2 * linux/mm/page_isolation.c
6 #include <linux/page-isolation.h>
7 #include <linux/pageblock-flags.h>
8 #include <linux/memory.h>
9 #include <linux/hugetlb.h>
12 int set_migratetype_isolate(struct page *page, bool skip_hwpoisoned_pages)
15 unsigned long flags, pfn;
16 struct memory_isolate_notify arg;
20 zone = page_zone(page);
22 spin_lock_irqsave(&zone->lock, flags);
24 pfn = page_to_pfn(page);
26 arg.nr_pages = pageblock_nr_pages;
30 * It may be possible to isolate a pageblock even if the
31 * migratetype is not MIGRATE_MOVABLE. The memory isolation
32 * notifier chain is used by balloon drivers to return the
33 * number of pages in a range that are held by the balloon
34 * driver to shrink memory. If all the pages are accounted for
35 * by balloons, are free, or on the LRU, isolation can continue.
36 * Later, for example, when memory hotplug notifier runs, these
37 * pages reported as "can be isolated" should be isolated(freed)
38 * by the balloon driver through the memory notifier chain.
40 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
41 notifier_ret = notifier_to_errno(notifier_ret);
45 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
46 * We just check MOVABLE pages.
48 if (!has_unmovable_pages(zone, page, arg.pages_found,
49 skip_hwpoisoned_pages))
53 * immobile means "not-on-lru" paes. If immobile is larger than
54 * removable-by-driver pages reported by notifier, we'll fail.
59 unsigned long nr_pages;
60 int migratetype = get_pageblock_migratetype(page);
62 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
63 zone->nr_isolate_pageblock++;
64 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE);
66 __mod_zone_freepage_state(zone, -nr_pages, migratetype);
69 spin_unlock_irqrestore(&zone->lock, flags);
71 drain_all_pages(zone);
75 void unset_migratetype_isolate(struct page *page, unsigned migratetype)
78 unsigned long flags, nr_pages;
79 struct page *isolated_page = NULL;
81 unsigned long page_idx, buddy_idx;
84 zone = page_zone(page);
85 spin_lock_irqsave(&zone->lock, flags);
86 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
90 * Because freepage with more than pageblock_order on isolated
91 * pageblock is restricted to merge due to freepage counting problem,
92 * it is possible that there is free buddy page.
93 * move_freepages_block() doesn't care of merge so we need other
94 * approach in order to merge them. Isolation and free will make
95 * these pages to be merged.
97 if (PageBuddy(page)) {
98 order = page_order(page);
99 if (order >= pageblock_order) {
100 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
101 buddy_idx = __find_buddy_index(page_idx, order);
102 buddy = page + (buddy_idx - page_idx);
104 if (pfn_valid_within(page_to_pfn(buddy)) &&
105 !is_migrate_isolate_page(buddy)) {
106 __isolate_free_page(page, order);
107 kernel_map_pages(page, (1 << order), 1);
108 set_page_refcounted(page);
109 isolated_page = page;
115 * If we isolate freepage with more than pageblock_order, there
116 * should be no freepage in the range, so we could avoid costly
117 * pageblock scanning for freepage moving.
119 if (!isolated_page) {
120 nr_pages = move_freepages_block(zone, page, migratetype);
121 __mod_zone_freepage_state(zone, nr_pages, migratetype);
123 set_pageblock_migratetype(page, migratetype);
124 zone->nr_isolate_pageblock--;
126 spin_unlock_irqrestore(&zone->lock, flags);
128 __free_pages(isolated_page, order);
131 static inline struct page *
132 __first_valid_page(unsigned long pfn, unsigned long nr_pages)
135 for (i = 0; i < nr_pages; i++)
136 if (pfn_valid_within(pfn + i))
138 if (unlikely(i == nr_pages))
140 return pfn_to_page(pfn + i);
144 * start_isolate_page_range() -- make page-allocation-type of range of pages
145 * to 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 * @migratetype: migrate type to set in error recovery.
150 * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
151 * the range will never be allocated. Any free pages and pages freed in the
152 * future will not be allocated again.
