memory-failure.c 46.4 KB
Newer Older
1 2 3 4 5 6 7 8 9
/*
 * Copyright (C) 2008, 2009 Intel Corporation
 * Authors: Andi Kleen, Fengguang Wu
 *
 * This software may be redistributed and/or modified under the terms of
 * the GNU General Public License ("GPL") version 2 only as published by the
 * Free Software Foundation.
 *
 * 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
11
 * failure.
12 13 14
 * 
 * In addition there is a "soft offline" entry point that allows stop using
 * not-yet-corrupted-by-suspicious pages without killing anything.
15 16
 *
 * Handles page cache pages in various states.	The tricky part
17 18 19 20 21 22 23 24 25 26 27 28 29
 * here is that we can access any page asynchronously in respect to 
 * other VM users, because memory failures could happen anytime and 
 * anywhere. This could violate some of their assumptions. This is why 
 * this code has to be extremely careful. Generally it tries to use 
 * normal locking rules, as in get the standard locks, even if that means 
 * the error handling takes potentially a long time.
 * 
 * There are several operations here with exponential complexity because
 * of unsuitable VM data structures. For example the operation to map back 
 * from RMAP chains to processes has to walk the complete process list and 
 * has non linear complexity with the number. But since memory corruptions
 * are rare we hope to get away with this. This avoids impacting the core 
 * VM.
30 31 32 33 34 35 36 37 38 39 40
 */

/*
 * Notebook:
 * - hugetlb needs more code
 * - kcore/oldmem/vmcore/mem/kmem check for hwpoison pages
 * - pass bad pages to kdump next kernel
 */
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/page-flags.h>
W
Wu Fengguang 已提交
41
#include <linux/kernel-page-flags.h>
42
#include <linux/sched.h>
H
Hugh Dickins 已提交
43
#include <linux/ksm.h>
44
#include <linux/rmap.h>
45
#include <linux/export.h>
46 47 48
#include <linux/pagemap.h>
#include <linux/swap.h>
#include <linux/backing-dev.h>
49 50 51
#include <linux/migrate.h>
#include <linux/page-isolation.h>
#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>
58 59 60 61 62 63
#include "internal.h"

int sysctl_memory_failure_early_kill __read_mostly = 0;

int sysctl_memory_failure_recovery __read_mostly = 1;

64
atomic_long_t num_poisoned_pages __read_mostly = ATOMIC_LONG_INIT(0);
65

66 67
#if defined(CONFIG_HWPOISON_INJECT) || defined(CONFIG_HWPOISON_INJECT_MODULE)

68
u32 hwpoison_filter_enable = 0;
W
Wu Fengguang 已提交
69 70
u32 hwpoison_filter_dev_major = ~0U;
u32 hwpoison_filter_dev_minor = ~0U;
W
Wu Fengguang 已提交
71 72
u64 hwpoison_filter_flags_mask;
u64 hwpoison_filter_flags_value;
73
EXPORT_SYMBOL_GPL(hwpoison_filter_enable);
W
Wu Fengguang 已提交
74 75
EXPORT_SYMBOL_GPL(hwpoison_filter_dev_major);
EXPORT_SYMBOL_GPL(hwpoison_filter_dev_minor);
W
Wu Fengguang 已提交
76 77
EXPORT_SYMBOL_GPL(hwpoison_filter_flags_mask);
EXPORT_SYMBOL_GPL(hwpoison_filter_flags_value);
W
Wu Fengguang 已提交
78 79 80 81 82 83 84 85 86 87 88

static int hwpoison_filter_dev(struct page *p)
{
	struct address_space *mapping;
	dev_t dev;

	if (hwpoison_filter_dev_major == ~0U &&
	    hwpoison_filter_dev_minor == ~0U)
		return 0;

	/*
89
	 * page_mapping() does not accept slab pages.
W
Wu Fengguang 已提交
90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108
	 */
	if (PageSlab(p))
		return -EINVAL;

	mapping = page_mapping(p);
	if (mapping == NULL || mapping->host == NULL)
		return -EINVAL;

	dev = mapping->host->i_sb->s_dev;
	if (hwpoison_filter_dev_major != ~0U &&
	    hwpoison_filter_dev_major != MAJOR(dev))
		return -EINVAL;
	if (hwpoison_filter_dev_minor != ~0U &&
	    hwpoison_filter_dev_minor != MINOR(dev))
		return -EINVAL;

	return 0;
}

W
Wu Fengguang 已提交
109 110 111 112 113 114 115 116 117 118 119 120
static int hwpoison_filter_flags(struct page *p)
{
	if (!hwpoison_filter_flags_mask)
		return 0;

	if ((stable_page_flags(p) & hwpoison_filter_flags_mask) ==
				    hwpoison_filter_flags_value)
		return 0;
	else
		return -EINVAL;
}

A
Andi Kleen 已提交
121 122 123 124 125 126 127 128 129 130
/*
 * This allows stress tests to limit test scope to a collection of tasks
 * by putting them under some memcg. This prevents killing unrelated/important
 * processes such as /sbin/init. Note that the target task may share clean
 * pages with init (eg. libc text), which is harmless. If the target task
 * share _dirty_ pages with another task B, the test scheme must make sure B
 * is also included in the memcg. At last, due to race conditions this filter
 * can only guarantee that the page either belongs to the memcg tasks, or is
 * a freed page.
 */
A
Andrew Morton 已提交
131
#ifdef	CONFIG_MEMCG_SWAP
A
Andi Kleen 已提交
132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163
u64 hwpoison_filter_memcg;
EXPORT_SYMBOL_GPL(hwpoison_filter_memcg);
static int hwpoison_filter_task(struct page *p)
{
	struct mem_cgroup *mem;
	struct cgroup_subsys_state *css;
	unsigned long ino;

	if (!hwpoison_filter_memcg)
		return 0;

	mem = try_get_mem_cgroup_from_page(p);
	if (!mem)
		return -EINVAL;

	css = mem_cgroup_css(mem);
	/* root_mem_cgroup has NULL dentries */
	if (!css->cgroup->dentry)
		return -EINVAL;

	ino = css->cgroup->dentry->d_inode->i_ino;
	css_put(css);

	if (ino != hwpoison_filter_memcg)
		return -EINVAL;

	return 0;
}
#else
static int hwpoison_filter_task(struct page *p) { return 0; }
#endif

W
Wu Fengguang 已提交
164 165
int hwpoison_filter(struct page *p)
{
166 167 168
	if (!hwpoison_filter_enable)
		return 0;

W
Wu Fengguang 已提交
169 170 171
	if (hwpoison_filter_dev(p))
		return -EINVAL;

W
Wu Fengguang 已提交
172 173 174
	if (hwpoison_filter_flags(p))
		return -EINVAL;

A
Andi Kleen 已提交
175 176 177
	if (hwpoison_filter_task(p))
		return -EINVAL;

W
Wu Fengguang 已提交
178 179
	return 0;
}
180 181 182 183 184 185 186
#else
int hwpoison_filter(struct page *p)
{
	return 0;
}
#endif

W
Wu Fengguang 已提交
187 188
EXPORT_SYMBOL_GPL(hwpoison_filter);

