Re: [PATCH] mm: thp: fix potential clearing to referenced flag in page_idle_clear_pte_refs_one()

[Kirill A. Shutemov]

On Thu, Feb 08, 2018 at 08:47:35PM -0800, Yang Shi wrote:
> 
> 
> On 2/8/18 8:33 PM, Kirill A. Shutemov wrote:
> > On Thu, Feb 08, 2018 at 02:39:26PM -0800, Andrew Morton wrote:
> > > On Tue,  6 Feb 2018 08:06:36 +0800 Yang Shi <yang.shi@xxxxxxxxxxxxxxxxx> wrote:
> > > 
> > > > For PTE-mapped THP, the compound THP has not been split to normal 4K
> > > > pages yet, the whole THP is considered referenced if any one of sub
> > > > page is referenced.
> > > > 
> > > > When walking PTE-mapped THP by pvmw, all relevant PTEs will be checked
> > > > to retrieve referenced bit. But, the current code just returns the
> > > > result of the last PTE. If the last PTE has not referenced, the
> > > > referenced flag will be cleared.
> > > > 
> > > > So, here just break pvmw walk once referenced PTE is found if the page
> > > > is a part of THP.
> > > > 
> > > > ...
> > > > 
> > > > --- a/mm/page_idle.c
> > > > +++ b/mm/page_idle.c
> > > > @@ -67,6 +67,14 @@ static bool page_idle_clear_pte_refs_one(struct page *page,
> > > >   		if (pvmw.pte) {
> > > >   			referenced = ptep_clear_young_notify(vma, addr,
> > > >   					pvmw.pte);
> > > > +			/*
> > > > +			 * For PTE-mapped THP, one sub page is referenced,
> > > > +			 * the whole THP is referenced.
> > > > +			 */
> > > > +			if (referenced && PageTransCompound(pvmw.page)) {
> > > > +				page_vma_mapped_walk_done(&pvmw);
> > > > +				break;
> > > > +			}
> > > This means that the function will no longer clear the referenced bits
> > > in all the ptes.  What effect does this have and should we document
> > > this in some fashion?
> > Yeah, the patch is wrong. We need to get all ptes for THP cleared.
> > 
> > What about something like this instead (untested):
> 
> Thanks, Kirill. It looks correct. All ptes should be cleared.
> 
> I'm going to prepare v2 patch.

Note, it should be ||=, not |= (although it would work correctly too).

I should really wake up properly before touching code. :-/

-- 
 Kirill A. Shutemov