[PATCH v7 02/24] KVM: arm64: Donate MMIO to the hypervisor

[Mostafa Saleh]

On Wed, Aug 05, 2026 at 12:47:03PM +0000, Sebastian Ene wrote:
> On Wed, Jul 15, 2026 at 11:58:43AM +0000, Mostafa Saleh wrote:
> > Add a function to donate MMIO to the hypervisor so IOMMU hypervisor
> > drivers can protect and access the MMIO of IOMMUs.
> > 
> > As donating MMIO is very rare, and we don’t need to encode the full
> > state, it’s reasonable to have a separate function to do this.
> > It will init the host s2 page table with an invalid leaf with the owner ID
> > to prevent the host from mapping the page on faults.
> > 
> > Also, prevent kvm_pgtable_stage2_unmap() from removing owner ID from
> > stage-2 PTEs, as this can be triggered from recycle logic under memory
> > pressure. There is no code relying on this, as all ownership changes is
> > done via kvm_pgtable_stage2_set_owner()
> > 
> > For the error path in IOMMU drivers, add a function to donate MMIO
> > back from hyp to host. However, that leaks the hypervisor virtual
> > address range which should be acceptable as this is quite rare and
> > it matches the behaviour of fix_map/block.
> > 
> > Signed-off-by: Mostafa Saleh <smostafa at google.com>
> > ---
> >  arch/arm64/kvm/hyp/include/nvhe/mem_protect.h |  7 ++
> >  arch/arm64/kvm/hyp/nvhe/mem_protect.c         | 91 ++++++++++++++++++-
> >  arch/arm64/kvm/hyp/pgtable.c                  | 11 +--
> >  3 files changed, 102 insertions(+), 7 deletions(-)
> > 
> > diff --git a/arch/arm64/kvm/hyp/include/nvhe/mem_protect.h b/arch/arm64/kvm/hyp/include/nvhe/mem_protect.h
> > index 29935c7da1de..51b0eb3844a9 100644
> > --- a/arch/arm64/kvm/hyp/include/nvhe/mem_protect.h
> > +++ b/arch/arm64/kvm/hyp/include/nvhe/mem_protect.h
> > @@ -36,6 +36,13 @@ int __pkvm_guest_share_host(struct pkvm_hyp_vcpu *vcpu, u64 gfn);
> >  int __pkvm_guest_unshare_host(struct pkvm_hyp_vcpu *vcpu, u64 gfn);
> >  int __pkvm_host_unshare_hyp(u64 pfn);
> >  int __pkvm_host_donate_hyp(u64 pfn, u64 nr_pages);
> > +/*
> > + * Donate MMIO range to the hypervisor, it will be mapped in the hypervisor's
> > + * private range and unmapped from the host stage-2.
> > + */
> > +int __pkvm_host_donate_hyp_mmio(phys_addr_t addr, size_t size, unsigned long *haddr);
> > +/* Remaps MMIO range in the host, typically used in error path. */
> > +int __pkvm_hyp_donate_host_mmio(phys_addr_t addr, size_t size);
> >  int __pkvm_hyp_donate_host(u64 pfn, u64 nr_pages);
> >  int __pkvm_host_share_ffa(u64 pfn, u64 nr_pages);
> >  int __pkvm_host_unshare_ffa(u64 pfn, u64 nr_pages);
> > diff --git a/arch/arm64/kvm/hyp/nvhe/mem_protect.c b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
> > index 4e329e39a695..d803b3dd4cb4 100644
> > --- a/arch/arm64/kvm/hyp/nvhe/mem_protect.c
> > +++ b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
> > @@ -378,7 +378,11 @@ static int host_stage2_unmap_dev_all(void)
> >  	u64 addr = 0;
> >  	int i, ret;
> >  
> > -	/* Unmap all non-memory regions to recycle the pages */
> > +	/*
> > +	 * Unmap all non-memory regions to recycle the pages.
> > +	 * That relies on kvm_pgtable_stage2_unmap() not clearing
> > +	 * counted PTEs which include hypervisor MMIO.
> > +	 */
> >  	for (i = 0; i < hyp_memblock_nr; i++, addr = reg->base + reg->size) {
> >  		reg = &hyp_memory[i];
> >  		ret = kvm_pgtable_stage2_unmap(pgt, addr, reg->base - addr);
> > @@ -1119,6 +1123,91 @@ int __pkvm_host_donate_hyp(u64 pfn, u64 nr_pages)
> >  	return ret;
> >  }
> >  
> > +int __pkvm_host_donate_hyp_mmio(phys_addr_t addr, size_t size, unsigned long *haddr)
> > +{
> > +	kvm_pte_t pte;
> > +	u64 offset;
> > +	int ret;
> > +
> > +	/* Only before de-privilege. */
> > +	if (static_branch_unlikely(&kvm_protected_mode_initialized))
> > +		return -EPERM;
> > +
> > +	if (!PAGE_ALIGNED(addr | size) ||
> > +	    !pfn_range_is_valid(hyp_phys_to_pfn(addr), size >> PAGE_SHIFT))
> > +		return -EINVAL;
> > +
> > +	ret = __pkvm_create_private_mapping(addr, size, PAGE_HYP_DEVICE, haddr);
> 
> > +	if (ret)
> > +		return ret;
> > +
> > +	host_lock_component();
> > +	for (offset = 0; offset < size; offset += PAGE_SIZE) {
> > +		if (addr_is_memory(addr + offset)) {
> > +			ret = -EINVAL;
> > +			goto unlock;
> 
> If this fails we are left with the mapping inside the hyp because the unlock
> doesn't destroy the private mapping. Is this intended ?
> 