189
/*
190 191 192
 * Send all the processes who have the page mapped a signal.
 * ``action optional'' if they are not immediately affected by the error
 * ``action required'' if error happened in current execution context
193
 */
194 195
static int kill_proc(struct task_struct *t, unsigned long addr, int trapno,
			unsigned long pfn, struct page *page, int flags)
196 197 198 199 200
{
	struct siginfo si;
	int ret;

	printk(KERN_ERR
201
		"MCE %#lx: Killing %s:%d due to hardware memory corruption\n",
202 203 204 205 206 207 208
		pfn, t->comm, t->pid);
	si.si_signo = SIGBUS;
	si.si_errno = 0;
	si.si_addr = (void *)addr;
#ifdef __ARCH_SI_TRAPNO
	si.si_trapno = trapno;
#endif
209
	si.si_addr_lsb = compound_order(compound_head(page)) + PAGE_SHIFT;
210 211 212 213 214 215 216 217 218 219 220 221 222 223

	if ((flags & MF_ACTION_REQUIRED) && t == current) {
		si.si_code = BUS_MCEERR_AR;
		ret = force_sig_info(SIGBUS, &si, t);
	} else {
		/*
		 * Don't use force here, it's convenient if the signal
		 * can be temporarily blocked.
		 * This could cause a loop when the user sets SIGBUS
		 * to SIG_IGN, but hopefully no one will do that?
		 */
		si.si_code = BUS_MCEERR_AO;
		ret = send_sig_info(SIGBUS, &si, t);  /* synchronous? */
	}
224 225 226 227 228 229
	if (ret < 0)
		printk(KERN_INFO "MCE: Error sending signal to %s:%d: %d\n",
		       t->comm, t->pid, ret);
	return ret;
}

230 231 232 233
/*
 * When a unknown page type is encountered drain as many buffers as possible
 * in the hope to turn the page into a LRU or free page, which we can handle.
 */
234
void shake_page(struct page *p, int access)
235 236 237 238 239 240 241 242 243
{
	if (!PageSlab(p)) {
		lru_add_drain_all();
		if (PageLRU(p))
			return;
		drain_all_pages();
		if (PageLRU(p) || is_free_buddy_page(p))
			return;
	}
244

245
	/*
246 247
	 * Only call shrink_slab here (which would also shrink other caches) if
	 * access is not potentially fatal.
248
	 */
249 250
	if (access) {
		int nr;
D
Dave Chinner 已提交
251
		int nid = page_to_nid(p);
252
		do {
253 254 255
			struct shrink_control shrink = {
				.gfp_mask = GFP_KERNEL,
			};
D
Dave Chinner 已提交
256
			node_set(nid, shrink.nodes_to_scan);
257

258
			nr = shrink_slab(&shrink, 1000, 1000);
259
			if (page_count(p) == 1)
260 261 262
				break;
		} while (nr > 10);
	}
263 264 265
}
EXPORT_SYMBOL_GPL(shake_page);

266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291
/*
 * Kill all processes that have a poisoned page mapped and then isolate
 * the page.
 *
 * General strategy:
 * Find all processes having the page mapped and kill them.
 * But we keep a page reference around so that the page is not
 * actually freed yet.
 * Then stash the page away
 *
 * There's no convenient way to get back to mapped processes
 * from the VMAs. So do a brute-force search over all
 * running processes.
 *
 * Remember that machine checks are not common (or rather
 * if they are common you have other problems), so this shouldn't
 * be a performance issue.
 *
 * Also there are some races possible while we get from the
 * error detection to actually handle it.
 */

struct to_kill {
	struct list_head nd;
	struct task_struct *tsk;
	unsigned long addr;
292
	char addr_valid;
293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332
};

/*
 * Failure handling: if we can't find or can't kill a process there's
 * not much we can do.	We just print a message and ignore otherwise.
 */

/*
 * Schedule a process for later kill.
 * Uses GFP_ATOMIC allocations to avoid potential recursions in the VM.
 * TBD would GFP_NOIO be enough?
 */
static void add_to_kill(struct task_struct *tsk, struct page *p,
		       struct vm_area_struct *vma,
		       struct list_head *to_kill,
		       struct to_kill **tkc)
{
	struct to_kill *tk;

	if (*tkc) {
		tk = *tkc;
		*tkc = NULL;
	} else {
		tk = kmalloc(sizeof(struct to_kill), GFP_ATOMIC);
		if (!tk) {
			printk(KERN_ERR
		"MCE: Out of memory while machine check handling\n");
			return;
		}
	}
	tk->addr = page_address_in_vma(p, vma);
	tk->addr_valid = 1;

	/*
	 * In theory we don't have to kill when the page was
	 * munmaped. But it could be also a mremap. Since that's
	 * likely very rare kill anyways just out of paranoia, but use
	 * a SIGKILL because the error is not contained anymore.
	 */
	if (tk->addr == -EFAULT) {
333
		pr_info("MCE: Unable to find user space address %lx in %s\n",
334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349
			page_to_pfn(p), tsk->comm);
		tk->addr_valid = 0;
	}
	get_task_struct(tsk);
	tk->tsk = tsk;
	list_add_tail(&tk->nd, to_kill);
}

/*
 * Kill the processes that have been collected earlier.
 *
 * Only do anything when DOIT is set, otherwise just free the list
 * (this is used for clean pages which do not need killing)
 * Also when FAIL is set do a force kill because something went
 * wrong earlier.
 */
350
static void kill_procs(struct list_head *to_kill, int forcekill, int trapno,
351 352
			  int fail, struct page *page, unsigned long pfn,
			  int flags)
353 354 355 356
{
	struct to_kill *tk, *next;

	list_for_each_entry_safe (tk, next, to_kill, nd) {
357
		if (forcekill) {
358
			/*
359
			 * In case something went wrong with munmapping
360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375
			 * make sure the process doesn't catch the
			 * signal and then access the memory. Just kill it.
			 */
			if (fail || tk->addr_valid == 0) {
				printk(KERN_ERR
		"MCE %#lx: forcibly killing %s:%d because of failure to unmap corrupted page\n",
					pfn, tk->tsk->comm, tk->tsk->pid);
				force_sig(SIGKILL, tk->tsk);
			}

			/*
			 * In theory the process could have mapped
			 * something else on the address in-between. We could
			 * check for that, but we need to tell the
			 * process anyways.
			 */
376 377
			else if (kill_proc(tk->tsk, tk->addr, trapno,
					      pfn, page, flags) < 0)
378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404
				printk(KERN_ERR
		"MCE %#lx: Cannot send advisory machine check signal to %s:%d\n",
					pfn, tk->tsk->comm, tk->tsk->pid);
		}
		put_task_struct(tk->tsk);
		kfree(tk);
	}
}

static int task_early_kill(struct task_struct *tsk)
{
	if (!tsk->mm)
		return 0;
	if (tsk->flags & PF_MCE_PROCESS)
		return !!(tsk->flags & PF_MCE_EARLY);
	return sysctl_memory_failure_early_kill;
}

/*
 * Collect processes when the error hit an anonymous page.
 */
static void collect_procs_anon(struct page *page, struct list_head *to_kill,
			      struct to_kill **tkc)
{
	struct vm_area_struct *vma;
	struct task_struct *tsk;
	struct anon_vma *av;
405
	pgoff_t pgoff;
406

407
	av = page_lock_anon_vma_read(page);
408
	if (av == NULL)	/* Not actually mapped anymore */
409 410
		return;

411
	pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
412
	read_lock(&tasklist_lock);
413
	for_each_process (tsk) {
414 415
		struct anon_vma_chain *vmac;

416 417
		if (!task_early_kill(tsk))
			continue;
418 419
		anon_vma_interval_tree_foreach(vmac, &av->rb_root,
					       pgoff, pgoff) {
420
			vma = vmac->vma;
421 422 423 424 425 426 427
			if (!page_mapped_in_vma(page, vma))
				continue;
			if (vma->vm_mm == tsk->mm)
				add_to_kill(tsk, page, vma, to_kill, tkc);
		}
	}
	read_unlock(&tasklist_lock);
428
	page_unlock_anon_vma_read(av);
429 430 431 432 433 434 435 436 437 438 439 440
}

/*
 * Collect processes when the error hit a file mapped page.
 */
static void collect_procs_file(struct page *page, struct list_head *to_kill,
			      struct to_kill **tkc)
{
	struct vm_area_struct *vma;
	struct task_struct *tsk;
	struct address_space *mapping = page->mapping;

