mm-only debug patch...
[mmotm.git] / mm / memory-failure.c
blob7fc2130d2737f034d0fc08995286ae5944c8b5e4
1 /*
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 2bit ECC memory or cache
11 * failure.
13 * Handles page cache pages in various states. The tricky part
14 * here is that we can access any page asynchronous to other VM
15 * users, because memory failures could happen anytime and anywhere,
16 * possibly violating some of their assumptions. This is why this code
17 * has to be extremely careful. Generally it tries to use normal locking
18 * rules, as in get the standard locks, even if that means the
19 * error handling takes potentially a long time.
21 * The operation to map back from RMAP chains to processes has to walk
22 * the complete process list and has non linear complexity with the number
23 * mappings. In short it can be quite slow. But since memory corruptions
24 * are rare we hope to get away with this.
28 * Notebook:
29 * - hugetlb needs more code
30 * - kcore/oldmem/vmcore/mem/kmem check for hwpoison pages
31 * - pass bad pages to kdump next kernel
33 #define DEBUG 1 /* remove me in 2.6.34 */
34 #include <linux/kernel.h>
35 #include <linux/mm.h>
36 #include <linux/page-flags.h>
37 #include <linux/sched.h>
38 #include <linux/ksm.h>
39 #include <linux/rmap.h>
40 #include <linux/pagemap.h>
41 #include <linux/swap.h>
42 #include <linux/backing-dev.h>
43 #include "internal.h"
45 int sysctl_memory_failure_early_kill __read_mostly = 0;
47 int sysctl_memory_failure_recovery __read_mostly = 1;
49 atomic_long_t mce_bad_pages __read_mostly = ATOMIC_LONG_INIT(0);
52 * Send all the processes who have the page mapped an ``action optional''
53 * signal.
55 static int kill_proc_ao(struct task_struct *t, unsigned long addr, int trapno,
56 unsigned long pfn)
58 struct siginfo si;
59 int ret;
61 printk(KERN_ERR
62 "MCE %#lx: Killing %s:%d early due to hardware memory corruption\n",
63 pfn, t->comm, t->pid);
64 si.si_signo = SIGBUS;
65 si.si_errno = 0;
66 si.si_code = BUS_MCEERR_AO;
67 si.si_addr = (void *)addr;
68 #ifdef __ARCH_SI_TRAPNO
69 si.si_trapno = trapno;
70 #endif
71 si.si_addr_lsb = PAGE_SHIFT;
73 * Don't use force here, it's convenient if the signal
74 * can be temporarily blocked.
75 * This could cause a loop when the user sets SIGBUS
76 * to SIG_IGN, but hopefully noone will do that?
78 ret = send_sig_info(SIGBUS, &si, t); /* synchronous? */
79 if (ret < 0)
80 printk(KERN_INFO "MCE: Error sending signal to %s:%d: %d\n",
81 t->comm, t->pid, ret);
82 return ret;
86 * Kill all processes that have a poisoned page mapped and then isolate
87 * the page.
89 * General strategy:
90 * Find all processes having the page mapped and kill them.
91 * But we keep a page reference around so that the page is not
92 * actually freed yet.
93 * Then stash the page away
95 * There's no convenient way to get back to mapped processes
96 * from the VMAs. So do a brute-force search over all
97 * running processes.
99 * Remember that machine checks are not common (or rather
100 * if they are common you have other problems), so this shouldn't
101 * be a performance issue.
103 * Also there are some races possible while we get from the
104 * error detection to actually handle it.
107 struct to_kill {
108 struct list_head nd;
109 struct task_struct *tsk;
110 unsigned long addr;
111 unsigned addr_valid:1;
115 * Failure handling: if we can't find or can't kill a process there's
116 * not much we can do. We just print a message and ignore otherwise.
120 * Schedule a process for later kill.
121 * Uses GFP_ATOMIC allocations to avoid potential recursions in the VM.
122 * TBD would GFP_NOIO be enough?
