Re: [PATCH]mmap: not merge cloned VMA

From: Shaohua Li
Date: Mon Mar 28 2011 - 20:53:37 EST


On Tue, 2011-03-29 at 00:57 +0800, Andi Kleen wrote:
> Shaohua Li <shaohua.li@xxxxxxxxx> writes:
>
> > Avoid merging a VMA with another VMA which is cloned from parent process. The
> > cloned VMA shares lock with parent process's VMA. If we do the merge, more vma
> > area (even the new range is only for current process) uses perent process's
> > anon_vma lock, so introduces scalability issues.
> > find_mergeable_anon_vma already considers this.
> In theory this could prevent quite some VMA merging, but I guess the
> tradeoff is worth it and that should be unlikely to hit anyways.
I'd expect there aren't too many VMAs cloned from parents.

> > static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
> > - struct anon_vma *anon_vma2)
> > + struct anon_vma *anon_vma2,
> > + struct vm_area_struct *vma)
> > {
> > - return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
> > + if ((!anon_vma1 || !anon_vma2) && (!vma ||
> > + list_is_singular(&vma->anon_vma_chain)))
> > + return 1;
>
> I think this if () needs a comment.
ok, updated the patch.

Avoid merging a VMA with another VMA which is cloned from parent process. The
cloned VMA shares lock with parent process's VMA. If we do the merge, more vma
area (even the new range is only for current process) uses perent process's
anon_vma lock, so introduces scalability issues.
find_mergeable_anon_vma already considers this.

Signed-off-by: Shaohua Li <shaohua.li@xxxxxxxxx>

---
mm/mmap.c | 18 +++++++++++++-----
1 file changed, 13 insertions(+), 5 deletions(-)

Index: linux/mm/mmap.c
===================================================================
--- linux.orig/mm/mmap.c 2011-03-29 08:22:01.000000000 +0800
+++ linux/mm/mmap.c 2011-03-29 08:30:12.000000000 +0800
@@ -699,9 +699,17 @@ static inline int is_mergeable_vma(struc
}

static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
- struct anon_vma *anon_vma2)
+ struct anon_vma *anon_vma2,
+ struct vm_area_struct *vma)
{
- return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
+ /*
+ * The list_is_singular() test is to avoid merging VMA cloned from
+ * parents. This can improve scalability caused by anon_vma lock.
+ */
+ if ((!anon_vma1 || !anon_vma2) && (!vma ||
+ list_is_singular(&vma->anon_vma_chain)))
+ return 1;
+ return anon_vma1 == anon_vma2;
}

/*
@@ -720,7 +728,7 @@ can_vma_merge_before(struct vm_area_stru
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)) {
+ is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
if (vma->vm_pgoff == vm_pgoff)
return 1;
}
@@ -739,7 +747,7 @@ can_vma_merge_after(struct vm_area_struc
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)) {
+ is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
pgoff_t vm_pglen;
vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
if (vma->vm_pgoff + vm_pglen == vm_pgoff)
@@ -817,7 +825,7 @@ struct vm_area_struct *vma_merge(struct
can_vma_merge_before(next, vm_flags,
anon_vma, file, pgoff+pglen) &&
is_mergeable_anon_vma(prev->anon_vma,
- next->anon_vma)) {
+ next->anon_vma, NULL)) {
/* cases 1, 6 */
err = vma_adjust(prev, prev->vm_start,
next->vm_end, prev->vm_pgoff, NULL);


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/