Re: [RFC 3/6] mm: mark dirty bit on swapped-in page

[Minchan Kim]

On Wed, Jun 10, 2015 at 10:23:05AM +0300, Cyrill Gorcunov wrote:
> On Wed, Jun 10, 2015 at 08:52:06AM +0900, Minchan Kim wrote:
> > > > +++ b/mm/memory.c
> > > > @@ -2557,9 +2557,11 @@ static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
> > > >  
> > > >  	inc_mm_counter_fast(mm, MM_ANONPAGES);
> > > >  	dec_mm_counter_fast(mm, MM_SWAPENTS);
> > > > -	pte = mk_pte(page, vma->vm_page_prot);
> > > > +
> > > > +	/* Mark dirty bit of page table because MADV_FREE relies on it */
> > > > +	pte = pte_mkdirty(mk_pte(page, vma->vm_page_prot));
> > > >  	if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
> > > > -		pte = maybe_mkwrite(pte_mkdirty(pte), vma);
> > > > +		pte = maybe_mkwrite(pte, vma);
> > > >  		flags &= ~FAULT_FLAG_WRITE;
> > > >  		ret |= VM_FAULT_WRITE;
> > > >  		exclusive = 1;
> > > 
> > > Hi Minchan! Really sorry for delay in reply. Look, I don't understand
> > > the moment -- if page has fault on read then before the patch the
> > > PTE won't carry the dirty flag but now we do set it up unconditionally
> > > and to me it looks somehow strange at least because this as well
> > > sets soft-dirty bit on pages which were not modified but only swapped
> > > out. Am I missing something obvious?
> > 
> > It's same one I sent a while ago and you said it's okay at that time. ;-)
> 
> Ah, I recall. If there is no way to escape dirtifying the page in pte itself
> maybe we should at least not make it softdirty on read faults?

You mean this? 

diff --git a/mm/memory.c b/mm/memory.c
index e1c45d0..c95340d 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2557,9 +2557,14 @@ static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
 
        inc_mm_counter_fast(mm, MM_ANONPAGES);
        dec_mm_counter_fast(mm, MM_SWAPENTS);
-       pte = mk_pte(page, vma->vm_page_prot);
+
+       /* Mark dirty bit of page table because MADV_FREE relies on it */
+       pte = pte_mkdirty(mk_pte(page, vma->vm_page_prot));
+       if (!flgas & FAULT_FLAG_WRITE)
+               pte = pte_clear_flags(pte, _PAGE_SOFT_DIRTY)
+
        if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
-               pte = maybe_mkwrite(pte_mkdirty(pte), vma);
+               pte = maybe_mkwrite(pte, vma);
                flags &= ~FAULT_FLAG_WRITE;
                ret |= VM_FAULT_WRITE;
                exclusive = 1;

It could be doable if everyone doesn't have strong objection
on this patchset.

I will wait more review.
Thanks.



> 
> > Okay, It might be lack of description compared to one I sent long time ago
> > because I moved some part of description to another patch and I didn't Cc
> > you. Sorry. I hope below will remind you.
> > 
> > https://www.mail-archive.com/linux-kernel%40vger.kernel.org/msg857827.html
> > 
> > In summary, the problem is that in MADV_FREE point of view,
> > clean anonymous page(ie, no dirty) in  page table entry has a problem
> > about sudden discarding under us by reclaimer. Otherwise, VM cannot
> > discard MADV_FREE hinted pages by PageDirty flag of page descriptor.
> > 
> > This patchset aims for solving the problem.
> > Please feel free to ask if you have questions without wasting your time
> > unless you can remind after reading above URL
> > 
> > Thanks for looking!

-- 
Kind regards,
Minchan Kim
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/