Date:   Thu, 13 Dec 2018 11:22:52 +0800
From:   Peter Xu <peterx@...hat.com>
To:     Konstantin Khlebnikov <koct9i@...il.com>
Cc:     Константин Хлебников 
        <khlebnikov@...dex-team.ru>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Andrea Arcangeli <aarcange@...hat.com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        "Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>,
        Matthew Wilcox <willy@...radead.org>,
        Michal Hocko <mhocko@...e.com>, dave.jiang@...el.com,
        "Aneesh Kumar K.V" <aneesh.kumar@...ux.vnet.ibm.com>,
        Souptick Joarder <jrdr.linux@...il.com>, linux-mm@...ck.org,
        Zi Yan <zi.yan@...rutgers.edu>
Subject: Re: [PATCH v2] mm: thp: fix flags for pmd migration when split

On Wed, Dec 12, 2018 at 04:51:38PM +0300, Konstantin Khlebnikov wrote:
> On Wed, Dec 12, 2018 at 8:15 AM Peter Xu <peterx@...hat.com> wrote:
> >
> > On Tue, Dec 11, 2018 at 11:21:44AM +0300, Konstantin Khlebnikov wrote:
> > > On 11.12.2018 8:12, Peter Xu wrote:
> > > > When splitting a huge migrating PMD, we'll transfer all the existing
> > > > PMD bits and apply them again onto the small PTEs.  However we are
> > > > fetching the bits unconditionally via pmd_soft_dirty(), pmd_write()
> > > > or pmd_yound() while actually they don't make sense at all when it's
> > > > a migration entry.  Fix them up by make it conditional.
> > > >
> > > > Note that if my understanding is correct about the problem then if
> > > > without the patch there is chance to lose some of the dirty bits in
> > > > the migrating pmd pages (on x86_64 we're fetching bit 11 which is part
> > > > of swap offset instead of bit 2) and it could potentially corrupt the
> > > > memory of an userspace program which depends on the dirty bit.
> > > >
> > > > CC: Andrea Arcangeli <aarcange@...hat.com>
> > > > CC: Andrew Morton <akpm@...ux-foundation.org>
> > > > CC: "Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>
> > > > CC: Matthew Wilcox <willy@...radead.org>
> > > > CC: Michal Hocko <mhocko@...e.com>
> > > > CC: Dave Jiang <dave.jiang@...el.com>
> > > > CC: "Aneesh Kumar K.V" <aneesh.kumar@...ux.vnet.ibm.com>
> > > > CC: Souptick Joarder <jrdr.linux@...il.com>
> > > > CC: Konstantin Khlebnikov <khlebnikov@...dex-team.ru>
> > > > CC: linux-mm@...ck.org
> > > > CC: linux-kernel@...r.kernel.org
> > > > Signed-off-by: Peter Xu <peterx@...hat.com>
> > > > ---
> > > > v2:
> > > > - fix it up for young/write/dirty bits too [Konstantin]
> > > > ---
> > > >   mm/huge_memory.c | 15 ++++++++++-----
> > > >   1 file changed, 10 insertions(+), 5 deletions(-)
> > > >
> > > > diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> > > > index f2d19e4fe854..b00941b3d342 100644
> > > > --- a/mm/huge_memory.c
> > > > +++ b/mm/huge_memory.c
> > > > @@ -2157,11 +2157,16 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,
> > > >             page = pmd_page(old_pmd);
> > > >     VM_BUG_ON_PAGE(!page_count(page), page);
> > > >     page_ref_add(page, HPAGE_PMD_NR - 1);
> > > > -   if (pmd_dirty(old_pmd))
> > > > -           SetPageDirty(page);
> > > > -   write = pmd_write(old_pmd);
> > > > -   young = pmd_young(old_pmd);
> > > > -   soft_dirty = pmd_soft_dirty(old_pmd);
> > > > +   if (unlikely(pmd_migration)) {
> > > > +           soft_dirty = pmd_swp_soft_dirty(old_pmd);
> > > > +           young = write = false;
> > > > +   } else {
> > > > +           if (pmd_dirty(old_pmd))
> > > > +                   SetPageDirty(page);
> > > > +           write = pmd_write(old_pmd);
> > > > +           young = pmd_young(old_pmd);
> > > > +           soft_dirty = pmd_soft_dirty(old_pmd);
> > > > +   }
> > >
> > > Write/read-only is encoded into migration entry.
> > > I suppose there should be something like this:
> > >
> > > --- a/mm/huge_memory.c
> > > +++ b/mm/huge_memory.c
> > > @@ -2151,16 +2151,21 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,
> > >
> > >                 entry = pmd_to_swp_entry(old_pmd);
> > >                 page = pfn_to_page(swp_offset(entry));
> > > +               write = is_write_migration_entry(entry);
> > > +               young = false;
> > > +               soft_dirty = pmd_swp_soft_dirty(old_pmd);
> > >         } else
> > >  #endif
> > > +       {
> > >                 page = pmd_page(old_pmd);
> > > +               if (pmd_dirty(old_pmd))
> > > +                       SetPageDirty(page);
> > > +               write = pmd_write(old_pmd);
> > > +               young = pmd_young(old_pmd);
> > > +               soft_dirty = pmd_soft_dirty(old_pmd);
> > > +       }
> > >         VM_BUG_ON_PAGE(!page_count(page), page);
> > >         page_ref_add(page, HPAGE_PMD_NR - 1);
> > > -       if (pmd_dirty(old_pmd))
> > > -               SetPageDirty(page);
> > > -       write = pmd_write(old_pmd);
> > > -       young = pmd_young(old_pmd);
> > > -       soft_dirty = pmd_soft_dirty(old_pmd);
> > >
> > >         /*
> > >          * Withdraw the table only after we mark the pmd entry invalid.
> > >
> >
> > Oops yes, I missed the write bit.  Thanks for pointing it out.
> >
> > Should I repost with your authorship and your sign-off?
> 
> Feel free to use this piece for your own patch.
> 
> > Or even I'll
> > consider to directly drop the CONFIG_ARCH_ENABLE_THP_MIGRATION if with
> > that since I don't see much gain to keep it:
> 
> Yep, this ifdef could be removed.
> Without CONFIG_ARCH_ENABLE_THP_MIGRATION
> is_pmd_migration_entry() is constant 0 so compiler should eliminate "if" branch.

Thank you, Konstantin.  I'll post v3 with the macro dropped.

Regards,

-- 
Peter Xu

Powered by blists - more mailing lists