Yes, there is no way to remove a private mapping at the moment, all
the callers to __pkvm_create_private_mapping() will leak it on failure.

> > +		}
> > +		ret = kvm_pgtable_get_leaf(&host_mmu.pgt, addr + offset, &pte, NULL);
> > +		if (ret)
> > +			goto unlock;
> > +		if (pte && !kvm_pte_valid(pte)) {
> > +			ret = -EPERM;
> > +			goto unlock;
> > +		}
> > +	}
> > +	/*
> > +	 * We set HYP as the owner of the MMIO pages in the host stage-2, for:
> > +	 * - host aborts: host_stage2_adjust_range() would fail for invalid non zero PTEs.
> > +	 * - recycle under memory pressure: host_stage2_unmap_dev_all() would call
> > +	 *   kvm_pgtable_stage2_unmap() which will not clear non zero invalid ptes (counted).
> > +	 * - other MMIO donation: Would fail as we check that the PTE is valid or empty.
> > +	 */
> > +	ret = host_stage2_try(kvm_pgtable_stage2_annotate, &host_mmu.pgt,
> > +			      addr, size, &host_s2_pool,
> > +			      KVM_HOST_INVALID_PTE_TYPE_DONATION,
> > +			      FIELD_PREP(KVM_HOST_DONATION_PTE_OWNER_MASK, PKVM_ID_HYP));
> > +unlock:
> > +	host_unlock_component();
> > +	return ret;
> > +}
> > +
> > +int __pkvm_hyp_donate_host_mmio(phys_addr_t addr, size_t size)
> 
> This function seems to only update the host stage-2 annotation but it
> doesn't destroy the hyp mapping.

Yes, as mentioned above there is no way to destroy it, and this was
not desgined for frequent use. Typicaly, __pkvm_host_donate_hyp_mmio()
is called at boot per area/device. And __pkvm_hyp_donate_host_mmio()
is only used for failures.

> 
> I was looking to make use of this patch in an upcoming posting for the
> v2 ITS hardening but in my case I don't need the private VA range
> creation.

I believe if you need to map MMIO, private range is the right way to
do it as the linear map was mainly designed around system memory.

Thanks,
Mostafa