1 // SPDX-License-Identifier: GPL-2.0
3 * linux/mm/page_isolation.c
7 #include <linux/page-isolation.h>
8 #include <linux/pageblock-flags.h>
9 #include <linux/memory.h>
10 #include <linux/hugetlb.h>
11 #include <linux/page_owner.h>
12 #include <linux/migrate.h>
15 #define CREATE_TRACE_POINTS
16 #include <trace/events/page_isolation.h>
19 * This function checks whether the range [start_pfn, end_pfn) includes
20 * unmovable pages or not. The range must fall into a single pageblock and
21 * consequently belong to a single zone.
23 * PageLRU check without isolation or lru_lock could race so that
24 * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
25 * check without lock_page also may miss some movable non-lru pages at
26 * race condition. So you can't expect this function should be exact.
28 * Returns a page without holding a reference. If the caller wants to
29 * dereference that page (e.g., dumping), it has to make sure that it
30 * cannot get removed (e.g., via memory unplug) concurrently.
33 static struct page *has_unmovable_pages(unsigned long start_pfn, unsigned long end_pfn,
34 int migratetype, int flags)
36 struct page *page = pfn_to_page(start_pfn);
37 struct zone *zone = page_zone(page);
40 VM_BUG_ON(pageblock_start_pfn(start_pfn) !=
41 pageblock_start_pfn(end_pfn - 1));
43 if (is_migrate_cma_page(page)) {
45 * CMA allocations (alloc_contig_range) really need to mark
46 * isolate CMA pageblocks even when they are not movable in fact
47 * so consider them movable here.
49 if (is_migrate_cma(migratetype))
55 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
56 page = pfn_to_page(pfn);
59 * Both, bootmem allocations and memory holes are marked
60 * PG_reserved and are unmovable. We can even have unmovable
61 * allocations inside ZONE_MOVABLE, for example when
62 * specifying "movablecore".
64 if (PageReserved(page))
68 * If the zone is movable and we have ruled out all reserved
69 * pages then it should be reasonably safe to assume the rest
72 if (zone_idx(zone) == ZONE_MOVABLE)
76 * Hugepages are not in LRU lists, but they're movable.
77 * THPs are on the LRU, but need to be counted as #small pages.
78 * We need not scan over tail pages because we don't
79 * handle each tail page individually in migration.
81 if (PageHuge(page) || PageTransCompound(page)) {
82 struct page *head = compound_head(page);
83 unsigned int skip_pages;
86 if (!hugepage_migration_supported(page_hstate(head)))
88 } else if (!PageLRU(head) && !__PageMovable(head)) {
92 skip_pages = compound_nr(head) - (page - head);
93 pfn += skip_pages - 1;
98 * We can't use page_count without pin a page
99 * because another CPU can free compound page.
100 * This check already skips compound tails of THP
101 * because their page->_refcount is zero at all time.
103 if (!page_ref_count(page)) {
105 pfn += (1 << buddy_order(page)) - 1;
110 * The HWPoisoned page may be not in buddy system, and
111 * page_count() is not 0.
113 if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
117 * We treat all PageOffline() pages as movable when offlining
118 * to give drivers a chance to decrement their reference count
119 * in MEM_GOING_OFFLINE in order to indicate that these pages
120 * can be offlined as there are no direct references anymore.
121 * For actually unmovable PageOffline() where the driver does
122 * not support this, we will fail later when trying to actually
123 * move these pages that still have a reference count > 0.
124 * (false negatives in this function only)
126 if ((flags & MEMORY_OFFLINE) && PageOffline(page))
129 if (__PageMovable(page) || PageLRU(page))
133 * If there are RECLAIMABLE pages, we need to check
134 * it. But now, memory offline itself doesn't call
135 * shrink_node_slabs() and it still to be fixed.
143 * This function set pageblock migratetype to isolate if no unmovable page is
144 * present in [start_pfn, end_pfn). The pageblock must intersect with
145 * [start_pfn, end_pfn).
