Re: [PATCH v6 15/49] KVM: arm64: gic-v5: Add resident/non-resident hyp calls

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] Early return in `vgic_v5_load` when `IN_WFI` is set breaks VPE hardware residency on CPU migration.
--

commit 5c761580bd9775ac4c98a890e1f6aab6f8768957
Author: Sascha Bischoff <Sascha.Bischoff@arm.com>

KVM: arm64: gic-v5: Add resident/non-resident hyp calls

This commit introduces hypercalls to manage GICv5 VPE residency and
extends vgic_v5_load and vgic_v5_put to make VPEs resident and
non-resident during the load-to-put interval. It also adds a check
for residency faults that marks the VM as dead on failure.

> diff --git a/arch/arm64/kvm/vgic/vgic-v5.c b/arch/arm64/kvm/vgic/vgic-v5.c
> index 5b599821b2da2..d7d6db5014269 100644
> --- a/arch/arm64/kvm/vgic/vgic-v5.c
> +++ b/arch/arm64/kvm/vgic/vgic-v5.c
[ ... ]
> @@ -1066,7 +1068,15 @@ void vgic_v5_load(struct kvm_vcpu *vcpu)
>  
>  	kvm_call_hyp(__vgic_v5_restore_vmcr_apr, cpu_if);
>  
> -	cpu_if->gicv5_vpe.resident = true;
> +	cpu_if->vgic_contextr = FIELD_PREP(ICH_CONTEXTR_EL2_V, true) |
> +				FIELD_PREP(ICH_CONTEXTR_EL2_VPE, vpe) |
> +				FIELD_PREP(ICH_CONTEXTR_EL2_VM, vm);
> +
> +	kvm_call_hyp(__vgic_v5_make_resident, cpu_if);

[Severity: High]
Does this break VPE hardware residency on CPU migration during WFI?

When a vCPU executes WFI and yields the CPU, the IN_WFI flag is set and
vgic_v5_put makes the VPE non-resident. When the thread is scheduled back in,
vgic_v5_load is called by kvm_sched_in. However, the pre-existing early
return in vgic_v5_load bypasses the rest of the function:

vgic_v5_load() {
    ...
    if (vcpu_get_flag(vcpu, IN_WFI))
        return;
    ...
}

Because the new hardware residency setup (__vgic_v5_make_resident) is placed
after this early return, it appears to be skipped completely. Even if
kvm_arch_vcpu_unblocking calls vgic_v5_load again later, it will also return
early since IN_WFI is only cleared after the blocking sequence finishes.

Could this result in the vCPU resuming execution after a WFI block without
its VPE being resident in the GIC hardware, breaking LPI and SPI interrupt
delivery?

> +
> +	/* Failed to make the VPE resident? Bang! */
> +	if (WARN_ON(!!FIELD_GET(ICH_CONTEXTR_EL2_F, cpu_if->vgic_contextr)))
> +		kvm_vm_dead(vcpu->kvm);
>  }

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260904113404.4051341-1-sascha.bischoff@arm.com?part=15