[PATCH] arm: mm: Fix definition of pmd_mknotpresent

[Kirill A. Shutemov]

On Wed, Mar 02, 2016 at 03:55:55PM +0000, Will Deacon wrote:
> On Wed, Mar 02, 2016 at 06:11:52PM +0300, Kirill A. Shutemov wrote:
> > On Wed, Mar 02, 2016 at 01:03:00AM +0000, Steve Capper wrote:
> > > Currently pmd_mknotpresent will use a zero entry to respresent an
> > > invalidated pmd.
> > > 
> > > Unfortunately this definition clashes with pmd_none, thus it is
> > > possible for a race condition to occur if zap_pmd_range sees pmd_none
> > > whilst __split_huge_pmd_locked is running too with pmdp_invalidate
> > > just called.
> > > 
> > > This patch fixes the race condition by modifying pmd_mknotpresent to
> > > create non-zero faulting entries (as is done in other architectures),
> > > removing the ambiguity with pmd_none.
> > > 
> > > Fixes: 8d9625070073 ("ARM: mm: Transparent huge page support for LPAE systems.")
> > > Reported-by: Kirill A. Shutemov <kirill.shutemov at linux.intel.com>
> > > Signed-off-by: Steve Capper <steve.capper at arm.com>
> > > Acked-by: Will Deacon <will.deacon at arm.com>
> > > 
> > > ---
> > > 
> > > This patch addresses an issue discussed at:
> > > http://lkml.iu.edu/hypermail/linux/kernel/1602.3/01846.html
> > > 
> > > ---
> > >  arch/arm/include/asm/pgtable-3level.h | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/arch/arm/include/asm/pgtable-3level.h b/arch/arm/include/asm/pgtable-3level.h
> > > index dc46398..451b6e7 100644
> > > --- a/arch/arm/include/asm/pgtable-3level.h
> > > +++ b/arch/arm/include/asm/pgtable-3level.h
> > > @@ -249,10 +249,10 @@ PMD_BIT_FUNC(mkyoung,   |= PMD_SECT_AF);
> > >  #define pfn_pmd(pfn,prot)	(__pmd(((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot)))
> > >  #define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
> > >  
> > > -/* represent a notpresent pmd by zero, this is used by pmdp_invalidate */
> > > +/* represent a notpresent pmd by faulting entry, this is used by pmdp_invalidate */
> > >  static inline pmd_t pmd_mknotpresent(pmd_t pmd)
> > >  {
> > > -	return __pmd(0);
> > > +	return __pmd(pmd_val(pmd) & ~PMD_TYPE_MASK);
> > 
> > I'm confused. Wouldn't it still be present?
> > 
> > arch/arm/include/asm/pgtable.h:#define pmd_present(pmd) (pmd_val(pmd))
> > 
> > Shouldn't pmd_present() be re-defined as (pmd_val(pmd) & PMD_TYPE_MASK) or
> > something?
> 
> Hmm, we have ths same issue on arm64 in that case. Do we also need to
> handle PROT_NONE for pmds?

I think so.

There's no use-cases where it would matter at the moment as far as I can
see. But if we would implement huge migration/swap entries, proper
pmd_present() will be required.

-- 
 Kirill A. Shutemov