124 static void add_to_kill(struct task_struct *tsk, struct page *p,
125 struct vm_area_struct *vma,
126 struct list_head *to_kill,
127 struct to_kill **tkc)
129 struct to_kill *tk;
131 if (*tkc) {
132 tk = *tkc;
133 *tkc = NULL;
134 } else {
135 tk = kmalloc(sizeof(struct to_kill), GFP_ATOMIC);
136 if (!tk) {
137 printk(KERN_ERR
138 "MCE: Out of memory while machine check handling\n");
139 return;
142 tk->addr = page_address_in_vma(p, vma);
143 tk->addr_valid = 1;
146 * In theory we don't have to kill when the page was
147 * munmaped. But it could be also a mremap. Since that's
148 * likely very rare kill anyways just out of paranoia, but use
149 * a SIGKILL because the error is not contained anymore.
151 if (tk->addr == -EFAULT) {
152 pr_debug("MCE: Unable to find user space address %lx in %s\n",
153 page_to_pfn(p), tsk->comm);
154 tk->addr_valid = 0;
156 get_task_struct(tsk);
157 tk->tsk = tsk;
158 list_add_tail(&tk->nd, to_kill);
162 * Kill the processes that have been collected earlier.
164 * Only do anything when DOIT is set, otherwise just free the list
165 * (this is used for clean pages which do not need killing)
166 * Also when FAIL is set do a force kill because something went
167 * wrong earlier.
169 static void kill_procs_ao(struct list_head *to_kill, int doit, int trapno,
170 int fail, unsigned long pfn)
172 struct to_kill *tk, *next;
174 list_for_each_entry_safe (tk, next, to_kill, nd) {
175 if (doit) {
177 * In case something went wrong with munmaping
178 * make sure the process doesn't catch the
179 * signal and then access the memory. Just kill it.
180 * the signal handlers
182 if (fail || tk->addr_valid == 0) {
183 printk(KERN_ERR
184 "MCE %#lx: forcibly killing %s:%d because of failure to unmap corrupted page\n",
185 pfn, tk->tsk->comm, tk->tsk->pid);
186 force_sig(SIGKILL, tk->tsk);
190 * In theory the process could have mapped
191 * something else on the address in-between. We could
192 * check for that, but we need to tell the
193 * process anyways.
195 else if (kill_proc_ao(tk->tsk, tk->addr, trapno,
196 pfn) < 0)
197 printk(KERN_ERR
198 "MCE %#lx: Cannot send advisory machine check signal to %s:%d\n",
199 pfn, tk->tsk->comm, tk->tsk->pid);
201 put_task_struct(tk->tsk);
202 kfree(tk);
206 static int task_early_kill(struct task_struct *tsk)
208 if (!tsk->mm)
209 return 0;
210 if (tsk->flags & PF_MCE_PROCESS)
211 return !!(tsk->flags & PF_MCE_EARLY);
212 return sysctl_memory_failure_early_kill;
216 * Collect processes when the error hit an anonymous page.
218 static void collect_procs_anon(struct page *page, struct list_head *to_kill,
219 struct to_kill **tkc)
221 struct vm_area_struct *vma;
222 struct task_struct *tsk;
223 struct anon_vma *av;
225 read_lock(&tasklist_lock);
226 av = page_lock_anon_vma(page);
227 if (av == NULL) /* Not actually mapped anymore */
228 goto out;
229 for_each_process (tsk) {
230 if (!task_early_kill(tsk))
231 continue;
232 list_for_each_entry (vma, &av->head, anon_vma_node) {
233 if (!page_mapped_in_vma(page, vma))
234 continue;
235 if (vma->vm_mm == tsk->mm)
236 add_to_kill(tsk, page, vma, to_kill, tkc);
239 page_unlock_anon_vma(av);
240 out:
241 read_unlock(&tasklist_lock);
245 * Collect processes when the error hit a file mapped page.
247 static void collect_procs_file(struct page *page, struct list_head *to_kill,
248 struct to_kill **tkc)
250 struct vm_area_struct *vma;
251 struct task_struct *tsk;
252 struct prio_tree_iter iter;
253 struct address_space *mapping = page->mapping;
256 * A note on the locking order between the two locks.
257 * We don't rely on this particular order.
258 * If you have some other code that needs a different order
259 * feel free to switch them around. Or add a reverse link
260 * from mm_struct to task_struct, then this could be all
261 * done without taking tasklist_lock and looping over all tasks.
264 read_lock(&tasklist_lock);
265 spin_lock(&mapping->i_mmap_lock);
266 for_each_process(tsk) {
267 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
269 if (!task_early_kill(tsk))
270 continue;
272 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff,
273 pgoff) {
275 * Send early kill signal to tasks where a vma covers
276 * the page but the corrupted page is not necessarily
277 * mapped it in its pte.
