From 40448665846ef861c2aca644e8f6ba8214ab6718 Mon Sep 17 00:00:00 2001 From: Mel Gorman Date: Wed, 19 Jun 2013 10:06:12 +1000 Subject: [PATCH] mm: Clear page active before releasing pages Active pages should not be freed to the page allocator as it triggers a bad page state warning. Fengguang Wu reported the following bug and bisected it to the patch "mm: remove lru parameter from __lru_cache_add and lru_cache_add_lru" which is currently in mmotm as mm-remove-lru-parameter-from-__lru_cache_add-and-lru_cache_add_lru.patch [ 84.212960] BUG: Bad page state in process rm pfn:0b0c9 [ 84.214682] page:ffff88000d646240 count:0 mapcount:0 mapping: (null) index:0x0 [ 84.216883] page flags: 0x20000000004c(referenced|uptodate|active) [ 84.218697] CPU: 1 PID: 283 Comm: rm Not tainted 3.10.0-rc4-04361-geeb9bfc #49 [ 84.220729] ffff88000d646240 ffff88000d179bb8 ffffffff82562956 ffff88000d179bd8 [ 84.223242] ffffffff811333f1 000020000000004c ffff88000d646240 ffff88000d179c28 [ 84.225387] ffffffff811346a4 ffff880000270000 0000000000000000 0000000000000006 [ 84.227294] Call Trace: [ 84.227867] [] dump_stack+0x27/0x30 [ 84.229045] [] bad_page+0x130/0x158 [ 84.230261] [] free_pages_prepare+0x8b/0x1e3 [ 84.231765] [] free_hot_cold_page+0x28/0x1cf [ 84.233171] [] ? _raw_spin_unlock_irqrestore+0x6b/0xc6 [ 84.234822] [] free_hot_cold_page_list+0x30/0x5a [ 84.236311] [] release_pages+0x251/0x267 [ 84.237653] [] ? delete_from_page_cache+0x48/0x9e [ 84.239142] [] __pagevec_release+0x2b/0x3d [ 84.240473] [] truncate_inode_pages_range+0x1b0/0x7ce [ 84.242032] [] ? put_lock_stats.isra.20+0x1c/0x53 [ 84.243480] [] ? lock_release_holdtime+0x113/0x11f [ 84.244935] [] truncate_inode_pages+0x14/0x1d [ 84.246337] [] evict+0x11f/0x232 [ 84.247501] [] iput+0x1a5/0x218 [ 84.248607] [] do_unlinkat+0x19b/0x25a [ 84.249828] [] ? trace_hardirqs_on_caller+0x210/0x2ce [ 84.251382] [] ? trace_hardirqs_on_thunk+0x3a/0x3f [ 84.252879] [] SyS_unlinkat+0x39/0x4c [ 84.254174] [] system_call_fastpath+0x1a/0x1f [ 84.255596] Disabling lock debugging due to kernel taint The problem was that a page marked for activation was released via pagevec. This patch clears the active bit before freeing in this case. Signed-off-by: Mel Gorman Reported-and-Tested-by: Fengguang Wu Tested-by: Michal Hocko Signed-off-by: Andrew Morton --- mm/swap.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/mm/swap.c b/mm/swap.c index ac23602b79aa..4a1d0d2c52fa 100644 --- a/mm/swap.c +++ b/mm/swap.c @@ -739,6 +739,9 @@ void release_pages(struct page **pages, int nr, int cold) del_page_from_lru_list(page, lruvec, page_off_lru(page)); } + /* Clear Active bit in case of parallel mark_page_accessed */ + ClearPageActive(page); + list_add(&page->lru, &pages_to_free); } if (zone) -- 2.39.5