441
	mutex_lock(&mapping->i_mmap_mutex);
442
	read_lock(&tasklist_lock);
443 444 445 446 447 448
	for_each_process(tsk) {
		pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);

		if (!task_early_kill(tsk))
			continue;

449
		vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff,
450 451 452 453 454 455 456 457 458 459 460 461 462
				      pgoff) {
			/*
			 * Send early kill signal to tasks where a vma covers
			 * the page but the corrupted page is not necessarily
			 * mapped it in its pte.
			 * Assume applications who requested early kill want
			 * to be informed of all such data corruptions.
			 */
			if (vma->vm_mm == tsk->mm)
				add_to_kill(tsk, page, vma, to_kill, tkc);
		}
	}
	read_unlock(&tasklist_lock);
463
	mutex_unlock(&mapping->i_mmap_mutex);
464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493
}

/*
 * Collect the processes who have the corrupted page mapped to kill.
 * This is done in two steps for locking reasons.
 * First preallocate one tokill structure outside the spin locks,
 * so that we can kill at least one process reasonably reliable.
 */
static void collect_procs(struct page *page, struct list_head *tokill)
{
	struct to_kill *tk;

	if (!page->mapping)
		return;

	tk = kmalloc(sizeof(struct to_kill), GFP_NOIO);
	if (!tk)
		return;
	if (PageAnon(page))
		collect_procs_anon(page, tokill, &tk);
	else
		collect_procs_file(page, tokill, &tk);
	kfree(tk);
}

/*
 * Error handlers for various types of pages.
 */

enum outcome {
494 495
	IGNORED,	/* Error: cannot be handled */
	FAILED,		/* Error: handling failed */
496 497 498 499 500
	DELAYED,	/* Will be handled later */
	RECOVERED,	/* Successfully recovered */
};

static const char *action_name[] = {
501
	[IGNORED] = "Ignored",
502 503 504 505 506
	[FAILED] = "Failed",
	[DELAYED] = "Delayed",
	[RECOVERED] = "Recovered",
};

507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530
/*
 * XXX: It is possible that a page is isolated from LRU cache,
 * and then kept in swap cache or failed to remove from page cache.
 * The page count will stop it from being freed by unpoison.
 * Stress tests should be aware of this memory leak problem.
 */
static int delete_from_lru_cache(struct page *p)
{
	if (!isolate_lru_page(p)) {
		/*
		 * Clear sensible page flags, so that the buddy system won't
		 * complain when the page is unpoison-and-freed.
		 */
		ClearPageActive(p);
		ClearPageUnevictable(p);
		/*
		 * drop the page count elevated by isolate_lru_page()
		 */
		page_cache_release(p);
		return 0;
	}
	return -EIO;
}

531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558
/*
 * Error hit kernel page.
 * Do nothing, try to be lucky and not touch this instead. For a few cases we
 * could be more sophisticated.
 */
static int me_kernel(struct page *p, unsigned long pfn)
{
	return IGNORED;
}

/*
 * Page in unknown state. Do nothing.
 */
static int me_unknown(struct page *p, unsigned long pfn)
{
	printk(KERN_ERR "MCE %#lx: Unknown page state\n", pfn);
	return FAILED;
}

/*
 * Clean (or cleaned) page cache page.
 */
static int me_pagecache_clean(struct page *p, unsigned long pfn)
{
	int err;
	int ret = FAILED;
	struct address_space *mapping;

559 560
	delete_from_lru_cache(p);

561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594
	/*
	 * For anonymous pages we're done the only reference left
	 * should be the one m_f() holds.
	 */
	if (PageAnon(p))
		return RECOVERED;

	/*
	 * Now truncate the page in the page cache. This is really
	 * more like a "temporary hole punch"
	 * Don't do this for block devices when someone else
	 * has a reference, because it could be file system metadata
	 * and that's not safe to truncate.
	 */
	mapping = page_mapping(p);
	if (!mapping) {
		/*
		 * Page has been teared down in the meanwhile
		 */
		return FAILED;
	}

	/*
	 * Truncation is a bit tricky. Enable it per file system for now.
	 *
	 * Open: to take i_mutex or not for this? Right now we don't.
	 */
	if (mapping->a_ops->error_remove_page) {
		err = mapping->a_ops->error_remove_page(mapping, p);
		if (err != 0) {
			printk(KERN_INFO "MCE %#lx: Failed to punch page: %d\n",
					pfn, err);
		} else if (page_has_private(p) &&
				!try_to_release_page(p, GFP_NOIO)) {
595
			pr_info("MCE %#lx: failed to release buffers\n", pfn);
596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613
		} else {
			ret = RECOVERED;
		}
	} else {
		/*
		 * If the file system doesn't support it just invalidate
		 * This fails on dirty or anything with private pages
		 */
		if (invalidate_inode_page(p))
			ret = RECOVERED;
		else
			printk(KERN_INFO "MCE %#lx: Failed to invalidate\n",
				pfn);
	}
	return ret;
}

/*
614
 * Dirty pagecache page
615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646
 * Issues: when the error hit a hole page the error is not properly
 * propagated.
 */
static int me_pagecache_dirty(struct page *p, unsigned long pfn)
{
	struct address_space *mapping = page_mapping(p);

	SetPageError(p);
	/* TBD: print more information about the file. */
	if (mapping) {
		/*
		 * IO error will be reported by write(), fsync(), etc.
		 * who check the mapping.
		 * This way the application knows that something went
		 * wrong with its dirty file data.
		 *
		 * There's one open issue:
		 *
		 * The EIO will be only reported on the next IO
		 * operation and then cleared through the IO map.
		 * Normally Linux has two mechanisms to pass IO error
		 * first through the AS_EIO flag in the address space
		 * and then through the PageError flag in the page.
		 * Since we drop pages on memory failure handling the
		 * only mechanism open to use is through AS_AIO.
		 *
		 * This has the disadvantage that it gets cleared on
		 * the first operation that returns an error, while
		 * the PageError bit is more sticky and only cleared
		 * when the page is reread or dropped.  If an
		 * application assumes it will always get error on
		 * fsync, but does other operations on the fd before
L
Lucas De Marchi 已提交
647
		 * and the page is dropped between then the error
648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689
		 * will not be properly reported.
		 *
		 * This can already happen even without hwpoisoned
		 * pages: first on metadata IO errors (which only
		 * report through AS_EIO) or when the page is dropped
		 * at the wrong time.
		 *
		 * So right now we assume that the application DTRT on
		 * the first EIO, but we're not worse than other parts
		 * of the kernel.
		 */
		mapping_set_error(mapping, EIO);
	}

	return me_pagecache_clean(p, pfn);
}

/*
 * Clean and dirty swap cache.
 *
 * Dirty swap cache page is tricky to handle. The page could live both in page
 * cache and swap cache(ie. page is freshly swapped in). So it could be
 * referenced concurrently by 2 types of PTEs:
 * normal PTEs and swap PTEs. We try to handle them consistently by calling
 * try_to_unmap(TTU_IGNORE_HWPOISON) to convert the normal PTEs to swap PTEs,
 * and then
 *      - clear dirty bit to prevent IO
 *      - remove from LRU
 *      - but keep in the swap cache, so that when we return to it on
 *        a later page fault, we know the application is accessing
 *        corrupted data and shall be killed (we installed simple
 *        interception code in do_swap_page to catch it).
 *
 * Clean swap cache pages can be directly isolated. A later page fault will
 * bring in the known good data from disk.
 */
static int me_swapcache_dirty(struct page *p, unsigned long pfn)
{
	ClearPageDirty(p);
	/* Trigger EIO in shmem: */
	ClearPageUptodate(p);

690 691 692 693
	if (!delete_from_lru_cache(p))
		return DELAYED;
	else
		return FAILED;
694 695 696 697 698
}

static int me_swapcache_clean(struct page *p, unsigned long pfn)
{
	delete_from_swap_cache(p);
699

