().

Make sense.

>
> static void kvm_load_xfeatures(struct kvm_vcpu *vcpu, bool load_guest)
> {
> 	u64 xcr0 = load_guest ? vcpu->arch.xcr0 : kvm_host.xcr0;
> 	u64 xss  = load_guest ? vcpu->arch.ia32_xss : kvm_host.xss;

Since they are only used once, I even want to open code them as:

static void kvm_load_xfeatures(struct kvm_vcpu *vcpu, bool load_guest)
{
         if (vcpu->arch.guest_state_protected)
                 return;

         if (kvm_is_cr4_bit_set(vcpu, X86_CR4_OSXSAVE)) {
                 if (vcpu->arch.xcr0 != kvm_host.xcr0)
                         xsetbv(XCR_XFEATURE_ENABLED_MASK,
                                load_guest ? vcpu->arch.xcr0 : kvm_host.xcr0);

                 if (guest_cpu_cap_has(vcpu, X86_FEATURE_XSAVES) &&
                     vcpu->arch.ia32_xss != kvm_host.xss)
                         wrmsrq(MSR_IA32_XSS,
                                load_guest ? vcpu->arch.ia32_xss : kvm_host.xss);
         }
}

>
> 	if (vcpu->arch.guest_state_protected)
> 		return;
>
>> 	/*
>> 	 * Sync xfd before calling handle_exit_irqoff() which may
>>
>> base-commit: a996dd2a5e1ec54dcf7d7b93915ea3f97e14e68a
>> prerequisite-patch-id: 9aafd634f0ab2033d7b032e227d356777469e046
>> prerequisite-patch-id: 656ce1f5aa97c77a9cf6125713707a5007b2c7ba
>> prerequisite-patch-id: d6328b8c0fdb8593bb534ab7378821edcf9f639d
>> prerequisite-patch-id: c7f36d1cedc4ae6416223d2225460944629b3d4f
>> -- 
>> 2.46.0
>>
>>

Re: [PATCH] KVM: x86: Add a help to dedup loading guest/host XCR0 and XSS

Posted by Sean Christopherson 1 month, 1 week ago

On Fri, Nov 07, 2025, Binbin Wu wrote:
> 
> 
> On 11/7/2025 11:14 AM, Chao Gao wrote:
> > s/help/helper in the subject.
> > 
> > On Thu, Nov 06, 2025 at 06:11:38PM +0800, Binbin Wu wrote:
> > > Add and use a helper, kvm_load_xfeatures(), to dedup the code that loads
> > > guest/host xfeatures by passing XCR0 and XSS values accordingly.
> > > 
> > > No functional change intended.
> > > 
> > > Signed-off-by: Binbin Wu <binbin.wu@linux.intel.com>
> > Reviewed-by: Chao Gao <chao.gao@intel.com>
> > 
> > <snip>
> > 
> > > @@ -11406,7 +11391,7 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
> > > 	vcpu->mode = OUTSIDE_GUEST_MODE;
> > > 	smp_wmb();
> > > 
> > > -	kvm_load_host_xfeatures(vcpu);
> > > +	kvm_load_xfeatures(vcpu, kvm_host.xcr0, kvm_host.xss);
> > Nit: given that xcr0/xss are either guest or host values, would it be slightly
> > better for this helper to accept a boolean (e.g., bool load_guest) to convey
> > that the API loads guest (or host) values rather than arbitrary xcr0/xss
> > values? like fpu_swap_kvm_fpstate().
> 
> Make sense.

I don't love passing true/false, but I Xiaoyao does make a good a point that
subtly requiring the caller to pass vcpu->arch.xcr0 vs. kvm_host.xcr0 is weird
and confusing.

> > static void kvm_load_xfeatures(struct kvm_vcpu *vcpu, bool load_guest)
> > {
> > 	u64 xcr0 = load_guest ? vcpu->arch.xcr0 : kvm_host.xcr0;
> > 	u64 xss  = load_guest ? vcpu->arch.ia32_xss : kvm_host.xss;
> 
> Since they are only used once, I even want to open code them as: 

+100, I find your version much more intuitive.

> static void kvm_load_xfeatures(struct kvm_vcpu *vcpu, bool load_guest)
> {
>         if (vcpu->arch.guest_state_protected)
>                 return;
> 
>         if (kvm_is_cr4_bit_set(vcpu, X86_CR4_OSXSAVE)) {
>                 if (vcpu->arch.xcr0 != kvm_host.xcr0)
>                         xsetbv(XCR_XFEATURE_ENABLED_MASK,
>                                load_guest ? vcpu->arch.xcr0 : kvm_host.xcr0);
> 
>                 if (guest_cpu_cap_has(vcpu, X86_FEATURE_XSAVES) &&
>                     vcpu->arch.ia32_xss != kvm_host.xss)
>                         wrmsrq(MSR_IA32_XSS,
>                                load_guest ? vcpu->arch.ia32_xss : kvm_host.xss);
>         }
> }
> 
> > 
> > 	if (vcpu->arch.guest_state_protected)
> > 		return;
> > 
> > > 	/*
> > > 	 * Sync xfd before calling handle_exit_irqoff() which may
> > > 
> > > base-commit: a996dd2a5e1ec54dcf7d7b93915ea3f97e14e68a
> > > prerequisite-patch-id: 9aafd634f0ab2033d7b032e227d356777469e046
> > > prerequisite-patch-id: 656ce1f5aa97c77a9cf6125713707a5007b2c7ba
> > > prerequisite-patch-id: d6328b8c0fdb8593bb534ab7378821edcf9f639d
> > > prerequisite-patch-id: c7f36d1cedc4ae6416223d2225460944629b3d4f
> > > -- 
> > > 2.46.0
> > > 
> > > 
>