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 * It can be very tempting to add handling for obscure cases here.
25 * In general any code for handling new cases should only be added iff:
26 * - You know how to test it.
27 * - You have a test that can be added to mce-test
28 * https://git.kernel.org/cgit/utils/cpu/mce/mce-test.git/
29 * - The case actually shows up as a frequent (top 10) page state in
30 * tools/vm/page-types when running a real workload.
32 * There are several operations here with exponential complexity because
33 * of unsuitable VM data structures. For example the operation to map back
34 * from RMAP chains to processes has to walk the complete process list and
35 * has non linear complexity with the number. But since memory corruptions
36 * are rare we hope to get away with this. This avoids impacting the core
39 #include <linux/kernel.h>
41 #include <linux/page-flags.h>
42 #include <linux/kernel-page-flags.h>
43 #include <linux/sched/signal.h>
44 #include <linux/sched/task.h>
45 #include <linux/ksm.h>
46 #include <linux/rmap.h>
47 #include <linux/export.h>
48 #include <linux/pagemap.h>
49 #include <linux/swap.h>
50 #include <linux/backing-dev.h>
51 #include <linux/migrate.h>
52 #include <linux/suspend.h>
53 #include <linux/slab.h>
54 #include <linux/swapops.h>
55 #include <linux/hugetlb.h>
56 #include <linux/memory_hotplug.h>
57 #include <linux/mm_inline.h>
58 #include <linux/kfifo.h>
59 #include <linux/ratelimit.h>
61 #include "ras/ras_event.h"
63 int sysctl_memory_failure_early_kill __read_mostly = 0;
65 int sysctl_memory_failure_recovery __read_mostly = 1;
67 atomic_long_t num_poisoned_pages __read_mostly = ATOMIC_LONG_INIT(0);
69 #if defined(CONFIG_HWPOISON_INJECT) || defined(CONFIG_HWPOISON_INJECT_MODULE)
71 u32 hwpoison_filter_enable = 0;
72 u32 hwpoison_filter_dev_major = ~0U;
73 u32 hwpoison_filter_dev_minor = ~0U;
74 u64 hwpoison_filter_flags_mask;
75 u64 hwpoison_filter_flags_value;
76 EXPORT_SYMBOL_GPL(hwpoison_filter_enable);
77 EXPORT_SYMBOL_GPL(hwpoison_filter_dev_major);
78 EXPORT_SYMBOL_GPL(hwpoison_filter_dev_minor);
79 EXPORT_SYMBOL_GPL(hwpoison_filter_flags_mask);
80 EXPORT_SYMBOL_GPL(hwpoison_filter_flags_value);
82 static int hwpoison_filter_dev(struct page *p)
84 struct address_space *mapping;
87 if (hwpoison_filter_dev_major == ~0U &&
88 hwpoison_filter_dev_minor == ~0U)
92 * page_mapping() does not accept slab pages.
97 mapping = page_mapping(p);
98 if (mapping == NULL || mapping->host == NULL)
101 dev = mapping->host->i_sb->s_dev;
102 if (hwpoison_filter_dev_major != ~0U &&
103 hwpoison_filter_dev_major != MAJOR(dev))
105 if (hwpoison_filter_dev_minor != ~0U &&
106 hwpoison_filter_dev_minor != MINOR(dev))
112 static int hwpoison_filter_flags(struct page *p)
114 if (!hwpoison_filter_flags_mask)
117 if ((stable_page_flags(p) & hwpoison_filter_flags_mask) ==
118 hwpoison_filter_flags_value)
125 * This allows stress tests to limit test scope to a collection of tasks
126 * by putting them under some memcg. This prevents killing unrelated/important
127 * processes such as /sbin/init. Note that the target task may share clean
128 * pages with init (eg. libc text), which is harmless. If the target task
129 * share _dirty_ pages with another task B, the test scheme must make sure B
130 * is also included in the memcg. At last, due to race conditions this filter
131 * can only guarantee that the page either belongs to the memcg tasks, or is
135 u64 hwpoison_filter_memcg;
136 EXPORT_SYMBOL_GPL(hwpoison_filter_memcg);
137 static int hwpoison_filter_task(struct page *p)
139 if (!hwpoison_filter_memcg)
142 if (page_cgroup_ino(p) != hwpoison_filter_memcg)
148 static int hwpoison_filter_task(struct page *p) { return 0; }
151 int hwpoison_filter(struct page *p)
153 if (!hwpoison_filter_enable)
156 if (hwpoison_filter_dev(p))
159 if (hwpoison_filter_flags(p))
162 if (hwpoison_filter_task(p))
168 int hwpoison_filter(struct page *p)
174 EXPORT_SYMBOL_GPL(hwpoison_filter);
177 * Send all the processes who have the page mapped a signal.
178 * ``action optional'' if they are not immediately affected by the error
179 * ``action required'' if error happened in current execution context
181 static int kill_proc(struct task_struct *t, unsigned long addr, int trapno,
182 unsigned long pfn, struct page *page, int flags)
187 pr_err("Memory failure: %#lx: Killing %s:%d due to hardware memory corruption\n",
188 pfn, t->comm, t->pid);
189 si.si_signo = SIGBUS;
191 si.si_addr = (void *)addr;
192 #ifdef __ARCH_SI_TRAPNO
193 si.si_trapno = trapno;
195 si.si_addr_lsb = compound_order(compound_head(page)) + PAGE_SHIFT;
197 if ((flags & MF_ACTION_REQUIRED) && t->mm == current->mm) {
198 si.si_code = BUS_MCEERR_AR;
199 ret = force_sig_info(SIGBUS, &si, current);
202 * Don't use force here, it's convenient if the signal
203 * can be temporarily blocked.
204 * This could cause a loop when the user sets SIGBUS
205 * to SIG_IGN, but hopefully no one will do that?
207 si.si_code = BUS_MCEERR_AO;
208 ret = send_sig_info(SIGBUS, &si, t); /* synchronous? */
211 pr_info("Memory failure: Error sending signal to %s:%d: %d\n",
212 t->comm, t->pid, ret);
217 * When a unknown page type is encountered drain as many buffers as possible
218 * in the hope to turn the page into a LRU or free page, which we can handle.