700 701 702 703
	if (!delete_from_lru_cache(p))
		return RECOVERED;
	else
		return FAILED;
704 705 706 707 708
}

/*
 * Huge pages. Needs work.
 * Issues:
709 710
 * - Error on hugepage is contained in hugepage unit (not in raw page unit.)
 *   To narrow down kill region to one page, we need to break up pmd.
711 712 713
 */
static int me_huge_page(struct page *p, unsigned long pfn)
{
714
	int res = 0;
715 716 717 718 719 720 721 722 723 724 725 726
	struct page *hpage = compound_head(p);
	/*
	 * We can safely recover from error on free or reserved (i.e.
	 * not in-use) hugepage by dequeuing it from freelist.
	 * To check whether a hugepage is in-use or not, we can't use
	 * page->lru because it can be used in other hugepage operations,
	 * such as __unmap_hugepage_range() and gather_surplus_pages().
	 * So instead we use page_mapping() and PageAnon().
	 * We assume that this function is called with page lock held,
	 * so there is no race between isolation and mapping/unmapping.
	 */
	if (!(page_mapping(hpage) || PageAnon(hpage))) {
727 728 729
		res = dequeue_hwpoisoned_huge_page(hpage);
		if (!res)
			return RECOVERED;
730 731
	}
	return DELAYED;
732 733 734 735 736 737 738 739 740
}

/*
 * Various page states we can handle.
 *
 * A page state is defined by its current page->flags bits.
 * The table matches them in order and calls the right handler.
 *
 * This is quite tricky because we can access page at any time
L
Lucas De Marchi 已提交
741
 * in its live cycle, so all accesses have to be extremely careful.
742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765
 *
 * This is not complete. More states could be added.
 * For any missing state don't attempt recovery.
 */

#define dirty		(1UL << PG_dirty)
#define sc		(1UL << PG_swapcache)
#define unevict		(1UL << PG_unevictable)
#define mlock		(1UL << PG_mlocked)
#define writeback	(1UL << PG_writeback)
#define lru		(1UL << PG_lru)
#define swapbacked	(1UL << PG_swapbacked)
#define head		(1UL << PG_head)
#define tail		(1UL << PG_tail)
#define compound	(1UL << PG_compound)
#define slab		(1UL << PG_slab)
#define reserved	(1UL << PG_reserved)

static struct page_state {
	unsigned long mask;
	unsigned long res;
	char *msg;
	int (*action)(struct page *p, unsigned long pfn);
} error_states[] = {
766
	{ reserved,	reserved,	"reserved kernel",	me_kernel },
767 768 769 770
	/*
	 * free pages are specially detected outside this table:
	 * PG_buddy pages only make a small fraction of all free pages.
	 */
771 772 773 774 775 776 777 778 779 780 781 782 783 784 785

	/*
	 * Could in theory check if slab page is free or if we can drop
	 * currently unused objects without touching them. But just
	 * treat it as standard kernel for now.
	 */
	{ slab,		slab,		"kernel slab",	me_kernel },

#ifdef CONFIG_PAGEFLAGS_EXTENDED
	{ head,		head,		"huge",		me_huge_page },
	{ tail,		tail,		"huge",		me_huge_page },
#else
	{ compound,	compound,	"huge",		me_huge_page },
#endif

786 787
	{ sc|dirty,	sc|dirty,	"dirty swapcache",	me_swapcache_dirty },
	{ sc|dirty,	sc,		"clean swapcache",	me_swapcache_clean },
788

789
	{ mlock|dirty,	mlock|dirty,	"dirty mlocked LRU",	me_pagecache_dirty },
790
	{ mlock|dirty,	mlock,		"clean mlocked LRU",	me_pagecache_clean },
791

792
	{ unevict|dirty, unevict|dirty,	"dirty unevictable LRU", me_pagecache_dirty },
793
	{ unevict|dirty, unevict,	"clean unevictable LRU", me_pagecache_clean },
794

795
	{ lru|dirty,	lru|dirty,	"dirty LRU",	me_pagecache_dirty },
796 797 798 799 800 801 802 803
	{ lru|dirty,	lru,		"clean LRU",	me_pagecache_clean },

	/*
	 * Catchall entry: must be at end.
	 */
	{ 0,		0,		"unknown page state",	me_unknown },
};

804 805 806 807 808 809 810 811 812 813 814 815 816
#undef dirty
#undef sc
#undef unevict
#undef mlock
#undef writeback
#undef lru
#undef swapbacked
#undef head
#undef tail
#undef compound
#undef slab
#undef reserved

817 818 819 820
/*
 * "Dirty/Clean" indication is not 100% accurate due to the possibility of
 * setting PG_dirty outside page lock. See also comment above set_page_dirty().
 */
821 822
static void action_result(unsigned long pfn, char *msg, int result)
{
823 824
	pr_err("MCE %#lx: %s page recovery: %s\n",
		pfn, msg, action_name[result]);
825 826 827
}

static int page_action(struct page_state *ps, struct page *p,
828
			unsigned long pfn)
829 830
{
	int result;
831
	int count;
832 833 834

	result = ps->action(p, pfn);
	action_result(pfn, ps->msg, result);
835

836
	count = page_count(p) - 1;
837 838 839
	if (ps->action == me_swapcache_dirty && result == DELAYED)
		count--;
	if (count != 0) {
840 841
		printk(KERN_ERR
		       "MCE %#lx: %s page still referenced by %d users\n",
842
		       pfn, ps->msg, count);
843 844
		result = FAILED;
	}
845 846 847 848 849 850

	/* Could do more checks here if page looks ok */
	/*
	 * Could adjust zone counters here to correct for the missing page.
	 */

851
	return (result == RECOVERED || result == DELAYED) ? 0 : -EBUSY;
852 853 854 855 856 857
}

/*
 * Do all that is necessary to remove user space mappings. Unmap
 * the pages and send SIGBUS to the processes if the data was dirty.
 */
W
Wu Fengguang 已提交
858
static int hwpoison_user_mappings(struct page *p, unsigned long pfn,
859
				  int trapno, int flags, struct page **hpagep)
860 861 862 863 864
{
	enum ttu_flags ttu = TTU_UNMAP | TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS;
	struct address_space *mapping;
	LIST_HEAD(tokill);
	int ret;
865
	int kill = 1, forcekill;
866
	struct page *hpage = *hpagep;
867
	struct page *ppage;
868

W
Wu Fengguang 已提交
869 870
	if (PageReserved(p) || PageSlab(p))
		return SWAP_SUCCESS;
871 872 873 874 875

	/*
	 * This check implies we don't kill processes if their pages
	 * are in the swap cache early. Those are always late kills.
	 */
876
	if (!page_mapped(hpage))
W
Wu Fengguang 已提交
877 878
		return SWAP_SUCCESS;

879
	if (PageKsm(p))
W
Wu Fengguang 已提交
880
		return SWAP_FAIL;
881 882 883 884 885 886 887 888 889 890

	if (PageSwapCache(p)) {
		printk(KERN_ERR
		       "MCE %#lx: keeping poisoned page in swap cache\n", pfn);
		ttu |= TTU_IGNORE_HWPOISON;
	}

	/*
	 * Propagate the dirty bit from PTEs to struct page first, because we
	 * need this to decide if we should kill or just drop the page.
891 892
	 * XXX: the dirty test could be racy: set_page_dirty() may not always
	 * be called inside page lock (it's recommended but not enforced).
893
	 */
894
	mapping = page_mapping(hpage);
895
	if (!(flags & MF_MUST_KILL) && !PageDirty(hpage) && mapping &&
896 897 898
	    mapping_cap_writeback_dirty(mapping)) {
		if (page_mkclean(hpage)) {
			SetPageDirty(hpage);
899 900 901 902 903 904 905 906 907
		} else {
			kill = 0;
			ttu |= TTU_IGNORE_HWPOISON;
			printk(KERN_INFO
	"MCE %#lx: corrupted page was clean: dropped without side effects\n",
				pfn);
		}
	}

