[PATCH v2 25/35] KVM: arm64: Reclaim faulting page from pKVM in spurious fault handler

[Will Deacon]

On Thu, Feb 12, 2026 at 05:22:28PM +0000, Alexandru Elisei wrote:
> Would be nice to merge this with the previous patch, that added the force
> reclaim function, as it would make reviewing easier.

I deliberately kept them separate as the previous patch isn't exactly
small and this makes the EL1 part easier to review in isolation.

> > diff --git a/arch/arm64/kvm/pkvm.c b/arch/arm64/kvm/pkvm.c
> > index 8be91051699e..d1926cb08c76 100644
> > --- a/arch/arm64/kvm/pkvm.c
> > +++ b/arch/arm64/kvm/pkvm.c
> > @@ -563,3 +563,10 @@ int pkvm_pgtable_stage2_split(struct kvm_pgtable *pgt, u64 addr, u64 size,
> >  	WARN_ON_ONCE(1);
> >  	return -EINVAL;
> >  }
> > +
> > +bool pkvm_reclaim_guest_page(phys_addr_t phys)
> > +{
> > +	int ret = kvm_call_hyp_nvhe(__pkvm_force_reclaim_guest_page, phys);
> 
> Nitpicking here, we have the functions __pkvm_reclaim_page_guest() and this
> function, pkvm_reclaim_guest_page(), which calls
> __pkvm_force_reclaim_guest_page, which in turn calls
> __pkvm_host_force_reclaim_page_guest(). I think having a bit of naming
> consistency would be really useful when navigating the source code.

Yuck, you're right! I'll make sure 'force' is used consistently on this
path.

> It might also be useful to document that callers of the hypercall
> __pkvm_force_reclaim_guest_page are not expected to unpin the page in case of
> success, but callers of __pkvm_reclaim_dying_guest_page are.

I'll add a comment, but this is the private hypercall interface with the
host and, as lwn might put it, it's rigorously undocumented. Unpinning
memory from panic context is an interesting idea ;)

Will