220 void shake_page(struct page *p, int access)
229 drain_all_pages(page_zone(p));
230 if (PageLRU(p) || is_free_buddy_page(p))
235 * Only call shrink_node_slabs here (which would also shrink
236 * other caches) if access is not potentially fatal.
239 drop_slab_node(page_to_nid(p));
241 EXPORT_SYMBOL_GPL(shake_page);
244 * Kill all processes that have a poisoned page mapped and then isolate
248 * Find all processes having the page mapped and kill them.
249 * But we keep a page reference around so that the page is not
250 * actually freed yet.
251 * Then stash the page away
253 * There's no convenient way to get back to mapped processes
254 * from the VMAs. So do a brute-force search over all
257 * Remember that machine checks are not common (or rather
258 * if they are common you have other problems), so this shouldn't
259 * be a performance issue.
261 * Also there are some races possible while we get from the
262 * error detection to actually handle it.
267 struct task_struct *tsk;
273 * Failure handling: if we can't find or can't kill a process there's
274 * not much we can do. We just print a message and ignore otherwise.
278 * Schedule a process for later kill.
279 * Uses GFP_ATOMIC allocations to avoid potential recursions in the VM.
280 * TBD would GFP_NOIO be enough?
282 static void add_to_kill(struct task_struct *tsk, struct page *p,
283 struct vm_area_struct *vma,
284 struct list_head *to_kill,
285 struct to_kill **tkc)
293 tk = kmalloc(sizeof(struct to_kill), GFP_ATOMIC);
295 pr_err("Memory failure: Out of memory while machine check handling\n");
299 tk->addr = page_address_in_vma(p, vma);
303 * In theory we don't have to kill when the page was
304 * munmaped. But it could be also a mremap. Since that's
305 * likely very rare kill anyways just out of paranoia, but use
306 * a SIGKILL because the error is not contained anymore.
308 if (tk->addr == -EFAULT) {
309 pr_info("Memory failure: Unable to find user space address %lx in %s\n",
310 page_to_pfn(p), tsk->comm);
313 get_task_struct(tsk);
315 list_add_tail(&tk->nd, to_kill);
319 * Kill the processes that have been collected earlier.
321 * Only do anything when DOIT is set, otherwise just free the list
322 * (this is used for clean pages which do not need killing)
323 * Also when FAIL is set do a force kill because something went
326 static void kill_procs(struct list_head *to_kill, int forcekill, int trapno,
327 bool fail, struct page *page, unsigned long pfn,
330 struct to_kill *tk, *next;
332 list_for_each_entry_safe (tk, next, to_kill, nd) {
335 * In case something went wrong with munmapping
336 * make sure the process doesn't catch the
337 * signal and then access the memory. Just kill it.
339 if (fail || tk->addr_valid == 0) {
340 pr_err("Memory failure: %#lx: forcibly killing %s:%d because of failure to unmap corrupted page\n",
341 pfn, tk->tsk->comm, tk->tsk->pid);
342 do_send_sig_info(SIGKILL, SEND_SIG_PRIV,
343 tk->tsk, PIDTYPE_PID);
347 * In theory the process could have mapped
348 * something else on the address in-between. We could
349 * check for that, but we need to tell the
352 else if (kill_proc(tk->tsk, tk->addr, trapno,
353 pfn, page, flags) < 0)
354 pr_err("Memory failure: %#lx: Cannot send advisory machine check signal to %s:%d\n",
355 pfn, tk->tsk->comm, tk->tsk->pid);
357 put_task_struct(tk->tsk);
363 * Find a dedicated thread which is supposed to handle SIGBUS(BUS_MCEERR_AO)
364 * on behalf of the thread group. Return task_struct of the (first found)
365 * dedicated thread if found, and return NULL otherwise.
367 * We already hold read_lock(&tasklist_lock) in the caller, so we don't
368 * have to call rcu_read_lock/unlock() in this function.
370 static struct task_struct *find_early_kill_thread(struct task_struct *tsk)
372 struct task_struct *t;
374 for_each_thread(tsk, t)
375 if ((t->flags & PF_MCE_PROCESS) && (t->flags & PF_MCE_EARLY))
381 * Determine whether a given process is "early kill" process which expects
382 * to be signaled when some page under the process is hwpoisoned.
383 * Return task_struct of the dedicated thread (main thread unless explicitly
384 * specified) if the process is "early kill," and otherwise returns NULL.
386 static struct task_struct *task_early_kill(struct task_struct *tsk,
389 struct task_struct *t;
394 t = find_early_kill_thread(tsk);
397 if (sysctl_memory_failure_early_kill)
403 * Collect processes when the error hit an anonymous page.
405 static void collect_procs_anon(struct page *page, struct list_head *to_kill,
406 struct to_kill **tkc, int force_early)
408 struct vm_area_struct *vma;
409 struct task_struct *tsk;
413 av = page_lock_anon_vma_read(page);
414 if (av == NULL) /* Not actually mapped anymore */
417 pgoff = page_to_pgoff(page);
418 read_lock(&tasklist_lock);
419 for_each_process (tsk) {
420 struct anon_vma_chain *vmac;
421 struct task_struct *t = task_early_kill(tsk, force_early);
425 anon_vma_interval_tree_foreach(vmac, &av->rb_root,
428 if (!page_mapped_in_vma(page, vma))
430 if (vma->vm_mm == t->mm)
431 add_to_kill(t, page, vma, to_kill, tkc);
434 read_unlock(&tasklist_lock);
435 page_unlock_anon_vma_read(av);
439 * Collect processes when the error hit a file mapped page.
441 static void collect_procs_file(struct page *page, struct list_head *to_kill,
442 struct to_kill **tkc, int force_early)
444 struct vm_area_struct *vma;
445 struct task_struct *tsk;
446 struct address_space *mapping = page->mapping;
448 i_mmap_lock_read(mapping);
449 read_lock(&tasklist_lock);
450 for_each_process(tsk) {
451 pgoff_t pgoff = page_to_pgoff(page);
452 struct task_struct *t = task_early_kill(tsk, force_early);
456 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff,
459 * Send early kill signal to tasks where a vma covers
460 * the page but the corrupted page is not necessarily
461 * mapped it in its pte.
462 * Assume applications who requested early kill want
463 * to be informed of all such data corruptions.
