blob: 04141a9bea7049513286047fe312161fe522dbc9 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07002/*
3 * linux/mm/page_isolation.c
4 */
5
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07006#include <linux/mm.h>
7#include <linux/page-isolation.h>
8#include <linux/pageblock-flags.h>
Minchan Kimee6f5092012-07-31 16:43:50 -07009#include <linux/memory.h>
Naoya Horiguchic8721bb2013-09-11 14:22:09 -070010#include <linux/hugetlb.h>
Joonsoo Kim83358ec2016-07-26 15:23:43 -070011#include <linux/page_owner.h>
Michal Hocko8b913232017-07-10 15:48:47 -070012#include <linux/migrate.h>
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -070013#include "internal.h"
14
Joonsoo Kim0f0848e2016-01-14 15:18:42 -080015#define CREATE_TRACE_POINTS
16#include <trace/events/page_isolation.h>
17
Zi Yanb48d8a82022-05-12 20:22:57 -070018/*
Zi Yan844fbae2022-05-12 20:22:58 -070019 * 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.
Zi Yanb48d8a82022-05-12 20:22:57 -070022 *
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.
27 *
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.
31 *
32 */
Zi Yan844fbae2022-05-12 20:22:58 -070033static struct page *has_unmovable_pages(unsigned long start_pfn, unsigned long end_pfn,
34 int migratetype, int flags)
Zi Yanb48d8a82022-05-12 20:22:57 -070035{
Zi Yan844fbae2022-05-12 20:22:58 -070036 struct page *page = pfn_to_page(start_pfn);
37 struct zone *zone = page_zone(page);
38 unsigned long pfn;
39
Kefeng Wang4f9bc692022-09-07 14:08:42 +080040 VM_BUG_ON(pageblock_start_pfn(start_pfn) !=
41 pageblock_start_pfn(end_pfn - 1));
Zi Yanb48d8a82022-05-12 20:22:57 -070042
43 if (is_migrate_cma_page(page)) {
44 /*
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.
48 */
49 if (is_migrate_cma(migratetype))
50 return NULL;
51
52 return page;
53 }
54
Zi Yan844fbae2022-05-12 20:22:58 -070055 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
56 page = pfn_to_page(pfn);
Zi Yanb48d8a82022-05-12 20:22:57 -070057
58 /*
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".
63 */
64 if (PageReserved(page))
65 return page;
66
67 /*
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
70 * is movable.
71 */
72 if (zone_idx(zone) == ZONE_MOVABLE)
73 continue;
74
75 /*
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.
80 */
81 if (PageHuge(page) || PageTransCompound(page)) {
82 struct page *head = compound_head(page);
83 unsigned int skip_pages;
84
85 if (PageHuge(page)) {
86 if (!hugepage_migration_supported(page_hstate(head)))
87 return page;
88 } else if (!PageLRU(head) && !__PageMovable(head)) {
89 return page;
90 }
91
92 skip_pages = compound_nr(head) - (page - head);
Zi Yan844fbae2022-05-12 20:22:58 -070093 pfn += skip_pages - 1;
Zi Yanb48d8a82022-05-12 20:22:57 -070094 continue;
95 }
96
97 /*
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.
102 */
103 if (!page_ref_count(page)) {
104 if (PageBuddy(page))
Zi Yan844fbae2022-05-12 20:22:58 -0700105 pfn += (1 << buddy_order(page)) - 1;
Zi Yanb48d8a82022-05-12 20:22:57 -0700106 continue;
107 }
108
109 /*
110 * The HWPoisoned page may be not in buddy system, and
111 * page_count() is not 0.
112 */
113 if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
114 continue;
115
116 /*
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)
125 */
126 if ((flags & MEMORY_OFFLINE) && PageOffline(page))
127 continue;
128
129 if (__PageMovable(page) || PageLRU(page))
130 continue;
131
132 /*
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.
136 */
137 return page;
138 }
139 return NULL;
140}
141
Zi Yan844fbae2022-05-12 20:22:58 -0700142/*
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).
