Re: [PATCH v3 01/17] KVM: arm64: Add pkvm_private_va_range_pa

On Mon, Jul 20, 2026 at 05:25:59PM +0000, [email protected] wrote:
> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> - [High] Incorrect physical address calculation for block mappings in pkvm_private_va_range_pa()
> --
> 
> --- Patch [1]: [PATCH v3 01/17] KVM: arm64: Add pkvm_private_va_range_pa ---
> commit 81aa726ac7516be8619c0cb2674493e48b6fca16
> Author: Vincent Donnefort <[email protected]>
> 
> KVM: arm64: Add pkvm_private_va_range_pa
> 
> This commit introduces pkvm_private_va_range_pa() to resolve physical
> addresses for the pKVM private range by walking the hypervisor page-table.
> This prepares for the upcoming pKVM heap allocator.
> 
> > diff --git a/arch/arm64/kvm/hyp/nvhe/mm.c b/arch/arm64/kvm/hyp/nvhe/mm.c
> > --- a/arch/arm64/kvm/hyp/nvhe/mm.c
> > +++ b/arch/arm64/kvm/hyp/nvhe/mm.c
> [ ... ]
> > +phys_addr_t pkvm_private_va_range_pa(void *va)
> > +{
> > +	kvm_pte_t pte = 0;
> > +
> > +	guard(hyp_spinlock)(&pkvm_pgd_lock);
> > +	WARN_ON(kvm_pgtable_get_leaf(&pkvm_pgtable, (u64)va, &pte, NULL));
> > +	WARN_ON(!kvm_pte_valid(pte));
> > +
> > +	return kvm_pte_to_phys(pte) + offset_in_page(va);
> 
> [Severity: High]
> Does this calculation handle block mappings correctly?

There's no risk to step into a block mapping with the users of that function.

I could WARN_ON() though...

> 
> If the virtual address is mapped via a block descriptor (such as a 2MB
> large page), kvm_pte_to_phys(pte) returns the base physical address of
> the block. Adding offset_in_page(va) will only preserve the lowest 12
> bits, discarding the intermediate offset bits between the page size and
> the block size.
> 
> Could this cause multiple virtual pages within the block mapping to
> incorrectly resolve to the same physical base page plus a small offset?
> Perhaps this should retrieve the mapping level from kvm_pgtable_get_leaf()
> and mask the virtual address with the corresponding granule size?
> 
> > +}
> 
> -- 
> Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1

-- 
-- 
Vincent