465 if (vma->vm_mm == t->mm)
466 add_to_kill(t, page, vma, to_kill, tkc);
469 read_unlock(&tasklist_lock);
470 i_mmap_unlock_read(mapping);
474 * Collect the processes who have the corrupted page mapped to kill.
475 * This is done in two steps for locking reasons.
476 * First preallocate one tokill structure outside the spin locks,
477 * so that we can kill at least one process reasonably reliable.
479 static void collect_procs(struct page *page, struct list_head *tokill,
487 tk = kmalloc(sizeof(struct to_kill), GFP_NOIO);
491 collect_procs_anon(page, tokill, &tk, force_early);
493 collect_procs_file(page, tokill, &tk, force_early);
497 static const char *action_name[] = {
498 [MF_IGNORED] = "Ignored",
499 [MF_FAILED] = "Failed",
500 [MF_DELAYED] = "Delayed",
501 [MF_RECOVERED] = "Recovered",
504 static const char * const action_page_types[] = {
505 [MF_MSG_KERNEL] = "reserved kernel page",
506 [MF_MSG_KERNEL_HIGH_ORDER] = "high-order kernel page",
507 [MF_MSG_SLAB] = "kernel slab page",
508 [MF_MSG_DIFFERENT_COMPOUND] = "different compound page after locking",
509 [MF_MSG_POISONED_HUGE] = "huge page already hardware poisoned",
510 [MF_MSG_HUGE] = "huge page",
511 [MF_MSG_FREE_HUGE] = "free huge page",
512 [MF_MSG_NON_PMD_HUGE] = "non-pmd-sized huge page",
513 [MF_MSG_UNMAP_FAILED] = "unmapping failed page",
514 [MF_MSG_DIRTY_SWAPCACHE] = "dirty swapcache page",
515 [MF_MSG_CLEAN_SWAPCACHE] = "clean swapcache page",
516 [MF_MSG_DIRTY_MLOCKED_LRU] = "dirty mlocked LRU page",
517 [MF_MSG_CLEAN_MLOCKED_LRU] = "clean mlocked LRU page",
518 [MF_MSG_DIRTY_UNEVICTABLE_LRU] = "dirty unevictable LRU page",
519 [MF_MSG_CLEAN_UNEVICTABLE_LRU] = "clean unevictable LRU page",
520 [MF_MSG_DIRTY_LRU] = "dirty LRU page",
521 [MF_MSG_CLEAN_LRU] = "clean LRU page",
522 [MF_MSG_TRUNCATED_LRU] = "already truncated LRU page",
523 [MF_MSG_BUDDY] = "free buddy page",
524 [MF_MSG_BUDDY_2ND] = "free buddy page (2nd try)",
525 [MF_MSG_UNKNOWN] = "unknown page",
529 * XXX: It is possible that a page is isolated from LRU cache,
530 * and then kept in swap cache or failed to remove from page cache.
531 * The page count will stop it from being freed by unpoison.
532 * Stress tests should be aware of this memory leak problem.
534 static int delete_from_lru_cache(struct page *p)
536 if (!isolate_lru_page(p)) {
538 * Clear sensible page flags, so that the buddy system won't
539 * complain when the page is unpoison-and-freed.
542 ClearPageUnevictable(p);
545 * Poisoned page might never drop its ref count to 0 so we have
546 * to uncharge it manually from its memcg.
548 mem_cgroup_uncharge(p);
551 * drop the page count elevated by isolate_lru_page()
559 static int truncate_error_page(struct page *p, unsigned long pfn,
560 struct address_space *mapping)
564 if (mapping->a_ops->error_remove_page) {
565 int err = mapping->a_ops->error_remove_page(mapping, p);
568 pr_info("Memory failure: %#lx: Failed to punch page: %d\n",
570 } else if (page_has_private(p) &&
571 !try_to_release_page(p, GFP_NOIO)) {
572 pr_info("Memory failure: %#lx: failed to release buffers\n",
579 * If the file system doesn't support it just invalidate
580 * This fails on dirty or anything with private pages
582 if (invalidate_inode_page(p))
585 pr_info("Memory failure: %#lx: Failed to invalidate\n",
593 * Error hit kernel page.
594 * Do nothing, try to be lucky and not touch this instead. For a few cases we
595 * could be more sophisticated.
597 static int me_kernel(struct page *p, unsigned long pfn)
603 * Page in unknown state. Do nothing.
605 static int me_unknown(struct page *p, unsigned long pfn)
607 pr_err("Memory failure: %#lx: Unknown page state\n", pfn);
612 * Clean (or cleaned) page cache page.
614 static int me_pagecache_clean(struct page *p, unsigned long pfn)
616 struct address_space *mapping;
618 delete_from_lru_cache(p);
621 * For anonymous pages we're done the only reference left
622 * should be the one m_f() holds.
628 * Now truncate the page in the page cache. This is really
629 * more like a "temporary hole punch"
630 * Don't do this for block devices when someone else
631 * has a reference, because it could be file system metadata
632 * and that's not safe to truncate.
634 mapping = page_mapping(p);
637 * Page has been teared down in the meanwhile
643 * Truncation is a bit tricky. Enable it per file system for now.
645 * Open: to take i_mutex or not for this? Right now we don't.
647 return truncate_error_page(p, pfn, mapping);
651 * Dirty pagecache page
652 * Issues: when the error hit a hole page the error is not properly
655 static int me_pagecache_dirty(struct page *p, unsigned long pfn)
657 struct address_space *mapping = page_mapping(p);
660 /* TBD: print more information about the file. */
663 * IO error will be reported by write(), fsync(), etc.
664 * who check the mapping.
665 * This way the application knows that something went
666 * wrong with its dirty file data.
668 * There's one open issue:
670 * The EIO will be only reported on the next IO
671 * operation and then cleared through the IO map.
672 * Normally Linux has two mechanisms to pass IO error
673 * first through the AS_EIO flag in the address space
674 * and then through the PageError flag in the page.
675 * Since we drop pages on memory failure handling the
676 * only mechanism open to use is through AS_AIO.
678 * This has the disadvantage that it gets cleared on
679 * the first operation that returns an error, while
680 * the PageError bit is more sticky and only cleared
681 * when the page is reread or dropped. If an
682 * application assumes it will always get error on
683 * fsync, but does other operations on the fd before
684 * and the page is dropped between then the error
685 * will not be properly reported.