146 */
147static int set_migratetype_isolate(struct page *page, int migratetype, int isol_flags,
148 unsigned long start_pfn, unsigned long end_pfn)
Minchan Kimee6f5092012-07-31 16:43:50 -0700149{
David Hildenbrand1c31cb42020-10-13 16:55:28 -0700150 struct zone *zone = page_zone(page);
151 struct page *unmovable;
David Hildenbrand3f9903b92020-01-30 22:14:01 -0800152 unsigned long flags;
Zi Yan844fbae2022-05-12 20:22:58 -0700153 unsigned long check_unmovable_start, check_unmovable_end;
Minchan Kimee6f5092012-07-31 16:43:50 -0700154
155 spin_lock_irqsave(&zone->lock, flags);
156
Mike Kravetz2c7452a2018-04-05 16:25:26 -0700157 /*
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
David Hildenbrand51030a52020-10-13 16:55:21 -0700160 * set it before us.
Mike Kravetz2c7452a2018-04-05 16:25:26 -0700161 */
David Hildenbrand51030a52020-10-13 16:55:21 -0700162 if (is_migrate_isolate_page(page)) {
163 spin_unlock_irqrestore(&zone->lock, flags);
164 return -EBUSY;
165 }
Mike Kravetz2c7452a2018-04-05 16:25:26 -0700166
Minchan Kimee6f5092012-07-31 16:43:50 -0700167 /*
168 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
169 * We just check MOVABLE pages.
Zi Yan844fbae2022-05-12 20:22:58 -0700170 *
171 * Pass the intersection of [start_pfn, end_pfn) and the page's pageblock
172 * to avoid redundant checks.
Minchan Kimee6f5092012-07-31 16:43:50 -0700173 */
Zi Yan844fbae2022-05-12 20:22:58 -0700174 check_unmovable_start = max(page_to_pfn(page), start_pfn);
Kefeng Wang4f9bc692022-09-07 14:08:42 +0800175 check_unmovable_end = min(pageblock_end_pfn(page_to_pfn(page)),
Zi Yan844fbae2022-05-12 20:22:58 -0700176 end_pfn);
177
178 unmovable = has_unmovable_pages(check_unmovable_start, check_unmovable_end,
179 migratetype, isol_flags);
Qian Cai4a55c042020-01-30 22:14:57 -0800180 if (!unmovable) {
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -0700181 unsigned long nr_pages;
Michal Hocko4da2ce22017-11-15 17:33:26 -0800182 int mt = get_pageblock_migratetype(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -0700183
Bartlomiej Zolnierkiewicza4584312013-01-04 15:35:08 -0800184 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
Joonsoo Kimad53f922014-11-13 15:19:11 -0800185 zone->nr_isolate_pageblock++;
Vlastimil Babka02aa0cd2017-05-08 15:54:40 -0700186 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE,
187 NULL);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -0700188
Michal Hocko4da2ce22017-11-15 17:33:26 -0800189 __mod_zone_freepage_state(zone, -nr_pages, mt);
David Hildenbrand1c31cb42020-10-13 16:55:28 -0700190 spin_unlock_irqrestore(&zone->lock, flags);
David Hildenbrand1c31cb42020-10-13 16:55:28 -0700191 return 0;
Minchan Kimee6f5092012-07-31 16:43:50 -0700192 }
193
194 spin_unlock_irqrestore(&zone->lock, flags);
David Hildenbrand1c31cb42020-10-13 16:55:28 -0700195 if (isol_flags & REPORT_FAILURE) {
David Hildenbrand48381d72020-10-13 16:55:24 -0700196 /*
197 * printk() with zone->lock held will likely trigger a
198 * lockdep splat, so defer it here.
199 */
200 dump_page(unmovable, "unmovable page");
Qian Cai3d680bd2020-01-30 22:15:01 -0800201 }
Qian Cai4a55c042020-01-30 22:14:57 -0800202
David Hildenbrand1c31cb42020-10-13 16:55:28 -0700203 return -EBUSY;
Minchan Kimee6f5092012-07-31 16:43:50 -0700204}
205
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700206static void unset_migratetype_isolate(struct page *page, int migratetype)
Minchan Kimee6f5092012-07-31 16:43:50 -0700207{
208 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -0700209 unsigned long flags, nr_pages;
Joonsoo Kime3a27132016-07-26 15:24:01 -0700210 bool isolated_page = false;
Joonsoo Kim3c605092014-11-13 15:19:21 -0800211 unsigned int order;
Joonsoo Kim3c605092014-11-13 15:19:21 -0800212 struct page *buddy;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -0700213
Minchan Kimee6f5092012-07-31 16:43:50 -0700214 zone = page_zone(page);
215 spin_lock_irqsave(&zone->lock, flags);
Xishi Qiubbf9ce92017-05-03 14:52:55 -0700216 if (!is_migrate_isolate_page(page))
Minchan Kimee6f5092012-07-31 16:43:50 -0700217 goto out;
Joonsoo Kim3c605092014-11-13 15:19:21 -0800218
219 /*
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.