147 static int set_migratetype_isolate(struct page *page, int migratetype, int isol_flags,
148 unsigned long start_pfn, unsigned long end_pfn)
150 struct zone *zone = page_zone(page);
151 struct page *unmovable;
153 unsigned long check_unmovable_start, check_unmovable_end;
155 spin_lock_irqsave(&zone->lock, flags);
158 * We assume the caller intended to SET migrate type to isolate.
159 * If it is already set, then someone else must have raced and
162 if (is_migrate_isolate_page(page)) {
163 spin_unlock_irqrestore(&zone->lock, flags);
168 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
169 * We just check MOVABLE pages.
171 * Pass the intersection of [start_pfn, end_pfn) and the page's pageblock
172 * to avoid redundant checks.
174 check_unmovable_start = max(page_to_pfn(page), start_pfn);
175 check_unmovable_end = min(pageblock_end_pfn(page_to_pfn(page)),
178 unmovable = has_unmovable_pages(check_unmovable_start, check_unmovable_end,
179 migratetype, isol_flags);
181 unsigned long nr_pages;
182 int mt = get_pageblock_migratetype(page);
184 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
185 zone->nr_isolate_pageblock++;
186 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE,
189 __mod_zone_freepage_state(zone, -nr_pages, mt);
190 spin_unlock_irqrestore(&zone->lock, flags);
194 spin_unlock_irqrestore(&zone->lock, flags);
195 if (isol_flags & REPORT_FAILURE) {
197 * printk() with zone->lock held will likely trigger a
198 * lockdep splat, so defer it here.
200 dump_page(unmovable, "unmovable page");
206 static void unset_migratetype_isolate(struct page *page, int migratetype)
209 unsigned long flags, nr_pages;
210 bool isolated_page = false;
214 zone = page_zone(page);
215 spin_lock_irqsave(&zone->lock, flags);
216 if (!is_migrate_isolate_page(page))
220 * Because freepage with more than pageblock_order on isolated
221 * pageblock is restricted to merge due to freepage counting problem,
222 * it is possible that there is free buddy page.
223 * move_freepages_block() doesn't care of merge so we need other
224 * approach in order to merge them. Isolation and free will make
225 * these pages to be merged.
227 if (PageBuddy(page)) {
228 order = buddy_order(page);
229 if (order >= pageblock_order && order < MAX_ORDER) {
230 buddy = find_buddy_page_pfn(page, page_to_pfn(page),
232 if (buddy && !is_migrate_isolate_page(buddy)) {
233 isolated_page = !!__isolate_free_page(page, order);
235 * Isolating a free page in an isolated pageblock
236 * is expected to always work as watermarks don't
239 VM_WARN_ON(!isolated_page);
245 * If we isolate freepage with more than pageblock_order, there
246 * should be no freepage in the range, so we could avoid costly
247 * pageblock scanning for freepage moving.
249 * We didn't actually touch any of the isolated pages, so place them
250 * to the tail of the freelist. This is an optimization for memory
251 * onlining - just onlined memory won't immediately be considered for
254 if (!isolated_page) {
255 nr_pages = move_freepages_block(zone, page, migratetype, NULL);
256 __mod_zone_freepage_state(zone, nr_pages, migratetype);
258 set_pageblock_migratetype(page, migratetype);
260 __putback_isolated_page(page, order, migratetype);
261 zone->nr_isolate_pageblock--;
263 spin_unlock_irqrestore(&zone->lock, flags);
266 static inline struct page *
267 __first_valid_page(unsigned long pfn, unsigned long nr_pages)
271 for (i = 0; i < nr_pages; i++) {
274 page = pfn_to_online_page(pfn + i);
283 * isolate_single_pageblock() -- tries to isolate a pageblock that might be
284 * within a free or in-use page.
285 * @boundary_pfn: pageblock-aligned pfn that a page might cross
286 * @flags: isolation flags
287 * @gfp_flags: GFP flags used for migrating pages
288 * @isolate_before: isolate the pageblock before the boundary_pfn
289 * @skip_isolation: the flag to skip the pageblock isolation in second
290 * isolate_single_pageblock()
291 * @migratetype: migrate type to set in error recovery.
