Re: [PATCH v3 1/3] KVM: arm64: Add pre_fault_memory implementation

[Thomson, Jack]

On 24/11/2025 12:54 pm, Marc Zyngier wrote:
> On Mon, 24 Nov 2025 11:34:38 +0000,
> Marc Zyngier <maz@xxxxxxxxxx> wrote:
> > On Wed, 19 Nov 2025 15:49:08 +0000,
> > Jack Thomson <jackabt.amazon@xxxxxxxxx> wrote:
> > >
>
> [...]
>
> > > +	fault_info->hpfar_el2 = HPFAR_EL2_NS |
> > > +		FIELD_PREP(HPFAR_EL2_FIPA, ipa >> 12);
> > > +
> > > +	if (kvm_slot_has_gmem(memslot)) {
> > > +		ret = gmem_abort(vcpu, ipa, NULL, memslot, false);
> > > +	} else {
> > > +		hva = gfn_to_hva_memslot_prot(memslot, gfn, NULL);
> > > +		if (kvm_is_error_hva(hva)) {
> > > +			ret = -EFAULT;
> > > +			goto out;
> > > +		}
> > > +		ret = user_mem_abort(vcpu, ipa, NULL, memslot, &page_size, hva,
> > > +				     false);
> > > +	}
>
> And thinking of it a bit more, this is completely broken. What happens
> if the vcpu is in a nested context? You just populate random pages in
> an IPA space that is not relevant at all, corrupting the guest state.
>
> You must correctly handle the context the vcpu is in, instead of
> assuming that this is the canonical context. This means going via the
> *guest's* S2 translation, just like handle_mem_abort() does.
>
> 	M.

Thanks Marc for taking a look, I'll update to fix these issues and
address the other comments.

--
Thanks,
Jack