226 */
227 if (PageBuddy(page)) {
Matthew Wilcox (Oracle)ab130f912020-10-15 20:10:15 -0700228 order = buddy_order(page);
Muchun Song2484be02020-12-14 19:12:27 -0800229 if (order >= pageblock_order && order < MAX_ORDER - 1) {
Zi Yan8170ac42022-04-28 23:16:01 -0700230 buddy = find_buddy_page_pfn(page, page_to_pfn(page),
231 order, NULL);
232 if (buddy && !is_migrate_isolate_page(buddy)) {
Miaohe Lina500cb32021-11-05 13:42:19 -0700233 isolated_page = !!__isolate_free_page(page, order);
234 /*
235 * Isolating a free page in an isolated pageblock
236 * is expected to always work as watermarks don't
237 * apply here.
238 */
239 VM_WARN_ON(!isolated_page);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800240 }
241 }
242 }
243
244 /*
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.
David Hildenbrand293ffa52020-10-15 20:09:30 -0700248 *
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
252 * allocation.
Joonsoo Kim3c605092014-11-13 15:19:21 -0800253 */
Chen Wanduna85468b2022-02-03 20:49:06 -0800254 if (!isolated_page) {
Vlastimil Babka02aa0cd2017-05-08 15:54:40 -0700255 nr_pages = move_freepages_block(zone, page, migratetype, NULL);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800256 __mod_zone_freepage_state(zone, nr_pages, migratetype);
257 }
Bartlomiej Zolnierkiewicza4584312013-01-04 15:35:08 -0800258 set_pageblock_migratetype(page, migratetype);
Alexander Duyck624f58d2020-04-06 20:04:53 -0700259 if (isolated_page)
260 __putback_isolated_page(page, order, migratetype);
Joonsoo Kimad53f922014-11-13 15:19:11 -0800261 zone->nr_isolate_pageblock--;
Minchan Kimee6f5092012-07-31 16:43:50 -0700262out:
263 spin_unlock_irqrestore(&zone->lock, flags);
264}
265
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700266static inline struct page *
267__first_valid_page(unsigned long pfn, unsigned long nr_pages)
268{
269 int i;
Michal Hocko2ce13642017-07-06 15:38:04 -0700270
271 for (i = 0; i < nr_pages; i++) {
272 struct page *page;
273
Michal Hocko2ce13642017-07-06 15:38:04 -0700274 page = pfn_to_online_page(pfn + i);
275 if (!page)
276 continue;
277 return page;
278 }
279 return NULL;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700280}
281
Qian Cai9b7ea462019-03-28 20:43:34 -0700282/**
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700283 * 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
Zi Yan88ee1342022-05-24 15:47:56 -0400286 * @flags: isolation flags
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700287 * @gfp_flags: GFP flags used for migrating pages
288 * @isolate_before: isolate the pageblock before the boundary_pfn
Yang Li04299932022-06-02 14:21:16 +0800289 * @skip_isolation: the flag to skip the pageblock isolation in second
290 * isolate_single_pageblock()
Zi Yan80e2b582022-09-13 22:39:13 -0400291 * @migratetype: migrate type to set in error recovery.
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700292 *
293 * Free and in-use pages can be as big as MAX_ORDER-1 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-1 = pageblock_order + 1, a MAX_ORDER-1 page has two pagelbocks.
297 * [ MAX_ORDER-1 ]
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.
304 */
Zi Yan88ee1342022-05-24 15:47:56 -0400305static int isolate_single_pageblock(unsigned long boundary_pfn, int flags,
Zi Yan80e2b582022-09-13 22:39:13 -0400306 gfp_t gfp_flags, bool isolate_before, bool skip_isolation,
307 int migratetype)
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700308{
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700309 unsigned long start_pfn;
310 unsigned long isolate_pageblock;
311 unsigned long pfn;
312 struct zone *zone;
Zi Yan88ee1342022-05-24 15:47:56 -0400313 int ret;
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700314
Kefeng Wangee0913c2022-09-07 14:08:44 +0800315 VM_BUG_ON(!pageblock_aligned(boundary_pfn));
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700316
317 if (isolate_before)
318 isolate_pageblock = boundary_pfn - pageblock_nr_pages;
319 else
320 isolate_pageblock = boundary_pfn;
321
322 /*
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.
