Re: [PATCH v7 19/20] fs/dax: Properly refcount fs dax pages

[David Hildenbrand]

>   
> -static inline unsigned long dax_page_share_put(struct page *page)
> +static inline unsigned long dax_folio_put(struct folio *folio)
>   {
> -	WARN_ON_ONCE(!page->share);
> -	return --page->share;
> +	unsigned long ref;
> +
> +	if (!dax_folio_is_shared(folio))
> +		ref = 0;
> +	else
> +		ref = --folio->share;
> +
> +	WARN_ON_ONCE(ref < 0);
> +	if (!ref) {
> +		folio->mapping = NULL;
> +		if (folio_order(folio)) {


I'd have made this easier to read by doing

if (ref)
	return ref;
folio->mapping = NULL;

order = folio_order(folio)
if (!order)
	return 0;

pgmap = page_pgmap(&folio->page);
for (i = 0; i < (1UL << order); i++) {
	// loop body see below
}
return 0;


In the context of similar split users and related discussions in the 
future with memdescs -- see 89a41a0263293856678189981e5407375261c4ff -- 
I would further do within the loop (avoiding messing with page-> 
completely):

...
	struct page *page = folio_page(folio, i);
	/* Careful: see __split_huge_page_tail() */
	struct folio *new_folio = (struct folio *)page;

	ClearPageHead(page);
	clear_compound_head(page);

	new_folio->mapping = NULL;	
	/*
	 * Reset pgmap which was over-written by
	 * prep_compound_page().
	 */
	new_folio->pgmap = pgmap;
	new_folio->share = 0;
	WARN_ON_ONCE(folio_ref_count(new_folio));
...


I scanned over the other stuff in here, but I'm not an expert on the 
pfn_t etc thingies.

--
Cheers,

David / dhildenb