Re: [PATCH] KVM: x86: use inlines instead of macros for is_sev_*guest

[Paolo Bonzini]

On 4/13/26 23:41, Sean Christopherson wrote:
> On Mon, Apr 13, 2026, Paolo Bonzini wrote:
> > This helps avoiding more embarrassment to this maintainer,
>
> Not building SEV support before pushing to kvm/queue is certainly a choice :-D

Fortunately, the part where I actually do more than a build smoke test 
and push to kvm/next has some automation... but the bots are too fast 
these days.

> As called out in commit 45d522d3ee9c ("KVM: SVM: Macrofy SEV=n versions of
> sev_xxx_guest()"), use of macros is intentional as gcc-12 at least generates
> suboptimal code if there's a proper function.
>
> That said, gcc-11 (I don't think I have gcc-12 laying around anywhere) generates
> identical code for all of kvm-amd.ko with and without macros, so maybe whatever I
> saw a transient issue?  I don't have a strong motivation for caring about any
> version of gcc, so I'm a-ok using __always_inline functions.

Yeah, it seems strange for the compiler to have this problem except at -O0.

>   #else
> -static __always_inline bool is_sev_guest(struct kvm_vcpu *vcpu)
> -{
> -       return false;
> -}
> -static __always_inline bool is_sev_es_guest(struct kvm_vcpu *vcpu)
> -{
> -       return false;
> -}
> -
> -static __always_inline bool is_sev_snp_guest(struct kvm_vcpu *vcpu)
> -{
> -       return false;
> -}
> +#define is_sev_guest(vcpu) ({ BUILD_BUG_ON_INVALID((vcpu)->kvm); false; })
> +#define is_sev_es_guest(vcpu) ({ BUILD_BUG_ON_INVALID((vcpu)->kvm); false; })
> +#define is_sev_snp_guest(vcpu) ({ BUILD_BUG_ON_INVALID((vcpu)->kvm); false; })

BUILD_BUG_ON_INVALID isn't great because it doesn't have type checking.

Paolo