Re: [PATCH 1/1] kvm: nVMX: Checks "VMCS shadowing" with VMCS link pointer for non-root mode VM{READ,WRITE}

From: Yuan Yao <yuan.yao@linux.intel.com>
To: Yuan Yao <yuan.yao@intel.com>
Cc: kvm@vger.kernel.org, linux-kernel@vger.kernel.org,
	Paolo Bonzini <pbonzini@redhat.com>,
	Sean Christopherson <seanjc@google.com>,
	Jim Mattson <jmattson@google.com>,
	Jon Cargille <jcargill@google.com>,
	Peter Shier <pshier@google.com>, Oliver Upton <oupton@google.com>
Subject: Re: [PATCH 1/1] kvm: nVMX: Checks "VMCS shadowing" with VMCS link pointer for non-root mode VM{READ,WRITE}
Date: Fri, 12 Aug 2022 10:02:06 +0800	[thread overview]
Message-ID: <20220812020206.foknky4hghgsadby@yy-desk-7060> (raw)
In-Reply-To: <20220812014706.43409-1-yuan.yao@intel.com>

On Fri, Aug 12, 2022 at 09:47:06AM +0800, Yuan Yao wrote:
> Add checking to VMCS12's "VMCS shadowing", make sure the checking of
> VMCS12's vmcs_link_pointer for non-root mode VM{READ,WRITE} happens
> only if VMCS12's "VMCS shadowing" is 1.
>
> SDM says that for non-root mode the VMCS's "VMCS shadowing" must be 1
> (and the corresponding bits in VMREAD/VMWRITE bitmap must be 0) when
> condition checking of [B] is reached(please refer [A]), which means
> checking to VMCS link pointer for non-root mode VM{READ,WRITE} should
> happen only when "VMCS shadowing" = 1.
>
> Description from SDM Vol3(April 2022) Chapter 30.3 VMREAD/VMWRITE:
>
> IF (not in VMX operation)
>    or (CR0.PE = 0)
>    or (RFLAGS.VM = 1)
>    or (IA32_EFER.LMA = 1 and CS.L = 0)
> THEN #UD;
> ELSIF in VMX non-root operation
>       AND (“VMCS shadowing” is 0 OR
>            source operand sets bits in range 63:15 OR
>            VMREAD bit corresponding to bits 14:0 of source
>            operand is 1)  <------[A]
> THEN VMexit;
> ELSIF CPL > 0
> THEN #GP(0);
> ELSIF (in VMX root operation AND current-VMCS pointer is not valid) OR
>       (in VMX non-root operation AND VMCS link pointer is not valid)
> THEN VMfailInvalid;  <------ [B]
> ...
>
> Fixes: dd2d6042b7f4 ("kvm: nVMX: VMWRITE checks VMCS-link pointer before VMCS field")
> Signed-off-by: Yuan Yao <yuan.yao@intel.com>
> ---
>  arch/x86/kvm/vmx/nested.c | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
> index ddd4367d4826..30685be54c5d 100644
> --- a/arch/x86/kvm/vmx/nested.c
> +++ b/arch/x86/kvm/vmx/nested.c
> @@ -5123,6 +5123,7 @@ static int handle_vmread(struct kvm_vcpu *vcpu)
>  		 */
>  		if (vmx->nested.current_vmptr == INVALID_GPA ||
>  		    (is_guest_mode(vcpu) &&
> +		     nested_cpu_has_shadow_vmcs(vcpu) &&

Oops, should be "nested_cpu_has_shadow_vmcs(get_vmcs12(vcpu))".

>  		     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
>  			return nested_vmx_failInvalid(vcpu);
>
> @@ -5233,6 +5234,7 @@ static int handle_vmwrite(struct kvm_vcpu *vcpu)
>  	 */
>  	if (vmx->nested.current_vmptr == INVALID_GPA ||
>  	    (is_guest_mode(vcpu) &&
> +	     nested_cpu_has_shadow_vmcs(vcpu) &&

Ditto.

>  	     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
>  		return nested_vmx_failInvalid(vcpu);
>
> --
> 2.27.0
>