687 * This can already happen even without hwpoisoned
688 * pages: first on metadata IO errors (which only
689 * report through AS_EIO) or when the page is dropped
692 * So right now we assume that the application DTRT on
693 * the first EIO, but we're not worse than other parts
696 mapping_set_error(mapping, -EIO);
699 return me_pagecache_clean(p, pfn);
703 * Clean and dirty swap cache.
705 * Dirty swap cache page is tricky to handle. The page could live both in page
706 * cache and swap cache(ie. page is freshly swapped in). So it could be
707 * referenced concurrently by 2 types of PTEs:
708 * normal PTEs and swap PTEs. We try to handle them consistently by calling
709 * try_to_unmap(TTU_IGNORE_HWPOISON) to convert the normal PTEs to swap PTEs,
711 * - clear dirty bit to prevent IO
713 * - but keep in the swap cache, so that when we return to it on
714 * a later page fault, we know the application is accessing
715 * corrupted data and shall be killed (we installed simple
716 * interception code in do_swap_page to catch it).
718 * Clean swap cache pages can be directly isolated. A later page fault will
719 * bring in the known good data from disk.
721 static int me_swapcache_dirty(struct page *p, unsigned long pfn)
724 /* Trigger EIO in shmem: */
725 ClearPageUptodate(p);
727 if (!delete_from_lru_cache(p))
733 static int me_swapcache_clean(struct page *p, unsigned long pfn)
735 delete_from_swap_cache(p);
737 if (!delete_from_lru_cache(p))
744 * Huge pages. Needs work.
746 * - Error on hugepage is contained in hugepage unit (not in raw page unit.)
747 * To narrow down kill region to one page, we need to break up pmd.
749 static int me_huge_page(struct page *p, unsigned long pfn)
752 struct page *hpage = compound_head(p);
753 struct address_space *mapping;
755 if (!PageHuge(hpage))
758 mapping = page_mapping(hpage);
760 res = truncate_error_page(hpage, pfn, mapping);
764 * migration entry prevents later access on error anonymous
765 * hugepage, so we can free and dissolve it into buddy to
766 * save healthy subpages.
770 dissolve_free_huge_page(p);
779 * Various page states we can handle.
781 * A page state is defined by its current page->flags bits.
782 * The table matches them in order and calls the right handler.
784 * This is quite tricky because we can access page at any time
785 * in its live cycle, so all accesses have to be extremely careful.
787 * This is not complete. More states could be added.
788 * For any missing state don't attempt recovery.
791 #define dirty (1UL << PG_dirty)
792 #define sc ((1UL << PG_swapcache) | (1UL << PG_swapbacked))
793 #define unevict (1UL << PG_unevictable)
794 #define mlock (1UL << PG_mlocked)
795 #define writeback (1UL << PG_writeback)
796 #define lru (1UL << PG_lru)
797 #define head (1UL << PG_head)
798 #define slab (1UL << PG_slab)
799 #define reserved (1UL << PG_reserved)
801 static struct page_state {
804 enum mf_action_page_type type;
805 int (*action)(struct page *p, unsigned long pfn);
807 { reserved, reserved, MF_MSG_KERNEL, me_kernel },
809 * free pages are specially detected outside this table:
810 * PG_buddy pages only make a small fraction of all free pages.
814 * Could in theory check if slab page is free or if we can drop
815 * currently unused objects without touching them. But just
816 * treat it as standard kernel for now.
818 { slab, slab, MF_MSG_SLAB, me_kernel },
820 { head, head, MF_MSG_HUGE, me_huge_page },
822 { sc|dirty, sc|dirty, MF_MSG_DIRTY_SWAPCACHE, me_swapcache_dirty },
823 { sc|dirty, sc, MF_MSG_CLEAN_SWAPCACHE, me_swapcache_clean },
825 { mlock|dirty, mlock|dirty, MF_MSG_DIRTY_MLOCKED_LRU, me_pagecache_dirty },
826 { mlock|dirty, mlock, MF_MSG_CLEAN_MLOCKED_LRU, me_pagecache_clean },
828 { unevict|dirty, unevict|dirty, MF_MSG_DIRTY_UNEVICTABLE_LRU, me_pagecache_dirty },
829 { unevict|dirty, unevict, MF_MSG_CLEAN_UNEVICTABLE_LRU, me_pagecache_clean },
831 { lru|dirty, lru|dirty, MF_MSG_DIRTY_LRU, me_pagecache_dirty },
832 { lru|dirty, lru, MF_MSG_CLEAN_LRU, me_pagecache_clean },
835 * Catchall entry: must be at end.
837 { 0, 0, MF_MSG_UNKNOWN, me_unknown },
851 * "Dirty/Clean" indication is not 100% accurate due to the possibility of
852 * setting PG_dirty outside page lock. See also comment above set_page_dirty().
854 static void action_result(unsigned long pfn, enum mf_action_page_type type,
855 enum mf_result result)
857 trace_memory_failure_event(pfn, type, result);
859 pr_err("Memory failure: %#lx: recovery action for %s: %s\n",
860 pfn, action_page_types[type], action_name[result]);
863 static int page_action(struct page_state *ps, struct page *p,
869 result = ps->action(p, pfn);
871 count = page_count(p) - 1;
872 if (ps->action == me_swapcache_dirty && result == MF_DELAYED)
875 pr_err("Memory failure: %#lx: %s still referenced by %d users\n",
876 pfn, action_page_types[ps->type], count);
879 action_result(pfn, ps->type, result);
881 /* Could do more checks here if page looks ok */
883 * Could adjust zone counters here to correct for the missing page.
886 return (result == MF_RECOVERED || result == MF_DELAYED) ? 0 : -EBUSY;
890 * get_hwpoison_page() - Get refcount for memory error handling:
891 * @page: raw error page (hit by memory error)
893 * Return: return 0 if failed to grab the refcount, otherwise true (some
896 int get_hwpoison_page(struct page *page)
898 struct page *head = compound_head(page);
900 if (!PageHuge(head) && PageTransHuge(head)) {
902 * Non anonymous thp exists only in allocation/free time. We
903 * can't handle such a case correctly, so let's give it up.
904 * This should be better than triggering BUG_ON when kernel
905 * tries to touch the "partially handled" page.
