Re: mm: kernel BUG at mm/huge_memory.c:3272!

["Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>]

On Mon, Nov 30, 2015 at 09:37:33AM -0500, Sasha Levin wrote:
> Hi Kirill,
> 
> I've hit the following while fuzzing with trinity on the latest -next kernel:
> 
> [  321.348184] page:ffffea0011a20080 count:1 mapcount:1 mapping:ffff8802d745f601 index:0x1802
> [  321.350607] flags: 0x320035c00040078(uptodate|dirty|lru|active|swapbacked)
> [  321.453706] page dumped because: VM_BUG_ON_PAGE(!PageLocked(page))
> [  321.455353] page->mem_cgroup:ffff880286620000

I think this should help:

>From aadc911f047b094c68b350550556dafabf05af13 Mon Sep 17 00:00:00 2001
From: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Date: Fri, 20 Nov 2015 12:20:00 +0200
Subject: [PATCH] thp: fix split_huge_page vs. deferred_split_scan race

Minchan[1] and Sasha[2] had reported crash in split_huge_page_to_list()
called from deferred_split_scan() due VM_BUG_ON_PAGE(!PageLocked(page)).

This can happen because race between deferred_split_scan() and
split_huge_page(). The result of the race is that the page can be split
under deferred_split_scan().

The patch prevents this by taking split_queue_lock in
split_huge_page_to_list() when we check if the page can be split.
If the page is suitable for splitting, we remove page from splitting
queue under the same lock, before splitting starts.

[1] http://lkml.kernel.org/g/20151117073539.GB32578@bbox
[2] http://lkml.kernel.org/g/565C5F2D.5060003@oracle.com

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reported-by: Minchan Kim <minchan@kernel.org>
Reported-by: Sasha Levin <sasha.levin@oracle.com>
---
 mm/huge_memory.c | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index dc2b947d4f85..7c0ad4d9110b 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -3186,13 +3186,6 @@ static void __split_huge_page(struct page *page, struct list_head *list)
 	spin_lock_irq(&zone->lru_lock);
 	lruvec = mem_cgroup_page_lruvec(head, zone);
 
-	spin_lock(&split_queue_lock);
-	if (!list_empty(page_deferred_list(head))) {
-		split_queue_len--;
-		list_del(page_deferred_list(head));
-	}
-	spin_unlock(&split_queue_lock);
-
 	/* complete memcg works before add pages to LRU */
 	mem_cgroup_split_huge_fixup(head);
 
@@ -3299,12 +3292,20 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
 	freeze_page(anon_vma, head);
 	VM_BUG_ON_PAGE(compound_mapcount(head), head);
 
+	/* Prevent deferred_split_scan() touching ->_count */
+	spin_lock(&split_queue_lock);
 	count = page_count(head);
 	mapcount = total_mapcount(head);
 	if (mapcount == count - 1) {
+		if (!list_empty(page_deferred_list(head))) {
+			split_queue_len--;
+			list_del(page_deferred_list(head));
+		}
+		spin_unlock(&split_queue_lock);
 		__split_huge_page(page, list);
 		ret = 0;
 	} else if (IS_ENABLED(CONFIG_DEBUG_VM) && mapcount > count - 1) {
+		spin_unlock(&split_queue_lock);
 		pr_alert("total_mapcount: %u, page_count(): %u\n",
 				mapcount, count);
 		if (PageTail(page))
@@ -3312,6 +3313,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
 		dump_page(page, "total_mapcount(head) > page_count(head) - 1");
 		BUG();
 	} else {
+		spin_unlock(&split_queue_lock);
 		unfreeze_page(anon_vma, head);
 		ret = -EBUSY;
 	}
-- 
2.6.2

-- 
 Kirill A. Shutemov