KVM: x86: Update the states size cpuid even if XCR0/IA32_XSS is reset

[PATCH] KVM: x86: Update the states size cpuid even if XCR0/IA32_XSS is reset

Posted by Like Xu 4 years, 5 months ago

From: Like Xu <likexu@tencent.com>

XCR0 is reset to 1 by RESET but not INIT and IA32_XSS is zeroed by
both RESET and INIT. In both cases, the size in bytes of the XSAVE
area containing all states enabled by XCR0 or (XCRO | IA32_XSS)
needs to be updated.

Fixes: a554d207dc46 ("KVM: X86: Processor States following Reset or INIT")
Signed-off-by: Like Xu <likexu@tencent.com>
---
 arch/x86/kvm/x86.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 76b4803dd3bd..5748a57e1cb7 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -11134,6 +11134,7 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
 	struct kvm_cpuid_entry2 *cpuid_0x1;
 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
 	unsigned long new_cr0;
+	bool need_update_cpuid = false;
 
 	/*
 	 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
@@ -11199,6 +11200,8 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
 
 		vcpu->arch.msr_misc_features_enables = 0;
 
+		if (vcpu->arch.xcr0 != XFEATURE_MASK_FP)
+			need_update_cpuid = true;
 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
 	}
 
@@ -11216,6 +11219,8 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
 	cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
 	kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
 
+	if (vcpu->arch.ia32_xss)
+		need_update_cpuid = true;
 	vcpu->arch.ia32_xss = 0;
 
 	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
@@ -11264,6 +11269,9 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
 	 */
 	if (init_event)
 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
+
+	if (need_update_cpuid)
+		kvm_update_cpuid_runtime(vcpu);
 }
 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
 
-- 
2.33.1

Re: [PATCH] KVM: x86: Update the states size cpuid even if XCR0/IA32_XSS is reset

Posted by Sean Christopherson 4 years, 5 months ago

On Mon, Jan 17, 2022, Like Xu wrote:
> From: Like Xu <likexu@tencent.com>
> 
> XCR0 is reset to 1 by RESET but not INIT and IA32_XSS is zeroed by
> both RESET and INIT. In both cases, the size in bytes of the XSAVE
> area containing all states enabled by XCR0 or (XCRO | IA32_XSS)
> needs to be updated.
> 
> Fixes: a554d207dc46 ("KVM: X86: Processor States following Reset or INIT")
> Signed-off-by: Like Xu <likexu@tencent.com>
> ---
>  arch/x86/kvm/x86.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 76b4803dd3bd..5748a57e1cb7 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -11134,6 +11134,7 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
>  	struct kvm_cpuid_entry2 *cpuid_0x1;
>  	unsigned long old_cr0 = kvm_read_cr0(vcpu);
>  	unsigned long new_cr0;
> +	bool need_update_cpuid = false;
>  
>  	/*
>  	 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
> @@ -11199,6 +11200,8 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
>  
>  		vcpu->arch.msr_misc_features_enables = 0;
>  
> +		if (vcpu->arch.xcr0 != XFEATURE_MASK_FP)
> +			need_update_cpuid = true;
>  		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
>  	}
>  
> @@ -11216,6 +11219,8 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
>  	cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
>  	kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
>  
> +	if (vcpu->arch.ia32_xss)
> +		need_update_cpuid = true;

This means that kvm_set_msr_common()'s handling of MSR_IA32_XSS also needs to
update kvm_update_cpuid_runtime().  And then for bnoth XCR0 and XSS, I would very
strongly prefer that use the helpers to write the values and let the helpers call
kvm_update_cpuid_runtime().  Yes, that will mean kvm_update_cpuid_runtime() may be
called multiple times during INIT, but that's already true (CR4), and this isn't
exactly a fast path.

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 55518b7d3b96..22d4b1d15e94 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -11256,7 +11256,7 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)

                vcpu->arch.msr_misc_features_enables = 0;

-               vcpu->arch.xcr0 = XFEATURE_MASK_FP;
+               __kvm_set_xcr(vcpu, 0, XFEATURE_MASK_FP);
        }

        /* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
@@ -11273,7 +11273,7 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
        cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
        kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);

-       vcpu->arch.ia32_xss = 0;
+       __kvm_set_msr(vcpu, MSR_IA32_XSS, 0, true);

        static_call(kvm_x86_vcpu_reset)(vcpu, init_event);

Re: [PATCH] KVM: x86: Update the states size cpuid even if XCR0/IA32_XSS is reset

Posted by Like Xu 4 years, 5 months ago

On 19/1/2022 2:30 am, Sean Christopherson wrote:
> On Mon, Jan 17, 2022, Like Xu wrote:
>> From: Like Xu <likexu@tencent.com>
>>
>> XCR0 is reset to 1 by RESET but not INIT and IA32_XSS is zeroed by
>> both RESET and INIT. In both cases, the size in bytes of the XSAVE
>> area containing all states enabled by XCR0 or (XCRO | IA32_XSS)
>> needs to be updated.
>>
>> Fixes: a554d207dc46 ("KVM: X86: Processor States following Reset or INIT")
>> Signed-off-by: Like Xu <likexu@tencent.com>
>> ---
>>   arch/x86/kvm/x86.c | 8 ++++++++
>>   1 file changed, 8 insertions(+)
>>
>> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
>> index 76b4803dd3bd..5748a57e1cb7 100644
>> --- a/arch/x86/kvm/x86.c
>> +++ b/arch/x86/kvm/x86.c
>> @@ -11134,6 +11134,7 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
>>   	struct kvm_cpuid_entry2 *cpuid_0x1;
>>   	unsigned long old_cr0 = kvm_read_cr0(vcpu);
>>   	unsigned long new_cr0;
>> +	bool need_update_cpuid = false;
>>   
>>   	/*
>>   	 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
>> @@ -11199,6 +11200,8 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
>>   
>>   		vcpu->arch.msr_misc_features_enables = 0;
>>   
>> +		if (vcpu->arch.xcr0 != XFEATURE_MASK_FP)
>> +			need_update_cpuid = true;
>>   		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
>>   	}
>>   
>> @@ -11216,6 +11219,8 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
>>   	cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
>>   	kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
>>   
>> +	if (vcpu->arch.ia32_xss)
>> +		need_update_cpuid = true;
> 
> This means that kvm_set_msr_common()'s handling of MSR_IA32_XSS also needs to
> update kvm_update_cpuid_runtime().  And then for bnoth XCR0 and XSS, I would very
> strongly prefer that use the helpers to write the values and let the helpers call

