Re: [RFC][PATCH v3 1/6] mm: Don't unmap gup()ed page

[Andrea Arcangeli <aarcange@redhat.com>]

On Wed, Apr 15, 2009 at 05:05:54PM +0900, KOSAKI Motohiro wrote:
> -	/*
> -	 * If the page is mlock()d, we cannot swap it out.
> -	 * If it's recently referenced (perhaps page_referenced
> -	 * skipped over this mm) then we should reactivate it.
> -	 */
>  	if (!migration) {
> +		if (PageSwapCache(page) &&
> +		    page_count(page) != page_mapcount(page) + 2) {
> +			ret = SWAP_FAIL;
> +			goto out_unmap;
> +		}
> +
> +		/*
> +		 * If the page is mlock()d, we cannot swap it out.
> +		 * If it's recently referenced (perhaps page_referenced
> +		 * skipped over this mm) then we should reactivate it.
> +		 */
>  		if (vma->vm_flags & VM_LOCKED) {
>  			ret = SWAP_MLOCK;
>  			goto out_unmap;
> @@ -790,7 +796,19 @@ static int try_to_unmap_one(struct page 
>  
>  	/* Nuke the page table entry. */
>  	flush_cache_page(vma, address, page_to_pfn(page));
> -	pteval = ptep_clear_flush_notify(vma, address, pte);
> +	pteval = ptep_clear_flush(vma, address, pte);
> +
> +	if (!migration) {
> +		/* re-check */
> +		if (PageSwapCache(page) &&
> +		    page_count(page) != page_mapcount(page) + 2) {
> +			/* We lose race against get_user_pages_fast() */
> +			set_pte_at(mm, address, pte, pteval);
> +			ret = SWAP_FAIL;
> +			goto out_unmap;
> +		}
> +	}
> +	mmu_notifier_invalidate_page(vma->vm_mm, address);

With regard to mmu notifier, this is the opposite of the right
ordering. One mmu_notifier_invalidate_page must run _before_ the first
check. The ptep_clear_flush_notify will then stay and there's no need
of a further mmu_notifier_invalidate_page after the second check.