blob: ce323e56b34d6bc43a9e16cc9054f35290bb0b63 [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
Michal Hockod381c542018-12-28 00:33:56 -080018static int set_migratetype_isolate(struct page *page, int migratetype, int isol_flags)
Minchan Kimee6f5092012-07-31 16:43:50 -070019{
20 struct zone *zone;
21 unsigned long flags, pfn;
22 struct memory_isolate_notify arg;
23 int notifier_ret;
24 int ret = -EBUSY;
25
26 zone = page_zone(page);
27
28 spin_lock_irqsave(&zone->lock, flags);
29
Mike Kravetz2c7452a2018-04-05 16:25:26 -070030 /*
31 * We assume the caller intended to SET migrate type to isolate.
32 * If it is already set, then someone else must have raced and
33 * set it before us. Return -EBUSY
34 */
35 if (is_migrate_isolate_page(page))
36 goto out;
37
Minchan Kimee6f5092012-07-31 16:43:50 -070038 pfn = page_to_pfn(page);
39 arg.start_pfn = pfn;
40 arg.nr_pages = pageblock_nr_pages;
41 arg.pages_found = 0;
42
43 /*
44 * It may be possible to isolate a pageblock even if the
45 * migratetype is not MIGRATE_MOVABLE. The memory isolation
46 * notifier chain is used by balloon drivers to return the
47 * number of pages in a range that are held by the balloon
48 * driver to shrink memory. If all the pages are accounted for
49 * by balloons, are free, or on the LRU, isolation can continue.
50 * Later, for example, when memory hotplug notifier runs, these
51 * pages reported as "can be isolated" should be isolated(freed)
52 * by the balloon driver through the memory notifier chain.
53 */
54 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
55 notifier_ret = notifier_to_errno(notifier_ret);
56 if (notifier_ret)
57 goto out;
58 /*
59 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
60 * We just check MOVABLE pages.
61 */
Michal Hockod381c542018-12-28 00:33:56 -080062 if (!has_unmovable_pages(zone, page, arg.pages_found, migratetype, flags))
Minchan Kimee6f5092012-07-31 16:43:50 -070063 ret = 0;
64
65 /*
Yisheng Xieac34dcd2016-10-07 17:01:16 -070066 * immobile means "not-on-lru" pages. If immobile is larger than
Minchan Kimee6f5092012-07-31 16:43:50 -070067 * removable-by-driver pages reported by notifier, we'll fail.
68 */
69
70out:
71 if (!ret) {
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -070072 unsigned long nr_pages;
Michal Hocko4da2ce22017-11-15 17:33:26 -080073 int mt = get_pageblock_migratetype(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -070074
Bartlomiej Zolnierkiewicza4584312013-01-04 15:35:08 -080075 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
Joonsoo Kimad53f922014-11-13 15:19:11 -080076 zone->nr_isolate_pageblock++;
Vlastimil Babka02aa0cd2017-05-08 15:54:40 -070077 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE,
78 NULL);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -070079
Michal Hocko4da2ce22017-11-15 17:33:26 -080080 __mod_zone_freepage_state(zone, -nr_pages, mt);
Minchan Kimee6f5092012-07-31 16:43:50 -070081 }
82
83 spin_unlock_irqrestore(&zone->lock, flags);
84 if (!ret)
Vlastimil Babkaec25af82014-12-10 15:43:04 -080085 drain_all_pages(zone);
Minchan Kimee6f5092012-07-31 16:43:50 -070086 return ret;
87}
88
Naoya Horiguchic5b4e1b2015-09-08 15:02:09 -070089static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
Minchan Kimee6f5092012-07-31 16:43:50 -070090{
91 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -070092 unsigned long flags, nr_pages;
Joonsoo Kime3a27132016-07-26 15:24:01 -070093 bool isolated_page = false;
Joonsoo Kim3c605092014-11-13 15:19:21 -080094 unsigned int order;
Vlastimil Babka76741e72017-02-22 15:41:48 -080095 unsigned long pfn, buddy_pfn;
Joonsoo Kim3c605092014-11-13 15:19:21 -080096 struct page *buddy;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -070097
Minchan Kimee6f5092012-07-31 16:43:50 -070098 zone = page_zone(page);
99 spin_lock_irqsave(&zone->lock, flags);
Xishi Qiubbf9ce92017-05-03 14:52:55 -0700100 if (!is_migrate_isolate_page(page))
Minchan Kimee6f5092012-07-31 16:43:50 -0700101 goto out;
Joonsoo Kim3c605092014-11-13 15:19:21 -0800102
103 /*
104 * Because freepage with more than pageblock_order on isolated
105 * pageblock is restricted to merge due to freepage counting problem,
106 * it is possible that there is free buddy page.
107 * move_freepages_block() doesn't care of merge so we need other
108 * approach in order to merge them. Isolation and free will make
109 * these pages to be merged.
