2 * Copyright (C) 2008, 2009 Intel Corporation
3 * Authors: Andi Kleen, Fengguang Wu
5 * This software may be redistributed and/or modified under the terms of
6 * the GNU General Public License ("GPL") version 2 only as published by the
7 * Free Software Foundation.
9 * High level machine check handler. Handles pages reported by the
10 * hardware as being corrupted usually due to a multi-bit ECC memory or cache
13 * In addition there is a "soft offline" entry point that allows stop using
14 * not-yet-corrupted-by-suspicious pages without killing anything.
16 * Handles page cache pages in various states. The tricky part
17 * here is that we can access any page asynchronously in respect to
18 * other VM users, because memory failures could happen anytime and
19 * anywhere. This could violate some of their assumptions. This is why
20 * this code has to be extremely careful. Generally it tries to use
21 * normal locking rules, as in get the standard locks, even if that means
22 * the error handling takes potentially a long time.
24 * There are several operations here with exponential complexity because
25 * of unsuitable VM data structures. For example the operation to map back
26 * from RMAP chains to processes has to walk the complete process list and
27 * has non linear complexity with the number. But since memory corruptions
28 * are rare we hope to get away with this. This avoids impacting the core
34 * - hugetlb needs more code
35 * - kcore/oldmem/vmcore/mem/kmem check for hwpoison pages
36 * - pass bad pages to kdump next kernel
38 #include <linux/kernel.h>
40 #include <linux/page-flags.h>
41 #include <linux/kernel-page-flags.h>
42 #include <linux/sched.h>
43 #include <linux/ksm.h>
44 #include <linux/rmap.h>
45 #include <linux/export.h>
46 #include <linux/pagemap.h>
47 #include <linux/swap.h>
48 #include <linux/backing-dev.h>
49 #include <linux/migrate.h>
50 #include <linux/page-isolation.h>
51 #include <linux/suspend.h>
52 #include <linux/slab.h>
53 #include <linux/swapops.h>
54 #include <linux/hugetlb.h>
55 #include <linux/memory_hotplug.h>
56 #include <linux/mm_inline.h>
57 #include <linux/kfifo.h>
60 int sysctl_memory_failure_early_kill __read_mostly
= 0;
62 int sysctl_memory_failure_recovery __read_mostly
= 1;
64 atomic_long_t num_poisoned_pages __read_mostly
= ATOMIC_LONG_INIT(0);
66 #if defined(CONFIG_HWPOISON_INJECT) || defined(CONFIG_HWPOISON_INJECT_MODULE)
68 u32 hwpoison_filter_enable
= 0;
69 u32 hwpoison_filter_dev_major
= ~0U;
70 u32 hwpoison_filter_dev_minor
= ~0U;
71 u64 hwpoison_filter_flags_mask
;
72 u64 hwpoison_filter_flags_value
;
73 EXPORT_SYMBOL_GPL(hwpoison_filter_enable
);
74 EXPORT_SYMBOL_GPL(hwpoison_filter_dev_major
);
75 EXPORT_SYMBOL_GPL(hwpoison_filter_dev_minor
);
76 EXPORT_SYMBOL_GPL(hwpoison_filter_flags_mask
);
77 EXPORT_SYMBOL_GPL(hwpoison_filter_flags_value
);
79 static int hwpoison_filter_dev(struct page
*p
)
81 struct address_space
*mapping
;
84 if (hwpoison_filter_dev_major
== ~0U &&
85 hwpoison_filter_dev_minor
== ~0U)
89 * page_mapping() does not accept slab pages.
94 mapping
= page_mapping(p
);
95 if (mapping
== NULL
|| mapping
->host
== NULL
)
98 dev
= mapping
->host
->i_sb
->s_dev
;
99 if (hwpoison_filter_dev_major
!= ~0U &&
100 hwpoison_filter_dev_major
!= MAJOR(dev
))
102 if (hwpoison_filter_dev_minor
!= ~0U &&
103 hwpoison_filter_dev_minor
!= MINOR(dev
))
109 static int hwpoison_filter_flags(struct page
*p
)
111 if (!hwpoison_filter_flags_mask
)
114 if ((stable_page_flags(p
) & hwpoison_filter_flags_mask
) ==
115 hwpoison_filter_flags_value
)
122 * This allows stress tests to limit test scope to a collection of tasks
123 * by putting them under some memcg. This prevents killing unrelated/important
124 * processes such as /sbin/init. Note that the target task may share clean
125 * pages with init (eg. libc text), which is harmless. If the target task
126 * share _dirty_ pages with another task B, the test scheme must make sure B
127 * is also included in the memcg. At last, due to race conditions this filter
128 * can only guarantee that the page either belongs to the memcg tasks, or is
131 #ifdef CONFIG_MEMCG_SWAP
132 u64 hwpoison_filter_memcg
;
133 EXPORT_SYMBOL_GPL(hwpoison_filter_memcg
);
134 static int hwpoison_filter_task(struct page
*p
)
136 struct mem_cgroup
*mem
;
137 struct cgroup_subsys_state
*css
;
140 if (!hwpoison_filter_memcg
)
143 mem
= try_get_mem_cgroup_from_page(p
);
147 css
= mem_cgroup_css(mem
);
148 ino
= cgroup_ino(css
->cgroup
);
151 if (!ino
|| ino
!= hwpoison_filter_memcg
)
157 static int hwpoison_filter_task(struct page
*p
) { return 0; }
160 int hwpoison_filter(struct page
*p
)
162 if (!hwpoison_filter_enable
)
165 if (hwpoison_filter_dev(p
))
168 if (hwpoison_filter_flags(p
))
171 if (hwpoison_filter_task(p
))
177 int hwpoison_filter(struct page
*p
)
183 EXPORT_SYMBOL_GPL(hwpoison_filter
);
186 * Send all the processes who have the page mapped a signal.
187 * ``action optional'' if they are not immediately affected by the error
188 * ``action required'' if error happened in current execution context
190 static int kill_proc(struct task_struct
*t
, unsigned long addr
, int trapno
,
191 unsigned long pfn
, struct page
*page
, int flags
)
197 "MCE %#lx: Killing %s:%d due to hardware memory corruption\n",
198 pfn
, t
->comm
, t
->pid
);
199 si
.si_signo
= SIGBUS
;
201 si
.si_addr
= (void *)addr
;
202 #ifdef __ARCH_SI_TRAPNO
203 si
.si_trapno
= trapno
;
205 si
.si_addr_lsb
= compound_order(compound_head(page
)) + PAGE_SHIFT
;
207 if ((flags
& MF_ACTION_REQUIRED
) && t
->mm
== current
->mm
) {
208 si
.si_code
= BUS_MCEERR_AR
;
209 ret
= force_sig_info(SIGBUS
, &si
, current
);
212 * Don't use force here, it's convenient if the signal
213 * can be temporarily blocked.
