Re: [RFC][PATCH 5/6] teach smaps_pte_range() about THP pmds

[Andrea Arcangeli <aarcange@redhat.com>]

On Tue, Feb 01, 2011 at 07:02:30AM -0800, Dave Hansen wrote:
> On Tue, 2011-02-01 at 11:11 +0100, Johannes Weiner wrote:
> > On Mon, Jan 31, 2011 at 04:34:03PM -0800, Dave Hansen wrote:
> > > +	if (pmd_trans_huge(*pmd)) {
> > > +		if (pmd_trans_splitting(*pmd)) {
> > > +			spin_unlock(&walk->mm->page_table_lock);
> > > +			wait_split_huge_page(vma->anon_vma, pmd);
> > > +			spin_lock(&walk->mm->page_table_lock);
> > > +			goto normal_ptes;
> > > +		}
> > > +		smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_SIZE, walk);
> > > +		return 0;
> > > +	}
> > > +normal_ptes:
> > >  	split_huge_page_pmd(walk->mm, pmd);
> > 
> > This line can go away now...?
> 
> I did this because I was unsure what keeps khugepaged away from the
> newly-split ptes between the wait_split_huge_page() and the
> reacquisition of the mm->page_table_lock.  mmap_sem, perhaps?

Any of mmap_sem read mode, PG_lock and anon_vma_lock keeps khugepaged
away.

> Looking at follow_page() and some of the other wait_split_huge_page(),
> it looks like this is unnecessary.  

When wait_split_huge_page returns after the pmd was splitting, the pmd
can't return huge under you as long as you hold any of the above.