[PATCH] ARM: KVM: Handle IPA unmapping on memory region deletion

[Marc Zyngier]

Hi Eric,

On 03/04/14 16:17, Eric Auger wrote:
> Currently when a KVM region is removed using
> kvm_vm_ioctl_set_memory_region (with memory region size equal to 0), the
> corresponding intermediate physical memory is not unmapped.
> 
> This patch unmaps the region's IPA range in
> kvm_arch_commit_memory_region using unmap_stage2_range.
> 
> The patch was tested on QEMU VFIO based use case where RAM memory region
> creation/deletion frequently happens for IRQ handling.
> 
> Notes:
> - the KVM_MR_MOVE case shall request some similar addition but I cannot test
>   this currently

I think you should try to handle it anyway. I'm sure you could hack QEMU
to do this test it, but even if you don't, better plug that hole right now.

> Signed-off-by: Eric Auger <eric.auger at linaro.org>
> ---
>  arch/arm/include/asm/kvm_mmu.h | 2 ++
>  arch/arm/kvm/arm.c             | 8 ++++++++
>  arch/arm/kvm/mmu.c             | 2 +-
>  3 files changed, 11 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/arm/include/asm/kvm_mmu.h b/arch/arm/include/asm/kvm_mmu.h
> index 2d122ad..a91c863 100644
> --- a/arch/arm/include/asm/kvm_mmu.h
> +++ b/arch/arm/include/asm/kvm_mmu.h
> @@ -52,6 +52,8 @@ void kvm_free_stage2_pgd(struct kvm *kvm);
>  int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa,
>  			  phys_addr_t pa, unsigned long size);
>  
> +void unmap_stage2_range(struct kvm *kvm, phys_addr_t guest_ipa, u64 size);
> +
>  int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run);
>  
>  void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu);
> diff --git a/arch/arm/kvm/arm.c b/arch/arm/kvm/arm.c
> index bd18bb8..9a4bc10 100644
> --- a/arch/arm/kvm/arm.c
> +++ b/arch/arm/kvm/arm.c
> @@ -241,6 +241,14 @@ void kvm_arch_commit_memory_region(struct kvm *kvm,
>  				   const struct kvm_memory_slot *old,
>  				   enum kvm_mr_change change)
>  {
> +	if (change == KVM_MR_DELETE) {
> +		gpa_t gpa = old->base_gfn << PAGE_SHIFT;
> +		u64 size = old->npages << PAGE_SHIFT;
> +
> +		spin_lock(&kvm->mmu_lock);
> +		unmap_stage2_range(kvm, gpa, size);
> +		spin_unlock(&kvm->mmu_lock);
> +	}
>  }

Just move the whole function to mmu.c, as it makes more sense to have it
there. And while you're at it, how about moving the other
memslot/memory_region stubs?

>  void kvm_arch_flush_shadow_all(struct kvm *kvm)
> diff --git a/arch/arm/kvm/mmu.c b/arch/arm/kvm/mmu.c
> index 7789857..e8580e2 100644
> --- a/arch/arm/kvm/mmu.c
> +++ b/arch/arm/kvm/mmu.c
> @@ -443,7 +443,7 @@ int kvm_alloc_stage2_pgd(struct kvm *kvm)
>   * destroying the VM), otherwise another faulting VCPU may come in and mess
>   * with things behind our backs.
>   */
> -static void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size)
> +void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size)
>  {
>  	unmap_range(kvm, kvm->arch.pgd, start, size);
>  }
> 

Looks sensible otherwise.

Thanks!

	M.
-- 
Jazz is not dead. It just smells funny...