154 * start_pfn/end_pfn must be aligned to pageblock_order.
155 * Returns 0 on success and -EBUSY if any part of range cannot be isolated.
157 int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
158 unsigned migratetype, bool skip_hwpoisoned_pages)
161 unsigned long undo_pfn;
164 BUG_ON((start_pfn) & (pageblock_nr_pages - 1));
165 BUG_ON((end_pfn) & (pageblock_nr_pages - 1));
167 for (pfn = start_pfn;
169 pfn += pageblock_nr_pages) {
170 page = __first_valid_page(pfn, pageblock_nr_pages);
172 set_migratetype_isolate(page, skip_hwpoisoned_pages)) {
179 for (pfn = start_pfn;
181 pfn += pageblock_nr_pages)
182 unset_migratetype_isolate(pfn_to_page(pfn), migratetype);
188 * Make isolated pages available again.
190 int undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
191 unsigned migratetype)
195 BUG_ON((start_pfn) & (pageblock_nr_pages - 1));
196 BUG_ON((end_pfn) & (pageblock_nr_pages - 1));
197 for (pfn = start_pfn;
199 pfn += pageblock_nr_pages) {
200 page = __first_valid_page(pfn, pageblock_nr_pages);
201 if (!page || get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
203 unset_migratetype_isolate(page, migratetype);
208 * Test all pages in the range is free(means isolated) or not.
209 * all pages in [start_pfn...end_pfn) must be in the same zone.
210 * zone->lock must be held before call this.
212 * Returns 1 if all pages in the range are isolated.
215 __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
216 bool skip_hwpoisoned_pages)
220 while (pfn < end_pfn) {
221 if (!pfn_valid_within(pfn)) {
225 page = pfn_to_page(pfn);
226 if (PageBuddy(page)) {
228 * If race between isolatation and allocation happens,
229 * some free pages could be in MIGRATE_MOVABLE list
230 * although pageblock's migratation type of the page
231 * is MIGRATE_ISOLATE. Catch it and move the page into
232 * MIGRATE_ISOLATE list.
234 if (get_freepage_migratetype(page) != MIGRATE_ISOLATE) {
235 struct page *end_page;
237 end_page = page + (1 << page_order(page)) - 1;
238 move_freepages(page_zone(page), page, end_page,
241 pfn += 1 << page_order(page);
243 else if (page_count(page) == 0 &&
244 get_freepage_migratetype(page) == MIGRATE_ISOLATE)
246 else if (skip_hwpoisoned_pages && PageHWPoison(page)) {
248 * The HWPoisoned page may be not in buddy
249 * system, and page_count() is not 0.
262 int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
263 bool skip_hwpoisoned_pages)
265 unsigned long pfn, flags;
271 * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
272 * are not aligned to pageblock_nr_pages.
273 * Then we just check migratetype first.
275 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
276 page = __first_valid_page(pfn, pageblock_nr_pages);
277 if (page && get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
280 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
281 if ((pfn < end_pfn) || !page)
283 /* Check all pages are free or marked as ISOLATED */
284 zone = page_zone(page);
285 spin_lock_irqsave(&zone->lock, flags);
286 ret = __test_page_isolated_in_pageblock(start_pfn, end_pfn,
287 skip_hwpoisoned_pages);
288 spin_unlock_irqrestore(&zone->lock, flags);
289 return ret ? 0 : -EBUSY;
292 struct page *alloc_migrate_target(struct page *page, unsigned long private,
295 gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
298 * TODO: allocate a destination hugepage from a nearest neighbor node,
299 * accordance with memory policy of the user process if possible. For
300 * now as a simple work-around, we use the next node for destination.
302 if (PageHuge(page)) {
303 nodemask_t src = nodemask_of_node(page_to_nid(page));
305 nodes_complement(dst, src);
306 return alloc_huge_page_node(page_hstate(compound_head(page)),
307 next_node(page_to_nid(page), dst));
310 if (PageHighMem(page))
311 gfp_mask |= __GFP_HIGHMEM;
313 return alloc_page(gfp_mask);