327 */
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);
331
Zi Yan80e2b582022-09-13 22:39:13 -0400332 if (skip_isolation) {
333 int mt = get_pageblock_migratetype(pfn_to_page(isolate_pageblock));
Zi Yan88ee1342022-05-24 15:47:56 -0400334
Zi Yan80e2b582022-09-13 22:39:13 -0400335 VM_BUG_ON(!is_migrate_isolate(mt));
336 } else {
337 ret = set_migratetype_isolate(pfn_to_page(isolate_pageblock), migratetype,
338 flags, isolate_pageblock, isolate_pageblock + pageblock_nr_pages);
Zi Yan9b209e52022-05-26 19:15:30 -0400339
340 if (ret)
341 return ret;
342 }
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700343
344 /*
345 * Bail out early when the to-be-isolated pageblock does not form
346 * a free or in-use page across boundary_pfn:
347 *
348 * 1. isolate before boundary_pfn: the page after is not online
349 * 2. isolate after boundary_pfn: the page before is not online
350 *
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
354 * below.
355 */
356 if (isolate_before) {
357 if (!pfn_to_online_page(boundary_pfn))
358 return 0;
359 } else {
360 if (!pfn_to_online_page(boundary_pfn - 1))
361 return 0;
362 }
363
364 for (pfn = start_pfn; pfn < boundary_pfn;) {
365 struct page *page = __first_valid_page(pfn, boundary_pfn - pfn);
366
367 VM_BUG_ON(!page);
368 pfn = page_to_pfn(page);
369 /*
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.
373 */
374 if (PageBuddy(page)) {
375 int order = buddy_order(page);
376
Zi Yan86d28b072022-05-26 19:15:31 -0400377 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))
380 continue;
381 }
382
383 pfn += 1UL << order;
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700384 continue;
385 }
386 /*
387 * migrate compound pages then let the free page handling code
388 * above do the rest. If migration is not possible, just fail.
389 */
390 if (PageCompound(page)) {
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700391 struct page *head = compound_head(page);
392 unsigned long head_pfn = page_to_pfn(head);
Zi Yan547be962022-05-30 22:44:50 -0400393 unsigned long nr_pages = compound_nr(head);
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700394
Zi Yan88ee1342022-05-24 15:47:56 -0400395 if (head_pfn + nr_pages <= boundary_pfn) {
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700396 pfn = head_pfn + nr_pages;
397 continue;
398 }
399#if defined CONFIG_COMPACTION || defined CONFIG_CMA
400 /*
401 * hugetlb, lru compound (THP), and movable compound pages
402 * can be migrated. Otherwise, fail the isolation.
403 */
404 if (PageHuge(page) || PageLRU(page) || __PageMovable(page)) {
405 int order;
406 unsigned long outer_pfn;
Zi Yan88ee1342022-05-24 15:47:56 -0400407 int page_mt = get_pageblock_migratetype(page);
408 bool isolate_page = !is_migrate_isolate_page(page);
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700409 struct compact_control cc = {
410 .nr_migratepages = 0,
411 .order = -1,
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,
418 };
419 INIT_LIST_HEAD(&cc.migratepages);
420
Zi Yan88ee1342022-05-24 15:47:56 -0400421 /*
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
426 * lists.
427 */
428 if (isolate_page) {
429 ret = set_migratetype_isolate(page, page_mt,
430 flags, head_pfn, head_pfn + nr_pages);
431 if (ret)
432 goto failed;
433 }
434
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700435 ret = __alloc_contig_migrate_range(&cc, head_pfn,
436 head_pfn + nr_pages);
437
Zi Yan88ee1342022-05-24 15:47:56 -0400438 /*
439 * restore the page's migratetype so that it can
440 * be split into separate migratetype free lists
441 * later.
442 */
443 if (isolate_page)
444 unset_migratetype_isolate(page, page_mt);
445
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700446 if (ret)
447 goto failed;
448 /*
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.
452 *
453 * head_pfn is not used here as a hugetlb page order
454 * can be bigger than MAX_ORDER-1, 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.
