lkml.org 
[lkml]   [2014]   [Jul]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH v4 13/13] mincore: apply page table walker on do_mincore()
On Wed, Jul 09, 2014 at 04:34:36PM +0300, Kirill A. Shutemov wrote:
> On Tue, Jul 01, 2014 at 01:07:31PM -0400, Naoya Horiguchi wrote:
> > This patch makes do_mincore() use walk_page_vma(), which reduces many lines
> > of code by using common page table walk code.
> >
> > ChangeLog v4:
> > - remove redundant vma
> >
> > ChangeLog v3:
> > - add NULL vma check in mincore_unmapped_range()
> > - don't use pte_entry()
> >
> > ChangeLog v2:
> > - change type of args of callbacks to void *
> > - move definition of mincore_walk to the start of the function to fix compiler
> > warning
> >
> > Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
>
> Trinity crases this implementation of mincore pretty easily:
>
> [ 42.775369] BUG: unable to handle kernel paging request at ffff88007bb61000
> [ 42.776656] IP: [<ffffffff81126f8f>] mincore_unmapped_range+0xdf/0x100

Thanks for your testing/reporting.

...
>
> Looks like 'vec' overflow. I don't see what could prevent do_mincore() to
> write more than PAGE_SIZE to 'vec'.

I found the miscalculation of walk->private (vec) on thp and hugetlbfs.
I confirmed that the reported problem is fixed (I checked that trinity
never triggers the reported BUG) with the following changes on this patch.

diff --git a/mm/mincore.c b/mm/mincore.c
index 3c64dcbcb3e2..9eb10d867a6f 100644
--- a/mm/mincore.c
+++ b/mm/mincore.c
@@ -34,7 +34,7 @@ static int mincore_hugetlb(pte_t *pte, unsigned long hmask, unsigned long addr,
present = pte && !huge_pte_none(huge_ptep_get(pte));
for (; addr != end; vec++, addr += PAGE_SIZE)
*vec = present;
- walk->private += (end - addr) >> PAGE_SHIFT;
+ walk->private = vec;
#else
BUG();
#endif
@@ -118,8 +118,10 @@ static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
return 0;
}

- if (pmd_trans_unstable(pmd))
+ if (pmd_trans_unstable(pmd)) {
+ walk->private += (end - addr) >> PAGE_SHIFT;
return 0;
+ }

ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
for (; addr != end; ptep++, addr += PAGE_SIZE) {
Thanks,
Naoya Horiguchi


\
 
 \ /
  Last update: 2014-07-10 00:22    [W:0.185 / U:4.276 seconds]
©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site