linux-kernel - Re: [PATCH RFC v1 08/20] KVM: VMX: Support extended register index in exit handling

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <670cf030-2ae7-4dd6-9324-bbfb7815e2e5@redhat.com>
Date: Fri, 14 Nov 2025 00:40:32 +0100
From: Paolo Bonzini <pbonzini@...hat.com>
To: "Chang S. Bae" <chang.seok.bae@...el.com>, kvm@...r.kernel.org,
 linux-kernel@...r.kernel.org
Cc: seanjc@...gle.com, chao.gao@...el.com, zhao1.liu@...el.com
Subject: Re: [PATCH RFC v1 08/20] KVM: VMX: Support extended register index in
 exit handling

On 11/14/25 00:22, Chang S. Bae wrote:
>>> @@ -415,7 +420,10 @@ static __always_inline unsigned long 
>>> vmx_get_exit_qual(struct kvm_vcpu *vcpu)
>>>   static inline int vmx_get_exit_qual_gpr(struct kvm_vcpu *vcpu)
>>>   {
>>> -    return (vmx_get_exit_qual(vcpu) >> 8) & 0xf;
>>> +    if (vmx_egpr_enabled(vcpu))
>>> +        return (vmx_get_exit_qual(vcpu) >> 8) & 0x1f;
>>> +    else
>>> +        return (vmx_get_exit_qual(vcpu) >> 8) & 0xf;
>>
>> Can this likewise mask against 0x1f, unconditionally?
> 
> It looks like the behavior of that previously-undefined bit is not
> guaranteed -- there's no architectural promise that the bit will always
> read as zero. So in this case, I think it's still safer to rely on the
> enumeration.
The manual says "VM Exit qualification is extended in-place"... This 
suggests that the wording of the SDM, "Not currently defined", really 
means "currently zero, but may not be *in the future*".

It is indeed safer, but it's a bit sad too. :)  Let's at least use 
either static_cpu_has() or a global __read_mostly variable, instead of a 
more expensive test using the vcpu.

Paolo