[PATCH 03/12] mm/pagewalk: Skip dax pages in pagewalk

[Alistair Popple]

On Thu, Jun 05, 2025 at 09:21:28AM -0700, Dan Williams wrote:
> David Hildenbrand wrote:
> > On 05.06.25 09:46, Christoph Hellwig wrote:
> > > On Wed, Jun 04, 2025 at 06:59:09PM -0700, Dan Williams wrote:
> > >> +/* return normal pages backed by the page allocator */
> > >> +static inline struct page *vm_normal_gfp_pmd(struct vm_area_struct *vma,
> > >> +					     unsigned long addr, pmd_t pmd)
> > >> +{
> > >> +	struct page *page = vm_normal_page_pmd(vma, addr, pmd);
> > >> +
> > >> +	if (!is_devdax_page(page) && !is_fsdax_page(page))
> > >> +		return page;
> > >> +	return NULL;
> > > 
> > > If you go for this make it more straight forward by having the
> > > normal path in the main flow:
> > > 
> > > 	if (is_devdax_page(page) || is_fsdax_page(page))
> > > 		return NULL;
> > > 	return page;
> > 
> > +1
> > 
> > But I'd defer introducing that for now if avoidable. I find the naming 
> > rather ... suboptimal :)
> 
> Agree, that was a "for lack of a better term" suggestion, but the
> suggestion is indeed lacking.

I don't like the naming either ... maybe that is motivation enough for me to
audit the callers and have them explicitly filter the pages they don't want.