907 if (!PageAnon(head)) {
908 pr_err("Memory failure: %#lx: non anonymous thp\n",
914 if (get_page_unless_zero(head)) {
915 if (head == compound_head(page))
918 pr_info("Memory failure: %#lx cannot catch tail\n",
925 EXPORT_SYMBOL_GPL(get_hwpoison_page);
928 * Do all that is necessary to remove user space mappings. Unmap
929 * the pages and send SIGBUS to the processes if the data was dirty.
931 static bool hwpoison_user_mappings(struct page *p, unsigned long pfn,
932 int trapno, int flags, struct page **hpagep)
934 enum ttu_flags ttu = TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS;
935 struct address_space *mapping;
938 int kill = 1, forcekill;
939 struct page *hpage = *hpagep;
940 bool mlocked = PageMlocked(hpage);
943 * Here we are interested only in user-mapped pages, so skip any
944 * other types of pages.
946 if (PageReserved(p) || PageSlab(p))
948 if (!(PageLRU(hpage) || PageHuge(p)))
952 * This check implies we don't kill processes if their pages
953 * are in the swap cache early. Those are always late kills.
955 if (!page_mapped(hpage))
959 pr_err("Memory failure: %#lx: can't handle KSM pages.\n", pfn);
963 if (PageSwapCache(p)) {
964 pr_err("Memory failure: %#lx: keeping poisoned page in swap cache\n",
966 ttu |= TTU_IGNORE_HWPOISON;
970 * Propagate the dirty bit from PTEs to struct page first, because we
971 * need this to decide if we should kill or just drop the page.
972 * XXX: the dirty test could be racy: set_page_dirty() may not always
973 * be called inside page lock (it's recommended but not enforced).
975 mapping = page_mapping(hpage);
976 if (!(flags & MF_MUST_KILL) && !PageDirty(hpage) && mapping &&
977 mapping_cap_writeback_dirty(mapping)) {
978 if (page_mkclean(hpage)) {
982 ttu |= TTU_IGNORE_HWPOISON;
983 pr_info("Memory failure: %#lx: corrupted page was clean: dropped without side effects\n",
989 * First collect all the processes that have the page
990 * mapped in dirty form. This has to be done before try_to_unmap,
991 * because ttu takes the rmap data structures down.
993 * Error handling: We ignore errors here because
994 * there's nothing that can be done.
997 collect_procs(hpage, &tokill, flags & MF_ACTION_REQUIRED);
999 unmap_success = try_to_unmap(hpage, ttu);
1001 pr_err("Memory failure: %#lx: failed to unmap page (mapcount=%d)\n",
1002 pfn, page_mapcount(hpage));
1005 * try_to_unmap() might put mlocked page in lru cache, so call
1006 * shake_page() again to ensure that it's flushed.
1009 shake_page(hpage, 0);
1012 * Now that the dirty bit has been propagated to the
1013 * struct page and all unmaps done we can decide if
1014 * killing is needed or not. Only kill when the page
1015 * was dirty or the process is not restartable,
1016 * otherwise the tokill list is merely
1017 * freed. When there was a problem unmapping earlier
1018 * use a more force-full uncatchable kill to prevent
1019 * any accesses to the poisoned memory.
1021 forcekill = PageDirty(hpage) || (flags & MF_MUST_KILL);
1022 kill_procs(&tokill, forcekill, trapno, !unmap_success, p, pfn, flags);
1024 return unmap_success;
1027 static int identify_page_state(unsigned long pfn, struct page *p,
1028 unsigned long page_flags)
1030 struct page_state *ps;
1033 * The first check uses the current page flags which may not have any
1034 * relevant information. The second check with the saved page flags is
1035 * carried out only if the first check can't determine the page status.
1037 for (ps = error_states;; ps++)
1038 if ((p->flags & ps->mask) == ps->res)
1041 page_flags |= (p->flags & (1UL << PG_dirty));
1044 for (ps = error_states;; ps++)
1045 if ((page_flags & ps->mask) == ps->res)
1047 return page_action(ps, p, pfn);
1050 static int memory_failure_hugetlb(unsigned long pfn, int trapno, int flags)
1052 struct page *p = pfn_to_page(pfn);
1053 struct page *head = compound_head(p);
1055 unsigned long page_flags;
1057 if (TestSetPageHWPoison(head)) {
1058 pr_err("Memory failure: %#lx: already hardware poisoned\n",
1063 num_poisoned_pages_inc();
1065 if (!(flags & MF_COUNT_INCREASED) && !get_hwpoison_page(p)) {
1067 * Check "filter hit" and "race with other subpage."
1070 if (PageHWPoison(head)) {
1071 if ((hwpoison_filter(p) && TestClearPageHWPoison(p))
1072 || (p != head && TestSetPageHWPoison(head))) {
1073 num_poisoned_pages_dec();
1079 dissolve_free_huge_page(p);
1080 action_result(pfn, MF_MSG_FREE_HUGE, MF_DELAYED);
1085 page_flags = head->flags;
1087 if (!PageHWPoison(head)) {
1088 pr_err("Memory failure: %#lx: just unpoisoned\n", pfn);
1089 num_poisoned_pages_dec();
1091 put_hwpoison_page(head);
1096 * TODO: hwpoison for pud-sized hugetlb doesn't work right now, so
1097 * simply disable it. In order to make it work properly, we need
1099 * - conversion of a pud that maps an error hugetlb into hwpoison
1100 * entry properly works, and
1101 * - other mm code walking over page table is aware of pud-aligned
1104 if (huge_page_size(page_hstate(head)) > PMD_SIZE) {
1105 action_result(pfn, MF_MSG_NON_PMD_HUGE, MF_IGNORED);
1110 if (!hwpoison_user_mappings(p, pfn, trapno, flags, &head)) {
1111 action_result(pfn, MF_MSG_UNMAP_FAILED, MF_IGNORED);
1116 res = identify_page_state(pfn, p, page_flags);
1123 * memory_failure - Handle memory failure of a page.
1124 * @pfn: Page Number of the corrupted page
1125 * @trapno: Trap number reported in the signal to user space.
1126 * @flags: fine tune action taken
1128 * This function is called by the low level machine check code
1129 * of an architecture when it detects hardware memory corruption
1130 * of a page. It tries its best to recover, which includes
1131 * dropping pages, killing processes etc.