908 909 910 911 912 913 914 915
	/*
	 * ppage: poisoned page
	 *   if p is regular page(4k page)
	 *        ppage == real poisoned page;
	 *   else p is hugetlb or THP, ppage == head page.
	 */
	ppage = hpage;

916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940
	if (PageTransHuge(hpage)) {
		/*
		 * Verify that this isn't a hugetlbfs head page, the check for
		 * PageAnon is just for avoid tripping a split_huge_page
		 * internal debug check, as split_huge_page refuses to deal with
		 * anything that isn't an anon page. PageAnon can't go away fro
		 * under us because we hold a refcount on the hpage, without a
		 * refcount on the hpage. split_huge_page can't be safely called
		 * in the first place, having a refcount on the tail isn't
		 * enough * to be safe.
		 */
		if (!PageHuge(hpage) && PageAnon(hpage)) {
			if (unlikely(split_huge_page(hpage))) {
				/*
				 * FIXME: if splitting THP is failed, it is
				 * better to stop the following operation rather
				 * than causing panic by unmapping. System might
				 * survive if the page is freed later.
				 */
				printk(KERN_INFO
					"MCE %#lx: failed to split THP\n", pfn);

				BUG_ON(!PageHWPoison(p));
				return SWAP_FAIL;
			}
941 942 943 944
			/*
			 * We pinned the head page for hwpoison handling,
			 * now we split the thp and we are interested in
			 * the hwpoisoned raw page, so move the refcount
945
			 * to it. Similarly, page lock is shifted.
946 947 948 949
			 */
			if (hpage != p) {
				put_page(hpage);
				get_page(p);
950 951 952
				lock_page(p);
				unlock_page(hpage);
				*hpagep = p;
953
			}
954 955
			/* THP is split, so ppage should be the real poisoned page. */
			ppage = p;
956 957 958
		}
	}

959 960 961 962 963 964 965 966 967
	/*
	 * First collect all the processes that have the page
	 * mapped in dirty form.  This has to be done before try_to_unmap,
	 * because ttu takes the rmap data structures down.
	 *
	 * Error handling: We ignore errors here because
	 * there's nothing that can be done.
	 */
	if (kill)
968
		collect_procs(ppage, &tokill);
969

970
	ret = try_to_unmap(ppage, ttu);
971 972
	if (ret != SWAP_SUCCESS)
		printk(KERN_ERR "MCE %#lx: failed to unmap page (mapcount=%d)\n",
973 974
				pfn, page_mapcount(ppage));

975 976 977 978
	/*
	 * Now that the dirty bit has been propagated to the
	 * struct page and all unmaps done we can decide if
	 * killing is needed or not.  Only kill when the page
979 980
	 * was dirty or the process is not restartable,
	 * otherwise the tokill list is merely
981 982 983 984
	 * freed.  When there was a problem unmapping earlier
	 * use a more force-full uncatchable kill to prevent
	 * any accesses to the poisoned memory.
	 */
985 986
	forcekill = PageDirty(ppage) || (flags & MF_MUST_KILL);
	kill_procs(&tokill, forcekill, trapno,
987
		      ret != SWAP_SUCCESS, p, pfn, flags);
W
Wu Fengguang 已提交
988 989

	return ret;
990 991
}

992 993 994
static void set_page_hwpoison_huge_page(struct page *hpage)
{
	int i;
995
	int nr_pages = 1 << compound_order(hpage);
996 997 998 999 1000 1001 1002
	for (i = 0; i < nr_pages; i++)
		SetPageHWPoison(hpage + i);
}

static void clear_page_hwpoison_huge_page(struct page *hpage)
{
	int i;
1003
	int nr_pages = 1 << compound_order(hpage);
1004 1005 1006 1007
	for (i = 0; i < nr_pages; i++)
		ClearPageHWPoison(hpage + i);
}

1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026
/**
 * memory_failure - Handle memory failure of a page.
 * @pfn: Page Number of the corrupted page
 * @trapno: Trap number reported in the signal to user space.
 * @flags: fine tune action taken
 *
 * This function is called by the low level machine check code
 * of an architecture when it detects hardware memory corruption
 * of a page. It tries its best to recover, which includes
 * dropping pages, killing processes etc.
 *
 * The function is primarily of use for corruptions that
 * happen outside the current execution context (e.g. when
 * detected by a background scrubber)
 *
 * Must run in process context (e.g. a work queue) with interrupts
 * enabled and no spinlocks hold.
 */
int memory_failure(unsigned long pfn, int trapno, int flags)
1027 1028 1029
{
	struct page_state *ps;
	struct page *p;
1030
	struct page *hpage;
1031
	int res;
1032
	unsigned int nr_pages;
1033
	unsigned long page_flags;
1034 1035 1036 1037 1038

	if (!sysctl_memory_failure_recovery)
		panic("Memory failure from trap %d on page %lx", trapno, pfn);

	if (!pfn_valid(pfn)) {
1039 1040 1041 1042
		printk(KERN_ERR
		       "MCE %#lx: memory outside kernel control\n",
		       pfn);
		return -ENXIO;
1043 1044 1045
	}

	p = pfn_to_page(pfn);
1046
	hpage = compound_head(p);
1047
	if (TestSetPageHWPoison(p)) {
1048
		printk(KERN_ERR "MCE %#lx: already hardware poisoned\n", pfn);
1049 1050 1051
		return 0;
	}

1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062
	/*
	 * Currently errors on hugetlbfs pages are measured in hugepage units,
	 * so nr_pages should be 1 << compound_order.  OTOH when errors are on
	 * transparent hugepages, they are supposed to be split and error
	 * measurement is done in normal page units.  So nr_pages should be one
	 * in this case.
	 */
	if (PageHuge(p))
		nr_pages = 1 << compound_order(hpage);
	else /* normal page or thp */
		nr_pages = 1;
1063
	atomic_long_add(nr_pages, &num_poisoned_pages);
1064 1065 1066 1067 1068

	/*
	 * We need/can do nothing about count=0 pages.
	 * 1) it's a free page, and therefore in safe hand:
	 *    prep_new_page() will be the gate keeper.
1069 1070 1071 1072
	 * 2) it's a free hugepage, which is also safe:
	 *    an affected hugepage will be dequeued from hugepage freelist,
	 *    so there's no concern about reusing it ever after.
	 * 3) it's part of a non-compound high order page.
1073 1074 1075 1076 1077 1078
	 *    Implies some kernel user: cannot stop them from
	 *    R/W the page; let's pray that the page has been
	 *    used and will be freed some time later.
	 * In fact it's dangerous to directly bump up page count from 0,
	 * that may make page_freeze_refs()/page_unfreeze_refs() mismatch.
	 */
1079
	if (!(flags & MF_COUNT_INCREASED) &&
1080
		!get_page_unless_zero(hpage)) {
1081 1082 1083
		if (is_free_buddy_page(p)) {
			action_result(pfn, "free buddy", DELAYED);
			return 0;
1084 1085 1086 1087 1088
		} else if (PageHuge(hpage)) {
			/*
			 * Check "just unpoisoned", "filter hit", and
			 * "race with other subpage."
			 */
J
Jens Axboe 已提交
1089
			lock_page(hpage);
1090 1091 1092
			if (!PageHWPoison(hpage)
			    || (hwpoison_filter(p) && TestClearPageHWPoison(p))
			    || (p != hpage && TestSetPageHWPoison(hpage))) {
1093
				atomic_long_sub(nr_pages, &num_poisoned_pages);
1094 1095 1096 1097 1098 1099 1100 1101
				return 0;
			}
			set_page_hwpoison_huge_page(hpage);
			res = dequeue_hwpoisoned_huge_page(hpage);
			action_result(pfn, "free huge",
				      res ? IGNORED : DELAYED);
			unlock_page(hpage);
			return res;
1102 1103 1104 1105
		} else {
			action_result(pfn, "high order kernel", IGNORED);
			return -EBUSY;
		}
1106 1107
	}

