On Fri, 20 Nov 2015 15:57:21 +0800 "Hillf Danton" <hillf...@alibaba-inc.com>
wrote:
> >
> > When dequeue_huge_page_vma() in alloc_huge_page() fails, we fall back to
> > alloc_buddy_huge_page() to directly create a hugepage from the buddy
> > allocator.
> > In that case, however, if alloc_buddy_huge_page() succeeds we don't
> > decrement
> > h->resv_huge_pages, which means that successful hugetlb_fault() returns
> > without
> > releasing the reserve count. As a result, subsequent hugetlb_fault() might
> > fail
> > despite that there are still free hugepages.
> >
> > This patch simply adds decrementing code on that code path.
> >
> > I reproduced this problem when testing v4.3 kernel in the following
> > situation:
> > - the test machine/VM is a NUMA system,
> > - hugepage overcommiting is enabled,
> > - most of hugepages are allocated and there's only one free hugepage
> > which is on node 0 (for example),
> > - another program, which calls set_mempolicy(MPOL_BIND) to bind itself to
> > node 1, tries to allocate a hugepage,
> > - the allocation should fail but the reserve count is still hold.
> >
> > Signed-off-by: Naoya Horiguchi <n-horigu...@ah.jp.nec.com>
> > Cc: <sta...@vger.kernel.org> [3.16+]
> > ---
> > - the reason why I set stable target to "3.16+" is that this patch can be
> > applied easily/automatically on these versions. But this bug seems to be
> > old one, so if you are interested in backporting to older kernels,
> > please let me know.
> > ---
> > mm/hugetlb.c | 5 ++++-
> > 1 files changed, 4 insertions(+), 1 deletions(-)
> >
> > diff --git v4.3/mm/hugetlb.c v4.3_patched/mm/hugetlb.c
> > index 9cc7734..77c518c 100644
> > --- v4.3/mm/hugetlb.c
> > +++ v4.3_patched/mm/hugetlb.c
> > @@ -1790,7 +1790,10 @@ struct page *alloc_huge_page(struct vm_area_struct
> > *vma,
> > page = alloc_buddy_huge_page(h, NUMA_NO_NODE);
> > if (!page)
> > goto out_uncharge_cgroup;
> > -
> > + if (!avoid_reserve && vma_has_reserves(vma, gbl_chg)) {
> > + SetPagePrivate(page);
> > + h->resv_huge_pages--;
> > + }
>
> I am wondering if this patch was prepared against the next tree.
It's against 4.3.
Here's the version I have, against current -linus:
---
a/mm/hugetlb.c~mm-hugetlb-fix-hugepage-memory-leak-caused-by-wrong-reserve-count
+++ a/mm/hugetlb.c
@@ -1886,7 +1886,10 @@ struct page *alloc_huge_page(struct vm_a
page = __alloc_buddy_huge_page_with_mpol(h, vma, addr);
if (!page)
goto out_uncharge_cgroup;
-
+ if (!avoid_reserve && vma_has_reserves(vma, gbl_chg)) {
+ SetPagePrivate(page);
+ h->resv_huge_pages--;
+ }
spin_lock(&hugetlb_lock);
list_move(&page->lru, &h->hugepage_activelist);
/* Fall through */
It needs a careful re-review and, preferably, retest please.
Probably when Greg comes to merge this he'll hit problems and we'll
need to provide him with the against-4.3 patch.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
