[PATCH v3 06/15] KVM: arm64: Implement kvm_pgtable_hyp_unmap() at EL2

[Quentin Perret]

Hi Andrew,

On Tuesday 07 Dec 2021 at 14:47:14 (+0000), Andrew Walbran wrote:
> On Wed, 1 Dec 2021 at 17:04, 'Quentin Perret' via kernel-team
> <kernel-team at android.com> wrote:
> >
> > From: Will Deacon <will at kernel.org>
> >
> > Implement kvm_pgtable_hyp_unmap() which can be used to remove hypervisor
> > stage-1 mappings at EL2.
> >
> > Signed-off-by: Will Deacon <will at kernel.org>
> > Signed-off-by: Quentin Perret <qperret at google.com>
> > ---
> >  arch/arm64/include/asm/kvm_pgtable.h | 21 ++++++++++
> >  arch/arm64/kvm/hyp/pgtable.c         | 63 ++++++++++++++++++++++++++++
> >  2 files changed, 84 insertions(+)
> >
> > diff --git a/arch/arm64/include/asm/kvm_pgtable.h b/arch/arm64/include/asm/kvm_pgtable.h
> > index 027783829584..9d076f36401d 100644
> > --- a/arch/arm64/include/asm/kvm_pgtable.h
> > +++ b/arch/arm64/include/asm/kvm_pgtable.h
> > @@ -251,6 +251,27 @@ void kvm_pgtable_hyp_destroy(struct kvm_pgtable *pgt);
> >  int kvm_pgtable_hyp_map(struct kvm_pgtable *pgt, u64 addr, u64 size, u64 phys,
> >                         enum kvm_pgtable_prot prot);
> >
> > +/**
> > + * kvm_pgtable_hyp_unmap() - Remove a mapping from a hypervisor stage-1 page-table.
> > + * @pgt:       Page-table structure initialised by kvm_pgtable_hyp_init().
> > + * @addr:      Virtual address from which to remove the mapping.
> > + * @size:      Size of the mapping.
> > + *
> > + * The offset of @addr within a page is ignored, @size is rounded-up to
> > + * the next page boundary and @phys is rounded-down to the previous page
> > + * boundary.
> > + *
> > + * TLB invalidation is performed for each page-table entry cleared during the
> > + * unmapping operation and the reference count for the page-table page
> > + * containing the cleared entry is decremented, with unreferenced pages being
> > + * freed. The unmapping operation will stop early if it encounters either an
> > + * invalid page-table entry or a valid block mapping which maps beyond the range
> > + * being unmapped.
> 
> How is the caller expected to break up the block mapping? Why not
> handle that within this function?

We don't really use block mappings for the hyp stage-1, since pretty
much forever (see the loop in pkvm_create_mappings_locked() for ex), so
handling it here would be somewhat unnecessary complexity. Handling this
in the pgtable code itself (which I assume would mean proactively
re-mapping the rest of the range with page-granularity mappings or
something along those lines) is tricky because of BBM and concurrency,
so I'd rather avoid handling same-level aborts at EL2 and all that mess
unless we have a good reason. Is there a use-case where you think that'd
be needed?

Cheers,
Quentin