457 */
458 order = 0;
459 outer_pfn = pfn;
460 while (!PageBuddy(pfn_to_page(outer_pfn))) {
Zi Yan88ee1342022-05-24 15:47:56 -0400461 /* stop if we cannot find the free page */
462 if (++order >= MAX_ORDER)
463 goto failed;
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700464 outer_pfn &= ~0UL << order;
465 }
466 pfn = outer_pfn;
467 continue;
468 } else
469#endif
470 goto failed;
471 }
472
473 pfn++;
474 }
475 return 0;
476failed:
477 /* restore the original migratetype */
Zi Yan9b209e52022-05-26 19:15:30 -0400478 if (!skip_isolation)
Zi Yan80e2b582022-09-13 22:39:13 -0400479 unset_migratetype_isolate(pfn_to_page(isolate_pageblock), migratetype);
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700480 return -EBUSY;
481}
482
483/**
Qian Cai9b7ea462019-03-28 20:43:34 -0700484 * start_isolate_page_range() - make page-allocation-type of range of pages to
485 * be MIGRATE_ISOLATE.
486 * @start_pfn: The lower PFN of the range to be isolated.
487 * @end_pfn: The upper PFN of the range to be isolated.
Qian Cai9b7ea462019-03-28 20:43:34 -0700488 * @migratetype: Migrate type to set in error recovery.
489 * @flags: The following flags are allowed (they can be combined in
490 * a bit mask)
David Hildenbrand756d25b2019-11-30 17:54:07 -0800491 * MEMORY_OFFLINE - isolate to offline (!allocate) memory
492 * e.g., skip over PageHWPoison() pages
David Hildenbrandaa2187952020-05-07 16:01:30 +0200493 * and PageOffline() pages.
Qian Cai9b7ea462019-03-28 20:43:34 -0700494 * REPORT_FAILURE - report details about the failure to
495 * isolate the range
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700496 * @gfp_flags: GFP flags used for migrating pages that sit across the
497 * range boundaries.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700498 *
499 * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
500 * the range will never be allocated. Any free pages and pages freed in the
Qian Cai9b7ea462019-03-28 20:43:34 -0700501 * future will not be allocated again. If specified range includes migrate types
502 * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all
503 * pages in the range finally, the caller have to free all pages in the range.
504 * test_page_isolated() can be used for test it.
Mike Kravetz2c7452a2018-04-05 16:25:26 -0700505 *
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700506 * The function first tries to isolate the pageblocks at the beginning and end
507 * of the range, since there might be pages across the range boundaries.
508 * Afterwards, it isolates the rest of the range.
509 *
Mike Kravetz2c7452a2018-04-05 16:25:26 -0700510 * There is no high level synchronization mechanism that prevents two threads
Qian Cai9b7ea462019-03-28 20:43:34 -0700511 * from trying to isolate overlapping ranges. If this happens, one thread
Mike Kravetz2c7452a2018-04-05 16:25:26 -0700512 * will notice pageblocks in the overlapping range already set to isolate.
513 * This happens in set_migratetype_isolate, and set_migratetype_isolate
Qian Cai9b7ea462019-03-28 20:43:34 -0700514 * returns an error. We then clean up by restoring the migration type on
515 * pageblocks we may have modified and return -EBUSY to caller. This
Mike Kravetz2c7452a2018-04-05 16:25:26 -0700516 * prevents two threads from simultaneously working on overlapping ranges.
Qian Cai9b7ea462019-03-28 20:43:34 -0700517 *
Pavel Tatashin96831822020-09-18 21:20:31 -0700518 * Please note that there is no strong synchronization with the page allocator
519 * either. Pages might be freed while their page blocks are marked ISOLATED.
Vlastimil Babka76129212020-12-14 19:10:56 -0800520 * A call to drain_all_pages() after isolation can flush most of them. However
521 * in some cases pages might still end up on pcp lists and that would allow
Pavel Tatashin96831822020-09-18 21:20:31 -0700522 * for their allocation even when they are in fact isolated already. Depending
Vlastimil Babkaec6e8c7e2020-12-14 19:10:59 -0800523 * on how strong of a guarantee the caller needs, zone_pcp_disable/enable()
524 * might be used to flush and disable pcplist before isolation and enable after
525 * unisolation.