293 * Free and in-use pages can be as big as MAX_ORDER and contain more than one
294 * pageblock. When not all pageblocks within a page are isolated at the same
295 * time, free page accounting can go wrong. For example, in the case of
296 * MAX_ORDER = pageblock_order + 1, a MAX_ORDER page has two pagelbocks.
298 * [ pageblock0 | pageblock1 ]
299 * When either pageblock is isolated, if it is a free page, the page is not
300 * split into separate migratetype lists, which is supposed to; if it is an
301 * in-use page and freed later, __free_one_page() does not split the free page
302 * either. The function handles this by splitting the free page or migrating
303 * the in-use page then splitting the free page.
305 static int isolate_single_pageblock(unsigned long boundary_pfn, int flags,
306 gfp_t gfp_flags, bool isolate_before, bool skip_isolation,
309 unsigned long start_pfn;
310 unsigned long isolate_pageblock;
315 VM_BUG_ON(!pageblock_aligned(boundary_pfn));
318 isolate_pageblock = boundary_pfn - pageblock_nr_pages;
320 isolate_pageblock = boundary_pfn;
323 * scan at the beginning of MAX_ORDER_NR_PAGES aligned range to avoid
324 * only isolating a subset of pageblocks from a bigger than pageblock
325 * free or in-use page. Also make sure all to-be-isolated pageblocks
326 * are within the same zone.
328 zone = page_zone(pfn_to_page(isolate_pageblock));
329 start_pfn = max(ALIGN_DOWN(isolate_pageblock, MAX_ORDER_NR_PAGES),
330 zone->zone_start_pfn);
332 if (skip_isolation) {
333 int mt __maybe_unused = get_pageblock_migratetype(pfn_to_page(isolate_pageblock));
335 VM_BUG_ON(!is_migrate_isolate(mt));
337 ret = set_migratetype_isolate(pfn_to_page(isolate_pageblock), migratetype,
338 flags, isolate_pageblock, isolate_pageblock + pageblock_nr_pages);
345 * Bail out early when the to-be-isolated pageblock does not form
346 * a free or in-use page across boundary_pfn:
348 * 1. isolate before boundary_pfn: the page after is not online
349 * 2. isolate after boundary_pfn: the page before is not online
351 * This also ensures correctness. Without it, when isolate after
352 * boundary_pfn and [start_pfn, boundary_pfn) are not online,
353 * __first_valid_page() will return unexpected NULL in the for loop
356 if (isolate_before) {
357 if (!pfn_to_online_page(boundary_pfn))
360 if (!pfn_to_online_page(boundary_pfn - 1))
364 for (pfn = start_pfn; pfn < boundary_pfn;) {
365 struct page *page = __first_valid_page(pfn, boundary_pfn - pfn);
368 pfn = page_to_pfn(page);
370 * start_pfn is MAX_ORDER_NR_PAGES aligned, if there is any
371 * free pages in [start_pfn, boundary_pfn), its head page will
372 * always be in the range.
374 if (PageBuddy(page)) {
375 int order = buddy_order(page);
377 if (pfn + (1UL << order) > boundary_pfn) {
378 /* free page changed before split, check it again */
379 if (split_free_page(page, order, boundary_pfn - pfn))
387 * migrate compound pages then let the free page handling code
388 * above do the rest. If migration is not possible, just fail.
390 if (PageCompound(page)) {
391 struct page *head = compound_head(page);
392 unsigned long head_pfn = page_to_pfn(head);
393 unsigned long nr_pages = compound_nr(head);
395 if (head_pfn + nr_pages <= boundary_pfn) {
396 pfn = head_pfn + nr_pages;
399 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
401 * hugetlb, lru compound (THP), and movable compound pages
402 * can be migrated. Otherwise, fail the isolation.
