Re: [PATCH v3 16/16] objtool: Validate kCFI calls

[Sean Christopherson]

On Fri, Jul 25, 2025, Xin Li wrote:
> On 7/24/2025 1:37 PM, Sean Christopherson wrote:
> > On Mon, Jul 14, 2025, Peter Zijlstra wrote:
> > > --- a/arch/x86/kvm/vmx/vmenter.S
> > > +++ b/arch/x86/kvm/vmx/vmenter.S
> > > @@ -361,6 +361,10 @@ SYM_FUNC_END(vmread_error_trampoline)
> > >   .section .text, "ax"
> > > +#ifndef CONFIG_X86_FRED
> > > +
> > >   SYM_FUNC_START(vmx_do_interrupt_irqoff)
> > >   	VMX_DO_EVENT_IRQOFF CALL_NOSPEC _ASM_ARG1
> > >   SYM_FUNC_END(vmx_do_interrupt_irqoff)
> > > +
> > > +#endif
> > 
> > This can go in the previous patch, "x86/fred: KVM: VMX: Always use FRED for IRQs
> > when CONFIG_X86_FRED=y".
> > 
> 
> I'm going to test patch 13~15, plus this change in patch 16.
> 
> BTW, there is a declaration for vmx_do_interrupt_irqoff() in
> arch/x86/kvm/vmx/vmx.c, so we'd better also do:
> 
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -6945,7 +6945,9 @@ void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64
> *eoi_exit_bitmap)
>         vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
>  }
> 
> +#ifndef CONFIG_X86_FRED
>  void vmx_do_interrupt_irqoff(unsigned long entry);
> +#endif

No, we want to keep the declaration.  Unconditionally decaring the symbol allows
KVM to use IS_ENABLED():

	if (IS_ENABLED(CONFIG_X86_FRED))
 		fred_entry_from_kvm(EVENT_TYPE_EXTINT, vector);

Hiding the declaration would require that to be a "proper" #ifdef, which would
be a net negative for readability.  The extra declaration won't hurt anything for
CONFIG_X86_FRED=n, as "bad" usage will still fail at link time.

>  void vmx_do_nmi_irqoff(void);
> 
>  static void handle_nm_fault_irqoff(struct kvm_vcpu *vcpu)