214 * This could cause a loop when the user sets SIGBUS
215 * to SIG_IGN, but hopefully no one will do that?
217 si
.si_code
= BUS_MCEERR_AO
;
218 ret
= send_sig_info(SIGBUS
, &si
, t
); /* synchronous? */
221 printk(KERN_INFO
"MCE: Error sending signal to %s:%d: %d\n",
222 t
->comm
, t
->pid
, ret
);
227 * When a unknown page type is encountered drain as many buffers as possible
228 * in the hope to turn the page into a LRU or free page, which we can handle.
230 void shake_page(struct page
*p
, int access
)
237 if (PageLRU(p
) || is_free_buddy_page(p
))
242 * Only call shrink_slab here (which would also shrink other caches) if
243 * access is not potentially fatal.
247 int nid
= page_to_nid(p
);
249 struct shrink_control shrink
= {
250 .gfp_mask
= GFP_KERNEL
,
252 node_set(nid
, shrink
.nodes_to_scan
);
254 nr
= shrink_slab(&shrink
, 1000, 1000);
255 if (page_count(p
) == 1)
260 EXPORT_SYMBOL_GPL(shake_page
);
263 * Kill all processes that have a poisoned page mapped and then isolate
267 * Find all processes having the page mapped and kill them.
268 * But we keep a page reference around so that the page is not
269 * actually freed yet.
270 * Then stash the page away
272 * There's no convenient way to get back to mapped processes
273 * from the VMAs. So do a brute-force search over all
276 * Remember that machine checks are not common (or rather
277 * if they are common you have other problems), so this shouldn't
278 * be a performance issue.
280 * Also there are some races possible while we get from the
281 * error detection to actually handle it.
286 struct task_struct
*tsk
;
292 * Failure handling: if we can't find or can't kill a process there's
293 * not much we can do. We just print a message and ignore otherwise.
297 * Schedule a process for later kill.
298 * Uses GFP_ATOMIC allocations to avoid potential recursions in the VM.
299 * TBD would GFP_NOIO be enough?
301 static void add_to_kill(struct task_struct
*tsk
, struct page
*p
,
302 struct vm_area_struct
*vma
,
303 struct list_head
*to_kill
,
304 struct to_kill
**tkc
)
312 tk
= kmalloc(sizeof(struct to_kill
), GFP_ATOMIC
);
315 "MCE: Out of memory while machine check handling\n");
319 tk
->addr
= page_address_in_vma(p
, vma
);
323 * In theory we don't have to kill when the page was
324 * munmaped. But it could be also a mremap. Since that's
325 * likely very rare kill anyways just out of paranoia, but use
326 * a SIGKILL because the error is not contained anymore.
328 if (tk
->addr
== -EFAULT
) {
329 pr_info("MCE: Unable to find user space address %lx in %s\n",
330 page_to_pfn(p
), tsk
->comm
);
333 get_task_struct(tsk
);
335 list_add_tail(&tk
->nd
, to_kill
);
339 * Kill the processes that have been collected earlier.
341 * Only do anything when DOIT is set, otherwise just free the list
342 * (this is used for clean pages which do not need killing)
343 * Also when FAIL is set do a force kill because something went
346 static void kill_procs(struct list_head
*to_kill
, int forcekill
, int trapno
,
347 int fail
, struct page
*page
, unsigned long pfn
,
350 struct to_kill
*tk
, *next
;
352 list_for_each_entry_safe (tk
, next
, to_kill
, nd
) {
355 * In case something went wrong with munmapping
356 * make sure the process doesn't catch the
357 * signal and then access the memory. Just kill it.
359 if (fail
|| tk
->addr_valid
== 0) {
361 "MCE %#lx: forcibly killing %s:%d because of failure to unmap corrupted page\n",
362 pfn
, tk
->tsk
->comm
, tk
->tsk
->pid
);
363 do_send_sig_info(SIGKILL
, SEND_SIG_PRIV
,
364 tk
->tsk
, PIDTYPE_PID
);
368 * In theory the process could have mapped
369 * something else on the address in-between. We could
370 * check for that, but we need to tell the
373 else if (kill_proc(tk
->tsk
, tk
->addr
, trapno
,
374 pfn
, page
, flags
) < 0)
376 "MCE %#lx: Cannot send advisory machine check signal to %s:%d\n",
377 pfn
, tk
->tsk
->comm
, tk
->tsk
->pid
);
379 put_task_struct(tk
->tsk
);
385 * Find a dedicated thread which is supposed to handle SIGBUS(BUS_MCEERR_AO)
386 * on behalf of the thread group. Return task_struct of the (first found)
387 * dedicated thread if found, and return NULL otherwise.
389 * We already hold read_lock(&tasklist_lock) in the caller, so we don't
390 * have to call rcu_read_lock/unlock() in this function.
392 static struct task_struct
*find_early_kill_thread(struct task_struct
*tsk
)
394 struct task_struct
*t
;
396 for_each_thread(tsk
, t
)
397 if ((t
->flags
& PF_MCE_PROCESS
) && (t
->flags
& PF_MCE_EARLY
))
403 * Determine whether a given process is "early kill" process which expects
404 * to be signaled when some page under the process is hwpoisoned.
405 * Return task_struct of the dedicated thread (main thread unless explicitly
406 * specified) if the process is "early kill," and otherwise returns NULL.
408 static struct task_struct
*task_early_kill(struct task_struct
*tsk
,
411 struct task_struct
*t
;
416 t
= find_early_kill_thread(tsk
);
419 if (sysctl_memory_failure_early_kill
)
425 * Collect processes when the error hit an anonymous page.