110 */
111 if (PageBuddy(page)) {
112 order = page_order(page);
113 if (order >= pageblock_order) {
Vlastimil Babka76741e72017-02-22 15:41:48 -0800114 pfn = page_to_pfn(page);
115 buddy_pfn = __find_buddy_pfn(pfn, order);
116 buddy = page + (buddy_pfn - pfn);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800117
Vlastimil Babka13ad59d2017-02-22 15:41:51 -0800118 if (pfn_valid_within(buddy_pfn) &&
Hui Zhu1ae70132015-05-14 15:17:04 -0700119 !is_migrate_isolate_page(buddy)) {
Joonsoo Kim3c605092014-11-13 15:19:21 -0800120 __isolate_free_page(page, order);
Joonsoo Kime3a27132016-07-26 15:24:01 -0700121 isolated_page = true;
Joonsoo Kim3c605092014-11-13 15:19:21 -0800122 }
123 }
124 }
125
126 /*
127 * If we isolate freepage with more than pageblock_order, there
128 * should be no freepage in the range, so we could avoid costly
129 * pageblock scanning for freepage moving.
130 */
131 if (!isolated_page) {
Vlastimil Babka02aa0cd2017-05-08 15:54:40 -0700132 nr_pages = move_freepages_block(zone, page, migratetype, NULL);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800133 __mod_zone_freepage_state(zone, nr_pages, migratetype);
134 }
Bartlomiej Zolnierkiewicza4584312013-01-04 15:35:08 -0800135 set_pageblock_migratetype(page, migratetype);
Joonsoo Kimad53f922014-11-13 15:19:11 -0800136 zone->nr_isolate_pageblock--;
Minchan Kimee6f5092012-07-31 16:43:50 -0700137out:
138 spin_unlock_irqrestore(&zone->lock, flags);
Joonsoo Kim83358ec2016-07-26 15:23:43 -0700139 if (isolated_page) {
Joonsoo Kim46f24fd2016-07-26 15:23:58 -0700140 post_alloc_hook(page, order, __GFP_MOVABLE);
Joonsoo Kime3a27132016-07-26 15:24:01 -0700141 __free_pages(page, order);
Joonsoo Kim83358ec2016-07-26 15:23:43 -0700142 }
Minchan Kimee6f5092012-07-31 16:43:50 -0700143}
144
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700145static inline struct page *
146__first_valid_page(unsigned long pfn, unsigned long nr_pages)
147{
148 int i;
Michal Hocko2ce13642017-07-06 15:38:04 -0700149
150 for (i = 0; i < nr_pages; i++) {
151 struct page *page;
152
153 if (!pfn_valid_within(pfn + i))
154 continue;
155 page = pfn_to_online_page(pfn + i);
156 if (!page)
157 continue;
158 return page;
159 }
160 return NULL;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700161}
162
163/*
164 * start_isolate_page_range() -- make page-allocation-type of range of pages
165 * to be MIGRATE_ISOLATE.
166 * @start_pfn: The lower PFN of the range to be isolated.
167 * @end_pfn: The upper PFN of the range to be isolated.
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +0200168 * @migratetype: migrate type to set in error recovery.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700169 *
170 * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
171 * the range will never be allocated. Any free pages and pages freed in the
172 * future will not be allocated again.
173 *
174 * start_pfn/end_pfn must be aligned to pageblock_order.
Mike Kravetz2c7452a2018-04-05 16:25:26 -0700175 * Return 0 on success and -EBUSY if any part of range cannot be isolated.
176 *
177 * There is no high level synchronization mechanism that prevents two threads
178 * from trying to isolate overlapping ranges. If this happens, one thread
179 * will notice pageblocks in the overlapping range already set to isolate.
180 * This happens in set_migratetype_isolate, and set_migratetype_isolate
181 * returns an error. We then clean up by restoring the migration type on
182 * pageblocks we may have modified and return -EBUSY to caller. This
183 * prevents two threads from simultaneously working on overlapping ranges.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700184 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +0200185int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
Michal Hockod381c542018-12-28 00:33:56 -0800186 unsigned migratetype, int flags)
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700187{
188 unsigned long pfn;
189 unsigned long undo_pfn;
190 struct page *page;
191
Naoya Horiguchifec174d2016-01-14 15:22:13 -0800192 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
193 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700194
195 for (pfn = start_pfn;
196 pfn < end_pfn;
197 pfn += pageblock_nr_pages) {
198 page = __first_valid_page(pfn, pageblock_nr_pages);
Wen Congyangb023f462012-12-11 16:00:45 -0800199 if (page &&
Michal Hockod381c542018-12-28 00:33:56 -0800200 set_migratetype_isolate(page, migratetype, flags)) {
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700201 undo_pfn = pfn;
202 goto undo;
203 }
204 }
205 return 0;
206undo:
207 for (pfn = start_pfn;
KAMEZAWA Hiroyukidbc0e4c2007-11-14 16:59:12 -0800208 pfn < undo_pfn;
Michal Hocko2ce13642017-07-06 15:38:04 -0700209 pfn += pageblock_nr_pages) {
210 struct page *page = pfn_to_online_page(pfn);
211 if (!page)
212 continue;
213 unset_migratetype_isolate(page, migratetype);
214 }
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700215
216 return -EBUSY;
217}
218
219/*
220 * Make isolated pages available again.
