Re: [PATCH v4 4/4] KVM: mmu: remove over-aggressive warnings

From: David Stevens <stevensd@chromium.org>
To: Sean Christopherson <seanjc@google.com>
Cc: Marc Zyngier <maz@kernel.org>,
	Paolo Bonzini <pbonzini@redhat.com>,
	James Morse <james.morse@arm.com>,
	Alexandru Elisei <alexandru.elisei@arm.com>,
	Suzuki K Poulose <suzuki.poulose@arm.com>,
	Will Deacon <will@kernel.org>, Wanpeng Li <wanpengli@tencent.com>,
	Jim Mattson <jmattson@google.com>, Joerg Roedel <joro@8bytes.org>,
	linux-arm-kernel@lists.infradead.org,
	kvmarm@lists.cs.columbia.edu,
	open list <linux-kernel@vger.kernel.org>,
	kvm@vger.kernel.org
Subject: Re: [PATCH v4 4/4] KVM: mmu: remove over-aggressive warnings
Date: Wed, 13 Oct 2021 12:29:36 +0900	[thread overview]
Message-ID: <CAD=HUj5HCdBBU2z=yJCCiAhTj0ARj-8XpLqdVbam7Kt9af+SSA@mail.gmail.com> (raw)
In-Reply-To: <YWYiJy1Z7VZ0SxAd@google.com>

On Wed, Oct 13, 2021 at 9:02 AM Sean Christopherson <seanjc@google.com> wrote:
>
> On Wed, Sep 29, 2021, David Stevens wrote:
> > From: David Stevens <stevensd@chromium.org>
> >
> > Remove two warnings that require ref counts for pages to be non-zero, as
> > mapped pfns from follow_pfn may not have an initialized ref count.
> >
> > Signed-off-by: David Stevens <stevensd@chromium.org>
> > ---
> >  arch/x86/kvm/mmu/mmu.c | 7 -------
> >  virt/kvm/kvm_main.c    | 2 +-
> >  2 files changed, 1 insertion(+), 8 deletions(-)
> >
> > diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> > index 5a1adcc9cfbc..3b469df63bcf 100644
> > --- a/arch/x86/kvm/mmu/mmu.c
> > +++ b/arch/x86/kvm/mmu/mmu.c
> > @@ -617,13 +617,6 @@ static int mmu_spte_clear_track_bits(struct kvm *kvm, u64 *sptep)
> >
> >       pfn = spte_to_pfn(old_spte);
> >
> > -     /*
> > -      * KVM does not hold the refcount of the page used by
> > -      * kvm mmu, before reclaiming the page, we should
> > -      * unmap it from mmu first.
> > -      */
> > -     WARN_ON(!kvm_is_reserved_pfn(pfn) && !page_count(pfn_to_page(pfn)));
>
> Have you actually observed false positives with this WARN?  I would expect anything
> without a struct page to get filtered out by !kvm_is_reserved_pfn(pfn).

Those are the type of pfns that were responsible for CVE-2021-22543
[1]. One specific example is that amdgpu uses ttm_pool, which makes
higher order, non-compound allocation. Without the head/tail metadata,
only the first base page in such an allocation has non-zero
page_count.

[1] https://github.com/google/security-research/security/advisories/GHSA-7wq5-phmq-m584

> If you have observed false positives, I would strongly prefer we find a way to
> keep the page_count() sanity check, it has proven very helpful in the past in
> finding/debugging bugs during MMU development.

When we see a refcount of zero, I think we can look up spte->(gfn,
slot)->hva->vma and determine whether or not the zero refcount is
okay, based on vm_flags. That's kind of heavy for a debug check,
although at least we'd only pay the cost for unusual mappings. But it
still might make sense to switch to a MMU_WARN_ON, in that case. Or we
could just ignore the cost, since at least from a superficial reading
and some basic tests, tdp_mmu doesn't seem to execute this code path.

Thoughts? I'd lean towards MMU_WARN_ON, but I'd like to know what the
maintainers' preferences are before sending an updated patch series.

-David

>
> > -
> >       if (is_accessed_spte(old_spte))
> >               kvm_set_pfn_accessed(pfn);
> >