427 static void collect_procs_anon(struct page
*page
, struct list_head
*to_kill
,
428 struct to_kill
**tkc
, int force_early
)
430 struct vm_area_struct
*vma
;
431 struct task_struct
*tsk
;
435 av
= page_lock_anon_vma_read(page
);
436 if (av
== NULL
) /* Not actually mapped anymore */
439 pgoff
= page_to_pgoff(page
);
440 read_lock(&tasklist_lock
);
441 for_each_process (tsk
) {
442 struct anon_vma_chain
*vmac
;
443 struct task_struct
*t
= task_early_kill(tsk
, force_early
);
447 anon_vma_interval_tree_foreach(vmac
, &av
->rb_root
,
450 if (!page_mapped_in_vma(page
, vma
))
452 if (vma
->vm_mm
== t
->mm
)
453 add_to_kill(t
, page
, vma
, to_kill
, tkc
);
456 read_unlock(&tasklist_lock
);
457 page_unlock_anon_vma_read(av
);
461 * Collect processes when the error hit a file mapped page.
463 static void collect_procs_file(struct page
*page
, struct list_head
*to_kill
,
464 struct to_kill
**tkc
, int force_early
)
466 struct vm_area_struct
*vma
;
467 struct task_struct
*tsk
;
468 struct address_space
*mapping
= page
->mapping
;
470 mutex_lock(&mapping
->i_mmap_mutex
);
471 read_lock(&tasklist_lock
);
472 for_each_process(tsk
) {
473 pgoff_t pgoff
= page_to_pgoff(page
);
474 struct task_struct
*t
= task_early_kill(tsk
, force_early
);
478 vma_interval_tree_foreach(vma
, &mapping
->i_mmap
, pgoff
,
481 * Send early kill signal to tasks where a vma covers
482 * the page but the corrupted page is not necessarily
483 * mapped it in its pte.
484 * Assume applications who requested early kill want
485 * to be informed of all such data corruptions.
487 if (vma
->vm_mm
== t
->mm
)
488 add_to_kill(t
, page
, vma
, to_kill
, tkc
);
491 read_unlock(&tasklist_lock
);
492 mutex_unlock(&mapping
->i_mmap_mutex
);
496 * Collect the processes who have the corrupted page mapped to kill.
497 * This is done in two steps for locking reasons.
498 * First preallocate one tokill structure outside the spin locks,
499 * so that we can kill at least one process reasonably reliable.
501 static void collect_procs(struct page
*page
, struct list_head
*tokill
,
509 tk
= kmalloc(sizeof(struct to_kill
), GFP_NOIO
);
513 collect_procs_anon(page
, tokill
, &tk
, force_early
);
515 collect_procs_file(page
, tokill
, &tk
, force_early
);
520 * Error handlers for various types of pages.
524 IGNORED
, /* Error: cannot be handled */
525 FAILED
, /* Error: handling failed */
526 DELAYED
, /* Will be handled later */
527 RECOVERED
, /* Successfully recovered */
530 static const char *action_name
[] = {
531 [IGNORED
] = "Ignored",
533 [DELAYED
] = "Delayed",
534 [RECOVERED
] = "Recovered",
538 * XXX: It is possible that a page is isolated from LRU cache,
539 * and then kept in swap cache or failed to remove from page cache.
540 * The page count will stop it from being freed by unpoison.
541 * Stress tests should be aware of this memory leak problem.
543 static int delete_from_lru_cache(struct page
*p
)
545 if (!isolate_lru_page(p
)) {
547 * Clear sensible page flags, so that the buddy system won't
548 * complain when the page is unpoison-and-freed.
551 ClearPageUnevictable(p
);
553 * drop the page count elevated by isolate_lru_page()
555 page_cache_release(p
);
562 * Error hit kernel page.
563 * Do nothing, try to be lucky and not touch this instead. For a few cases we
564 * could be more sophisticated.
566 static int me_kernel(struct page
*p
, unsigned long pfn
)
572 * Page in unknown state. Do nothing.
574 static int me_unknown(struct page
*p
, unsigned long pfn
)
576 printk(KERN_ERR
"MCE %#lx: Unknown page state\n", pfn
);
581 * Clean (or cleaned) page cache page.
583 static int me_pagecache_clean(struct page
*p
, unsigned long pfn
)
587 struct address_space
*mapping
;
589 delete_from_lru_cache(p
);
592 * For anonymous pages we're done the only reference left
593 * should be the one m_f() holds.
599 * Now truncate the page in the page cache. This is really
600 * more like a "temporary hole punch"
601 * Don't do this for block devices when someone else
602 * has a reference, because it could be file system metadata
603 * and that's not safe to truncate.
605 mapping
= page_mapping(p
);
608 * Page has been teared down in the meanwhile
614 * Truncation is a bit tricky. Enable it per file system for now.
616 * Open: to take i_mutex or not for this? Right now we don't.
618 if (mapping
->a_ops
->error_remove_page
) {
619 err
= mapping
->a_ops
->error_remove_page(mapping
, p
);
621 printk(KERN_INFO
"MCE %#lx: Failed to punch page: %d\n",
623 } else if (page_has_private(p
) &&
624 !try_to_release_page(p
, GFP_NOIO
)) {
625 pr_info("MCE %#lx: failed to release buffers\n", pfn
);
631 * If the file system doesn't support it just invalidate
632 * This fails on dirty or anything with private pages
634 if (invalidate_inode_page(p
))
637 printk(KERN_INFO
"MCE %#lx: Failed to invalidate\n",
644 * Dirty pagecache page
645 * Issues: when the error hit a hole page the error is not properly
648 static int me_pagecache_dirty(struct page
*p
, unsigned long pfn
)
650 struct address_space
*mapping
= page_mapping(p
);
653 /* TBD: print more information about the file. */
656 * IO error will be reported by write(), fsync(), etc.
657 * who check the mapping.
658 * This way the application knows that something went
659 * wrong with its dirty file data.
661 * There's one open issue:
663 * The EIO will be only reported on the next IO
664 * operation and then cleared through the IO map.
665 * Normally Linux has two mechanisms to pass IO error
666 * first through the AS_EIO flag in the address space
667 * and then through the PageError flag in the page.
668 * Since we drop pages on memory failure handling the
669 * only mechanism open to use is through AS_AIO.
671 * This has the disadvantage that it gets cleared on
672 * the first operation that returns an error, while
673 * the PageError bit is more sticky and only cleared
674 * when the page is reread or dropped. If an
675 * application assumes it will always get error on
676 * fsync, but does other operations on the fd before
677 * and the page is dropped between then the error
678 * will not be properly reported.
680 * This can already happen even without hwpoisoned
681 * pages: first on metadata IO errors (which only
682 * report through AS_EIO) or when the page is dropped
685 * So right now we assume that the application DTRT on
686 * the first EIO, but we're not worse than other parts
689 mapping_set_error(mapping
, EIO
);
692 return me_pagecache_clean(p
, pfn
);
696 * Clean and dirty swap cache.