1108 1109 1110 1111 1112 1113 1114 1115
	/*
	 * We ignore non-LRU pages for good reasons.
	 * - PG_locked is only well defined for LRU pages and a few others
	 * - to avoid races with __set_page_locked()
	 * - to avoid races with __SetPageSlab*() (and more non-atomic ops)
	 * The check (unnecessarily) ignores LRU pages being isolated and
	 * walked by the page reclaim code, however that's not a big loss.
	 */
1116
	if (!PageHuge(p) && !PageTransTail(p)) {
1117 1118 1119 1120 1121 1122 1123
		if (!PageLRU(p))
			shake_page(p, 0);
		if (!PageLRU(p)) {
			/*
			 * shake_page could have turned it free.
			 */
			if (is_free_buddy_page(p)) {
1124 1125 1126 1127
				if (flags & MF_COUNT_INCREASED)
					action_result(pfn, "free buddy", DELAYED);
				else
					action_result(pfn, "free buddy, 2nd try", DELAYED);
1128 1129 1130 1131 1132
				return 0;
			}
			action_result(pfn, "non LRU", IGNORED);
			put_page(p);
			return -EBUSY;
1133
		}
1134 1135
	}

1136 1137 1138 1139 1140
	/*
	 * Lock the page and wait for writeback to finish.
	 * It's very difficult to mess with pages currently under IO
	 * and in many cases impossible, so we just avoid it here.
	 */
J
Jens Axboe 已提交
1141
	lock_page(hpage);
W
Wu Fengguang 已提交
1142

1143 1144 1145 1146 1147 1148 1149 1150 1151
	/*
	 * We use page flags to determine what action should be taken, but
	 * the flags can be modified by the error containment action.  One
	 * example is an mlocked page, where PG_mlocked is cleared by
	 * page_remove_rmap() in try_to_unmap_one(). So to determine page status
	 * correctly, we save a copy of the page flags at this time.
	 */
	page_flags = p->flags;

W
Wu Fengguang 已提交
1152 1153 1154 1155
	/*
	 * unpoison always clear PG_hwpoison inside page lock
	 */
	if (!PageHWPoison(p)) {
1156
		printk(KERN_ERR "MCE %#lx: just unpoisoned\n", pfn);
W
Wu Fengguang 已提交
1157 1158 1159
		res = 0;
		goto out;
	}
W
Wu Fengguang 已提交
1160 1161
	if (hwpoison_filter(p)) {
		if (TestClearPageHWPoison(p))
1162
			atomic_long_sub(nr_pages, &num_poisoned_pages);
1163 1164
		unlock_page(hpage);
		put_page(hpage);
W
Wu Fengguang 已提交
1165 1166
		return 0;
	}
W
Wu Fengguang 已提交
1167

1168 1169 1170 1171
	/*
	 * For error on the tail page, we should set PG_hwpoison
	 * on the head page to show that the hugepage is hwpoisoned
	 */
1172
	if (PageHuge(p) && PageTail(p) && TestSetPageHWPoison(hpage)) {
1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187
		action_result(pfn, "hugepage already hardware poisoned",
				IGNORED);
		unlock_page(hpage);
		put_page(hpage);
		return 0;
	}
	/*
	 * Set PG_hwpoison on all pages in an error hugepage,
	 * because containment is done in hugepage unit for now.
	 * Since we have done TestSetPageHWPoison() for the head page with
	 * page lock held, we can safely set PG_hwpoison bits on tail pages.
	 */
	if (PageHuge(p))
		set_page_hwpoison_huge_page(hpage);

1188 1189 1190 1191
	wait_on_page_writeback(p);

	/*
	 * Now take care of user space mappings.
1192
	 * Abort on fail: __delete_from_page_cache() assumes unmapped page.
1193 1194 1195
	 *
	 * When the raw error page is thp tail page, hpage points to the raw
	 * page after thp split.
1196
	 */
1197 1198
	if (hwpoison_user_mappings(p, pfn, trapno, flags, &hpage)
	    != SWAP_SUCCESS) {
W
Wu Fengguang 已提交
1199 1200 1201 1202
		printk(KERN_ERR "MCE %#lx: cannot unmap page, give up\n", pfn);
		res = -EBUSY;
		goto out;
	}
1203 1204 1205 1206

	/*
	 * Torn down by someone else?
	 */
1207
	if (PageLRU(p) && !PageSwapCache(p) && p->mapping == NULL) {
1208
		action_result(pfn, "already truncated LRU", IGNORED);
1209
		res = -EBUSY;
1210 1211 1212 1213
		goto out;
	}

	res = -EBUSY;
1214 1215 1216 1217 1218 1219 1220
	/*
	 * The first check uses the current page flags which may not have any
	 * relevant information. The second check with the saved page flagss is
	 * carried out only if the first check can't determine the page status.
	 */
	for (ps = error_states;; ps++)
		if ((p->flags & ps->mask) == ps->res)
1221
			break;
1222 1223 1224

	page_flags |= (p->flags & (1UL << PG_dirty));

1225 1226 1227 1228 1229
	if (!ps->mask)
		for (ps = error_states;; ps++)
			if ((page_flags & ps->mask) == ps->res)
				break;
	res = page_action(ps, p, pfn);
1230
out:
1231
	unlock_page(hpage);
1232 1233
	return res;
}
1234
EXPORT_SYMBOL_GPL(memory_failure);
W
Wu Fengguang 已提交
1235

1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282
#define MEMORY_FAILURE_FIFO_ORDER	4
#define MEMORY_FAILURE_FIFO_SIZE	(1 << MEMORY_FAILURE_FIFO_ORDER)

struct memory_failure_entry {
	unsigned long pfn;
	int trapno;
	int flags;
};

struct memory_failure_cpu {
	DECLARE_KFIFO(fifo, struct memory_failure_entry,
		      MEMORY_FAILURE_FIFO_SIZE);
	spinlock_t lock;
	struct work_struct work;
};

static DEFINE_PER_CPU(struct memory_failure_cpu, memory_failure_cpu);

/**
 * memory_failure_queue - Schedule handling memory failure of a page.
 * @pfn: Page Number of the corrupted page
 * @trapno: Trap number reported in the signal to user space.
 * @flags: Flags for memory failure handling
 *
 * This function is called by the low level hardware error handler
 * when it detects hardware memory corruption of a page. It schedules
 * the recovering of error page, including dropping pages, killing
 * processes etc.
 *
 * The function is primarily of use for corruptions that
 * happen outside the current execution context (e.g. when
 * detected by a background scrubber)
 *
 * Can run in IRQ context.
 */
void memory_failure_queue(unsigned long pfn, int trapno, int flags)
{
	struct memory_failure_cpu *mf_cpu;
	unsigned long proc_flags;
	struct memory_failure_entry entry = {
		.pfn =		pfn,
		.trapno =	trapno,
		.flags =	flags,
	};

	mf_cpu = &get_cpu_var(memory_failure_cpu);
	spin_lock_irqsave(&mf_cpu->lock, proc_flags);
S
Stefani Seibold 已提交
1283
	if (kfifo_put(&mf_cpu->fifo, entry))
1284 1285
		schedule_work_on(smp_processor_id(), &mf_cpu->work);
	else
J
Joe Perches 已提交
1286
		pr_err("Memory failure: buffer overflow when queuing memory failure at %#lx\n",
1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306
		       pfn);
	spin_unlock_irqrestore(&mf_cpu->lock, proc_flags);
	put_cpu_var(memory_failure_cpu);
}
EXPORT_SYMBOL_GPL(memory_failure_queue);

