Re: [PATCH v2] LoongArch: KVM: Add hypercall crash support

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] kvm: new guest-visible KVM_FEATURE_CRASH capability is silently enabled by default
--

commit 6d0e897e0c46a79eb02651baa4b7271884b07abb
Author: Bibo Mao <maobibo@loongson.cn>
LoongArch: KVM: Add hypercall crash support

This commit adds hypercall crash support (KVM_HCALL_CRASH) to LoongArch KVM
to provide a more direct, low-level crash notification method. It allows the
VMM to capture CPU crash context during exception entry even if the CPU
hardware context or kernel stack is corrupted.

> diff --git a/arch/loongarch/kvm/vm.c b/arch/loongarch/kvm/vm.c
> index 6dabb227a7321..9bf4b3bb8f25a 100644
> --- a/arch/loongarch/kvm/vm.c
> +++ b/arch/loongarch/kvm/vm.c

[ ... ]

> @@ -49,8 +49,9 @@ static void kvm_vm_init_features(struct kvm *kvm)
>  		kvm->arch.kvm_features |= BIT(KVM_LOONGARCH_VM_FEAT_PMU);
>  
>  	/* Enable all PV features by default */
> -	kvm->arch.pv_features |= BIT(KVM_FEATURE_IPI);
> +	kvm->arch.pv_features |= BIT(KVM_FEATURE_IPI) | BIT(KVM_FEATURE_CRASH);
>  	kvm->arch.kvm_features |= BIT(KVM_LOONGARCH_VM_FEAT_PV_IPI);
> +	kvm->arch.kvm_features |= BIT(KVM_LOONGARCH_VM_FEAT_CRASH);

[Severity: High]
Does unconditionally setting BIT(KVM_FEATURE_CRASH) and
BIT(KVM_LOONGARCH_VM_FEAT_CRASH) here break live migration and backward
compatibility?

If a VM is migrated to a newer kernel with this patch, an older VMM unaware
of the capability will not explicitly disable it. If the guest discovers
and triggers the new KVM_HCALL_CRASH hypercall, the older VMM will receive
an unexpected KVM_EXIT_SYSTEM_EVENT and will likely terminate the VM.

Should new guest-visible capabilities be left off by default and require
explicit VMM opt-in through capability negotiation?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261016034245.3002859-1-maobibo@loongson.cn?part=1