698 * Dirty swap cache page is tricky to handle. The page could live both in page
699 * cache and swap cache(ie. page is freshly swapped in). So it could be
700 * referenced concurrently by 2 types of PTEs:
701 * normal PTEs and swap PTEs. We try to handle them consistently by calling
702 * try_to_unmap(TTU_IGNORE_HWPOISON) to convert the normal PTEs to swap PTEs,
704 * - clear dirty bit to prevent IO
706 * - but keep in the swap cache, so that when we return to it on
707 * a later page fault, we know the application is accessing
708 * corrupted data and shall be killed (we installed simple
709 * interception code in do_swap_page to catch it).
711 * Clean swap cache pages can be directly isolated. A later page fault will
712 * bring in the known good data from disk.
714 static int me_swapcache_dirty(struct page
*p
, unsigned long pfn
)
717 /* Trigger EIO in shmem: */
718 ClearPageUptodate(p
);
720 if (!delete_from_lru_cache(p
))
726 static int me_swapcache_clean(struct page
*p
, unsigned long pfn
)
728 delete_from_swap_cache(p
);
730 if (!delete_from_lru_cache(p
))
737 * Huge pages. Needs work.
739 * - Error on hugepage is contained in hugepage unit (not in raw page unit.)
740 * To narrow down kill region to one page, we need to break up pmd.
742 static int me_huge_page(struct page
*p
, unsigned long pfn
)
745 struct page
*hpage
= compound_head(p
);
747 * We can safely recover from error on free or reserved (i.e.
748 * not in-use) hugepage by dequeuing it from freelist.
749 * To check whether a hugepage is in-use or not, we can't use
750 * page->lru because it can be used in other hugepage operations,
751 * such as __unmap_hugepage_range() and gather_surplus_pages().
752 * So instead we use page_mapping() and PageAnon().
753 * We assume that this function is called with page lock held,
754 * so there is no race between isolation and mapping/unmapping.
756 if (!(page_mapping(hpage
) || PageAnon(hpage
))) {
757 res
= dequeue_hwpoisoned_huge_page(hpage
);
765 * Various page states we can handle.
767 * A page state is defined by its current page->flags bits.
768 * The table matches them in order and calls the right handler.
770 * This is quite tricky because we can access page at any time
771 * in its live cycle, so all accesses have to be extremely careful.
773 * This is not complete. More states could be added.
774 * For any missing state don't attempt recovery.
777 #define dirty (1UL << PG_dirty)
778 #define sc (1UL << PG_swapcache)
779 #define unevict (1UL << PG_unevictable)
780 #define mlock (1UL << PG_mlocked)
781 #define writeback (1UL << PG_writeback)
782 #define lru (1UL << PG_lru)
783 #define swapbacked (1UL << PG_swapbacked)
784 #define head (1UL << PG_head)
785 #define tail (1UL << PG_tail)
786 #define compound (1UL << PG_compound)
787 #define slab (1UL << PG_slab)
788 #define reserved (1UL << PG_reserved)
790 static struct page_state
{
794 int (*action
)(struct page
*p
, unsigned long pfn
);
796 { reserved
, reserved
, "reserved kernel", me_kernel
},
798 * free pages are specially detected outside this table:
799 * PG_buddy pages only make a small fraction of all free pages.
803 * Could in theory check if slab page is free or if we can drop
804 * currently unused objects without touching them. But just
805 * treat it as standard kernel for now.
807 { slab
, slab
, "kernel slab", me_kernel
},
809 #ifdef CONFIG_PAGEFLAGS_EXTENDED
810 { head
, head
, "huge", me_huge_page
},
811 { tail
, tail
, "huge", me_huge_page
},
813 { compound
, compound
, "huge", me_huge_page
},
816 { sc
|dirty
, sc
|dirty
, "dirty swapcache", me_swapcache_dirty
},
817 { sc
|dirty
, sc
, "clean swapcache", me_swapcache_clean
},
819 { mlock
|dirty
, mlock
|dirty
, "dirty mlocked LRU", me_pagecache_dirty
},
820 { mlock
|dirty
, mlock
, "clean mlocked LRU", me_pagecache_clean
},
822 { unevict
|dirty
, unevict
|dirty
, "dirty unevictable LRU", me_pagecache_dirty
},
823 { unevict
|dirty
, unevict
, "clean unevictable LRU", me_pagecache_clean
},
825 { lru
|dirty
, lru
|dirty
, "dirty LRU", me_pagecache_dirty
},
826 { lru
|dirty
, lru
, "clean LRU", me_pagecache_clean
},
829 * Catchall entry: must be at end.
831 { 0, 0, "unknown page state", me_unknown
},
848 * "Dirty/Clean" indication is not 100% accurate due to the possibility of
849 * setting PG_dirty outside page lock. See also comment above set_page_dirty().
851 static void action_result(unsigned long pfn
, char *msg
, int result
)
853 pr_err("MCE %#lx: %s page recovery: %s\n",
854 pfn
, msg
, action_name
[result
]);
857 static int page_action(struct page_state
*ps
, struct page
*p
,
863 result
= ps
->action(p
, pfn
);
864 action_result(pfn
, ps
->msg
, result
);
866 count
= page_count(p
) - 1;
867 if (ps
->action
== me_swapcache_dirty
&& result
== DELAYED
)
871 "MCE %#lx: %s page still referenced by %d users\n",
872 pfn
, ps
->msg
, count
);
876 /* Could do more checks here if page looks ok */
878 * Could adjust zone counters here to correct for the missing page.
881 return (result
== RECOVERED
|| result
== DELAYED
) ? 0 : -EBUSY
;
885 * Do all that is necessary to remove user space mappings. Unmap
886 * the pages and send SIGBUS to the processes if the data was dirty.
888 static int hwpoison_user_mappings(struct page
*p
, unsigned long pfn
,
889 int trapno
, int flags
, struct page
**hpagep
)
891 enum ttu_flags ttu
= TTU_UNMAP
| TTU_IGNORE_MLOCK
| TTU_IGNORE_ACCESS
;
892 struct address_space
*mapping
;
895 int kill
= 1, forcekill
;
896 struct page
*hpage
= *hpagep
;
900 * Here we are interested only in user-mapped pages, so skip any
901 * other types of pages.
