Re: [PATCH v3] mm/mempolicy,hugetlb: Checking hstate for hugetlbfs page in vma_migratable

[Mike Kravetz <mike.kravetz@oracle.com>]

On 1/15/20 5:39 AM, Li Xinhai wrote:
> Checking hstate at early phase when isolating page, instead of during
> unmap and move phase, to avoid useless isolation.
> 
> Signed-off-by: Li Xinhai <lixinhai.lxh@gmail.com>
> Cc: Michal Hocko <mhocko@suse.com>
> Cc: Mike Kravetz <mike.kravetz@oracle.com>
> ---
>  include/linux/hugetlb_inline.h | 7 +++++++
>  include/linux/mempolicy.h      | 5 ++---
>  mm/hugetlb.c                   | 9 +++++++++
>  3 files changed, 18 insertions(+), 3 deletions(-)
> 
> diff --git a/include/linux/hugetlb_inline.h b/include/linux/hugetlb_inline.h
> index 0660a03..fc07139 100644
> --- a/include/linux/hugetlb_inline.h
> +++ b/include/linux/hugetlb_inline.h
> @@ -11,6 +11,8 @@ static inline bool is_vm_hugetlb_page(struct vm_area_struct *vma)
>  	return !!(vma->vm_flags & VM_HUGETLB);
>  }
>  
> +extern bool vm_hugepage_migration_supported(struct vm_area_struct *vma);
> +
>  #else
>  
>  static inline bool is_vm_hugetlb_page(struct vm_area_struct *vma)
> @@ -18,6 +20,11 @@ static inline bool is_vm_hugetlb_page(struct vm_area_struct *vma)
>  	return false;
>  }
>  
> +static inline bool vm_hugepage_migration_supported(struct vm_area_struct *vma)
> +{
> +	return false;
> +}
> +
>  #endif
>  
>  #endif
> diff --git a/include/linux/mempolicy.h b/include/linux/mempolicy.h
> index 5228c62..e9ed49e 100644
> --- a/include/linux/mempolicy.h
> +++ b/include/linux/mempolicy.h
> @@ -185,10 +185,9 @@ static inline bool vma_migratable(struct vm_area_struct *vma)
>  	if (vma_is_dax(vma))
>  		return false;
>  
> -#ifndef CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION
> -	if (vma->vm_flags & VM_HUGETLB)
> +	if (is_vm_hugetlb_page(vma) &&
> +		!vm_hugepage_migration_supported(vma))
>  		return false;
> -#endif
>  
>  	/*
>  	 * Migration allocates pages in the highest zone. If we cannot

There are not many ways to work around the hugepage_migration_supported
dependency issue.  I suppose you could have moved vma_migratable() into
mm/mempolicy.c instead of being an inline in the header file.  However,
it is probably best to do what you have done and hide the ugly stuff in
the hugetlbfs files.

> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index dd8737a..8ce4a98 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -1316,6 +1316,15 @@ int PageHeadHuge(struct page *page_head)
>  	return get_compound_page_dtor(page_head) == free_huge_page;
>  }
>  
> +bool vm_hugepage_migration_supported(struct vm_area_struct *vma)
> +{
> +	VM_BUG_ON_VMA(!is_vm_hugetlb_page(vma), vma);

This VM_BUG_ON_VMA() seems unnecessary.  The only caller is the one
added with this patch which is only called if is_vm_hugetlb_page(vma).
Even though VM_BUG_ON() is only enabled if CONFIG_DEBUG_VM, many distros
(including the one running my desktop) enable CONFIG_DEBUG_VM in their
default kernel.

-- 
Mike Kravetz

> +	if (hugepage_migration_supported(hstate_vma(vma)))
> +		return true;
> +
> +	return false;
> +}
> +
>  pgoff_t __basepage_index(struct page *page)
>  {
>  	struct page *page_head = compound_head(page);
>