278 * Assume applications who requested early kill want
279 * to be informed of all such data corruptions.
281 if (vma->vm_mm == tsk->mm)
282 add_to_kill(tsk, page, vma, to_kill, tkc);
285 spin_unlock(&mapping->i_mmap_lock);
286 read_unlock(&tasklist_lock);
290 * Collect the processes who have the corrupted page mapped to kill.
291 * This is done in two steps for locking reasons.
292 * First preallocate one tokill structure outside the spin locks,
293 * so that we can kill at least one process reasonably reliable.
295 static void collect_procs(struct page *page, struct list_head *tokill)
297 struct to_kill *tk;
299 if (!page->mapping)
300 return;
302 tk = kmalloc(sizeof(struct to_kill), GFP_NOIO);
303 if (!tk)
304 return;
305 if (PageAnon(page))
306 collect_procs_anon(page, tokill, &tk);
307 else
308 collect_procs_file(page, tokill, &tk);
309 kfree(tk);
313 * Error handlers for various types of pages.
316 enum outcome {
317 FAILED, /* Error handling failed */
318 DELAYED, /* Will be handled later */
319 IGNORED, /* Error safely ignored */
320 RECOVERED, /* Successfully recovered */
323 static const char *action_name[] = {
324 [FAILED] = "Failed",
325 [DELAYED] = "Delayed",
326 [IGNORED] = "Ignored",
327 [RECOVERED] = "Recovered",
331 * Error hit kernel page.
332 * Do nothing, try to be lucky and not touch this instead. For a few cases we
333 * could be more sophisticated.
335 static int me_kernel(struct page *p, unsigned long pfn)
337 return DELAYED;
341 * Already poisoned page.
343 static int me_ignore(struct page *p, unsigned long pfn)
345 return IGNORED;
349 * Page in unknown state. Do nothing.
351 static int me_unknown(struct page *p, unsigned long pfn)
353 printk(KERN_ERR "MCE %#lx: Unknown page state\n", pfn);
354 return FAILED;
358 * Free memory
360 static int me_free(struct page *p, unsigned long pfn)
362 return DELAYED;
366 * Clean (or cleaned) page cache page.
368 static int me_pagecache_clean(struct page *p, unsigned long pfn)
370 int err;
371 int ret = FAILED;
372 struct address_space *mapping;
374 if (!isolate_lru_page(p))
375 page_cache_release(p);
378 * For anonymous pages we're done the only reference left
379 * should be the one m_f() holds.
381 if (PageAnon(p))
382 return RECOVERED;
385 * Now truncate the page in the page cache. This is really
386 * more like a "temporary hole punch"
387 * Don't do this for block devices when someone else
388 * has a reference, because it could be file system metadata
389 * and that's not safe to truncate.
391 mapping = page_mapping(p);
392 if (!mapping) {
394 * Page has been teared down in the meanwhile
396 return FAILED;
400 * Truncation is a bit tricky. Enable it per file system for now.
402 * Open: to take i_mutex or not for this? Right now we don't.
404 if (mapping->a_ops->error_remove_page) {
405 err = mapping->a_ops->error_remove_page(mapping, p);
406 if (err != 0) {
407 printk(KERN_INFO "MCE %#lx: Failed to punch page: %d\n",
408 pfn, err);
409 } else if (page_has_private(p) &&
410 !try_to_release_page(p, GFP_NOIO)) {
411 pr_debug("MCE %#lx: failed to release buffers\n", pfn);
412 } else {
413 ret = RECOVERED;
415 } else {
417 * If the file system doesn't support it just invalidate
418 * This fails on dirty or anything with private pages
420 if (invalidate_inode_page(p))
421 ret = RECOVERED;
422 else
423 printk(KERN_INFO "MCE %#lx: Failed to invalidate\n",
424 pfn);
426 return ret;
430 * Dirty cache page page
431 * Issues: when the error hit a hole page the error is not properly
432 * propagated.
434 static int me_pagecache_dirty(struct page *p, unsigned long pfn)
436 struct address_space *mapping = page_mapping(p);
438 SetPageError(p);
439 /* TBD: print more information about the file. */
440 if (mapping) {
442 * IO error will be reported by write(), fsync(), etc.