903 if (PageReserved(p
) || PageSlab(p
))
905 if (!(PageLRU(hpage
) || PageHuge(p
)))
909 * This check implies we don't kill processes if their pages
910 * are in the swap cache early. Those are always late kills.
912 if (!page_mapped(hpage
))
916 pr_err("MCE %#lx: can't handle KSM pages.\n", pfn
);
920 if (PageSwapCache(p
)) {
922 "MCE %#lx: keeping poisoned page in swap cache\n", pfn
);
923 ttu
|= TTU_IGNORE_HWPOISON
;
927 * Propagate the dirty bit from PTEs to struct page first, because we
928 * need this to decide if we should kill or just drop the page.
929 * XXX: the dirty test could be racy: set_page_dirty() may not always
930 * be called inside page lock (it's recommended but not enforced).
932 mapping
= page_mapping(hpage
);
933 if (!(flags
& MF_MUST_KILL
) && !PageDirty(hpage
) && mapping
&&
934 mapping_cap_writeback_dirty(mapping
)) {
935 if (page_mkclean(hpage
)) {
939 ttu
|= TTU_IGNORE_HWPOISON
;
941 "MCE %#lx: corrupted page was clean: dropped without side effects\n",
947 * ppage: poisoned page
948 * if p is regular page(4k page)
949 * ppage == real poisoned page;
950 * else p is hugetlb or THP, ppage == head page.
954 if (PageTransHuge(hpage
)) {
956 * Verify that this isn't a hugetlbfs head page, the check for
957 * PageAnon is just for avoid tripping a split_huge_page
958 * internal debug check, as split_huge_page refuses to deal with
959 * anything that isn't an anon page. PageAnon can't go away fro
960 * under us because we hold a refcount on the hpage, without a
961 * refcount on the hpage. split_huge_page can't be safely called
962 * in the first place, having a refcount on the tail isn't
963 * enough * to be safe.
965 if (!PageHuge(hpage
) && PageAnon(hpage
)) {
966 if (unlikely(split_huge_page(hpage
))) {
968 * FIXME: if splitting THP is failed, it is
969 * better to stop the following operation rather
970 * than causing panic by unmapping. System might
971 * survive if the page is freed later.
974 "MCE %#lx: failed to split THP\n", pfn
);
976 BUG_ON(!PageHWPoison(p
));
980 * We pinned the head page for hwpoison handling,
981 * now we split the thp and we are interested in
982 * the hwpoisoned raw page, so move the refcount
983 * to it. Similarly, page lock is shifted.
986 if (!(flags
& MF_COUNT_INCREASED
)) {
994 /* THP is split, so ppage should be the real poisoned page. */
1000 * First collect all the processes that have the page
1001 * mapped in dirty form. This has to be done before try_to_unmap,
1002 * because ttu takes the rmap data structures down.
1004 * Error handling: We ignore errors here because
1005 * there's nothing that can be done.
1008 collect_procs(ppage
, &tokill
, flags
& MF_ACTION_REQUIRED
);
1010 ret
= try_to_unmap(ppage
, ttu
);
1011 if (ret
!= SWAP_SUCCESS
)
1012 printk(KERN_ERR
"MCE %#lx: failed to unmap page (mapcount=%d)\n",
1013 pfn
, page_mapcount(ppage
));
1016 * Now that the dirty bit has been propagated to the
1017 * struct page and all unmaps done we can decide if
1018 * killing is needed or not. Only kill when the page
1019 * was dirty or the process is not restartable,
1020 * otherwise the tokill list is merely
1021 * freed. When there was a problem unmapping earlier
1022 * use a more force-full uncatchable kill to prevent
1023 * any accesses to the poisoned memory.
1025 forcekill
= PageDirty(ppage
) || (flags
& MF_MUST_KILL
);
1026 kill_procs(&tokill
, forcekill
, trapno
,
1027 ret
!= SWAP_SUCCESS
, p
, pfn
, flags
);
1032 static void set_page_hwpoison_huge_page(struct page
*hpage
)
1035 int nr_pages
= 1 << compound_order(hpage
);
1036 for (i
= 0; i
< nr_pages
; i
++)
1037 SetPageHWPoison(hpage
+ i
);
1040 static void clear_page_hwpoison_huge_page(struct page
*hpage
)
1043 int nr_pages
= 1 << compound_order(hpage
);
1044 for (i
= 0; i
< nr_pages
; i
++)
1045 ClearPageHWPoison(hpage
+ i
);
1049 * memory_failure - Handle memory failure of a page.
1050 * @pfn: Page Number of the corrupted page
1051 * @trapno: Trap number reported in the signal to user space.
1052 * @flags: fine tune action taken
1054 * This function is called by the low level machine check code
1055 * of an architecture when it detects hardware memory corruption
1056 * of a page. It tries its best to recover, which includes
1057 * dropping pages, killing processes etc.
1059 * The function is primarily of use for corruptions that
1060 * happen outside the current execution context (e.g. when
1061 * detected by a background scrubber)
1063 * Must run in process context (e.g. a work queue) with interrupts
1064 * enabled and no spinlocks hold.
1066 int memory_failure(unsigned long pfn
, int trapno
, int flags
)
1068 struct page_state
*ps
;
1072 unsigned int nr_pages
;
1073 unsigned long page_flags
;
1075 if (!sysctl_memory_failure_recovery
)
1076 panic("Memory failure from trap %d on page %lx", trapno
, pfn
);
1078 if (!pfn_valid(pfn
)) {
1080 "MCE %#lx: memory outside kernel control\n",
1085 p
= pfn_to_page(pfn
);
1086 hpage
= compound_head(p
);
1087 if (TestSetPageHWPoison(p
)) {
1088 printk(KERN_ERR
"MCE %#lx: already hardware poisoned\n", pfn
);
1093 * Currently errors on hugetlbfs pages are measured in hugepage units,
1094 * so nr_pages should be 1 << compound_order. OTOH when errors are on
1095 * transparent hugepages, they are supposed to be split and error
1096 * measurement is done in normal page units. So nr_pages should be one
1100 nr_pages
= 1 << compound_order(hpage
);
1101 else /* normal page or thp */
1103 atomic_long_add(nr_pages
, &num_poisoned_pages
);
1106 * We need/can do nothing about count=0 pages.
1107 * 1) it's a free page, and therefore in safe hand:
1108 * prep_new_page() will be the gate keeper.
1109 * 2) it's a free hugepage, which is also safe:
1110 * an affected hugepage will be dequeued from hugepage freelist,
1111 * so there's no concern about reusing it ever after.
