[PATCH V2] powerpc/mm: Initialize kernel pagetable memory for PTE fragments

[Aneesh Kumar K.V]

On 06/20/2018 02:52 PM, Anshuman Khandual wrote:
> Kernel pagetable pages for PTE fragments never go through the standard init
> sequence which can cause inaccuracies in utilization statistics reported at
> places like /proc and /sysfs interfaces etc. Also the allocated page misses
> out on pagetable lock and page flag init as well. Fix it by making sure all
> pages allocated for either user process or kernel PTE fragments go through
> same initialization.
> 
> Signed-off-by: Anshuman Khandual <khandual at linux.vnet.ibm.com>
> ---
> Changes in V2:
> 
> - Call the destructor function during free for all cases
> 
>   arch/powerpc/mm/pgtable-book3s64.c | 27 ++++++++++++---------------
>   1 file changed, 12 insertions(+), 15 deletions(-)
> 
> diff --git a/arch/powerpc/mm/pgtable-book3s64.c b/arch/powerpc/mm/pgtable-book3s64.c
> index c1f4ca4..a820ee6 100644
> --- a/arch/powerpc/mm/pgtable-book3s64.c
> +++ b/arch/powerpc/mm/pgtable-book3s64.c
> @@ -335,23 +335,21 @@ static pte_t *get_pte_from_cache(struct mm_struct *mm)
>   
>   static pte_t *__alloc_for_ptecache(struct mm_struct *mm, int kernel)
>   {
> +	gfp_t gfp_mask = PGALLOC_GFP;
>   	void *ret = NULL;
>   	struct page *page;
>   
> -	if (!kernel) {
> -		page = alloc_page(PGALLOC_GFP | __GFP_ACCOUNT);
> -		if (!page)
> -			return NULL;
> -		if (!pgtable_page_ctor(page)) {
> -			__free_page(page);
> -			return NULL;
> -		}
> -	} else {
> -		page = alloc_page(PGALLOC_GFP);
> -		if (!page)
> -			return NULL;
> -	}
> +	if (!kernel)
> +		gfp_mask |= __GFP_ACCOUNT;
>   
> +	page = alloc_page(gfp_mask);
> +	if (!page)
> +		return NULL;
> +
> +	if (!pgtable_page_ctor(page)) {
> +		__free_page(page);
> +		return NULL;
> +	}
>   
>   	ret = page_address(page);
>   	/*
> @@ -391,8 +389,7 @@ void pte_fragment_free(unsigned long *table, int kernel)
>   	struct page *page = virt_to_page(table);
>   
>   	if (put_page_testzero(page)) {
> -		if (!kernel)
> -			pgtable_page_dtor(page);
> +		pgtable_page_dtor(page);
>   		free_unref_page(page);
>   	}
>   }
> 

If we really are going to do this, may be we can  do it as 
alloc_for_pmdcache, kill the kernel arg and compare against init_mm?

-aneesh