1133 * The function is primarily of use for corruptions that
1134 * happen outside the current execution context (e.g. when
1135 * detected by a background scrubber)
1137 * Must run in process context (e.g. a work queue) with interrupts
1138 * enabled and no spinlocks hold.
1140 int memory_failure(unsigned long pfn, int trapno, int flags)
1144 struct page *orig_head;
1146 unsigned long page_flags;
1148 if (!sysctl_memory_failure_recovery)
1149 panic("Memory failure from trap %d on page %lx", trapno, pfn);
1151 if (!pfn_valid(pfn)) {
1152 pr_err("Memory failure: %#lx: memory outside kernel control\n",
1157 p = pfn_to_page(pfn);
1159 return memory_failure_hugetlb(pfn, trapno, flags);
1160 if (TestSetPageHWPoison(p)) {
1161 pr_err("Memory failure: %#lx: already hardware poisoned\n",
1166 orig_head = hpage = compound_head(p);
1167 num_poisoned_pages_inc();
1170 * We need/can do nothing about count=0 pages.
1171 * 1) it's a free page, and therefore in safe hand:
1172 * prep_new_page() will be the gate keeper.
1173 * 2) it's part of a non-compound high order page.
1174 * Implies some kernel user: cannot stop them from
1175 * R/W the page; let's pray that the page has been
1176 * used and will be freed some time later.
1177 * In fact it's dangerous to directly bump up page count from 0,
1178 * that may make page_freeze_refs()/page_unfreeze_refs() mismatch.
1180 if (!(flags & MF_COUNT_INCREASED) && !get_hwpoison_page(p)) {
1181 if (is_free_buddy_page(p)) {
1182 action_result(pfn, MF_MSG_BUDDY, MF_DELAYED);
1185 action_result(pfn, MF_MSG_KERNEL_HIGH_ORDER, MF_IGNORED);
1190 if (PageTransHuge(hpage)) {
1192 if (!PageAnon(p) || unlikely(split_huge_page(p))) {
1195 pr_err("Memory failure: %#lx: non anonymous thp\n",
1198 pr_err("Memory failure: %#lx: thp split failed\n",
1200 if (TestClearPageHWPoison(p))
1201 num_poisoned_pages_dec();
1202 put_hwpoison_page(p);
1206 VM_BUG_ON_PAGE(!page_count(p), p);
1207 hpage = compound_head(p);
1211 * We ignore non-LRU pages for good reasons.
1212 * - PG_locked is only well defined for LRU pages and a few others
1213 * - to avoid races with __SetPageLocked()
1214 * - to avoid races with __SetPageSlab*() (and more non-atomic ops)
1215 * The check (unnecessarily) ignores LRU pages being isolated and
1216 * walked by the page reclaim code, however that's not a big loss.
1219 /* shake_page could have turned it free. */
1220 if (!PageLRU(p) && is_free_buddy_page(p)) {
1221 if (flags & MF_COUNT_INCREASED)
1222 action_result(pfn, MF_MSG_BUDDY, MF_DELAYED);
1224 action_result(pfn, MF_MSG_BUDDY_2ND, MF_DELAYED);
1231 * The page could have changed compound pages during the locking.
1232 * If this happens just bail out.
1234 if (PageCompound(p) && compound_head(p) != orig_head) {
1235 action_result(pfn, MF_MSG_DIFFERENT_COMPOUND, MF_IGNORED);
1241 * We use page flags to determine what action should be taken, but
1242 * the flags can be modified by the error containment action. One
1243 * example is an mlocked page, where PG_mlocked is cleared by
1244 * page_remove_rmap() in try_to_unmap_one(). So to determine page status
1245 * correctly, we save a copy of the page flags at this time.
1248 page_flags = hpage->flags;
1250 page_flags = p->flags;
1253 * unpoison always clear PG_hwpoison inside page lock
1255 if (!PageHWPoison(p)) {
1256 pr_err("Memory failure: %#lx: just unpoisoned\n", pfn);
1257 num_poisoned_pages_dec();
1259 put_hwpoison_page(p);
1262 if (hwpoison_filter(p)) {
1263 if (TestClearPageHWPoison(p))
1264 num_poisoned_pages_dec();
1266 put_hwpoison_page(p);
1271 * __munlock_pagevec may clear a writeback page's LRU flag without
1272 * page_lock. We need wait writeback completion for this page or it
1273 * may trigger vfs BUG while evict inode.
1275 if (!PageTransTail(p) && !PageLRU(p) && !PageWriteback(p))
1276 goto identify_page_state;
1279 * It's very difficult to mess with pages currently under IO
1280 * and in many cases impossible, so we just avoid it here.
1282 wait_on_page_writeback(p);
1285 * Now take care of user space mappings.
1286 * Abort on fail: __delete_from_page_cache() assumes unmapped page.
1288 * When the raw error page is thp tail page, hpage points to the raw
1289 * page after thp split.
1291 if (!hwpoison_user_mappings(p, pfn, trapno, flags, &hpage)) {
1292 action_result(pfn, MF_MSG_UNMAP_FAILED, MF_IGNORED);
1298 * Torn down by someone else?
1300 if (PageLRU(p) && !PageSwapCache(p) && p->mapping == NULL) {
1301 action_result(pfn, MF_MSG_TRUNCATED_LRU, MF_IGNORED);
1306 identify_page_state:
1307 res = identify_page_state(pfn, p, page_flags);
1312 EXPORT_SYMBOL_GPL(memory_failure);
1314 #define MEMORY_FAILURE_FIFO_ORDER 4
1315 #define MEMORY_FAILURE_FIFO_SIZE (1 << MEMORY_FAILURE_FIFO_ORDER)
1317 struct memory_failure_entry {
1323 struct memory_failure_cpu {
1324 DECLARE_KFIFO(fifo, struct memory_failure_entry,
1325 MEMORY_FAILURE_FIFO_SIZE);
1327 struct work_struct work;
1330 static DEFINE_PER_CPU(struct memory_failure_cpu, memory_failure_cpu);
1333 * memory_failure_queue - Schedule handling memory failure of a page.
1334 * @pfn: Page Number of the corrupted page
1335 * @trapno: Trap number reported in the signal to user space.