1112 * 3) it's part of a non-compound high order page.
1113 * Implies some kernel user: cannot stop them from
1114 * R/W the page; let's pray that the page has been
1115 * used and will be freed some time later.
1116 * In fact it's dangerous to directly bump up page count from 0,
1117 * that may make page_freeze_refs()/page_unfreeze_refs() mismatch.
1119 if (!(flags
& MF_COUNT_INCREASED
) &&
1120 !get_page_unless_zero(hpage
)) {
1121 if (is_free_buddy_page(p
)) {
1122 action_result(pfn
, "free buddy", DELAYED
);
1124 } else if (PageHuge(hpage
)) {
1126 * Check "filter hit" and "race with other subpage."
1129 if (PageHWPoison(hpage
)) {
1130 if ((hwpoison_filter(p
) && TestClearPageHWPoison(p
))
1131 || (p
!= hpage
&& TestSetPageHWPoison(hpage
))) {
1132 atomic_long_sub(nr_pages
, &num_poisoned_pages
);
1137 set_page_hwpoison_huge_page(hpage
);
1138 res
= dequeue_hwpoisoned_huge_page(hpage
);
1139 action_result(pfn
, "free huge",
1140 res
? IGNORED
: DELAYED
);
1144 action_result(pfn
, "high order kernel", IGNORED
);
1150 * We ignore non-LRU pages for good reasons.
1151 * - PG_locked is only well defined for LRU pages and a few others
1152 * - to avoid races with __set_page_locked()
1153 * - to avoid races with __SetPageSlab*() (and more non-atomic ops)
1154 * The check (unnecessarily) ignores LRU pages being isolated and
1155 * walked by the page reclaim code, however that's not a big loss.
1158 if (!PageLRU(hpage
))
1159 shake_page(hpage
, 0);
1160 if (!PageLRU(hpage
)) {
1162 * shake_page could have turned it free.
1164 if (is_free_buddy_page(p
)) {
1165 if (flags
& MF_COUNT_INCREASED
)
1166 action_result(pfn
, "free buddy", DELAYED
);
1168 action_result(pfn
, "free buddy, 2nd try", DELAYED
);
1177 * We use page flags to determine what action should be taken, but
1178 * the flags can be modified by the error containment action. One
1179 * example is an mlocked page, where PG_mlocked is cleared by
1180 * page_remove_rmap() in try_to_unmap_one(). So to determine page status
1181 * correctly, we save a copy of the page flags at this time.
1184 page_flags
= hpage
->flags
;
1186 page_flags
= p
->flags
;
1189 * unpoison always clear PG_hwpoison inside page lock
1191 if (!PageHWPoison(p
)) {
1192 printk(KERN_ERR
"MCE %#lx: just unpoisoned\n", pfn
);
1193 atomic_long_sub(nr_pages
, &num_poisoned_pages
);
1198 if (hwpoison_filter(p
)) {
1199 if (TestClearPageHWPoison(p
))
1200 atomic_long_sub(nr_pages
, &num_poisoned_pages
);
1206 if (!PageHuge(p
) && !PageTransTail(p
) && !PageLRU(p
))
1207 goto identify_page_state
;
1210 * For error on the tail page, we should set PG_hwpoison
1211 * on the head page to show that the hugepage is hwpoisoned
1213 if (PageHuge(p
) && PageTail(p
) && TestSetPageHWPoison(hpage
)) {
1214 action_result(pfn
, "hugepage already hardware poisoned",
1221 * Set PG_hwpoison on all pages in an error hugepage,
1222 * because containment is done in hugepage unit for now.
1223 * Since we have done TestSetPageHWPoison() for the head page with
1224 * page lock held, we can safely set PG_hwpoison bits on tail pages.
1227 set_page_hwpoison_huge_page(hpage
);
1230 * It's very difficult to mess with pages currently under IO
1231 * and in many cases impossible, so we just avoid it here.
1233 wait_on_page_writeback(p
);
1236 * Now take care of user space mappings.
1237 * Abort on fail: __delete_from_page_cache() assumes unmapped page.
1239 * When the raw error page is thp tail page, hpage points to the raw
1240 * page after thp split.
1242 if (hwpoison_user_mappings(p
, pfn
, trapno
, flags
, &hpage
)
1244 action_result(pfn
, "unmapping failed", IGNORED
);
1250 * Torn down by someone else?
1252 if (PageLRU(p
) && !PageSwapCache(p
) && p
->mapping
== NULL
) {
1253 action_result(pfn
, "already truncated LRU", IGNORED
);
1258 identify_page_state
:
1261 * The first check uses the current page flags which may not have any
1262 * relevant information. The second check with the saved page flagss is
1263 * carried out only if the first check can't determine the page status.
1265 for (ps
= error_states
;; ps
++)
1266 if ((p
->flags
& ps
->mask
) == ps
->res
)
1269 page_flags
|= (p
->flags
& (1UL << PG_dirty
));
1272 for (ps
= error_states
;; ps
++)
1273 if ((page_flags
& ps
->mask
) == ps
->res
)
1275 res
= page_action(ps
, p
, pfn
);
1280 EXPORT_SYMBOL_GPL(memory_failure
);
1282 #define MEMORY_FAILURE_FIFO_ORDER 4
1283 #define MEMORY_FAILURE_FIFO_SIZE (1 << MEMORY_FAILURE_FIFO_ORDER)
1285 struct memory_failure_entry
{
1291 struct memory_failure_cpu
{
1292 DECLARE_KFIFO(fifo
, struct memory_failure_entry
,
1293 MEMORY_FAILURE_FIFO_SIZE
);
1295 struct work_struct work
;
1298 static DEFINE_PER_CPU(struct memory_failure_cpu
, memory_failure_cpu
);
1301 * memory_failure_queue - Schedule handling memory failure of a page.
1302 * @pfn: Page Number of the corrupted page
1303 * @trapno: Trap number reported in the signal to user space.
1304 * @flags: Flags for memory failure handling
1306 * This function is called by the low level hardware error handler
1307 * when it detects hardware memory corruption of a page. It schedules
1308 * the recovering of error page, including dropping pages, killing
1311 * The function is primarily of use for corruptions that
1312 * happen outside the current execution context (e.g. when
1313 * detected by a background scrubber)
1315 * Can run in IRQ context.
