[PATCH v2 01/22] KVM: arm64: Prefault host stage-2 entries on block split

[Vincent Donnefort]

On Thu, Sep 24, 2026 at 05:38:49PM +0000, Mostafa Saleh wrote:
> On Fri, Sep 11, 2026 at 02:50:32PM +0100, 'Vincent Donnefort' via kernel-team wrote:
> > From: Keir Fraser <keirf at google.com>
> > 
> > For the host stage-2 that is idmap, we can avoid lazy-mapping on block
> > splitting by propagating existing valid mappings to neighbouring entries
> > of the new table.
> > 
> > Signed-off-by: Keir Fraser <keirf at google.com>
> > Co-developed-by: Quentin Perret <qperret at google.com>
> > Signed-off-by: Quentin Perret <qperret at google.com>
> > Signed-off-by: Vincent Donnefort <vdonnefort at google.com>
> > ---
> >  arch/arm64/kvm/hyp/pgtable.c | 32 ++++++++++++++++++++++++++++++++
> >  1 file changed, 32 insertions(+)
> > 
> > diff --git a/arch/arm64/kvm/hyp/pgtable.c b/arch/arm64/kvm/hyp/pgtable.c
> > index b74dd5ce1efd..34f78a58cded 100644
> > --- a/arch/arm64/kvm/hyp/pgtable.c
> > +++ b/arch/arm64/kvm/hyp/pgtable.c
> > @@ -1030,10 +1030,39 @@ static int stage2_map_walk_table_pre(const struct kvm_pgtable_visit_ctx *ctx,
> >  	return 0;
> >  }
> >  
> > +static void stage2_map_prefault_idmap(const struct kvm_pgtable_visit_ctx *ctx, kvm_pte_t *ptep)
> > +{
> > +	kvm_pte_t block_pte = ctx->old;
> > +	u64 pa;
> > +	int i;
> > +
> > +	if (!kvm_pte_valid(block_pte))
> > +		return;
> > +
> > +	pa = ALIGN_DOWN(ctx->addr, kvm_granule_size(ctx->level));
> > +	for (i = 0; i < PTRS_PER_PTE; ++i, ++ptep, pa += kvm_granule_size(ctx->level + 1)) {
> > +		kvm_pte_t pte = kvm_init_valid_leaf_pte(pa, block_pte, ctx->level + 1);
> > +
> > +		/*
> > +		 * Skip ptes in the range being modified by the caller if we're
> > +		 * installing last level entries. Otherwise, we need to
> > +		 * temporarily put in a valid mapping to make sure the
> > +		 * prefaulting logic is triggered on the next
> > +		 * stage2_map_walk_leaf(). This adds an unnecessary TLBI as
> > +		 * we'll presumably re-break the freshly installed block, but
> > +		 * that should happen very infrequently.
> > +		 */
> > +		if ((ctx->level < (KVM_PGTABLE_LAST_LEVEL - 1)) ||
> > +		    (pa < ctx->addr) || (pa >= ctx->end))
> > +			*ptep = pte;
> 
> Do we need now to call get_page() on the table as we are installing a
> new PTE?

Ouch, I really start mixing android branches with upstream... I thought this
wasn't an issue because the host stage-2 doesn't refcount valid PTE... but this
is only true on Android branches, not in the current upstream...

So yes, you are right (and so were Wei-Lin who reported it as well on v1), I
should move the if (counted) get_page() from PATCH2 into PATCH1.

> 
> Also, I think we need a WRITE_ONCE() for the PTE write.

This one though I am pretty sure it isn't needed because of the
smp_store_release() when replacing the PTE with a table.
> 
> Thanks,
> Mostafa
> 
> > +	}
> > +}
> > +
> >  static int stage2_map_walk_leaf(const struct kvm_pgtable_visit_ctx *ctx,
> >  				struct stage2_map_data *data)
> >  {
> >  	struct kvm_pgtable_mm_ops *mm_ops = ctx->mm_ops;
> > +	struct kvm_pgtable *pgt = data->mmu->pgt;
> >  	kvm_pte_t *childp, new;
> >  	int ret;
> >  
> > @@ -1051,6 +1080,9 @@ static int stage2_map_walk_leaf(const struct kvm_pgtable_visit_ctx *ctx,
> >  	if (!childp)
> >  		return -ENOMEM;
> >  
> > +	if (pgt->flags & KVM_PGTABLE_S2_IDMAP)
> > +		stage2_map_prefault_idmap(ctx, childp);
> > +
> >  	if (!stage2_try_break_pte(ctx, data->mmu)) {
> >  		mm_ops->put_page(childp);
> >  		return -EAGAIN;
> > -- 
> > 2.55.0.1007.g17ff1f9808-goog
> > 
> > To unsubscribe from this group and stop receiving emails from it, send an email to kernel-team+unsubscribe at android.com.
> > 

-- 
Vincent