Re: [PATCH v2] smaps should deal with huge zero page exactly same as normal zero page.

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

On Tue, Oct 28, 2014 at 11:18:38PM +0800, Fengwei Yin wrote:
> On Mon, Oct 27, 2014 at 03:17:48PM -0700, Andrew Morton wrote:
> > On Mon, 27 Oct 2014 23:02:13 +0800 Fengwei Yin <yfw.kernel@gmail.com> wrote:
> > 
> > > We could see following memory info in /proc/xxxx/smaps with THP enabled.
> > >   7bea458b3000-7fea458b3000 r--p 00000000 00:13 39989  /dev/zero
> > >   Size:           4294967296 kB
> > >   Rss:            10612736 kB
> > >   Pss:            10612736 kB
> > >   Shared_Clean:          0 kB
> > >   Shared_Dirty:          0 kB
> > >   Private_Clean:  10612736 kB
> > >   Private_Dirty:         0 kB
> > >   Referenced:     10612736 kB
> > >   Anonymous:             0 kB
> > >   AnonHugePages:  10612736 kB
> > >   Swap:                  0 kB
> > >   KernelPageSize:        4 kB
> > >   MMUPageSize:           4 kB
> > >   Locked:                0 kB
> > >   VmFlags: rd mr mw me
> > > which is wrong becuase just huge_zero_page/normal_zero_page is used for
> > > /dev/zero. Most of the value should be 0.
> > > 
> > > This patch detects huge_zero_page (original implementation just detect
> > > normal_zero_page) and avoids to update the wrong value for huge_zero_page.
> > > 
> > > ...
> > >
> > > --- a/mm/memory.c
> > > +++ b/mm/memory.c
> > > @@ -41,6 +41,7 @@
> > >  #include <linux/kernel_stat.h>
> > >  #include <linux/mm.h>
> > >  #include <linux/hugetlb.h>
> > > +#include <linux/huge_mm.h>
> > >  #include <linux/mman.h>
> > >  #include <linux/swap.h>
> > >  #include <linux/highmem.h>
> > > @@ -787,6 +788,9 @@ check_pfn:
> > >  		return NULL;
> > >  	}
> > >  
> > > +	if (is_huge_zero_pfn(pfn))
> > > +		return NULL;
> > > +
> > 
> > Why this change?
> > 
> I suppose the huge zero page should have same behavior as normal zero
> page. vm_normal_page will return NULL if the pte is for normal zero
> page. This change make it return NULL for huge zero page.
> 
> > What effect does it have upon vm_normal_page()'s many existing callers?
> This is good question. I suppose it will not impact existing caller.

vm_normal_page() is designed to handle pte. We only get there due hack
with pmd to pte cast in smaps_pte_range(). Let's try to get rid of it
instead.

Could you test the patch below? I think it's a better fix.