Re: [PATCH 07/10] mbind: add hugepage migration code to mbind()

[Michal Hocko <mhocko@suse.cz>]

On Fri 05-04-13 18:23:07, KOSAKI Motohiro wrote:
> >> -	if (!new_hpage)
> >> +	/*
> >> +	 * Getting a new hugepage with alloc_huge_page() (which can happen
> >> +	 * when migration is caused by mbind()) can return ERR_PTR value,
> >> +	 * so we need take care of the case here.
> >> +	 */
> >> +	if (!new_hpage || IS_ERR_VALUE(new_hpage))
> >>  		return -ENOMEM;
> > 
> > Please no. get_new_page returns NULL or a page. You are hooking a wrong
> > callback here. The error value doesn't make any sense here. IMO you
> > should just wrap alloc_huge_page by something that returns NULL or page.
> 
> I suggest just opposite way. new_vma_page() always return ENOMEM, ENOSPC etc instad 
> of NULL. and caller propegate it to userland.
> I guess userland want to distingush why mbind was failed.

Sure, and I wasn't suggesting to change alloc_huge_page. I was just
pointing out that new_page_t used to return page or NULL and hugetlb
unmap_and_move shouldn't be any different in that direction so using
alloc_huge_page is not a good fit here.

> Anyway, If new_vma_page() have a change to return both NULL and
> -ENOMEM. That's a bug.

-- 
Michal Hocko
SUSE Labs