shake_page() is called before going into core error handling code in order to ensure that the error page is flushed from lru_cache lists where pages stay during transferring among LRU lists. But currently it's not fully functional because when the page is linked to lru_cache by calling activate_page(), its PageLRU flag is set and shake_page() is skipped. The result is to fail error handling with "still referenced by 1 users" message. When the page is linked to lru_cache by isolate_lru_page(), its PageLRU is clear, so that's fine.
This patch makes shake_page() unconditionally called to avoild the failure. Link: http://lkml.kernel.org/r/20170417055948.GM31394@yexl-desktop Reported-by: kernel test robot <[email protected]> Signed-off-by: Naoya Horiguchi <[email protected]> --- mm/hwpoison-inject.c | 3 +-- mm/memory-failure.c | 27 +++++++++++---------------- 2 files changed, 12 insertions(+), 18 deletions(-) diff --git v4.11-rc6-mmotm-2017-04-13-14-50/mm/hwpoison-inject.c v4.11-rc6-mmotm-2017-04-13-14-50_patched/mm/hwpoison-inject.c index 9d26fd9..356df05 100644 --- v4.11-rc6-mmotm-2017-04-13-14-50/mm/hwpoison-inject.c +++ v4.11-rc6-mmotm-2017-04-13-14-50_patched/mm/hwpoison-inject.c @@ -34,8 +34,7 @@ static int hwpoison_inject(void *data, u64 val) if (!hwpoison_filter_enable) goto inject; - if (!PageLRU(hpage) && !PageHuge(p)) - shake_page(hpage, 0); + shake_page(hpage, 0); /* * This implies unable to support non-LRU pages. */ diff --git v4.11-rc6-mmotm-2017-04-13-14-50/mm/memory-failure.c v4.11-rc6-mmotm-2017-04-13-14-50_patched/mm/memory-failure.c index 8c02811..77cf9c3 100644 --- v4.11-rc6-mmotm-2017-04-13-14-50/mm/memory-failure.c +++ v4.11-rc6-mmotm-2017-04-13-14-50_patched/mm/memory-failure.c @@ -220,6 +220,9 @@ static int kill_proc(struct task_struct *t, unsigned long addr, int trapno, */ void shake_page(struct page *p, int access) { + if (PageHuge(p)) + return; + if (!PageSlab(p)) { lru_add_drain_all(); if (PageLRU(p)) @@ -1140,22 +1143,14 @@ int memory_failure(unsigned long pfn, int trapno, int flags) * The check (unnecessarily) ignores LRU pages being isolated and * walked by the page reclaim code, however that's not a big loss. */ - if (!PageHuge(p)) { - if (!PageLRU(p)) - shake_page(p, 0); - if (!PageLRU(p)) { - /* - * shake_page could have turned it free. - */ - if (is_free_buddy_page(p)) { - if (flags & MF_COUNT_INCREASED) - action_result(pfn, MF_MSG_BUDDY, MF_DELAYED); - else - action_result(pfn, MF_MSG_BUDDY_2ND, - MF_DELAYED); - return 0; - } - } + shake_page(p, 0); + /* shake_page could have turned it free. */ + if (!PageLRU(p) && is_free_buddy_page(p)) { + if (flags & MF_COUNT_INCREASED) + action_result(pfn, MF_MSG_BUDDY, MF_DELAYED); + else + action_result(pfn, MF_MSG_BUDDY_2ND, MF_DELAYED); + return 0; } lock_page(hpage); -- 2.7.0