static void memory_failure_work_func(struct work_struct *work)
{
	struct memory_failure_cpu *mf_cpu;
	struct memory_failure_entry entry = { 0, };
	unsigned long proc_flags;
	int gotten;

	mf_cpu = &__get_cpu_var(memory_failure_cpu);
	for (;;) {
		spin_lock_irqsave(&mf_cpu->lock, proc_flags);
		gotten = kfifo_get(&mf_cpu->fifo, &entry);
		spin_unlock_irqrestore(&mf_cpu->lock, proc_flags);
		if (!gotten)
			break;
1307 1308 1309 1310
		if (entry.flags & MF_SOFT_OFFLINE)
			soft_offline_page(pfn_to_page(entry.pfn), entry.flags);
		else
			memory_failure(entry.pfn, entry.trapno, entry.flags);
1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329
	}
}

static int __init memory_failure_init(void)
{
	struct memory_failure_cpu *mf_cpu;
	int cpu;

	for_each_possible_cpu(cpu) {
		mf_cpu = &per_cpu(memory_failure_cpu, cpu);
		spin_lock_init(&mf_cpu->lock);
		INIT_KFIFO(mf_cpu->fifo);
		INIT_WORK(&mf_cpu->work, memory_failure_work_func);
	}

	return 0;
}
core_initcall(memory_failure_init);

W
Wu Fengguang 已提交
1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346
/**
 * unpoison_memory - Unpoison a previously poisoned page
 * @pfn: Page number of the to be unpoisoned page
 *
 * Software-unpoison a page that has been poisoned by
 * memory_failure() earlier.
 *
 * This is only done on the software-level, so it only works
 * for linux injected failures, not real hardware failures
 *
 * Returns 0 for success, otherwise -errno.
 */
int unpoison_memory(unsigned long pfn)
{
	struct page *page;
	struct page *p;
	int freeit = 0;
1347
	unsigned int nr_pages;
W
Wu Fengguang 已提交
1348 1349 1350 1351 1352 1353 1354 1355

	if (!pfn_valid(pfn))
		return -ENXIO;

	p = pfn_to_page(pfn);
	page = compound_head(p);

	if (!PageHWPoison(p)) {
1356
		pr_info("MCE: Page was already unpoisoned %#lx\n", pfn);
W
Wu Fengguang 已提交
1357 1358 1359
		return 0;
	}

1360 1361 1362 1363 1364
	/*
	 * unpoison_memory() can encounter thp only when the thp is being
	 * worked by memory_failure() and the page lock is not held yet.
	 * In such case, we yield to memory_failure() and make unpoison fail.
	 */
1365
	if (!PageHuge(page) && PageTransHuge(page)) {
1366 1367 1368 1369
		pr_info("MCE: Memory failure is now running on %#lx\n", pfn);
			return 0;
	}

1370
	nr_pages = 1 << compound_order(page);
1371

W
Wu Fengguang 已提交
1372
	if (!get_page_unless_zero(page)) {
1373 1374 1375 1376 1377 1378 1379
		/*
		 * Since HWPoisoned hugepage should have non-zero refcount,
		 * race between memory failure and unpoison seems to happen.
		 * In such case unpoison fails and memory failure runs
		 * to the end.
		 */
		if (PageHuge(page)) {
1380
			pr_info("MCE: Memory failure is now running on free hugepage %#lx\n", pfn);
1381 1382
			return 0;
		}
W
Wu Fengguang 已提交
1383
		if (TestClearPageHWPoison(p))
1384
			atomic_long_dec(&num_poisoned_pages);
1385
		pr_info("MCE: Software-unpoisoned free page %#lx\n", pfn);
W
Wu Fengguang 已提交
1386 1387 1388
		return 0;
	}

J
Jens Axboe 已提交
1389
	lock_page(page);
W
Wu Fengguang 已提交
1390 1391 1392 1393 1394 1395
	/*
	 * This test is racy because PG_hwpoison is set outside of page lock.
	 * That's acceptable because that won't trigger kernel panic. Instead,
	 * the PG_hwpoison page will be caught and isolated on the entrance to
	 * the free buddy page pool.
	 */
1396
	if (TestClearPageHWPoison(page)) {
1397
		pr_info("MCE: Software-unpoisoned page %#lx\n", pfn);
1398
		atomic_long_sub(nr_pages, &num_poisoned_pages);
W
Wu Fengguang 已提交
1399
		freeit = 1;
1400 1401
		if (PageHuge(page))
			clear_page_hwpoison_huge_page(page);
W
Wu Fengguang 已提交
1402 1403 1404 1405
	}
	unlock_page(page);

	put_page(page);
1406
	if (freeit && !(pfn == my_zero_pfn(0) && page_count(p) == 1))
W
Wu Fengguang 已提交
1407 1408 1409 1410 1411
		put_page(page);

	return 0;
}
EXPORT_SYMBOL(unpoison_memory);
1412 1413 1414

static struct page *new_page(struct page *p, unsigned long private, int **x)
{
1415
	int nid = page_to_nid(p);
1416 1417 1418 1419 1420
	if (PageHuge(p))
		return alloc_huge_page_node(page_hstate(compound_head(p)),
						   nid);
	else
		return alloc_pages_exact_node(nid, GFP_HIGHUSER_MOVABLE, 0);
1421 1422 1423 1424 1425 1426 1427 1428
}

/*
 * Safely get reference count of an arbitrary page.
 * Returns 0 for a free page, -EIO for a zero refcount page
 * that is not free, and 1 for any other page type.
 * For 1 the page is returned with increased page count, otherwise not.
 */
1429
static int __get_any_page(struct page *p, unsigned long pfn, int flags)
1430 1431 1432 1433 1434 1435
{
	int ret;

	if (flags & MF_COUNT_INCREASED)
		return 1;

1436 1437 1438 1439
	/*
	 * When the target page is a free hugepage, just remove it
	 * from free hugepage list.
	 */
1440
	if (!get_page_unless_zero(compound_head(p))) {
1441
		if (PageHuge(p)) {
1442
			pr_info("%s: %#lx free huge page\n", __func__, pfn);
1443
			ret = 0;
1444
		} else if (is_free_buddy_page(p)) {
1445
			pr_info("%s: %#lx free buddy page\n", __func__, pfn);
1446 1447
			ret = 0;
		} else {
1448 1449
			pr_info("%s: %#lx: unknown zero refcount page type %lx\n",
				__func__, pfn, p->flags);
1450 1451 1452 1453 1454 1455 1456 1457 1458
			ret = -EIO;
		}
	} else {
		/* Not a free page */
		ret = 1;
	}
	return ret;
}

1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482
static int get_any_page(struct page *page, unsigned long pfn, int flags)
{
	int ret = __get_any_page(page, pfn, flags);

	if (ret == 1 && !PageHuge(page) && !PageLRU(page)) {
		/*
		 * Try to free it.
		 */
		put_page(page);
		shake_page(page, 1);

		/*
		 * Did it turn free?
		 */
		ret = __get_any_page(page, pfn, 0);
		if (!PageLRU(page)) {
			pr_info("soft_offline: %#lx: unknown non LRU page type %lx\n",
				pfn, page->flags);
			return -EIO;
		}
	}
	return ret;
}

1483 1484 1485 1486 1487
static int soft_offline_huge_page(struct page *page, int flags)
{
	int ret;
	unsigned long pfn = page_to_pfn(page);
	struct page *hpage = compound_head(page);
1488
	LIST_HEAD(pagelist);
1489