443 * who check the mapping.
444 * This way the application knows that something went
445 * wrong with its dirty file data.
447 * There's one open issue:
449 * The EIO will be only reported on the next IO
450 * operation and then cleared through the IO map.
451 * Normally Linux has two mechanisms to pass IO error
452 * first through the AS_EIO flag in the address space
453 * and then through the PageError flag in the page.
454 * Since we drop pages on memory failure handling the
455 * only mechanism open to use is through AS_AIO.
457 * This has the disadvantage that it gets cleared on
458 * the first operation that returns an error, while
459 * the PageError bit is more sticky and only cleared
460 * when the page is reread or dropped. If an
461 * application assumes it will always get error on
462 * fsync, but does other operations on the fd before
463 * and the page is dropped inbetween then the error
464 * will not be properly reported.
466 * This can already happen even without hwpoisoned
467 * pages: first on metadata IO errors (which only
468 * report through AS_EIO) or when the page is dropped
469 * at the wrong time.
471 * So right now we assume that the application DTRT on
472 * the first EIO, but we're not worse than other parts
473 * of the kernel.
475 mapping_set_error(mapping, EIO);
478 return me_pagecache_clean(p, pfn);
482 * Clean and dirty swap cache.
484 * Dirty swap cache page is tricky to handle. The page could live both in page
485 * cache and swap cache(ie. page is freshly swapped in). So it could be
486 * referenced concurrently by 2 types of PTEs:
487 * normal PTEs and swap PTEs. We try to handle them consistently by calling
488 * try_to_unmap(TTU_IGNORE_HWPOISON) to convert the normal PTEs to swap PTEs,
489 * and then
490 * - clear dirty bit to prevent IO
491 * - remove from LRU
492 * - but keep in the swap cache, so that when we return to it on
493 * a later page fault, we know the application is accessing
494 * corrupted data and shall be killed (we installed simple
495 * interception code in do_swap_page to catch it).
497 * Clean swap cache pages can be directly isolated. A later page fault will
498 * bring in the known good data from disk.
500 static int me_swapcache_dirty(struct page *p, unsigned long pfn)
502 int ret = FAILED;
504 ClearPageDirty(p);
505 /* Trigger EIO in shmem: */
506 ClearPageUptodate(p);
508 if (!isolate_lru_page(p)) {
509 page_cache_release(p);
510 ret = DELAYED;
513 return ret;
516 static int me_swapcache_clean(struct page *p, unsigned long pfn)
518 int ret = FAILED;
520 if (!isolate_lru_page(p)) {
521 page_cache_release(p);
522 ret = RECOVERED;
524 delete_from_swap_cache(p);
525 return ret;
529 * Huge pages. Needs work.
530 * Issues:
531 * No rmap support so we cannot find the original mapper. In theory could walk
532 * all MMs and look for the mappings, but that would be non atomic and racy.
533 * Need rmap for hugepages for this. Alternatively we could employ a heuristic,
534 * like just walking the current process and hoping it has it mapped (that
535 * should be usually true for the common "shared database cache" case)
536 * Should handle free huge pages and dequeue them too, but this needs to
537 * handle huge page accounting correctly.
539 static int me_huge_page(struct page *p, unsigned long pfn)
541 return FAILED;
545 * Various page states we can handle.
547 * A page state is defined by its current page->flags bits.
548 * The table matches them in order and calls the right handler.
550 * This is quite tricky because we can access page at any time
551 * in its live cycle, so all accesses have to be extremly careful.
553 * This is not complete. More states could be added.
554 * For any missing state don't attempt recovery.
557 #define dirty (1UL << PG_dirty)
558 #define sc (1UL << PG_swapcache)
559 #define unevict (1UL << PG_unevictable)
560 #define mlock (1UL << PG_mlocked)
561 #define writeback (1UL << PG_writeback)
562 #define lru (1UL << PG_lru)
563 #define swapbacked (1UL << PG_swapbacked)
564 #define head (1UL << PG_head)
565 #define tail (1UL << PG_tail)
566 #define compound (1UL << PG_compound)
567 #define slab (1UL << PG_slab)
568 #define buddy (1UL << PG_buddy)
569 #define reserved (1UL << PG_reserved)
571 static struct page_state {
572 unsigned long mask;
573 unsigned long res;
574 char *msg;
575 int (*action)(struct page *p, unsigned long pfn);
576 } error_states[] = {
577 { reserved, reserved, "reserved kernel", me_ignore },
578 { buddy, buddy, "free kernel", me_free },
581 * Could in theory check if slab page is free or if we can drop
582 * currently unused objects without touching them. But just
583 * treat it as standard kernel for now.
