Re: [PATCH v3 1/4] LoongArch: KVM: Preserve auto-enabled PV features on userspace override

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] The patch unconditionally forces auto-enabled PV features ON, preventing userspace from disabling them, which breaks VM capability negotiation and live migration.
--

commit 2e3dfd10f9606ff82713655e6b87be567a31ec59
Author: Tao Cui <cuitao@kylinos.cn>

LoongArch: KVM: Preserve auto-enabled PV features on userspace override

This commit adds a pv_auto_features field to struct kvm_arch to save the
kernel-default PV features. It then alters kvm_loongarch_cpucfg_set_attr()
to automatically OR these saved features back into the user-provided feature
mask during KVM_SET_DEVICE_ATTR.

> diff --git a/arch/loongarch/kvm/vcpu.c b/arch/loongarch/kvm/vcpu.c
> index e28084c49e682..47a59ce9e5616 100644
> --- a/arch/loongarch/kvm/vcpu.c
> +++ b/arch/loongarch/kvm/vcpu.c
> @@ -1165,6 +1165,13 @@ static int kvm_loongarch_cpucfg_set_attr(struct kvm_vcpu *vcpu,
>  		if (val & ~valid)
>  			return -EINVAL;
>  
> +		/*
> +		 * Auto-enabled features (e.g. PV TLB flush) that userspace
> +		 * is not aware of are preserved so that no QEMU changes are
> +		 * needed for pure kernel-internal optimisations.
> +		 */
> +		val = (val | kvm->arch.pv_auto_features) & valid;

[Severity: High]
Does unconditionally forcing auto-enabled PV features on prevent userspace
from disabling them?

If a VMM configures a VM for migration to an older host that lacks features
like KVM_FEATURE_STEAL_TIME, it will clear the corresponding bits in the
feature mask.

By forcing these features back on even when userspace explicitly omitted
them, could this cause the guest to observe a feature set unsupported on
the migration destination, resulting in migration failures?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260602021819.2373404-1-cui.tao@linux.dev?part=1