[PATCH v4 10/28] KVM: arm64: iommu: Shadow host stage-2 page table

[Will Deacon]

On Tue, Aug 19, 2025 at 09:51:38PM +0000, Mostafa Saleh wrote:
> Create a shadow page table for the IOMMU that shadows the
> host CPU stage-2 into the IOMMUs to establish DMA isolation.
> 
> An initial snapshot is created after the driver init, then
> on every permission change a callback would be called for
> the IOMMU driver to update the page table.
> 
> For some cases, an SMMUv3 may be able to share the same page
> table used with the host CPU stage-2 directly.
> However, this is too strict and requires changes to the core hypervisor
> page table code, plus it would require the hypervisor to handle IOMMU
> page faults. This can be added later as an optimization for SMMUV3.
> 
> Signed-off-by: Mostafa Saleh <smostafa at google.com>
> ---
>  arch/arm64/kvm/hyp/include/nvhe/iommu.h |  4 ++
>  arch/arm64/kvm/hyp/nvhe/iommu/iommu.c   | 83 ++++++++++++++++++++++++-
>  arch/arm64/kvm/hyp/nvhe/mem_protect.c   |  5 ++
>  3 files changed, 90 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/kvm/hyp/include/nvhe/iommu.h b/arch/arm64/kvm/hyp/include/nvhe/iommu.h
> index 1ac70cc28a9e..219363045b1c 100644
> --- a/arch/arm64/kvm/hyp/include/nvhe/iommu.h
> +++ b/arch/arm64/kvm/hyp/include/nvhe/iommu.h
> @@ -3,11 +3,15 @@
>  #define __ARM64_KVM_NVHE_IOMMU_H__
>  
>  #include <asm/kvm_host.h>
> +#include <asm/kvm_pgtable.h>
>  
>  struct kvm_iommu_ops {
>  	int (*init)(void);
> +	void (*host_stage2_idmap)(phys_addr_t start, phys_addr_t end, int prot);
>  };
>  
>  int kvm_iommu_init(void);
>  
> +void kvm_iommu_host_stage2_idmap(phys_addr_t start, phys_addr_t end,
> +				 enum kvm_pgtable_prot prot);
>  #endif /* __ARM64_KVM_NVHE_IOMMU_H__ */
> diff --git a/arch/arm64/kvm/hyp/nvhe/iommu/iommu.c b/arch/arm64/kvm/hyp/nvhe/iommu/iommu.c
> index a01c036c55be..f7d1c8feb358 100644
> --- a/arch/arm64/kvm/hyp/nvhe/iommu/iommu.c
> +++ b/arch/arm64/kvm/hyp/nvhe/iommu/iommu.c
> @@ -4,15 +4,94 @@
>   *
>   * Copyright (C) 2022 Linaro Ltd.
>   */
> +#include <linux/iommu.h>
> +
>  #include <nvhe/iommu.h>
> +#include <nvhe/mem_protect.h>
> +#include <nvhe/spinlock.h>
>  
>  /* Only one set of ops supported */
>  struct kvm_iommu_ops *kvm_iommu_ops;
>  
> +/* Protected by host_mmu.lock */
> +static bool kvm_idmap_initialized;
> +
> +static inline int pkvm_to_iommu_prot(enum kvm_pgtable_prot prot)
> +{
> +	int iommu_prot = 0;
> +
> +	if (prot & KVM_PGTABLE_PROT_R)
> +		iommu_prot |= IOMMU_READ;
> +	if (prot & KVM_PGTABLE_PROT_W)
> +		iommu_prot |= IOMMU_WRITE;
> +	if (prot == PKVM_HOST_MMIO_PROT)
> +		iommu_prot |= IOMMU_MMIO;

This looks a little odd to me.

On the CPU side, the only different between PKVM_HOST_MEM_PROT and
PKVM_HOST_MMIO_PROT is that the former has execute permission. Both are
mapped as cacheable at stage-2 because it's the job of the host to set
the more restrictive memory type at stage-1.

Carrying that over to the SMMU would suggest that we don't care about
IOMMU_MMIO at stage-2 at all, so why do we need to set it here?

> +	/* We don't understand that, might be dangerous. */
> +	WARN_ON(prot & ~PKVM_HOST_MEM_PROT);
> +	return iommu_prot;
> +}
> +
> +static int __snapshot_host_stage2(const struct kvm_pgtable_visit_ctx *ctx,
> +				  enum kvm_pgtable_walk_flags visit)
> +{
> +	u64 start = ctx->addr;
> +	kvm_pte_t pte = *ctx->ptep;
> +	u32 level = ctx->level;
> +	u64 end = start + kvm_granule_size(level);
> +	int prot =  IOMMU_READ | IOMMU_WRITE;
> +
> +	/* Keep unmapped. */
> +	if (pte && !kvm_pte_valid(pte))
> +		return 0;
> +
> +	if (kvm_pte_valid(pte))
> +		prot = pkvm_to_iommu_prot(kvm_pgtable_stage2_pte_prot(pte));
> +	else if (!addr_is_memory(start))
> +		prot |= IOMMU_MMIO;

Why do we need to map MMIO regions pro-actively here? I'd have thought
we could just do:

	if (!kvm_pte_valid(pte))
		return 0;

	prot = pkvm_to_iommu_prot(kvm_pgtable_stage2_pte_prot(pte);
	kvm_iommu_ops->host_stage2_idmap(start, end, prot);
	return 0;

but I think that IOMMU_MMIO is throwing me again...

Will