Re: 4.0-rc1/PARISC: BUG: non-zero nr_pmds on freeing mm

["Kirill A. Shutemov" <kirill@shutemov.name>]

On Wed, Feb 25, 2015 at 01:31:40PM -0800, Andrew Morton wrote:
> On Wed, 25 Feb 2015 22:47:43 +0200 "Kirill A. Shutemov" <kirill@shutemov.name> wrote:
> 
> > > > If not, I can prepare a patchset which only adds missing
> > > > __PAGETABLE_PUD_FOLDED and __PAGETABLE_PMD_FOLDED.
> > > 
> > > Something simple would be preferred, but I don't know how much simpler
> > > the above would be?
> > 
> > Not much simplier: __PAGETABLE_PMD_FOLDED is missing in frv, m32r, m68k,
> > mn10300, parisc and s390.
> 
> I don't really know what's going on here.  Let's rewind a bit, please. 
> What is the bug, what causes it, which commit caused it and why the
> heck does it require a massive patchset to fix 4.0?

PMD accounting happens in __pmd_alloc() and free_pmd_range(). PMD
accounting only makes sense on architectures with 3 or more page tables
levels. We use __PAGETABLE_PMD_FOLDED to check whether the PMD page table
level exists.

Unfortunately, some architectures don't use <asm-generic/pgtable-nopmd.h>
to indicate that PMD level doesn't exists and fold it in a custom way.
Some of them don't define __PAGETABLE_PMD_FOLDED as pgtable-nopmd.h does.

Missing __PAGETABLE_PMD_FOLDED causes undeflow of mm->nr_pmds:
__pmd_alloc() is never called, but we decrement mm->nr_pmds in
free_pmd_range().

These architecures need to be fixed to define __PAGETABLE_PMD_FOLDED too.

I can do in one patch if you want. Or split per-arch. After that
CONFIG_PGTABLE_LEVELS patchset will require rebasing.

-- 
 Kirill A. Shutemov

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>