[PATCH 6/6] arm/arm64: KVM: map MMIO regions at creation time

[christoffer.dall@linaro.org (Christoffer Dall)]

On Wed, Sep 17, 2014 at 02:56:21PM -0700, Ard Biesheuvel wrote:
> There is really no point in faulting in memory regions page by page
> if they are not backed by demand paged system RAM but by a linear
> passthrough mapping of a host MMIO region.
> 
> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
> ---
>  arch/arm/kvm/mmu.c | 50 +++++++++++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 49 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/arm/kvm/mmu.c b/arch/arm/kvm/mmu.c
> index fe53c3a30383..b153ef0c6d9f 100644
> --- a/arch/arm/kvm/mmu.c
> +++ b/arch/arm/kvm/mmu.c
> @@ -1162,7 +1162,55 @@ void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
>  int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
>  			    unsigned long npages)
>  {
> -	return 0;
> +	hva_t hva = slot->userspace_addr;
> +	phys_addr_t gpa = slot->base_gfn << PAGE_SHIFT;
> +	phys_addr_t size = slot->npages << PAGE_SHIFT;
> +	int ret = 0;
> +
> +	/*
> +	 * A memslot could potentially cover multiple VMAs, so iterate
> +	 * over all of them to find out if we can map any of them right now.
> +	 *
> +	 *     +--------------------------------------------+
> +	 * +---+---------+-------------------+--------------+----+
> +	 * |   : VMA 1   |       VMA 2       |       VMA 3  :    |
> +	 * +---+---------+-------------------+--------------+----+
> +	 *     |                   memslot                  |
> +	 *     +--------------------------------------------+
> +	 */
> +	do {
> +		struct vm_area_struct *vma = find_vma(current->mm, hva);
> +		hva_t start, end;
> +
> +		if (!vma || vma->vm_start > hva) {
> +			ret = -EFAULT;
> +			break;
> +		}
> +
> +		start = max(slot->userspace_addr, vma->vm_start);
> +		end = min((hva_t)(slot->userspace_addr + size), vma->vm_end);
> +
> +		if (vma->vm_flags & VM_PFNMAP) {
> +			phys_addr_t pa = (vma->vm_pgoff << PAGE_SHIFT) + start -
> +					 vma->vm_start;
> +			bool writable = vma->vm_flags & VM_WRITE &&
> +					!(slot->flags & KVM_MEM_READONLY);
> +
> +			ret = kvm_phys_addr_ioremap(kvm, gpa, pa, end - start,
> +						    writable);
> +			if (ret)
> +				break;
> +		}
> +		hva += end - start;
> +		gpa += end - start;
> +	} while (hva < slot->userspace_addr + size);
> +
> +	if (ret) {
> +		spin_lock(&kvm->mmu_lock);
> +		unmap_stage2_range(kvm, slot->base_gfn << PAGE_SHIFT, size);
> +		spin_unlock(&kvm->mmu_lock);
> +	}
> +	return ret;
>  }
>  
>  void kvm_arch_memslots_updated(struct kvm *kvm)
> -- 
> 1.8.3.2
> 

Looks really good!  But we should handle moving a memslot as well, which
also tells me we should probably move this logic to
kvm_arch_prepare_memory_region() instead...

Thanks,
-Christoffer