Looks good to me and let me apply it in the next version.

> kvm_update_cpuid_runtime().  Yes, that will mean kvm_update_cpuid_runtime() may be
> called multiple times during INIT, but that's already true (CR4), and this isn't
> exactly a fast path.

An undisclosed lazy mechanism is under analyzed for performance gains.

> 
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 55518b7d3b96..22d4b1d15e94 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -11256,7 +11256,7 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
> 
>                  vcpu->arch.msr_misc_features_enables = 0;
> 
> -               vcpu->arch.xcr0 = XFEATURE_MASK_FP;
> +               __kvm_set_xcr(vcpu, 0, XFEATURE_MASK_FP);
>          }
> 
>          /* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
> @@ -11273,7 +11273,7 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
>          cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
>          kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
> 
> -       vcpu->arch.ia32_xss = 0;
> +       __kvm_set_msr(vcpu, MSR_IA32_XSS, 0, true);
> 
>          static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
> 
>

Re: [PATCH] KVM: x86: Update the states size cpuid even if XCR0/IA32_XSS is reset

Posted by Paolo Bonzini 4 years, 5 months ago

On 1/17/22 09:26, Like Xu wrote:
> From: Like Xu <likexu@tencent.com>
> 
> XCR0 is reset to 1 by RESET but not INIT and IA32_XSS is zeroed by
> both RESET and INIT. In both cases, the size in bytes of the XSAVE
> area containing all states enabled by XCR0 or (XCRO | IA32_XSS)
> needs to be updated.
> 
> Fixes: a554d207dc46 ("KVM: X86: Processor States following Reset or INIT")
> Signed-off-by: Like Xu <likexu@tencent.com>

Can you write a test case please?

Thanks,

Paolo

> ---
>   arch/x86/kvm/x86.c | 8 ++++++++
>   1 file changed, 8 insertions(+)
> 
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 76b4803dd3bd..5748a57e1cb7 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -11134,6 +11134,7 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
>   	struct kvm_cpuid_entry2 *cpuid_0x1;
>   	unsigned long old_cr0 = kvm_read_cr0(vcpu);
>   	unsigned long new_cr0;
> +	bool need_update_cpuid = false;
>   
>   	/*
>   	 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
> @@ -11199,6 +11200,8 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
>   
>   		vcpu->arch.msr_misc_features_enables = 0;
>   
> +		if (vcpu->arch.xcr0 != XFEATURE_MASK_FP)
> +			need_update_cpuid = true;
>   		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
>   	}
>   
> @@ -11216,6 +11219,8 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
>   	cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
>   	kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
>   
> +	if (vcpu->arch.ia32_xss)
> +		need_update_cpuid = true;
>   	vcpu->arch.ia32_xss = 0;
>   
>   	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
> @@ -11264,6 +11269,9 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
>   	 */
>   	if (init_event)
>   		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
> +
> +	if (need_update_cpuid)
> +		kvm_update_cpuid_runtime(vcpu);
>   }
>   EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
>

[PATCH v2] KVM: x86: Sync the states size with the XCR0/IA32_XSS at any time

Posted by Like Xu 4 years, 5 months ago

From: Like Xu <likexu@tencent.com>

XCR0 is reset to 1 by RESET but not INIT and IA32_XSS is zeroed by
both RESET and INIT. The kvm_set_msr_common()'s handling of MSR_IA32_XSS
also needs to update kvm_update_cpuid_runtime(). In the above cases, the
size in bytes of the XSAVE area containing all states enabled by XCR0 or
(XCRO | IA32_XSS) needs to be updated.

For simplicity and consistency, legacy helpers are used to write values
and call kvm_update_cpuid_runtime(), and it's not exactly a fast path.

Fixes: a554d207dc46 ("KVM: X86: Processor States following Reset or INIT")
Signed-off-by: Like Xu <likexu@tencent.com>
---
v1 -> v2 Changelog:
- Strongly prefer that use the helpers to write the values; (Sean)
- Postpone IA32_XSS test cases once non-zero values are supported; (Paolo)
- User space may call SET_CPUID2 after kvm_vcpu_reset(init_event=false); (Paolo)

  arch/x86/kvm/x86.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 55518b7d3b96..22d4b1d15e94 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -11256,7 +11256,7 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)

  		vcpu->arch.msr_misc_features_enables = 0;

-		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
+		__kvm_set_xcr(vcpu, 0, XFEATURE_MASK_FP);
  	}

  	/* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
@@ -11273,7 +11273,7 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
  	cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
  	kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);

-	vcpu->arch.ia32_xss = 0;
+	__kvm_set_msr(vcpu, MSR_IA32_XSS, 0, true);

  	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);

-- 
2.33.1