585 { slab, slab, "kernel slab", me_kernel },
587 #ifdef CONFIG_PAGEFLAGS_EXTENDED
588 { head, head, "huge", me_huge_page },
589 { tail, tail, "huge", me_huge_page },
590 #else
591 { compound, compound, "huge", me_huge_page },
592 #endif
594 { sc|dirty, sc|dirty, "swapcache", me_swapcache_dirty },
595 { sc|dirty, sc, "swapcache", me_swapcache_clean },
597 { unevict|dirty, unevict|dirty, "unevictable LRU", me_pagecache_dirty},
598 { unevict, unevict, "unevictable LRU", me_pagecache_clean},
600 #ifdef CONFIG_HAVE_MLOCKED_PAGE_BIT
601 { mlock|dirty, mlock|dirty, "mlocked LRU", me_pagecache_dirty },
602 { mlock, mlock, "mlocked LRU", me_pagecache_clean },
603 #endif
605 { lru|dirty, lru|dirty, "LRU", me_pagecache_dirty },
606 { lru|dirty, lru, "clean LRU", me_pagecache_clean },
607 { swapbacked, swapbacked, "anonymous", me_pagecache_clean },
610 * Catchall entry: must be at end.
612 { 0, 0, "unknown page state", me_unknown },
615 #undef lru
617 static void action_result(unsigned long pfn, char *msg, int result)
619 struct page *page = NULL;
620 if (pfn_valid(pfn))
621 page = pfn_to_page(pfn);
623 printk(KERN_ERR "MCE %#lx: %s%s page recovery: %s\n",
624 pfn,
625 page && PageDirty(page) ? "dirty " : "",
626 msg, action_name[result]);
629 static int page_action(struct page_state *ps, struct page *p,
630 unsigned long pfn, int ref)
632 int result;
634 result = ps->action(p, pfn);
635 action_result(pfn, ps->msg, result);
636 if (page_count(p) != 1 + ref)
637 printk(KERN_ERR
638 "MCE %#lx: %s page still referenced by %d users\n",
639 pfn, ps->msg, page_count(p) - 1);
641 /* Could do more checks here if page looks ok */
643 * Could adjust zone counters here to correct for the missing page.
646 return result == RECOVERED ? 0 : -EBUSY;
649 #define N_UNMAP_TRIES 5
652 * Do all that is necessary to remove user space mappings. Unmap
653 * the pages and send SIGBUS to the processes if the data was dirty.
655 static void hwpoison_user_mappings(struct page *p, unsigned long pfn,
656 int trapno)
658 enum ttu_flags ttu = TTU_UNMAP | TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS;
659 struct address_space *mapping;
660 LIST_HEAD(tokill);
661 int ret;
662 int i;
663 int kill = 1;
665 if (PageReserved(p) || PageCompound(p) || PageSlab(p) || PageKsm(p))
666 return;
668 if (!PageLRU(p))
669 lru_add_drain_all();
672 * This check implies we don't kill processes if their pages
673 * are in the swap cache early. Those are always late kills.
675 if (!page_mapped(p))
676 return;
678 if (PageSwapCache(p)) {
679 printk(KERN_ERR
680 "MCE %#lx: keeping poisoned page in swap cache\n", pfn);
681 ttu |= TTU_IGNORE_HWPOISON;
685 * Propagate the dirty bit from PTEs to struct page first, because we
686 * need this to decide if we should kill or just drop the page.
688 mapping = page_mapping(p);
689 if (!PageDirty(p) && mapping && mapping_cap_writeback_dirty(mapping)) {
690 if (page_mkclean(p)) {
691 SetPageDirty(p);
692 } else {
693 kill = 0;
694 ttu |= TTU_IGNORE_HWPOISON;
695 printk(KERN_INFO
696 "MCE %#lx: corrupted page was clean: dropped without side effects\n",
697 pfn);
702 * First collect all the processes that have the page
703 * mapped in dirty form. This has to be done before try_to_unmap,
704 * because ttu takes the rmap data structures down.