1336 * @flags: Flags for memory failure handling
1338 * This function is called by the low level hardware error handler
1339 * when it detects hardware memory corruption of a page. It schedules
1340 * the recovering of error page, including dropping pages, killing
1343 * The function is primarily of use for corruptions that
1344 * happen outside the current execution context (e.g. when
1345 * detected by a background scrubber)
1347 * Can run in IRQ context.
1349 void memory_failure_queue(unsigned long pfn, int trapno, int flags)
1351 struct memory_failure_cpu *mf_cpu;
1352 unsigned long proc_flags;
1353 struct memory_failure_entry entry = {
1359 mf_cpu = &get_cpu_var(memory_failure_cpu);
1360 spin_lock_irqsave(&mf_cpu->lock, proc_flags);
1361 if (kfifo_put(&mf_cpu->fifo, entry))
1362 schedule_work_on(smp_processor_id(), &mf_cpu->work);
1364 pr_err("Memory failure: buffer overflow when queuing memory failure at %#lx\n",
1366 spin_unlock_irqrestore(&mf_cpu->lock, proc_flags);
1367 put_cpu_var(memory_failure_cpu);
1369 EXPORT_SYMBOL_GPL(memory_failure_queue);
1371 static void memory_failure_work_func(struct work_struct *work)
1373 struct memory_failure_cpu *mf_cpu;
1374 struct memory_failure_entry entry = { 0, };
1375 unsigned long proc_flags;
1378 mf_cpu = this_cpu_ptr(&memory_failure_cpu);
1380 spin_lock_irqsave(&mf_cpu->lock, proc_flags);
1381 gotten = kfifo_get(&mf_cpu->fifo, &entry);
1382 spin_unlock_irqrestore(&mf_cpu->lock, proc_flags);
1385 if (entry.flags & MF_SOFT_OFFLINE)
1386 soft_offline_page(pfn_to_page(entry.pfn), entry.flags);
1388 memory_failure(entry.pfn, entry.trapno, entry.flags);
1392 static int __init memory_failure_init(void)
1394 struct memory_failure_cpu *mf_cpu;
1397 for_each_possible_cpu(cpu) {
1398 mf_cpu = &per_cpu(memory_failure_cpu, cpu);
1399 spin_lock_init(&mf_cpu->lock);
1400 INIT_KFIFO(mf_cpu->fifo);
1401 INIT_WORK(&mf_cpu->work, memory_failure_work_func);
1406 core_initcall(memory_failure_init);
1408 #define unpoison_pr_info(fmt, pfn, rs) \
1410 if (__ratelimit(rs)) \
1411 pr_info(fmt, pfn); \
1415 * unpoison_memory - Unpoison a previously poisoned page
1416 * @pfn: Page number of the to be unpoisoned page
1418 * Software-unpoison a page that has been poisoned by
1419 * memory_failure() earlier.
1421 * This is only done on the software-level, so it only works
1422 * for linux injected failures, not real hardware failures
1424 * Returns 0 for success, otherwise -errno.
1426 int unpoison_memory(unsigned long pfn)
1431 static DEFINE_RATELIMIT_STATE(unpoison_rs, DEFAULT_RATELIMIT_INTERVAL,
1432 DEFAULT_RATELIMIT_BURST);
1434 if (!pfn_valid(pfn))
1437 p = pfn_to_page(pfn);
1438 page = compound_head(p);
1440 if (!PageHWPoison(p)) {
1441 unpoison_pr_info("Unpoison: Page was already unpoisoned %#lx\n",
1446 if (page_count(page) > 1) {
1447 unpoison_pr_info("Unpoison: Someone grabs the hwpoison page %#lx\n",
1452 if (page_mapped(page)) {
1453 unpoison_pr_info("Unpoison: Someone maps the hwpoison page %#lx\n",
1458 if (page_mapping(page)) {
1459 unpoison_pr_info("Unpoison: the hwpoison page has non-NULL mapping %#lx\n",
1465 * unpoison_memory() can encounter thp only when the thp is being
1466 * worked by memory_failure() and the page lock is not held yet.
1467 * In such case, we yield to memory_failure() and make unpoison fail.
1469 if (!PageHuge(page) && PageTransHuge(page)) {
1470 unpoison_pr_info("Unpoison: Memory failure is now running on %#lx\n",
1475 if (!get_hwpoison_page(p)) {
1476 if (TestClearPageHWPoison(p))
1477 num_poisoned_pages_dec();
1478 unpoison_pr_info("Unpoison: Software-unpoisoned free page %#lx\n",
1485 * This test is racy because PG_hwpoison is set outside of page lock.
1486 * That's acceptable because that won't trigger kernel panic. Instead,
1487 * the PG_hwpoison page will be caught and isolated on the entrance to
1488 * the free buddy page pool.
1490 if (TestClearPageHWPoison(page)) {
1491 unpoison_pr_info("Unpoison: Software-unpoisoned page %#lx\n",
1493 num_poisoned_pages_dec();
1498 put_hwpoison_page(page);
1499 if (freeit && !(pfn == my_zero_pfn(0) && page_count(p) == 1))
1500 put_hwpoison_page(page);
1504 EXPORT_SYMBOL(unpoison_memory);
1506 static struct page *new_page(struct page *p, unsigned long private, int **x)
1508 int nid = page_to_nid(p);
1510 return new_page_nodemask(p, nid, &node_states[N_MEMORY]);
1514 * Safely get reference count of an arbitrary page.
1515 * Returns 0 for a free page, -EIO for a zero refcount page
1516 * that is not free, and 1 for any other page type.
1517 * For 1 the page is returned with increased page count, otherwise not.
1519 static int __get_any_page(struct page *p, unsigned long pfn, int flags)
1523 if (flags & MF_COUNT_INCREASED)
1527 * When the target page is a free hugepage, just remove it
1528 * from free hugepage list.
