Re: [RFC 04/11] mm: remove SWAP_MLOCK check for SWAP_SUCCESS in ttu

[Minchan Kim]

On Tue, Mar 07, 2017 at 05:26:43PM +0300, Kirill A. Shutemov wrote:
> On Thu, Mar 02, 2017 at 03:39:18PM +0900, Minchan Kim wrote:
> > If the page is mapped and rescue in ttuo, page_mapcount(page) == 0 cannot
> > be true so page_mapcount check in ttu is enough to return SWAP_SUCCESS.
> > IOW, SWAP_MLOCK check is redundant so remove it.
> > 
> > Signed-off-by: Minchan Kim <minchan@xxxxxxxxxx>
> > ---
> >  mm/rmap.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/mm/rmap.c b/mm/rmap.c
> > index 3a14013..0a48958 100644
> > --- a/mm/rmap.c
> > +++ b/mm/rmap.c
> > @@ -1523,7 +1523,7 @@ int try_to_unmap(struct page *page, enum ttu_flags flags)
> >  	else
> >  		ret = rmap_walk(page, &rwc);
> >  
> > -	if (ret != SWAP_MLOCK && !page_mapcount(page))
> > +	if (!page_mapcount(page))
> 
> Hm. I think there's bug in current code.
> It should be !total_mapcount(page) otherwise it can be false-positive if
> there's THP mapped with PTEs.

Hmm, I lost THP thesedays totally so I can miss something easily.
When I look at that, it seems every pages passed try_to_unmap is already
splited by split split_huge_page_to_list which calls freeze_page which
split pmd. So I guess it's no problem. Right?

Anyway, it's out of scope in this patch so if it's really problem,
I'd like to handle it separately.

One asking:

When we should use total_mapcount instead of page_mapcount?
If total_mapcount has some lengthy description, it would be very helpful
for one who not is faimilar with that.

> 
> And in this case ret != SWAP_MLOCK is helpful to cut down some cost.
> Althouth it should be fine to remove it, I guess.

Sure but be hard to measure it, I think. As well, later patch removes
SWAP_MLOCK.