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);
59 else if ((isol_flags & REPORT_FAILURE) && unmovable)
61 * printk() with zone->lock held will guarantee to trigger a
62 * lockdep splat, so defer it here.
64 dump_page(unmovable, "unmovable page");
69 static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
72 unsigned long flags, nr_pages;
73 bool isolated_page = false;
75 unsigned long pfn, buddy_pfn;
78 zone = page_zone(page);
79 spin_lock_irqsave(&zone->lock, flags);
80 if (!is_migrate_isolate_page(page))
84 * Because freepage with more than pageblock_order on isolated
85 * pageblock is restricted to merge due to freepage counting problem,
86 * it is possible that there is free buddy page.
87 * move_freepages_block() doesn't care of merge so we need other
88 * approach in order to merge them. Isolation and free will make
89 * these pages to be merged.
91 if (PageBuddy(page)) {
92 order = page_order(page);
93 if (order >= pageblock_order) {
94 pfn = page_to_pfn(page);
95 buddy_pfn = __find_buddy_pfn(pfn, order);
96 buddy = page + (buddy_pfn - pfn);
98 if (pfn_valid_within(buddy_pfn) &&
99 !is_migrate_isolate_page(buddy)) {
100 __isolate_free_page(page, order);
101 isolated_page = true;
107 * If we isolate freepage with more than pageblock_order, there
108 * should be no freepage in the range, so we could avoid costly
109 * pageblock scanning for freepage moving.
111 if (!isolated_page) {
112 nr_pages = move_freepages_block(zone, page, migratetype, NULL);
113 __mod_zone_freepage_state(zone, nr_pages, migratetype);
115 set_pageblock_migratetype(page, migratetype);
116 zone->nr_isolate_pageblock--;
118 spin_unlock_irqrestore(&zone->lock, flags);
120 post_alloc_hook(page, order, __GFP_MOVABLE);
121 __free_pages(page, order);
125 static inline struct page *
126 __first_valid_page(unsigned long pfn, unsigned long nr_pages)
130 for (i = 0; i < nr_pages; i++) {
133 page = pfn_to_online_page(pfn + i);
142 * start_isolate_page_range() - make page-allocation-type of range of pages to
143 * be MIGRATE_ISOLATE.
144 * @start_pfn: The lower PFN of the range to be isolated.
145 * @end_pfn: The upper PFN of the range to be isolated.
146 * start_pfn/end_pfn must be aligned to pageblock_order.
147 * @migratetype: Migrate type to set in error recovery.
148 * @flags: The following flags are allowed (they can be combined in
150 * MEMORY_OFFLINE - isolate to offline (!allocate) memory
151 * e.g., skip over PageHWPoison() pages
152 * REPORT_FAILURE - report details about the failure to
155 * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
156 * the range will never be allocated. Any free pages and pages freed in the
157 * future will not be allocated again. If specified range includes migrate types
158 * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all
159 * pages in the range finally, the caller have to free all pages in the range.
160 * test_page_isolated() can be used for test it.
162 * There is no high level synchronization mechanism that prevents two threads
163 * from trying to isolate overlapping ranges. If this happens, one thread
164 * will notice pageblocks in the overlapping range already set to isolate.
165 * This happens in set_migratetype_isolate, and set_migratetype_isolate
166 * returns an error. We then clean up by restoring the migration type on
167 * pageblocks we may have modified and return -EBUSY to caller. This
168 * prevents two threads from simultaneously working on overlapping ranges.
170 * Return: the number of isolated pageblocks on success and -EBUSY if any part
171 * of range cannot be isolated.
173 int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
174 unsigned migratetype, int flags)
177 unsigned long undo_pfn;
179 int nr_isolate_pageblock = 0;
181 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
182 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
184 for (pfn = start_pfn;
186 pfn += pageblock_nr_pages) {
187 page = __first_valid_page(pfn, pageblock_nr_pages);
189 if (set_migratetype_isolate(page, migratetype, flags)) {
193 nr_isolate_pageblock++;
196 return nr_isolate_pageblock;
198 for (pfn = start_pfn;
200 pfn += pageblock_nr_pages) {
201 struct page *page = pfn_to_online_page(pfn);
204 unset_migratetype_isolate(page, migratetype);
211 * Make isolated pages available again.
213 void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
214 unsigned migratetype)
219 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
220 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
222 for (pfn = start_pfn;
224 pfn += pageblock_nr_pages) {
225 page = __first_valid_page(pfn, pageblock_nr_pages);
226 if (!page || !is_migrate_isolate_page(page))
228 unset_migratetype_isolate(page, migratetype);
232 * Test all pages in the range is free(means isolated) or not.
233 * all pages in [start_pfn...end_pfn) must be in the same zone.
234 * zone->lock must be held before call this.
236 * Returns the last tested pfn.
239 __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
244 while (pfn < end_pfn) {
245 if (!pfn_valid_within(pfn)) {
249 page = pfn_to_page(pfn);
252 * If the page is on a free list, it has to be on
253 * the correct MIGRATE_ISOLATE freelist. There is no
254 * simple way to verify that as VM_BUG_ON(), though.
256 pfn += 1 << page_order(page);
257 else if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
258 /* A HWPoisoned page cannot be also PageBuddy */
267 /* Caller should ensure that requested range is in a single zone */
268 int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
271 unsigned long pfn, flags;
276 * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
277 * are not aligned to pageblock_nr_pages.
278 * Then we just check migratetype first.
280 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
281 page = __first_valid_page(pfn, pageblock_nr_pages);
282 if (page && !is_migrate_isolate_page(page))
285 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
286 if ((pfn < end_pfn) || !page)
288 /* Check all pages are free or marked as ISOLATED */
289 zone = page_zone(page);
290 spin_lock_irqsave(&zone->lock, flags);
291 pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, isol_flags);
292 spin_unlock_irqrestore(&zone->lock, flags);
294 trace_test_pages_isolated(start_pfn, end_pfn, pfn);
296 return pfn < end_pfn ? -EBUSY : 0;
299 struct page *alloc_migrate_target(struct page *page, unsigned long private)
301 return new_page_nodemask(page, numa_node_id(), &node_states[N_MEMORY]);