1317 void memory_failure_queue(unsigned long pfn
, int trapno
, int flags
)
1319 struct memory_failure_cpu
*mf_cpu
;
1320 unsigned long proc_flags
;
1321 struct memory_failure_entry entry
= {
1327 mf_cpu
= &get_cpu_var(memory_failure_cpu
);
1328 spin_lock_irqsave(&mf_cpu
->lock
, proc_flags
);
1329 if (kfifo_put(&mf_cpu
->fifo
, entry
))
1330 schedule_work_on(smp_processor_id(), &mf_cpu
->work
);
1332 pr_err("Memory failure: buffer overflow when queuing memory failure at %#lx\n",
1334 spin_unlock_irqrestore(&mf_cpu
->lock
, proc_flags
);
1335 put_cpu_var(memory_failure_cpu
);
1337 EXPORT_SYMBOL_GPL(memory_failure_queue
);
1339 static void memory_failure_work_func(struct work_struct
*work
)
1341 struct memory_failure_cpu
*mf_cpu
;
1342 struct memory_failure_entry entry
= { 0, };
1343 unsigned long proc_flags
;
1346 mf_cpu
= this_cpu_ptr(&memory_failure_cpu
);
1348 spin_lock_irqsave(&mf_cpu
->lock
, proc_flags
);
1349 gotten
= kfifo_get(&mf_cpu
->fifo
, &entry
);
1350 spin_unlock_irqrestore(&mf_cpu
->lock
, proc_flags
);
1353 if (entry
.flags
& MF_SOFT_OFFLINE
)
1354 soft_offline_page(pfn_to_page(entry
.pfn
), entry
.flags
);
1356 memory_failure(entry
.pfn
, entry
.trapno
, entry
.flags
);
1360 static int __init
memory_failure_init(void)
1362 struct memory_failure_cpu
*mf_cpu
;
1365 for_each_possible_cpu(cpu
) {
1366 mf_cpu
= &per_cpu(memory_failure_cpu
, cpu
);
1367 spin_lock_init(&mf_cpu
->lock
);
1368 INIT_KFIFO(mf_cpu
->fifo
);
1369 INIT_WORK(&mf_cpu
->work
, memory_failure_work_func
);
1374 core_initcall(memory_failure_init
);
1377 * unpoison_memory - Unpoison a previously poisoned page
1378 * @pfn: Page number of the to be unpoisoned page
1380 * Software-unpoison a page that has been poisoned by
1381 * memory_failure() earlier.
1383 * This is only done on the software-level, so it only works
1384 * for linux injected failures, not real hardware failures
1386 * Returns 0 for success, otherwise -errno.
1388 int unpoison_memory(unsigned long pfn
)
1393 unsigned int nr_pages
;
1395 if (!pfn_valid(pfn
))
1398 p
= pfn_to_page(pfn
);
1399 page
= compound_head(p
);
1401 if (!PageHWPoison(p
)) {
1402 pr_info("MCE: Page was already unpoisoned %#lx\n", pfn
);
1407 * unpoison_memory() can encounter thp only when the thp is being
1408 * worked by memory_failure() and the page lock is not held yet.
1409 * In such case, we yield to memory_failure() and make unpoison fail.
1411 if (!PageHuge(page
) && PageTransHuge(page
)) {
1412 pr_info("MCE: Memory failure is now running on %#lx\n", pfn
);
1416 nr_pages
= 1 << compound_order(page
);
1418 if (!get_page_unless_zero(page
)) {
1420 * Since HWPoisoned hugepage should have non-zero refcount,
1421 * race between memory failure and unpoison seems to happen.
1422 * In such case unpoison fails and memory failure runs
1425 if (PageHuge(page
)) {
1426 pr_info("MCE: Memory failure is now running on free hugepage %#lx\n", pfn
);
1429 if (TestClearPageHWPoison(p
))
1430 atomic_long_dec(&num_poisoned_pages
);
1431 pr_info("MCE: Software-unpoisoned free page %#lx\n", pfn
);
1437 * This test is racy because PG_hwpoison is set outside of page lock.
1438 * That's acceptable because that won't trigger kernel panic. Instead,
1439 * the PG_hwpoison page will be caught and isolated on the entrance to
1440 * the free buddy page pool.
1442 if (TestClearPageHWPoison(page
)) {
1443 pr_info("MCE: Software-unpoisoned page %#lx\n", pfn
);
1444 atomic_long_sub(nr_pages
, &num_poisoned_pages
);
1447 clear_page_hwpoison_huge_page(page
);
1452 if (freeit
&& !(pfn
== my_zero_pfn(0) && page_count(p
) == 1))
1457 EXPORT_SYMBOL(unpoison_memory
);
1459 static struct page
*new_page(struct page
*p
, unsigned long private, int **x
)
1461 int nid
= page_to_nid(p
);
1463 return alloc_huge_page_node(page_hstate(compound_head(p
)),
1466 return alloc_pages_exact_node(nid
, GFP_HIGHUSER_MOVABLE
, 0);
1470 * Safely get reference count of an arbitrary page.
1471 * Returns 0 for a free page, -EIO for a zero refcount page
1472 * that is not free, and 1 for any other page type.
1473 * For 1 the page is returned with increased page count, otherwise not.
1475 static int __get_any_page(struct page
*p
, unsigned long pfn
, int flags
)
1479 if (flags
& MF_COUNT_INCREASED
)
1483 * When the target page is a free hugepage, just remove it
1484 * from free hugepage list.
1486 if (!get_page_unless_zero(compound_head(p
))) {
1488 pr_info("%s: %#lx free huge page\n", __func__
, pfn
);
1490 } else if (is_free_buddy_page(p
)) {
1491 pr_info("%s: %#lx free buddy page\n", __func__
, pfn
);
1494 pr_info("%s: %#lx: unknown zero refcount page type %lx\n",
1495 __func__
, pfn
, p
->flags
);
1499 /* Not a free page */
1505 static int get_any_page(struct page
*page
, unsigned long pfn
, int flags
)
1507 int ret
= __get_any_page(page
, pfn
, flags
);
1509 if (ret
== 1 && !PageHuge(page
) && !PageLRU(page
)) {
1514 shake_page(page
, 1);
1519 ret
= __get_any_page(page
, pfn
, 0);
1520 if (ret
== 1 && !PageLRU(page
)) {
1521 /* Drop page reference which is from __get_any_page() */
1523 pr_info("soft_offline: %#lx: unknown non LRU page type %lx\n",
1531 static int soft_offline_huge_page(struct page
*page
, int flags
)
1534 unsigned long pfn
= page_to_pfn(page
);
1535 struct page
*hpage
= compound_head(page
);
1536 LIST_HEAD(pagelist
);
1539 * This double-check of PageHWPoison is to avoid the race with
1540 * memory_failure(). See also comment in __soft_offline_page().
