Re: [PATCH v2 1/2] x86/kvm/vmx: Move IRQ/NMI dispatch from KVM into x86 core

[Binbin Wu <binbin.wu@linux.intel.com>]

On 5/2/2026 4:37 AM, Peter Zijlstra wrote:
[...]

> --- /dev/null
> +++ b/arch/x86/entry/common.c
> @@ -0,0 +1,48 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +
> +#include <linux/entry-common.h>
> +#include <linux/kvm_types.h>
> +#include <asm/fred.h>
> +#include <asm/desc.h>
> +
> +#if IS_ENABLED(CONFIG_KVM_INTEL)
> +/*
> + * On VMX, NMIs and IRQs (as configured by KVM) are acknowledged by hardware as
> + * part of the VM-Exit, i.e. the event itself is consumed as part the VM-Exit.
> + * x86_entry_from_kvm() is invoked by KVM to effectively forward NMIs and IRQs
> + * to the kernel for servicing.  On SVM, a.k.a. AMD, the NMI/IRQ VM-Exit is
> + * purely a signal that an NMI/IRQ is pending, i.e. the event that triggered
> + * the VM-Exit is held pending until it's unblocked in the host.
> + */
> +noinstr void x86_entry_from_kvm(unsigned int event_type, unsigned int vector)
> +{
> +	if (event_type == EVENT_TYPE_EXTINT) {
> +#ifdef CONFIG_X86_64
> +		/*
> +		 * Use FRED dispatch, even when running IDT. The dispatch
> +		 * tables are kept in sync between FRED and IDT, and the FRED
> +		 * dispatch works well with CFI.
> +		 */
> +		fred_entry_from_kvm(event_type, vector);
> +#else
> +		idt_entry_from_kvm(vector);
> +#endif
> +		return;
> +	}
> +
> +	WARN_ON_ONCE(event_type != EVENT_TYPE_NMI);

Not sure if it's OK to use WARN_ON_ONCE() here.
If the warning is triggered, it could unblock NMI due to handling of #UD.

> +
> +#ifdef CONFIG_X86_64
> +	if (cpu_feature_enabled(X86_FEATURE_FRED))
> +		return fred_entry_from_kvm(event_type, vector);
> +#endif
> +
> +	/*
> +	 * Notably, we must use IDT dispatch for NMI when running in IDT mode.
> +	 * The FRED NMI context is significantly different and will not work
> +	 * right (speficially FRED fixed the NMI recursion issue).
> +	 */
> +	idt_entry_from_kvm(vector);
> +}
> +EXPORT_SYMBOL_FOR_KVM(x86_entry_from_kvm);
> +#endif

[...]


> --- a/arch/x86/include/asm/desc.h
> +++ b/arch/x86/include/asm/desc.h
> @@ -438,6 +438,10 @@ extern void idt_setup_traps(void);
>  extern void idt_setup_apic_and_irq_gates(void);
>  extern bool idt_is_f00f_address(unsigned long address);
>  
> +extern void idt_do_interrupt_irqoff(unsigned int vector);

In idt_entry_from_kvm() below, gate_offset() returns 'unsigned long', but here
it uses 'unsigned int'. It's not safe since there is no guarantee that the
address is within 32 bits for x86_64.

Also, the argument is not a vector.

[...]
> +noinstr void idt_entry_from_kvm(unsigned int vector)
> +{
> +	if (vector == NMI_VECTOR)
> +		return idt_do_nmi_irqoff();
> +
> +	/*
> +	 * Only the NMI path requires noinstr.
> +	 */
> +	instrumentation_begin();
> +	idt_do_interrupt_irqoff(gate_offset(idt_table + vector));
> +	instrumentation_end();
> +}
> +