404 if (PageHuge(page) || PageLRU(page) || __PageMovable(page)) {
406 unsigned long outer_pfn;
407 int page_mt = get_pageblock_migratetype(page);
408 bool isolate_page = !is_migrate_isolate_page(page);
409 struct compact_control cc = {
410 .nr_migratepages = 0,
412 .zone = page_zone(pfn_to_page(head_pfn)),
413 .mode = MIGRATE_SYNC,
414 .ignore_skip_hint = true,
415 .no_set_skip_hint = true,
416 .gfp_mask = gfp_flags,
417 .alloc_contig = true,
419 INIT_LIST_HEAD(&cc.migratepages);
422 * XXX: mark the page as MIGRATE_ISOLATE so that
423 * no one else can grab the freed page after migration.
424 * Ideally, the page should be freed as two separate
425 * pages to be added into separate migratetype free
429 ret = set_migratetype_isolate(page, page_mt,
430 flags, head_pfn, head_pfn + nr_pages);
435 ret = __alloc_contig_migrate_range(&cc, head_pfn,
436 head_pfn + nr_pages);
439 * restore the page's migratetype so that it can
440 * be split into separate migratetype free lists
444 unset_migratetype_isolate(page, page_mt);
449 * reset pfn to the head of the free page, so
450 * that the free page handling code above can split
451 * the free page to the right migratetype list.
453 * head_pfn is not used here as a hugetlb page order
454 * can be bigger than MAX_ORDER, but after it is
455 * freed, the free page order is not. Use pfn within
456 * the range to find the head of the free page.
460 while (!PageBuddy(pfn_to_page(outer_pfn))) {
461 /* stop if we cannot find the free page */
462 if (++order > MAX_ORDER)
464 outer_pfn &= ~0UL << order;
477 /* restore the original migratetype */
479 unset_migratetype_isolate(pfn_to_page(isolate_pageblock), migratetype);
484 * start_isolate_page_range() - mark page range MIGRATE_ISOLATE
485 * @start_pfn: The first PFN of the range to be isolated.
486 * @end_pfn: The last PFN of the range to be isolated.
487 * @migratetype: Migrate type to set in error recovery.
488 * @flags: The following flags are allowed (they can be combined in
490 * MEMORY_OFFLINE - isolate to offline (!allocate) memory
491 * e.g., skip over PageHWPoison() pages
492 * and PageOffline() pages.
493 * REPORT_FAILURE - report details about the failure to
495 * @gfp_flags: GFP flags used for migrating pages that sit across the
498 * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
499 * the range will never be allocated. Any free pages and pages freed in the
500 * future will not be allocated again. If specified range includes migrate types
501 * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all
502 * pages in the range finally, the caller have to free all pages in the range.
503 * test_page_isolated() can be used for test it.
505 * The function first tries to isolate the pageblocks at the beginning and end
506 * of the range, since there might be pages across the range boundaries.
507 * Afterwards, it isolates the rest of the range.
509 * There is no high level synchronization mechanism that prevents two threads
510 * from trying to isolate overlapping ranges. If this happens, one thread
511 * will notice pageblocks in the overlapping range already set to isolate.
512 * This happens in set_migratetype_isolate, and set_migratetype_isolate
513 * returns an error. We then clean up by restoring the migration type on
514 * pageblocks we may have modified and return -EBUSY to caller. This
515 * prevents two threads from simultaneously working on overlapping ranges.
517 * Please note that there is no strong synchronization with the page allocator
518 * either. Pages might be freed while their page blocks are marked ISOLATED.
519 * A call to drain_all_pages() after isolation can flush most of them. However
520 * in some cases pages might still end up on pcp lists and that would allow
521 * for their allocation even when they are in fact isolated already. Depending
522 * on how strong of a guarantee the caller needs, zone_pcp_disable/enable()
523 * might be used to flush and disable pcplist before isolation and enable after
526 * Return: 0 on success and -EBUSY if any part of range cannot be isolated.
