[PATCH v2] KVM: arm64: Fix protected VM fault on system with pages larger than 4K

[Vincent Donnefort]

On Mon, Sep 14, 2026 at 09:32:47AM +0100, Marc Zyngier wrote:
> On Mon, 14 Sep 2026 08:58:39 +0100,
> Vincent Donnefort <vdonnefort at google.com> wrote:
> > 
> > Just like commit 08f97454b7fa ("KVM: arm64: Fix protected mode handling
> > of pages larger than 4kB") fixed the boot of non-protected VMs on system
> > larger than 4K pages, align the fault IPA down to the page-size for
> > protected VMs.
> > 
> > To paraphrase Marc, pkvm_pgtable_stage2_map() assumes the address passed
> > as a parameter is aligned to the size of the intended mapping, while
> > HPFAR_EL2 gives the IPA minus the bottom 12 bits, regardless of the
> > system page size configuration.
> > 
> > Add a check at the start of pkvm_pgtable_stage2_map() as we do not
> > support !PAGE_ALIGNED arguments and use the gfn as an argument in all
> > callers.
> > 
> > Fixes: ea03466e806f ("KVM: arm64: Handle aborts from protected VMs")
> > Signed-off-by: Vincent Donnefort <vdonnefort at google.com>
> > ---
> > 
> > Changelog:
> > 
> > v2:
> >   - Use gfn_to_gpa(gfn)
> >   - Drop "phys" from the PAGE_ALIGNED check, it isn't a requirement.
> >   - Fix gmem_abort() as well (Sashiko)
> > 
> > v1: https://lore.kernel.org/all/20260913173516.3122436-1-vdonnefort@google.com/
> > 
> >  arch/arm64/kvm/mmu.c  | 7 ++++---
> >  arch/arm64/kvm/pkvm.c | 3 +++
> >  2 files changed, 7 insertions(+), 3 deletions(-)
> > 
> > diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
> > index 9ba86450fe4a..ee78201a44df 100644
> > --- a/arch/arm64/kvm/mmu.c
> > +++ b/arch/arm64/kvm/mmu.c
> > @@ -1672,10 +1672,10 @@ static int gmem_abort(const struct kvm_s2_fault_desc *s2fd)
> >  		 * PTE, which will be preserved.
> >  		 */
> 
> Context:
> 
> 	if (s2fd->nested)
> 		gfn = kvm_s2_trans_output(s2fd->nested) >> PAGE_SHIFT;
> 	else
> 		gfn = s2fd->fault_ipa >> PAGE_SHIFT;
> 
> [...]
> 
> >  		prot &= ~KVM_NV_GUEST_MAP_SZ;
> > -		ret = KVM_PGT_FN(kvm_pgtable_stage2_relax_perms)(pgt, s2fd->fault_ipa,
> > +		ret = KVM_PGT_FN(kvm_pgtable_stage2_relax_perms)(pgt, gfn_to_gpa(gfn),
> >  								 prot, flags);
> >  	} else {
> > -		ret = KVM_PGT_FN(kvm_pgtable_stage2_map)(pgt, s2fd->fault_ipa, PAGE_SIZE,
> > +		ret = KVM_PGT_FN(kvm_pgtable_stage2_map)(pgt, gfn_to_gpa(gfn), PAGE_SIZE,
> >  							 __pfn_to_phys(pfn), prot,
> >  							 memcache, flags);
> 
> So 'gfn' is not related to the faulting IPA in the nested case, but is
> the output address of the guest's own S2. You probably end-up with a
> livelock handling the same fault forever, or worse.
> 
> It is a bit annoying that neither gmem nor pkvm are using
> kvm_s2_fault_vma_info, which has the ability to reports all the
> important stuff, as it would avoid this sort of bugs.
> 
> Thanks,
> 
> 	M.
> 
> -- 
> Without deviation from the norm, progress is not possible.

Using kvm_s2_fault_vma_info would surely help the THP support for protected VMs
[1]. I am thinking now about bringing it as part of that series.

However for a fix, that's a bit much?

[1] https://lore.kernel.org/all/20260911135053.146435-23-vdonnefort@google.com/

-- 
Vincent