Re: [PATCH 2/3] KVM: x86: Reject enabling KVM_CAP_X86_BUS_LOCK_EXIT when !kvm_caps.has_bus_lock_exit

[Xiaoyao Li <xiaoyao.li@intel.com>]

On 8/6/2026 7:35 PM, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> 
> Pre-existing issues:
> - [High] Modifying the VM-wide state `kvm->arch.bus_lock_detection_enabled` lacks protection from `kvm->lock` and does not verify that no vCPUs have been created yet.
> --
> 
> --- Patch [2]: [PATCH 2/3] KVM: x86: Reject enabling KVM_CAP_X86_BUS_LOCK_EXIT when !kvm_caps.has_bus_lock_exit ---
> commit e0edfcdc943c7835281cad8477ff3399dfe81a05
> Author: Xiaoyao Li <xiaoyao.li@intel.com>
> 
> KVM: x86: Reject enabling KVM_CAP_X86_BUS_LOCK_EXIT when !kvm_caps.has_bus_lock_exit
> 
> This commit returns -EINVAL to reject enabling KVM_CAP_X86_BUS_LOCK_EXIT
> or KVM_BUS_LOCK_DETECTION_OFF from userspace when kvm_caps.has_bus_lock_exit
> is false. It aims to keep the API behavior consistent when KVM does not
> support bus lock exit.
> 
>> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
>> index afcac1042947a..91895a3d6f857 100644
>> --- a/arch/x86/kvm/x86.c
>> +++ b/arch/x86/kvm/x86.c
>> @@ -6835,8 +6835,10 @@ int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
>>   		    (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
>>   			break;
>>   
>> -		if (kvm_caps.has_bus_lock_exit &&
>> -		    cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
>> +		if (!kvm_caps.has_bus_lock_exit)
>> +			break;
>> +
>> +		if (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
>>   			kvm->arch.bus_lock_detection_enabled = true;
> 
> [Severity: High]
> This is a pre-existing issue, but does setting the VM-wide flag
> bus_lock_detection_enabled here in kvm_vm_ioctl_enable_cap() require
> kvm->lock protection and a check for !kvm->created_vcpus?
> 
> If userspace calls KVM_ENABLE_CAP for KVM_CAP_X86_BUS_LOCK_EXIT concurrently
> with or after KVM_CREATE_VCPU, existing vCPUs might miss the update. This
> could lead to an inconsistent state where some vCPUs mitigate bus lock DoS
> and others do not, leaving the host vulnerable.

This is addressed by the next patch.

>>   		r = 0;
>>   		break;
>