221 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +0200222int undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
223 unsigned migratetype)
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700224{
225 unsigned long pfn;
226 struct page *page;
Wang Xiaoqiang6f8d2b82016-01-15 16:57:13 -0800227
228 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
229 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
230
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700231 for (pfn = start_pfn;
232 pfn < end_pfn;
233 pfn += pageblock_nr_pages) {
234 page = __first_valid_page(pfn, pageblock_nr_pages);
Xishi Qiubbf9ce92017-05-03 14:52:55 -0700235 if (!page || !is_migrate_isolate_page(page))
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700236 continue;
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +0200237 unset_migratetype_isolate(page, migratetype);
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700238 }
239 return 0;
240}
241/*
242 * Test all pages in the range is free(means isolated) or not.
243 * all pages in [start_pfn...end_pfn) must be in the same zone.
244 * zone->lock must be held before call this.
245 *
Neil Zhangec3b6882016-04-01 14:31:37 -0700246 * Returns the last tested pfn.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700247 */
Joonsoo Kimfea85cf2016-01-14 15:18:39 -0800248static unsigned long
Wen Congyangb023f462012-12-11 16:00:45 -0800249__test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
250 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700251{
252 struct page *page;
253
254 while (pfn < end_pfn) {
255 if (!pfn_valid_within(pfn)) {
256 pfn++;
257 continue;
258 }
259 page = pfn_to_page(pfn);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -0700260 if (PageBuddy(page))
Minchan Kim435b4052012-10-08 16:32:16 -0700261 /*
Vlastimil Babkaaa016d12015-09-08 15:01:22 -0700262 * If the page is on a free list, it has to be on
263 * the correct MIGRATE_ISOLATE freelist. There is no
264 * simple way to verify that as VM_BUG_ON(), though.
Minchan Kim435b4052012-10-08 16:32:16 -0700265 */
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700266 pfn += 1 << page_order(page);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -0700267 else if (skip_hwpoisoned_pages && PageHWPoison(page))
268 /* A HWPoisoned page cannot be also PageBuddy */
Wen Congyangb023f462012-12-11 16:00:45 -0800269 pfn++;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700270 else
271 break;
272 }
Joonsoo Kimfea85cf2016-01-14 15:18:39 -0800273
274 return pfn;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700275}
276
Joonsoo Kimb9eb6312016-05-19 17:12:06 -0700277/* Caller should ensure that requested range is in a single zone */
Wen Congyangb023f462012-12-11 16:00:45 -0800278int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
279 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700280{
Gerald Schaefer6c1b7f62008-10-02 14:50:16 -0700281 unsigned long pfn, flags;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700282 struct page *page;
Gerald Schaefer6c1b7f62008-10-02 14:50:16 -0700283 struct zone *zone;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700284
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700285 /*
Tang Chen85dbe702013-06-20 18:10:19 +0800286 * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
287 * are not aligned to pageblock_nr_pages.
288 * Then we just check migratetype first.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700289 */
290 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
291 page = __first_valid_page(pfn, pageblock_nr_pages);
Xishi Qiubbf9ce92017-05-03 14:52:55 -0700292 if (page && !is_migrate_isolate_page(page))
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700293 break;
294 }
Gerald Schaefera70dcb92008-11-06 12:53:36 -0800295 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
296 if ((pfn < end_pfn) || !page)
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700297 return -EBUSY;
Tang Chen85dbe702013-06-20 18:10:19 +0800298 /* Check all pages are free or marked as ISOLATED */
Gerald Schaefera70dcb92008-11-06 12:53:36 -0800299 zone = page_zone(page);
Gerald Schaefer6c1b7f62008-10-02 14:50:16 -0700300 spin_lock_irqsave(&zone->lock, flags);
Joonsoo Kimfea85cf2016-01-14 15:18:39 -0800301 pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn,
Wen Congyangb023f462012-12-11 16:00:45 -0800302 skip_hwpoisoned_pages);
Gerald Schaefer6c1b7f62008-10-02 14:50:16 -0700303 spin_unlock_irqrestore(&zone->lock, flags);
Joonsoo Kimfea85cf2016-01-14 15:18:39 -0800304
Joonsoo Kim0f0848e2016-01-14 15:18:42 -0800305 trace_test_pages_isolated(start_pfn, end_pfn, pfn);
306
Joonsoo Kimfea85cf2016-01-14 15:18:39 -0800307 return pfn < end_pfn ? -EBUSY : 0;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -0700308}
Minchan Kim723a0642012-10-08 16:32:52 -0700309
Michal Hocko666feb22018-04-10 16:30:03 -0700310struct page *alloc_migrate_target(struct page *page, unsigned long private)
Minchan Kim723a0642012-10-08 16:32:52 -0700311{
Michal Hocko8b913232017-07-10 15:48:47 -0700312 return new_page_nodemask(page, numa_node_id(), &node_states[N_MEMORY]);
Minchan Kim723a0642012-10-08 16:32:52 -0700313}