Pavel Tatashin96831822020-09-18 21:20:31 -0700526 *
David Hildenbrand3fa0c7c2020-10-15 20:08:07 -0700527 * Return: 0 on success and -EBUSY if any part of range cannot be isolated.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700528 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +0200529int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700530 int migratetype, int flags, gfp_t gfp_flags)
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700531{
532 unsigned long pfn;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700533 struct page *page;
Zi Yan6e263ff2022-05-12 20:22:58 -0700534 /* isolation is done at page block granularity */
Kefeng Wang4f9bc692022-09-07 14:08:42 +0800535 unsigned long isolate_start = pageblock_start_pfn(start_pfn);
Kefeng Wang5f7fa132022-09-07 14:08:43 +0800536 unsigned long isolate_end = pageblock_align(end_pfn);
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700537 int ret;
Zi Yan9b209e52022-05-26 19:15:30 -0400538 bool skip_isolation = false;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700539
Zi Yan6e263ff2022-05-12 20:22:58 -0700540 /* isolate [isolate_start, isolate_start + pageblock_nr_pages) pageblock */
Zi Yan80e2b582022-09-13 22:39:13 -0400541 ret = isolate_single_pageblock(isolate_start, flags, gfp_flags, false,
542 skip_isolation, migratetype);
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700543 if (ret)
544 return ret;
545
Zi Yan9b209e52022-05-26 19:15:30 -0400546 if (isolate_start == isolate_end - pageblock_nr_pages)
547 skip_isolation = true;
548
Zi Yan6e263ff2022-05-12 20:22:58 -0700549 /* isolate [isolate_end - pageblock_nr_pages, isolate_end) pageblock */
Zi Yan80e2b582022-09-13 22:39:13 -0400550 ret = isolate_single_pageblock(isolate_end, flags, gfp_flags, true,
551 skip_isolation, migratetype);
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700552 if (ret) {
Zi Yan6e263ff2022-05-12 20:22:58 -0700553 unset_migratetype_isolate(pfn_to_page(isolate_start), migratetype);
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700554 return ret;
555 }
556
557 /* skip isolated pageblocks at the beginning and end */
Zi Yan6e263ff2022-05-12 20:22:58 -0700558 for (pfn = isolate_start + pageblock_nr_pages;
559 pfn < isolate_end - pageblock_nr_pages;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700560 pfn += pageblock_nr_pages) {
561 page = __first_valid_page(pfn, pageblock_nr_pages);
Zi Yan844fbae2022-05-12 20:22:58 -0700562 if (page && set_migratetype_isolate(page, migratetype, flags,
563 start_pfn, end_pfn)) {
Zi Yan6e263ff2022-05-12 20:22:58 -0700564 undo_isolate_page_range(isolate_start, pfn, migratetype);
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700565 unset_migratetype_isolate(
Zi Yan6e263ff2022-05-12 20:22:58 -0700566 pfn_to_page(isolate_end - pageblock_nr_pages),
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700567 migratetype);
Miaohe Line1d8c962021-11-05 13:42:16 -0700568 return -EBUSY;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700569 }
570 }
David Hildenbrand3fa0c7c2020-10-15 20:08:07 -0700571 return 0;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700572}
573
574/*
575 * Make isolated pages available again.
576 */
Pingfan Liu1fcf0a52019-07-11 20:54:49 -0700577void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
Zi Yanb2c9e2f2022-05-12 20:22:58 -0700578 int migratetype)
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700579{
580 unsigned long pfn;
581 struct page *page;
Kefeng Wang4f9bc692022-09-07 14:08:42 +0800582 unsigned long isolate_start = pageblock_start_pfn(start_pfn);
Kefeng Wang5f7fa132022-09-07 14:08:43 +0800583 unsigned long isolate_end = pageblock_align(end_pfn);
Wang Xiaoqiang6f8d2b82016-01-15 16:57:13 -0800584
Zi Yan6e263ff2022-05-12 20:22:58 -0700585 for (pfn = isolate_start;
586 pfn < isolate_end;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700587 pfn += pageblock_nr_pages) {
588 page = __first_valid_page(pfn, pageblock_nr_pages);
Xishi Qiubbf9ce92017-05-03 14:52:55 -0700589 if (!page || !is_migrate_isolate_page(page))
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700590 continue;
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +0200591 unset_migratetype_isolate(page, migratetype);
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700592 }
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700593}
594/*
595 * Test all pages in the range is free(means isolated) or not.
596 * all pages in [start_pfn...end_pfn) must be in the same zone.
597 * zone->lock must be held before call this.
