On 08/27/2011 05:23 AM, Andrew Morton wrote:

On Mon,  1 Aug 2011 23:28:55 +0800
Caspar Zhang[off-list ref]  wrote:

quoted

commit 9d8cebd4bcd7c3878462fdfda34bbcdeb4df7ef4 didn't real fix the
mbind vma merge problem due to wrong pgoff value passing to vma_merge(),
which made vma_merge() always return NULL.

Re-tested the patched kernel with the reproducer provided in commit
9d8cebd, got correct result like below:

addr = 0x7ffa5aaa2000
[snip]
7ffa5aaa2000-7ffa5aaa6000 rw-p 00000000 00:00 0
7fffd556f000-7fffd5584000 rw-p 00000000 00:00 0                          [stack]

Please also describe the output before the patch is applied, and tell
us what you believe is wrong with it?

Sure. Before the patch applied, we are getting a result like:

addr = 0x7fa58f00c000
[snip]
7fa58f00c000-7fa58f00d000 rw-p 00000000 00:00 0
7fa58f00d000-7fa58f00e000 rw-p 00000000 00:00 0
7fa58f00e000-7fa58f00f000 rw-p 00000000 00:00 0

here 7fa58f00c000->7fa58f00f000 we get 3 VMAs which are expected to be 
merged described in commit 9d8cebd.

commit 9d8cebd says:

-------------------------------------------------------------------
commit 9d8cebd4bcd7c3878462fdfda34bbcdeb4df7ef4
Author: KOSAKI Motohiro [off-list ref]
Date:   Fri Mar 5 13:41:57 2010 -0800

     mm: fix mbind vma merge problem

     Strangely, current mbind() doesn't merge vma with neighbor vma 
although it's possible.
     Unfortunately, many vma can reduce performance...

     This patch fixes it.
-------------------------------------------------------------------

mbind_range() invokes vma_range() to merge current VMA with its successor:

	prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
			  vma->anon_vma, vma->vm_file, pgoff, new_pol);

in vma_merge(), it should go into:

	if (next && end == next->vm_start &&
  			mpol_equal(policy, vma_policy(next)) &&
			can_vma_merge_before(next, vm_flags,
					anon_vma, file, pgoff+pglen)) {
	...
	}

while in can_vma_merge_before(), we have:

static int
can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
	struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
{
	if (is_mergeable_vma(vma, file, vm_flags) &&
	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
		if (vma->vm_pgoff == vm_pgoff)
			return 1;
	}
	return 0;
}

to make can_vma_merge_before() return 1, we should have (next->vm_pgoff 
== pgoff + pglen) in vma_merge(), i.e., pgoff should be the page offset 
of *current* VMA.

However from the current codes in mbind_range(), we get:

	pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);

start is the beginning addr of the range, so here pgoff will always be 
the page offset of the beginning VMA, that makes merging VMAs impossible 
in mbind_range().

To solve this bug, we simply need to pass vma->pgoff instead of the 
calculated pgoff value to vma_merge().

Thanks,
Caspar

quoted

--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c

@@ -636,7 +636,6 @@ static int mbind_range(struct mm_struct *mm, unsigned long start,
  	struct vm_area_struct *prev;
  	struct vm_area_struct *vma;
  	int err = 0;
-	pgoff_t pgoff;
  	unsigned long vmstart;
  	unsigned long vmend;

@@ -649,9 +648,9 @@ static int mbind_range(struct mm_struct *mm, unsigned long start,
  		vmstart = max(start, vma->vm_start);
  		vmend   = min(end, vma->vm_end);

-		pgoff = vma->vm_pgoff + ((start - vma->vm_start)>>  PAGE_SHIFT);
  		prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
-				  vma->anon_vma, vma->vm_file, pgoff, new_pol);
+				  vma->anon_vma, vma->vm_file, vma->vm_pgoff,
+				  new_pol);
  		if (prev) {
  			vma = prev;
  			next = vma->vm_next;
--

1.7.6

--
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/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>