528 int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
529 int migratetype, int flags, gfp_t gfp_flags)
533 /* isolation is done at page block granularity */
534 unsigned long isolate_start = pageblock_start_pfn(start_pfn);
535 unsigned long isolate_end = pageblock_align(end_pfn);
537 bool skip_isolation = false;
539 /* isolate [isolate_start, isolate_start + pageblock_nr_pages) pageblock */
540 ret = isolate_single_pageblock(isolate_start, flags, gfp_flags, false,
541 skip_isolation, migratetype);
545 if (isolate_start == isolate_end - pageblock_nr_pages)
546 skip_isolation = true;
548 /* isolate [isolate_end - pageblock_nr_pages, isolate_end) pageblock */
549 ret = isolate_single_pageblock(isolate_end, flags, gfp_flags, true,
550 skip_isolation, migratetype);
552 unset_migratetype_isolate(pfn_to_page(isolate_start), migratetype);
556 /* skip isolated pageblocks at the beginning and end */
557 for (pfn = isolate_start + pageblock_nr_pages;
558 pfn < isolate_end - pageblock_nr_pages;
559 pfn += pageblock_nr_pages) {
560 page = __first_valid_page(pfn, pageblock_nr_pages);
561 if (page && set_migratetype_isolate(page, migratetype, flags,
562 start_pfn, end_pfn)) {
563 undo_isolate_page_range(isolate_start, pfn, migratetype);
564 unset_migratetype_isolate(
565 pfn_to_page(isolate_end - pageblock_nr_pages),
574 * undo_isolate_page_range - undo effects of start_isolate_page_range()
575 * @start_pfn: The first PFN of the isolated range
576 * @end_pfn: The last PFN of the isolated range
577 * @migratetype: New migrate type to set on the range
579 * This finds every MIGRATE_ISOLATE page block in the given range
580 * and switches it to @migratetype.
582 void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
587 unsigned long isolate_start = pageblock_start_pfn(start_pfn);
588 unsigned long isolate_end = pageblock_align(end_pfn);
590 for (pfn = isolate_start;
592 pfn += pageblock_nr_pages) {
593 page = __first_valid_page(pfn, pageblock_nr_pages);
594 if (!page || !is_migrate_isolate_page(page))
596 unset_migratetype_isolate(page, migratetype);
600 * Test all pages in the range is free(means isolated) or not.
601 * all pages in [start_pfn...end_pfn) must be in the same zone.
602 * zone->lock must be held before call this.
604 * Returns the last tested pfn.
607 __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
612 while (pfn < end_pfn) {
613 page = pfn_to_page(pfn);
616 * If the page is on a free list, it has to be on
617 * the correct MIGRATE_ISOLATE freelist. There is no
618 * simple way to verify that as VM_BUG_ON(), though.
620 pfn += 1 << buddy_order(page);
621 else if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
622 /* A HWPoisoned page cannot be also PageBuddy */
624 else if ((flags & MEMORY_OFFLINE) && PageOffline(page) &&
627 * The responsible driver agreed to skip PageOffline()
628 * pages when offlining memory by dropping its
629 * reference in MEM_GOING_OFFLINE.
640 * test_pages_isolated - check if pageblocks in range are isolated
641 * @start_pfn: The first PFN of the isolated range
642 * @end_pfn: The first PFN *after* the isolated range
643 * @isol_flags: Testing mode flags
645 * This tests if all in the specified range are free.
647 * If %MEMORY_OFFLINE is specified in @flags, it will consider
648 * poisoned and offlined pages free as well.
650 * Caller must ensure the requested range doesn't span zones.
652 * Returns 0 if true, -EBUSY if one or more pages are in use.
654 int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
657 unsigned long pfn, flags;
663 * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
664 * are not aligned to pageblock_nr_pages.
665 * Then we just check migratetype first.
667 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
668 page = __first_valid_page(pfn, pageblock_nr_pages);
669 if (page && !is_migrate_isolate_page(page))
672 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
673 if ((pfn < end_pfn) || !page) {
678 /* Check all pages are free or marked as ISOLATED */
679 zone = page_zone(page);
680 spin_lock_irqsave(&zone->lock, flags);
681 pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, isol_flags);
682 spin_unlock_irqrestore(&zone->lock, flags);
684 ret = pfn < end_pfn ? -EBUSY : 0;
687 trace_test_pages_isolated(start_pfn, end_pfn, pfn);