Date:   Fri, 10 Dec 2021 14:34:16 +0000
From:   Quentin Perret <qperret@google.com>
To:     Will Deacon <will@kernel.org>
Cc:     Marc Zyngier <maz@kernel.org>, James Morse <james.morse@arm.com>,
        Alexandru Elisei <alexandru.elisei@arm.com>,
        Suzuki K Poulose <suzuki.poulose@arm.com>,
        Catalin Marinas <catalin.marinas@arm.com>,
        linux-arm-kernel@lists.infradead.org, kvmarm@lists.cs.columbia.edu,
        linux-kernel@vger.kernel.org, kernel-team@android.com
Subject: Re: [PATCH v3 03/15] KVM: arm64: Refcount hyp stage-1 pgtable pages
Message-ID: <YbNlaIczPLXwSCcZ@google.com>
References: <20211201170411.1561936-1-qperret@google.com>
 <20211201170411.1561936-4-qperret@google.com>
 <20211209102924.GC1833@willie-the-truck>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20211209102924.GC1833@willie-the-truck>
Precedence: bulk

On Thursday 09 Dec 2021 at 10:29:24 (+0000), Will Deacon wrote:
> On Wed, Dec 01, 2021 at 05:03:57PM +0000, Quentin Perret wrote:
> > To prepare the ground for allowing hyp stage-1 mappings to be removed at
> > run-time, update the KVM page-table code to maintain a correct refcount
> > using the ->{get,put}_page() function callbacks.
> > 
> > Signed-off-by: Quentin Perret <qperret@google.com>
> > ---
> >  arch/arm64/kvm/hyp/pgtable.c | 17 ++++++++++++++---
> >  1 file changed, 14 insertions(+), 3 deletions(-)
> > 
> > diff --git a/arch/arm64/kvm/hyp/pgtable.c b/arch/arm64/kvm/hyp/pgtable.c
> > index f8ceebe4982e..768a58835153 100644
> > --- a/arch/arm64/kvm/hyp/pgtable.c
> > +++ b/arch/arm64/kvm/hyp/pgtable.c
> > @@ -408,8 +408,10 @@ static bool hyp_map_walker_try_leaf(u64 addr, u64 end, u32 level,
> >  		return false;
> >  
> >  	new = kvm_init_valid_leaf_pte(phys, data->attr, level);
> > -	if (hyp_pte_needs_update(old, new))
> > +	if (hyp_pte_needs_update(old, new)) {
> >  		smp_store_release(ptep, new);
> > +		data->mm_ops->get_page(ptep);
> 
> In the case where we're just updating software bits for a valid pte, doesn't
> this result in us taking a spurious reference to the page?

Ahem, yes, that is the case. I ended up with the below diff to fix it,
which I intend to fold in the next version:

diff --git a/arch/arm64/kvm/hyp/pgtable.c b/arch/arm64/kvm/hyp/pgtable.c
index 6ad4cb2d6947..e2047d3f05a2 100644
--- a/arch/arm64/kvm/hyp/pgtable.c
+++ b/arch/arm64/kvm/hyp/pgtable.c
@@ -383,21 +383,6 @@ enum kvm_pgtable_prot kvm_pgtable_hyp_pte_prot(kvm_pte_t pte)
        return prot;
 }

-static bool hyp_pte_needs_update(kvm_pte_t old, kvm_pte_t new)
-{
-       /*
-        * Tolerate KVM recreating the exact same mapping, or changing software
-        * bits if the existing mapping was valid.
-        */
-       if (old == new)
-               return false;
-
-       if (!kvm_pte_valid(old))
-               return true;
-
-       return !WARN_ON((old ^ new) & ~KVM_PTE_LEAF_ATTR_HI_SW);
-}
-
 static bool hyp_map_walker_try_leaf(u64 addr, u64 end, u32 level,
                                    kvm_pte_t *ptep, struct hyp_map_data *data)
 {
@@ -407,13 +392,16 @@ static bool hyp_map_walker_try_leaf(u64 addr, u64 end, u32 level,
        if (!kvm_block_mapping_supported(addr, end, phys, level))
                return false;

+       data->phys += granule;
        new = kvm_init_valid_leaf_pte(phys, data->attr, level);
-       if (hyp_pte_needs_update(old, new)) {
-               smp_store_release(ptep, new);
+       if (old == new)
+               return true;
+       else if (!kvm_pte_valid(old))
                data->mm_ops->get_page(ptep);
-       }
+       else if (WARN_ON((old ^ new) & ~KVM_PTE_LEAF_ATTR_HI_SW))
+               return false;

-       data->phys += granule;
+       smp_store_release(ptep, new);
        return true;
 }