Re: [PATCH 0/2] Fix migration races in rmap_walk() V3

[Andrea Arcangeli <aarcange@redhat.com>]

On Sat, May 01, 2010 at 03:51:10PM +0200, Johannes Weiner wrote:
> On Fri, Apr 30, 2010 at 08:28:53PM +0200, Andrea Arcangeli wrote:
> > Subject: adapt mprotect to anon_vma chain semantics
> > 
> > From: Andrea Arcangeli <aarcange@redhat.com>
> > 
> > wait_split_huge_page interface changed.
> > 
> > Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
> > ---
> > 
> > diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> > --- a/mm/huge_memory.c
> > +++ b/mm/huge_memory.c
> > @@ -929,7 +929,7 @@ int change_huge_pmd(struct vm_area_struc
> >  	if (likely(pmd_trans_huge(*pmd))) {
> >  		if (unlikely(pmd_trans_splitting(*pmd))) {
> >  			spin_unlock(&mm->page_table_lock);
> > -			wait_split_huge_page(vma->anon_vma, pmd);
> > +			wait_split_huge_page(mm, pmd);
> 
> That makes mprotect-vma-arg obsolete, I guess.

Well it's needed for flush_tlb_range. Also normally we could run a
single invlpg on x86 to invalidate huge pmd tlbs, but I read some
errata for some x86, and I didn't want to take risks plus this is
common code so I can't just run a common code flush_tlb_page. In
mincore_huge_pmd probably we could pass vma->vm_mm instead of vma (as
there is not flush_tlb_range), I can change it if you prefer.