Re: [PATCH v2] KVM: x86/cpuid: Exclude unpermitted xfeatures sizes at KVM_GET_SUPPORTED_CPUID

[Paolo Bonzini]

On 1/25/22 12:52, Like Xu wrote:
> From: Like Xu <likexu@xxxxxxxxxxx>
>
> With the help of xstate_get_guest_group_perm(), KVM can exclude unpermitted
> xfeatures in cpuid.0xd.0.eax, in which case the corresponding xfeatures
> sizes should also be matched to the permitted xfeatures.
>
> To fix this inconsistency, the permitted_xcr0 and permitted_xss are defined
> consistently, which implies 'supported' plus certain permissions for this
> task, and it also fixes cpuid.0xd.1.ebx and later leaf-by-leaf queries.
>
> Fixes: 445ecdf79be0 ("kvm: x86: Exclude unpermitted xfeatures at KVM_GET_SUPPORTED_CPUID")
> Signed-off-by: Like Xu <likexu@xxxxxxxxxxx>
> ---
> v1 -> v2 Changelog:
> - Drop the use of shadow variable; (Paolo)
> - Define permitted_xss consistently; (Kevin)
>
> Previous:
> https://lore.kernel.org/kvm/20220124080251.60558-1-likexu@xxxxxxxxxxx/
>
>   arch/x86/kvm/cpuid.c | 25 +++++++++++++------------
>   1 file changed, 13 insertions(+), 12 deletions(-)
>
> diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
> index 3902c28fb6cb..07844d15dfdf 100644
> --- a/arch/x86/kvm/cpuid.c
> +++ b/arch/x86/kvm/cpuid.c
> @@ -887,13 +887,14 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
>   		}
>   		break;
>   	case 0xd: {
> -		u64 guest_perm = xstate_get_guest_group_perm();
> +		u64 permitted_xcr0 = supported_xcr0 & xstate_get_guest_group_perm();
> +		u64 permitted_xss = supported_xss;
>   
> -		entry->eax &= supported_xcr0 & guest_perm;
> -		entry->ebx = xstate_required_size(supported_xcr0, false);
> +		entry->eax &= permitted_xcr0;
> +		entry->ebx = xstate_required_size(permitted_xcr0, false);
>   		entry->ecx = entry->ebx;
> -		entry->edx &= (supported_xcr0 & guest_perm) >> 32;
> -		if (!supported_xcr0)
> +		entry->edx &= permitted_xcr0 >> 32;
> +		if (!permitted_xcr0)
>   			break;
>   
>   		entry = do_host_cpuid(array, function, 1);
> @@ -902,20 +903,20 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
>   
>   		cpuid_entry_override(entry, CPUID_D_1_EAX);
>   		if (entry->eax & (F(XSAVES)|F(XSAVEC)))
> -			entry->ebx = xstate_required_size(supported_xcr0 | supported_xss,
> +			entry->ebx = xstate_required_size(permitted_xcr0 | permitted_xss,
>   							  true);
>   		else {
> -			WARN_ON_ONCE(supported_xss != 0);
> +			WARN_ON_ONCE(permitted_xss != 0);
>   			entry->ebx = 0;
>   		}
> -		entry->ecx &= supported_xss;
> -		entry->edx &= supported_xss >> 32;
> +		entry->ecx &= permitted_xss;
> +		entry->edx &= permitted_xss >> 32;
>   
>   		for (i = 2; i < 64; ++i) {
>   			bool s_state;
> -			if (supported_xcr0 & BIT_ULL(i))
> +			if (permitted_xcr0 & BIT_ULL(i))
>   				s_state = false;
> -			else if (supported_xss & BIT_ULL(i))
> +			else if (permitted_xss & BIT_ULL(i))
>   				s_state = true;
>   			else
>   				continue;
> @@ -929,7 +930,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
>   			 * invalid sub-leafs.  Only valid sub-leafs should
>   			 * reach this point, and they should have a non-zero
>   			 * save state size.  Furthermore, check whether the
> -			 * processor agrees with supported_xcr0/supported_xss
> +			 * processor agrees with permitted_xcr0/permitted_xss
>   			 * on whether this is an XCR0- or IA32_XSS-managed area.
>   			 */
>   			if (WARN_ON_ONCE(!entry->eax || (entry->ecx & 0x1) != s_state)) {

Queued, thanks.

Paolo