598 *
Neil Zhangec3b6882016-04-01 14:31:37 -0700599 * Returns the last tested pfn.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700600 */
Joonsoo Kimfea85cf2016-01-14 15:18:39 -0800601static unsigned long
Wen Congyangb023f462012-12-11 16:00:45 -0800602__test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
David Hildenbrand756d25b2019-11-30 17:54:07 -0800603 int flags)
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700604{
605 struct page *page;
606
607 while (pfn < end_pfn) {
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700608 page = pfn_to_page(pfn);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -0700609 if (PageBuddy(page))
Minchan Kim435b4052012-10-08 16:32:16 -0700610 /*
Vlastimil Babkaaa016d12015-09-08 15:01:22 -0700611 * If the page is on a free list, it has to be on
612 * the correct MIGRATE_ISOLATE freelist. There is no
613 * simple way to verify that as VM_BUG_ON(), though.
Minchan Kim435b4052012-10-08 16:32:16 -0700614 */
Matthew Wilcox (Oracle)ab130f912020-10-15 20:10:15 -0700615 pfn += 1 << buddy_order(page);
David Hildenbrand756d25b2019-11-30 17:54:07 -0800616 else if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
Vlastimil Babkaaa016d12015-09-08 15:01:22 -0700617 /* A HWPoisoned page cannot be also PageBuddy */
Wen Congyangb023f462012-12-11 16:00:45 -0800618 pfn++;
David Hildenbrandaa2187952020-05-07 16:01:30 +0200619 else if ((flags & MEMORY_OFFLINE) && PageOffline(page) &&
620 !page_count(page))
621 /*
622 * The responsible driver agreed to skip PageOffline()
623 * pages when offlining memory by dropping its
624 * reference in MEM_GOING_OFFLINE.
625 */
626 pfn++;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700627 else
628 break;
629 }
Joonsoo Kimfea85cf2016-01-14 15:18:39 -0800630
631 return pfn;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700632}
633
Joonsoo Kimb9eb6312016-05-19 17:12:06 -0700634/* Caller should ensure that requested range is in a single zone */
Wen Congyangb023f462012-12-11 16:00:45 -0800635int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
David Hildenbrand756d25b2019-11-30 17:54:07 -0800636 int isol_flags)
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700637{
Gerald Schaefer6c1b7f62008-10-02 14:50:16 -0700638 unsigned long pfn, flags;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700639 struct page *page;
Gerald Schaefer6c1b7f62008-10-02 14:50:16 -0700640 struct zone *zone;
George G. Davis1d095102021-09-02 14:58:16 -0700641 int ret;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700642
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700643 /*
Tang Chen85dbe702013-06-20 18:10:19 +0800644 * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
645 * are not aligned to pageblock_nr_pages.
646 * Then we just check migratetype first.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700647 */
648 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
649 page = __first_valid_page(pfn, pageblock_nr_pages);
Xishi Qiubbf9ce92017-05-03 14:52:55 -0700650 if (page && !is_migrate_isolate_page(page))
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700651 break;
652 }
Gerald Schaefera70dcb92008-11-06 12:53:36 -0800653 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
George G. Davis1d095102021-09-02 14:58:16 -0700654 if ((pfn < end_pfn) || !page) {
655 ret = -EBUSY;
656 goto out;
657 }
658
Tang Chen85dbe702013-06-20 18:10:19 +0800659 /* Check all pages are free or marked as ISOLATED */
Gerald Schaefera70dcb92008-11-06 12:53:36 -0800660 zone = page_zone(page);
Gerald Schaefer6c1b7f62008-10-02 14:50:16 -0700661 spin_lock_irqsave(&zone->lock, flags);
David Hildenbrand756d25b2019-11-30 17:54:07 -0800662 pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, isol_flags);
Gerald Schaefer6c1b7f62008-10-02 14:50:16 -0700663 spin_unlock_irqrestore(&zone->lock, flags);
Joonsoo Kimfea85cf2016-01-14 15:18:39 -0800664
George G. Davis1d095102021-09-02 14:58:16 -0700665 ret = pfn < end_pfn ? -EBUSY : 0;
666
667out:
Joonsoo Kim0f0848e2016-01-14 15:18:42 -0800668 trace_test_pages_isolated(start_pfn, end_pfn, pfn);
669
George G. Davis1d095102021-09-02 14:58:16 -0700670 return ret;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700671}