1530 if (!get_hwpoison_page(p)) {
1532 pr_info("%s: %#lx free huge page\n", __func__, pfn);
1534 } else if (is_free_buddy_page(p)) {
1535 pr_info("%s: %#lx free buddy page\n", __func__, pfn);
1538 pr_info("%s: %#lx: unknown zero refcount page type %lx\n",
1539 __func__, pfn, p->flags);
1543 /* Not a free page */
1549 static int get_any_page(struct page *page, unsigned long pfn, int flags)
1551 int ret = __get_any_page(page, pfn, flags);
1553 if (ret == 1 && !PageHuge(page) &&
1554 !PageLRU(page) && !__PageMovable(page)) {
1558 put_hwpoison_page(page);
1559 shake_page(page, 1);
1564 ret = __get_any_page(page, pfn, 0);
1565 if (ret == 1 && !PageLRU(page)) {
1566 /* Drop page reference which is from __get_any_page() */
1567 put_hwpoison_page(page);
1568 pr_info("soft_offline: %#lx: unknown non LRU page type %lx (%pGp)\n",
1569 pfn, page->flags, &page->flags);
1576 static int soft_offline_huge_page(struct page *page, int flags)
1579 unsigned long pfn = page_to_pfn(page);
1580 struct page *hpage = compound_head(page);
1581 LIST_HEAD(pagelist);
1584 * This double-check of PageHWPoison is to avoid the race with
1585 * memory_failure(). See also comment in __soft_offline_page().
1588 if (PageHWPoison(hpage)) {
1590 put_hwpoison_page(hpage);
1591 pr_info("soft offline: %#lx hugepage already poisoned\n", pfn);
1596 ret = isolate_huge_page(hpage, &pagelist);
1598 * get_any_page() and isolate_huge_page() takes a refcount each,
1599 * so need to drop one here.
1601 put_hwpoison_page(hpage);
1603 pr_info("soft offline: %#lx hugepage failed to isolate\n", pfn);
1607 ret = migrate_pages(&pagelist, new_page, NULL, MPOL_MF_MOVE_ALL,
1608 MIGRATE_SYNC, MR_MEMORY_FAILURE);
1610 pr_info("soft offline: %#lx: migration failed %d, type %lx (%pGp)\n",
1611 pfn, ret, page->flags, &page->flags);
1612 if (!list_empty(&pagelist))
1613 putback_movable_pages(&pagelist);
1618 dissolve_free_huge_page(page);
1623 static int __soft_offline_page(struct page *page, int flags)
1626 unsigned long pfn = page_to_pfn(page);
1629 * Check PageHWPoison again inside page lock because PageHWPoison
1630 * is set by memory_failure() outside page lock. Note that
1631 * memory_failure() also double-checks PageHWPoison inside page lock,
1632 * so there's no race between soft_offline_page() and memory_failure().
1635 wait_on_page_writeback(page);
1636 if (PageHWPoison(page)) {
1638 put_hwpoison_page(page);
1639 pr_info("soft offline: %#lx page already poisoned\n", pfn);
1643 * Try to invalidate first. This should work for
1644 * non dirty unmapped page cache pages.
1646 ret = invalidate_inode_page(page);
1649 * RED-PEN would be better to keep it isolated here, but we
1650 * would need to fix isolation locking first.
1653 put_hwpoison_page(page);
1654 pr_info("soft_offline: %#lx: invalidated\n", pfn);
1655 SetPageHWPoison(page);
1656 num_poisoned_pages_inc();
1661 * Simple invalidation didn't work.
1662 * Try to migrate to a new page instead. migrate.c
1663 * handles a large number of cases for us.
1666 ret = isolate_lru_page(page);
1668 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1670 * Drop page reference which is came from get_any_page()
1671 * successful isolate_lru_page() already took another one.
1673 put_hwpoison_page(page);
1675 LIST_HEAD(pagelist);
1677 * After isolated lru page, the PageLRU will be cleared,
1678 * so use !__PageMovable instead for LRU page's mapping
1679 * cannot have PAGE_MAPPING_MOVABLE.
1681 if (!__PageMovable(page))
1682 inc_node_page_state(page, NR_ISOLATED_ANON +
1683 page_is_file_cache(page));
1684 list_add(&page->lru, &pagelist);
1685 ret = migrate_pages(&pagelist, new_page, NULL, MPOL_MF_MOVE_ALL,
1686 MIGRATE_SYNC, MR_MEMORY_FAILURE);
1688 if (!list_empty(&pagelist))
1689 putback_movable_pages(&pagelist);
1691 pr_info("soft offline: %#lx: migration failed %d, type %lx (%pGp)\n",
1692 pfn, ret, page->flags, &page->flags);
1697 pr_info("soft offline: %#lx: isolation failed: %d, page count %d, type %lx (%pGp)\n",
1698 pfn, ret, page_count(page), page->flags, &page->flags);
1703 static int soft_offline_in_use_page(struct page *page, int flags)
1706 struct page *hpage = compound_head(page);
1708 if (!PageHuge(page) && PageTransHuge(hpage)) {
1710 if (!PageAnon(page) || unlikely(split_huge_page(page))) {
1712 if (!PageAnon(page))
1713 pr_info("soft offline: %#lx: non anonymous thp\n", page_to_pfn(page));
1715 pr_info("soft offline: %#lx: thp split failed\n", page_to_pfn(page));
1716 put_hwpoison_page(page);
1723 ret = soft_offline_huge_page(page, flags);
1725 ret = __soft_offline_page(page, flags);
1730 static void soft_offline_free_page(struct page *page)
1732 struct page *head = compound_head(page);
1734 if (!TestSetPageHWPoison(head)) {
1735 num_poisoned_pages_inc();
1737 dissolve_free_huge_page(page);
1742 * soft_offline_page - Soft offline a page.
1743 * @page: page to offline
1744 * @flags: flags. Same as memory_failure().
1746 * Returns 0 on success, otherwise negated errno.
1748 * Soft offline a page, by migration or invalidation,
1749 * without killing anything. This is for the case when
1750 * a page is not corrupted yet (so it's still valid to access),
1751 * but has had a number of corrected errors and is better taken
1754 * The actual policy on when to do that is maintained by
1757 * This should never impact any application or cause data loss,
1758 * however it might take some time.
1760 * This is not a 100% solution for all memory, but tries to be
1761 * ``good enough'' for the majority of memory.
1763 int soft_offline_page(struct page *page, int flags)
1766 unsigned long pfn = page_to_pfn(page);
1768 if (PageHWPoison(page)) {
1769 pr_info("soft offline: %#lx page already poisoned\n", pfn);
1770 if (flags & MF_COUNT_INCREASED)
1771 put_hwpoison_page(page);
1776 ret = get_any_page(page, pfn, flags);
1780 ret = soft_offline_in_use_page(page, flags);
1782 soft_offline_free_page(page);