1490 1491 1492 1493 1494
	/*
	 * This double-check of PageHWPoison is to avoid the race with
	 * memory_failure(). See also comment in __soft_offline_page().
	 */
	lock_page(hpage);
1495
	if (PageHWPoison(hpage)) {
1496 1497
		unlock_page(hpage);
		put_page(hpage);
1498
		pr_info("soft offline: %#lx hugepage already poisoned\n", pfn);
1499
		return -EBUSY;
1500
	}
1501
	unlock_page(hpage);
1502 1503

	/* Keep page count to indicate a given hugepage is isolated. */
1504 1505 1506
	list_move(&hpage->lru, &pagelist);
	ret = migrate_pages(&pagelist, new_page, MPOL_MF_MOVE_ALL,
				MIGRATE_SYNC, MR_MEMORY_FAILURE);
1507
	if (ret) {
1508 1509
		pr_info("soft offline: %#lx: migration failed %d, type %lx\n",
			pfn, ret, page->flags);
1510 1511 1512 1513 1514 1515 1516 1517
		/*
		 * We know that soft_offline_huge_page() tries to migrate
		 * only one hugepage pointed to by hpage, so we need not
		 * run through the pagelist here.
		 */
		putback_active_hugepage(hpage);
		if (ret > 0)
			ret = -EIO;
1518
	} else {
1519 1520 1521 1522 1523 1524 1525 1526 1527 1528
		/* overcommit hugetlb page will be freed to buddy */
		if (PageHuge(page)) {
			set_page_hwpoison_huge_page(hpage);
			dequeue_hwpoisoned_huge_page(hpage);
			atomic_long_add(1 << compound_order(hpage),
					&num_poisoned_pages);
		} else {
			SetPageHWPoison(page);
			atomic_long_inc(&num_poisoned_pages);
		}
1529 1530 1531 1532
	}
	return ret;
}

1533 1534 1535 1536
static int __soft_offline_page(struct page *page, int flags)
{
	int ret;
	unsigned long pfn = page_to_pfn(page);
1537 1538

	/*
1539 1540 1541 1542
	 * Check PageHWPoison again inside page lock because PageHWPoison
	 * is set by memory_failure() outside page lock. Note that
	 * memory_failure() also double-checks PageHWPoison inside page lock,
	 * so there's no race between soft_offline_page() and memory_failure().
1543
	 */
1544 1545
	lock_page(page);
	wait_on_page_writeback(page);
1546 1547 1548 1549 1550 1551
	if (PageHWPoison(page)) {
		unlock_page(page);
		put_page(page);
		pr_info("soft offline: %#lx page already poisoned\n", pfn);
		return -EBUSY;
	}
1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562
	/*
	 * Try to invalidate first. This should work for
	 * non dirty unmapped page cache pages.
	 */
	ret = invalidate_inode_page(page);
	unlock_page(page);
	/*
	 * RED-PEN would be better to keep it isolated here, but we
	 * would need to fix isolation locking first.
	 */
	if (ret == 1) {
1563
		put_page(page);
1564
		pr_info("soft_offline: %#lx: invalidated\n", pfn);
1565 1566 1567
		SetPageHWPoison(page);
		atomic_long_inc(&num_poisoned_pages);
		return 0;
1568 1569 1570 1571 1572 1573 1574 1575
	}

	/*
	 * Simple invalidation didn't work.
	 * Try to migrate to a new page instead. migrate.c
	 * handles a large number of cases for us.
	 */
	ret = isolate_lru_page(page);
1576 1577 1578 1579 1580
	/*
	 * Drop page reference which is came from get_any_page()
	 * successful isolate_lru_page() already took another one.
	 */
	put_page(page);
1581 1582
	if (!ret) {
		LIST_HEAD(pagelist);
1583
		inc_zone_page_state(page, NR_ISOLATED_ANON +
1584
					page_is_file_cache(page));
1585
		list_add(&page->lru, &pagelist);
1586
		ret = migrate_pages(&pagelist, new_page, MPOL_MF_MOVE_ALL,
1587
					MIGRATE_SYNC, MR_MEMORY_FAILURE);
1588
		if (ret) {
1589 1590 1591 1592 1593 1594 1595
			if (!list_empty(&pagelist)) {
				list_del(&page->lru);
				dec_zone_page_state(page, NR_ISOLATED_ANON +
						page_is_file_cache(page));
				putback_lru_page(page);
			}

1596
			pr_info("soft offline: %#lx: migration failed %d, type %lx\n",
1597 1598 1599
				pfn, ret, page->flags);
			if (ret > 0)
				ret = -EIO;
1600
		} else {
1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612
			/*
			 * After page migration succeeds, the source page can
			 * be trapped in pagevec and actual freeing is delayed.
			 * Freeing code works differently based on PG_hwpoison,
			 * so there's a race. We need to make sure that the
			 * source page should be freed back to buddy before
			 * setting PG_hwpoison.
			 */
			if (!is_free_buddy_page(page))
				lru_add_drain_all();
			if (!is_free_buddy_page(page))
				drain_all_pages();
1613
			SetPageHWPoison(page);
1614 1615 1616
			if (!is_free_buddy_page(page))
				pr_info("soft offline: %#lx: page leaked\n",
					pfn);
1617
			atomic_long_inc(&num_poisoned_pages);
1618 1619
		}
	} else {
1620
		pr_info("soft offline: %#lx: isolation failed: %d, page count %d, type %lx\n",
1621
			pfn, ret, page_count(page), page->flags);
1622 1623 1624
	}
	return ret;
}
1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665

/**
 * soft_offline_page - Soft offline a page.
 * @page: page to offline
 * @flags: flags. Same as memory_failure().
 *
 * Returns 0 on success, otherwise negated errno.
 *
 * Soft offline a page, by migration or invalidation,
 * without killing anything. This is for the case when
 * a page is not corrupted yet (so it's still valid to access),
 * but has had a number of corrected errors and is better taken
 * out.
 *
 * The actual policy on when to do that is maintained by
 * user space.
 *
 * This should never impact any application or cause data loss,
 * however it might take some time.
 *
 * This is not a 100% solution for all memory, but tries to be
 * ``good enough'' for the majority of memory.
 */
int soft_offline_page(struct page *page, int flags)
{
	int ret;
	unsigned long pfn = page_to_pfn(page);
	struct page *hpage = compound_trans_head(page);

	if (PageHWPoison(page)) {
		pr_info("soft offline: %#lx page already poisoned\n", pfn);
		return -EBUSY;
	}
	if (!PageHuge(page) && PageTransHuge(hpage)) {
		if (PageAnon(hpage) && unlikely(split_huge_page(hpage))) {
			pr_info("soft offline: %#lx: failed to split THP\n",
				pfn);
			return -EBUSY;
		}
	}

1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679
	/*
	 * The lock_memory_hotplug prevents a race with memory hotplug.
	 * This is a big hammer, a better would be nicer.
	 */
	lock_memory_hotplug();

	/*
	 * Isolate the page, so that it doesn't get reallocated if it
	 * was free. This flag should be kept set until the source page
	 * is freed and PG_hwpoison on it is set.
	 */
	if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
		set_migratetype_isolate(page, true);

1680
	ret = get_any_page(page, pfn, flags);
1681 1682
	unlock_memory_hotplug();
	if (ret > 0) { /* for in-use pages */
1683 1684 1685 1686
		if (PageHuge(page))
			ret = soft_offline_huge_page(page, flags);
		else
			ret = __soft_offline_page(page, flags);
1687
	} else if (ret == 0) { /* for free pages */
1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700
		if (PageHuge(page)) {
			set_page_hwpoison_huge_page(hpage);
			dequeue_hwpoisoned_huge_page(hpage);
			atomic_long_add(1 << compound_order(hpage),
					&num_poisoned_pages);
		} else {
			SetPageHWPoison(page);
			atomic_long_inc(&num_poisoned_pages);
		}
	}
	unset_migratetype_isolate(page, MIGRATE_MOVABLE);
	return ret;
}