1543 if (PageHWPoison(hpage
)) {
1546 pr_info("soft offline: %#lx hugepage already poisoned\n", pfn
);
1551 /* Keep page count to indicate a given hugepage is isolated. */
1552 list_move(&hpage
->lru
, &pagelist
);
1553 ret
= migrate_pages(&pagelist
, new_page
, NULL
, MPOL_MF_MOVE_ALL
,
1554 MIGRATE_SYNC
, MR_MEMORY_FAILURE
);
1556 pr_info("soft offline: %#lx: migration failed %d, type %lx\n",
1557 pfn
, ret
, page
->flags
);
1558 if (!list_empty(&pagelist
))
1559 putback_movable_pages(&pagelist
);
1563 /* overcommit hugetlb page will be freed to buddy */
1564 if (PageHuge(page
)) {
1565 set_page_hwpoison_huge_page(hpage
);
1566 dequeue_hwpoisoned_huge_page(hpage
);
1567 atomic_long_add(1 << compound_order(hpage
),
1568 &num_poisoned_pages
);
1570 SetPageHWPoison(page
);
1571 atomic_long_inc(&num_poisoned_pages
);
1577 static int __soft_offline_page(struct page
*page
, int flags
)
1580 unsigned long pfn
= page_to_pfn(page
);
1583 * Check PageHWPoison again inside page lock because PageHWPoison
1584 * is set by memory_failure() outside page lock. Note that
1585 * memory_failure() also double-checks PageHWPoison inside page lock,
1586 * so there's no race between soft_offline_page() and memory_failure().
1589 wait_on_page_writeback(page
);
1590 if (PageHWPoison(page
)) {
1593 pr_info("soft offline: %#lx page already poisoned\n", pfn
);
1597 * Try to invalidate first. This should work for
1598 * non dirty unmapped page cache pages.
1600 ret
= invalidate_inode_page(page
);
1603 * RED-PEN would be better to keep it isolated here, but we
1604 * would need to fix isolation locking first.
1608 pr_info("soft_offline: %#lx: invalidated\n", pfn
);
1609 SetPageHWPoison(page
);
1610 atomic_long_inc(&num_poisoned_pages
);
1615 * Simple invalidation didn't work.
1616 * Try to migrate to a new page instead. migrate.c
1617 * handles a large number of cases for us.
1619 ret
= isolate_lru_page(page
);
1621 * Drop page reference which is came from get_any_page()
1622 * successful isolate_lru_page() already took another one.
1626 LIST_HEAD(pagelist
);
1627 inc_zone_page_state(page
, NR_ISOLATED_ANON
+
1628 page_is_file_cache(page
));
1629 list_add(&page
->lru
, &pagelist
);
1630 ret
= migrate_pages(&pagelist
, new_page
, NULL
, MPOL_MF_MOVE_ALL
,
1631 MIGRATE_SYNC
, MR_MEMORY_FAILURE
);
1633 if (!list_empty(&pagelist
)) {
1634 list_del(&page
->lru
);
1635 dec_zone_page_state(page
, NR_ISOLATED_ANON
+
1636 page_is_file_cache(page
));
1637 putback_lru_page(page
);
1640 pr_info("soft offline: %#lx: migration failed %d, type %lx\n",
1641 pfn
, ret
, page
->flags
);
1646 * After page migration succeeds, the source page can
1647 * be trapped in pagevec and actual freeing is delayed.
1648 * Freeing code works differently based on PG_hwpoison,
1649 * so there's a race. We need to make sure that the
1650 * source page should be freed back to buddy before
1651 * setting PG_hwpoison.
1653 if (!is_free_buddy_page(page
))
1655 SetPageHWPoison(page
);
1656 if (!is_free_buddy_page(page
))
1657 pr_info("soft offline: %#lx: page leaked\n",
1659 atomic_long_inc(&num_poisoned_pages
);
1662 pr_info("soft offline: %#lx: isolation failed: %d, page count %d, type %lx\n",
1663 pfn
, ret
, page_count(page
), page
->flags
);
1669 * soft_offline_page - Soft offline a page.
1670 * @page: page to offline
1671 * @flags: flags. Same as memory_failure().
1673 * Returns 0 on success, otherwise negated errno.
1675 * Soft offline a page, by migration or invalidation,
1676 * without killing anything. This is for the case when
1677 * a page is not corrupted yet (so it's still valid to access),
1678 * but has had a number of corrected errors and is better taken
1681 * The actual policy on when to do that is maintained by
1684 * This should never impact any application or cause data loss,
1685 * however it might take some time.
1687 * This is not a 100% solution for all memory, but tries to be
1688 * ``good enough'' for the majority of memory.
1690 int soft_offline_page(struct page
*page
, int flags
)
1693 unsigned long pfn
= page_to_pfn(page
);
1694 struct page
*hpage
= compound_head(page
);
1696 if (PageHWPoison(page
)) {
1697 pr_info("soft offline: %#lx page already poisoned\n", pfn
);
1700 if (!PageHuge(page
) && PageTransHuge(hpage
)) {
1701 if (PageAnon(hpage
) && unlikely(split_huge_page(hpage
))) {
1702 pr_info("soft offline: %#lx: failed to split THP\n",
1711 * Isolate the page, so that it doesn't get reallocated if it
1712 * was free. This flag should be kept set until the source page
1713 * is freed and PG_hwpoison on it is set.
1715 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
1716 set_migratetype_isolate(page
, true);
1718 ret
= get_any_page(page
, pfn
, flags
);
1720 if (ret
> 0) { /* for in-use pages */
1722 ret
= soft_offline_huge_page(page
, flags
);
1724 ret
= __soft_offline_page(page
, flags
);
1725 } else if (ret
== 0) { /* for free pages */
1726 if (PageHuge(page
)) {
1727 set_page_hwpoison_huge_page(hpage
);
1728 if (!dequeue_hwpoisoned_huge_page(hpage
))
1729 atomic_long_add(1 << compound_order(hpage
),
1730 &num_poisoned_pages
);
1732 if (!TestSetPageHWPoison(page
))
1733 atomic_long_inc(&num_poisoned_pages
);
1736 unset_migratetype_isolate(page
, MIGRATE_MOVABLE
);