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>
18 static int set_migratetype_isolate(struct page *page, int migratetype, int isol_flags)
20 struct page *unmovable = NULL;
25 zone = page_zone(page);
27 spin_lock_irqsave(&zone->lock, flags);
30 * We assume the caller intended to SET migrate type to isolate.
31 * If it is already set, then someone else must have raced and
32 * set it before us. Return -EBUSY
34 if (is_migrate_isolate_page(page))
38 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
39 * We just check MOVABLE pages.
41 unmovable = has_unmovable_pages(zone, page, migratetype, isol_flags);
43 unsigned long nr_pages;
44 int mt = get_pageblock_migratetype(page);
46 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
47 zone->nr_isolate_pageblock++;
48 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE,
51 __mod_zone_freepage_state(zone, -nr_pages, mt);
56 spin_unlock_irqrestore(&zone->lock, flags);
58 drain_all_pages(zone);
60 WARN_ON_ONCE(zone_idx(zone) == ZONE_MOVABLE);
62 if ((isol_flags & REPORT_FAILURE) && unmovable)
64 * printk() with zone->lock held will likely trigger a
65 * lockdep splat, so defer it here.
67 dump_page(unmovable, "unmovable page");
73 static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
76 unsigned long flags, nr_pages;
77 bool isolated_page = false;
79 unsigned long pfn, buddy_pfn;
82 zone = page_zone(page);
83 spin_lock_irqsave(&zone->lock, flags);
84 if (!is_migrate_isolate_page(page))
88 * Because freepage with more than pageblock_order on isolated
89 * pageblock is restricted to merge due to freepage counting problem,
90 * it is possible that there is free buddy page.
91 * move_freepages_block() doesn't care of merge so we need other
92 * approach in order to merge them. Isolation and free will make
93 * these pages to be merged.
95 if (PageBuddy(page)) {
96 order = page_order(page);
97 if (order >= pageblock_order) {
98 pfn = page_to_pfn(page);
99 buddy_pfn = __find_buddy_pfn(pfn, order);
100 buddy = page + (buddy_pfn - pfn);
102 if (pfn_valid_within(buddy_pfn) &&
103 !is_migrate_isolate_page(buddy)) {
104 __isolate_free_page(page, order);
105 isolated_page = true;
111 * If we isolate freepage with more than pageblock_order, there
112 * should be no freepage in the range, so we could avoid costly
113 * pageblock scanning for freepage moving.
115 if (!isolated_page) {
116 nr_pages = move_freepages_block(zone, page, migratetype, NULL);
117 __mod_zone_freepage_state(zone, nr_pages, migratetype);
119 set_pageblock_migratetype(page, migratetype);
121 __putback_isolated_page(page, order, migratetype);
122 zone->nr_isolate_pageblock--;
124 spin_unlock_irqrestore(&zone->lock, flags);
127 static inline struct page *
128 __first_valid_page(unsigned long pfn, unsigned long nr_pages)
132 for (i = 0; i < nr_pages; i++) {
135 page = pfn_to_online_page(pfn + i);
144 * start_isolate_page_range() - make page-allocation-type of range of pages to
145 * be MIGRATE_ISOLATE.
146 * @start_pfn: The lower PFN of the range to be isolated.
147 * @end_pfn: The upper PFN of the range to be isolated.
148 * start_pfn/end_pfn must be aligned to pageblock_order.
149 * @migratetype: Migrate type to set in error recovery.
150 * @flags: The following flags are allowed (they can be combined in
152 * MEMORY_OFFLINE - isolate to offline (!allocate) memory
153 * e.g., skip over PageHWPoison() pages
154 * and PageOffline() pages.
155 * REPORT_FAILURE - report details about the failure to
158 * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
159 * the range will never be allocated. Any free pages and pages freed in the
160 * future will not be allocated again. If specified range includes migrate types
161 * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all
162 * pages in the range finally, the caller have to free all pages in the range.
163 * test_page_isolated() can be used for test it.
165 * There is no high level synchronization mechanism that prevents two threads
166 * from trying to isolate overlapping ranges. If this happens, one thread
167 * will notice pageblocks in the overlapping range already set to isolate.
168 * This happens in set_migratetype_isolate, and set_migratetype_isolate
169 * returns an error. We then clean up by restoring the migration type on
170 * pageblocks we may have modified and return -EBUSY to caller. This
171 * prevents two threads from simultaneously working on overlapping ranges.
173 * Return: the number of isolated pageblocks on success and -EBUSY if any part
174 * of range cannot be isolated.
176 int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
177 unsigned migratetype, int flags)
180 unsigned long undo_pfn;
182 int nr_isolate_pageblock = 0;
184 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
185 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
187 for (pfn = start_pfn;
189 pfn += pageblock_nr_pages) {
190 page = __first_valid_page(pfn, pageblock_nr_pages);
192 if (set_migratetype_isolate(page, migratetype, flags)) {
196 nr_isolate_pageblock++;
199 return nr_isolate_pageblock;
201 for (pfn = start_pfn;
203 pfn += pageblock_nr_pages) {
204 struct page *page = pfn_to_online_page(pfn);
207 unset_migratetype_isolate(page, migratetype);
214 * Make isolated pages available again.
216 void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
217 unsigned migratetype)
222 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
223 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
225 for (pfn = start_pfn;
227 pfn += pageblock_nr_pages) {
228 page = __first_valid_page(pfn, pageblock_nr_pages);
229 if (!page || !is_migrate_isolate_page(page))
231 unset_migratetype_isolate(page, migratetype);
235 * Test all pages in the range is free(means isolated) or not.
236 * all pages in [start_pfn...end_pfn) must be in the same zone.
237 * zone->lock must be held before call this.
239 * Returns the last tested pfn.
242 __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
247 while (pfn < end_pfn) {
248 if (!pfn_valid_within(pfn)) {
252 page = pfn_to_page(pfn);
255 * If the page is on a free list, it has to be on
256 * the correct MIGRATE_ISOLATE freelist. There is no
257 * simple way to verify that as VM_BUG_ON(), though.
259 pfn += 1 << page_order(page);
260 else if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
261 /* A HWPoisoned page cannot be also PageBuddy */
263 else if ((flags & MEMORY_OFFLINE) && PageOffline(page) &&
266 * The responsible driver agreed to skip PageOffline()
267 * pages when offlining memory by dropping its
268 * reference in MEM_GOING_OFFLINE.
278 /* Caller should ensure that requested range is in a single zone */
279 int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
282 unsigned long pfn, flags;
287 * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
288 * are not aligned to pageblock_nr_pages.
289 * Then we just check migratetype first.
291 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
292 page = __first_valid_page(pfn, pageblock_nr_pages);
293 if (page && !is_migrate_isolate_page(page))
296 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
297 if ((pfn < end_pfn) || !page)
299 /* Check all pages are free or marked as ISOLATED */
300 zone = page_zone(page);
301 spin_lock_irqsave(&zone->lock, flags);
302 pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, isol_flags);
303 spin_unlock_irqrestore(&zone->lock, flags);
305 trace_test_pages_isolated(start_pfn, end_pfn, pfn);
307 return pfn < end_pfn ? -EBUSY : 0;
310 struct page *alloc_migrate_target(struct page *page, unsigned long private)
312 return new_page_nodemask(page, numa_node_id(), &node_states[N_MEMORY]);