[PATCH] mm: numa: disable change protection for vma(VM_HUGETLB)

From: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
To: Mel Gorman <mgorman@suse.de>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>,
	"linux-mm@kvack.org" <linux-mm@kvack.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Hugh Dickins <hughd@google.com>,
	"Kirill A. Shutemov" <kirill@shutemov.name>,
	"David Rientjes" <rientjes@google.com>,
	Rik van Riel <riel@redhat.com>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: [PATCH] mm: numa: disable change protection for vma(VM_HUGETLB)
Date: Tue, 31 Mar 2015 01:45:55 +0000	[thread overview]
Message-ID: <20150331014554.GA8128@hori1.linux.bs1.fc.nec.co.jp> (raw)
In-Reply-To: <20150330115901.GR4701@suse.de>

On Mon, Mar 30, 2015 at 12:59:01PM +0100, Mel Gorman wrote:
> On Mon, Mar 30, 2015 at 07:42:13PM +0900, Naoya Horiguchi wrote:
...
> 
> I note now that the patch was too hasty. By rights, that check
> should be covered by vma_migratable() but it's only checked if
> CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION which means it's x86-only. If you
> are seeing this problem on any other arch then a more correct fix might be
> to remove the CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION check in vma_migratable.

Changing vma_migratable() affects other usecases of hugepage migration like
mbind(), so simply removing the ifdef doesn't work for such usecases.
I didn't test other archs, but I guess that this problem could happen on all
archs enabling numa balancing, whether it supports CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION.

So I'd like pick/push your first suggestion. It passed my testing.

Thanks,
Naoya Horiguchi
---
From: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Subject: [PATCH] mm: numa: disable change protection for vma(VM_HUGETLB)

Currently when a process accesses to hugetlb range protected with PROTNONE,
unexpected COWs are triggered, which finally put hugetlb subsystem into
broken/uncontrollable state, where for example h->resv_huge_pages is subtracted
too much and wrapped around to a very large number, and free hugepage pool
is no longer maintainable.

This patch simply stops changing protection for vma(VM_HUGETLB) to fix the
problem. And this also allows us to avoid useless overhead of minor faults.

Suggested-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
---
 kernel/sched/fair.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 7ce18f3c097a..6ad0d570f38e 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -2161,8 +2161,10 @@ void task_numa_work(struct callback_head *work)
 		vma = mm->mmap;
 	}
 	for (; vma; vma = vma->vm_next) {
-		if (!vma_migratable(vma) || !vma_policy_mof(vma))
+		if (!vma_migratable(vma) || !vma_policy_mof(vma) ||
+			is_vm_hugetlb_page(vma)) {
 			continue;
+		}
 
 		/*
 		 * Shared library pages mapped by multiple processes are not
-- 
1.9.3