706 * Error handling: We ignore errors here because
707 * there's nothing that can be done.
709 if (kill)
710 collect_procs(p, &tokill);
713 * try_to_unmap can fail temporarily due to races.
714 * Try a few times (RED-PEN better strategy?)
716 for (i = 0; i < N_UNMAP_TRIES; i++) {
717 ret = try_to_unmap(p, ttu);
718 if (ret == SWAP_SUCCESS)
719 break;
720 pr_debug("MCE %#lx: try_to_unmap retry needed %d\n", pfn, ret);
723 if (ret != SWAP_SUCCESS)
724 printk(KERN_ERR "MCE %#lx: failed to unmap page (mapcount=%d)\n",
725 pfn, page_mapcount(p));
728 * Now that the dirty bit has been propagated to the
729 * struct page and all unmaps done we can decide if
730 * killing is needed or not. Only kill when the page
731 * was dirty, otherwise the tokill list is merely
732 * freed. When there was a problem unmapping earlier
733 * use a more force-full uncatchable kill to prevent
734 * any accesses to the poisoned memory.
736 kill_procs_ao(&tokill, !!PageDirty(p), trapno,
737 ret != SWAP_SUCCESS, pfn);
740 int __memory_failure(unsigned long pfn, int trapno, int ref)
742 struct page_state *ps;
743 struct page *p;
744 int res;
746 if (!sysctl_memory_failure_recovery)
747 panic("Memory failure from trap %d on page %lx", trapno, pfn);
749 if (!pfn_valid(pfn)) {
750 action_result(pfn, "memory outside kernel control", IGNORED);
751 return -EIO;
754 p = pfn_to_page(pfn);
755 if (TestSetPageHWPoison(p)) {
756 action_result(pfn, "already hardware poisoned", IGNORED);
757 return 0;
760 atomic_long_add(1, &mce_bad_pages);
763 * We need/can do nothing about count=0 pages.
764 * 1) it's a free page, and therefore in safe hand:
765 * prep_new_page() will be the gate keeper.
766 * 2) it's part of a non-compound high order page.
767 * Implies some kernel user: cannot stop them from
768 * R/W the page; let's pray that the page has been
769 * used and will be freed some time later.
770 * In fact it's dangerous to directly bump up page count from 0,
771 * that may make page_freeze_refs()/page_unfreeze_refs() mismatch.
773 if (!get_page_unless_zero(compound_head(p))) {
774 action_result(pfn, "free or high order kernel", IGNORED);
775 return PageBuddy(compound_head(p)) ? 0 : -EBUSY;
779 * Lock the page and wait for writeback to finish.
780 * It's very difficult to mess with pages currently under IO
781 * and in many cases impossible, so we just avoid it here.
783 lock_page_nosync(p);
784 wait_on_page_writeback(p);
787 * Now take care of user space mappings.
789 hwpoison_user_mappings(p, pfn, trapno);
792 * Torn down by someone else?
794 if (PageLRU(p) && !PageSwapCache(p) && p->mapping == NULL) {
795 action_result(pfn, "already truncated LRU", IGNORED);
796 res = 0;
797 goto out;
800 res = -EBUSY;
801 for (ps = error_states;; ps++) {
802 if ((p->flags & ps->mask) == ps->res) {
803 res = page_action(ps, p, pfn, ref);
804 break;
807 out:
808 unlock_page(p);
809 return res;
811 EXPORT_SYMBOL_GPL(__memory_failure);
814 * memory_failure - Handle memory failure of a page.
815 * @pfn: Page Number of the corrupted page
816 * @trapno: Trap number reported in the signal to user space.
818 * This function is called by the low level machine check code
819 * of an architecture when it detects hardware memory corruption
820 * of a page. It tries its best to recover, which includes
821 * dropping pages, killing processes etc.
823 * The function is primarily of use for corruptions that
824 * happen outside the current execution context (e.g. when
825 * detected by a background scrubber)
827 * Must run in process context (e.g. a work queue) with interrupts
828 * enabled and no spinlocks hold.
830 void memory_failure(unsigned long pfn, int trapno)
832 __memory_failure(pfn, trapno, 0);