[PATCH v2 2/6] KVM: arm64: Walk userspace page tables to compute the THP mapping size

[Alexandru Elisei]

Hi Marc,

On 7/26/21 4:35 PM, Marc Zyngier wrote:
> We currently rely on the kvm_is_transparent_hugepage() helper to
> discover whether a given page has the potential to be mapped as
> a block mapping.
>
> However, this API doesn't really give un everything we want:
> - we don't get the size: this is not crucial today as we only
>   support PMD-sized THPs, but we'd like to have larger sizes
>   in the future
> - we're the only user left of the API, and there is a will
>   to remove it altogether
>
> To address the above, implement a simple walker using the existing
> page table infrastructure, and plumb it into transparent_hugepage_adjust().
> No new page sizes are supported in the process.
>
> Signed-off-by: Marc Zyngier <maz at kernel.org>
> ---
>  arch/arm64/kvm/mmu.c | 34 ++++++++++++++++++++++++++++++----
>  1 file changed, 30 insertions(+), 4 deletions(-)
>
> diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
> index 3155c9e778f0..0adc1617c557 100644
> --- a/arch/arm64/kvm/mmu.c
> +++ b/arch/arm64/kvm/mmu.c
> @@ -433,6 +433,32 @@ int create_hyp_exec_mappings(phys_addr_t phys_addr, size_t size,
>  	return 0;
>  }
>  
> +static struct kvm_pgtable_mm_ops kvm_user_mm_ops = {
> +	/* We shouldn't need any other callback to walk the PT */

That looks correct to me, mm_ops is used in __kvm_pgtable_visit(), and then only
the phys_to_virt field callback is used. kvm_host_va() is also the callback used
by kvm_s2_mm_ops, which looks right to me.

> +	.phys_to_virt		= kvm_host_va,
> +};
> +
> +static int get_user_mapping_size(struct kvm *kvm, u64 addr)
> +{
> +	struct kvm_pgtable pgt = {
> +		.pgd		= (kvm_pte_t *)kvm->mm->pgd,
> +		.ia_bits	= VA_BITS,
> +		.start_level	= (KVM_PGTABLE_MAX_LEVELS -
> +				   CONFIG_PGTABLE_LEVELS),
> +		.mm_ops		= &kvm_user_mm_ops,
> +	};
> +	kvm_pte_t pte = 0;	/* Keep GCC quiet... */
> +	u32 level = ~0;
> +	int ret;
> +
> +	ret = kvm_pgtable_get_leaf(&pgt, addr, &pte, &level);
> +	VM_BUG_ON(ret);
> +	VM_BUG_ON(level >= KVM_PGTABLE_MAX_LEVELS);
> +	VM_BUG_ON(!(pte & PTE_VALID));
> +
> +	return BIT(ARM64_HW_PGTABLE_LEVEL_SHIFT(level));
> +}
> +
>  static struct kvm_pgtable_mm_ops kvm_s2_mm_ops = {
>  	.zalloc_page		= stage2_memcache_zalloc_page,
>  	.zalloc_pages_exact	= kvm_host_zalloc_pages_exact,
> @@ -780,7 +806,7 @@ static bool fault_supports_stage2_huge_mapping(struct kvm_memory_slot *memslot,
>   * Returns the size of the mapping.
>   */
>  static unsigned long
> -transparent_hugepage_adjust(struct kvm_memory_slot *memslot,
> +transparent_hugepage_adjust(struct kvm *kvm, struct kvm_memory_slot *memslot,
>  			    unsigned long hva, kvm_pfn_t *pfnp,
>  			    phys_addr_t *ipap)
>  {
> @@ -791,8 +817,8 @@ transparent_hugepage_adjust(struct kvm_memory_slot *memslot,
>  	 * sure that the HVA and IPA are sufficiently aligned and that the
>  	 * block map is contained within the memslot.
>  	 */
> -	if (kvm_is_transparent_hugepage(pfn) &&
> -	    fault_supports_stage2_huge_mapping(memslot, hva, PMD_SIZE)) {
> +	if (fault_supports_stage2_huge_mapping(memslot, hva, PMD_SIZE) &&
> +	    get_user_mapping_size(kvm, hva) >= PMD_SIZE) {
>  		/*
>  		 * The address we faulted on is backed by a transparent huge
>  		 * page.  However, because we map the compound huge page and
> @@ -1051,7 +1077,7 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
>  	 * backed by a THP and thus use block mapping if possible.
>  	 */
>  	if (vma_pagesize == PAGE_SIZE && !(force_pte || device))
> -		vma_pagesize = transparent_hugepage_adjust(memslot, hva,
> +		vma_pagesize = transparent_hugepage_adjust(kvm, memslot, hva,
>  							   &pfn, &fault_ipa);
>  
>  	if (fault_status != FSC_PERM && !device && kvm_has_mte(kvm)) {

Sean explained well why holding the mmap lock isn't needed here. The patch looks
correct to me:

Reviewed-by: Alexandru Elisei <alexandru.elisei at arm.com>

Thanks,

Alex