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 static int set_migratetype_isolate(struct page *page,
13 bool skip_hwpoisoned_pages)
16 unsigned long flags, pfn;
17 struct memory_isolate_notify arg;
21 zone = page_zone(page);
23 spin_lock_irqsave(&zone->lock, flags);
25 pfn = page_to_pfn(page);
27 arg.nr_pages = pageblock_nr_pages;
31 * It may be possible to isolate a pageblock even if the
32 * migratetype is not MIGRATE_MOVABLE. The memory isolation
33 * notifier chain is used by balloon drivers to return the
34 * number of pages in a range that are held by the balloon
35 * driver to shrink memory. If all the pages are accounted for
36 * by balloons, are free, or on the LRU, isolation can continue.
37 * Later, for example, when memory hotplug notifier runs, these
38 * pages reported as "can be isolated" should be isolated(freed)
39 * by the balloon driver through the memory notifier chain.
41 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
42 notifier_ret = notifier_to_errno(notifier_ret);
46 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
47 * We just check MOVABLE pages.
49 if (!has_unmovable_pages(zone, page, arg.pages_found,
50 skip_hwpoisoned_pages))
54 * immobile means "not-on-lru" paes. If immobile is larger than
55 * removable-by-driver pages reported by notifier, we'll fail.
60 unsigned long nr_pages;
61 int migratetype = get_pageblock_migratetype(page);
63 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
64 zone->nr_isolate_pageblock++;
65 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE);
67 __mod_zone_freepage_state(zone, -nr_pages, migratetype);
70 spin_unlock_irqrestore(&zone->lock, flags);
72 drain_all_pages(zone);
76 static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
79 unsigned long flags, nr_pages;
80 struct page *isolated_page = NULL;
82 unsigned long page_idx, buddy_idx;
85 zone = page_zone(page);
86 spin_lock_irqsave(&zone->lock, flags);
87 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
91 * Because freepage with more than pageblock_order on isolated
92 * pageblock is restricted to merge due to freepage counting problem,
93 * it is possible that there is free buddy page.
94 * move_freepages_block() doesn't care of merge so we need other
95 * approach in order to merge them. Isolation and free will make
96 * these pages to be merged.
98 if (PageBuddy(page)) {
99 order = page_order(page);
100 if (order >= pageblock_order) {
101 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
102 buddy_idx = __find_buddy_index(page_idx, order);
103 buddy = page + (buddy_idx - page_idx);
105 if (pfn_valid_within(page_to_pfn(buddy)) &&
106 !is_migrate_isolate_page(buddy)) {
107 __isolate_free_page(page, order);
108 kernel_map_pages(page, (1 << order), 1);
109 set_page_refcounted(page);
110 isolated_page = page;
116 * If we isolate freepage with more than pageblock_order, there
117 * should be no freepage in the range, so we could avoid costly
118 * pageblock scanning for freepage moving.
120 if (!isolated_page) {
121 nr_pages = move_freepages_block(zone, page, migratetype);
122 __mod_zone_freepage_state(zone, nr_pages, migratetype);
124 set_pageblock_migratetype(page, migratetype);
125 zone->nr_isolate_pageblock--;
127 spin_unlock_irqrestore(&zone->lock, flags);
129 __free_pages(isolated_page, order);
132 static inline struct page *
133 __first_valid_page(unsigned long pfn, unsigned long nr_pages)
136 for (i = 0; i < nr_pages; i++)
137 if (pfn_valid_within(pfn + i))
139 if (unlikely(i == nr_pages))
141 return pfn_to_page(pfn + i);
145 * start_isolate_page_range() -- make page-allocation-type of range of pages
146 * to be MIGRATE_ISOLATE.
147 * @start_pfn: The lower PFN of the range to be isolated.
148 * @end_pfn: The upper PFN of the range to be isolated.
149 * @migratetype: migrate type to set in error recovery.
151 * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
152 * the range will never be allocated. Any free pages and pages freed in the
153 * future will not be allocated again.
155 * start_pfn/end_pfn must be aligned to pageblock_order.
156 * Returns 0 on success and -EBUSY if any part of range cannot be isolated.
158 int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
159 unsigned migratetype, bool skip_hwpoisoned_pages)
162 unsigned long undo_pfn;
165 BUG_ON((start_pfn) & (pageblock_nr_pages - 1));
166 BUG_ON((end_pfn) & (pageblock_nr_pages - 1));
168 for (pfn = start_pfn;
170 pfn += pageblock_nr_pages) {
171 page = __first_valid_page(pfn, pageblock_nr_pages);
173 set_migratetype_isolate(page, skip_hwpoisoned_pages)) {
180 for (pfn = start_pfn;
182 pfn += pageblock_nr_pages)
183 unset_migratetype_isolate(pfn_to_page(pfn), migratetype);
189 * Make isolated pages available again.
191 int undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
192 unsigned migratetype)
196 BUG_ON((start_pfn) & (pageblock_nr_pages - 1));
197 BUG_ON((end_pfn) & (pageblock_nr_pages - 1));
198 for (pfn = start_pfn;
200 pfn += pageblock_nr_pages) {
201 page = __first_valid_page(pfn, pageblock_nr_pages);
202 if (!page || get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
204 unset_migratetype_isolate(page, migratetype);
209 * Test all pages in the range is free(means isolated) or not.
210 * all pages in [start_pfn...end_pfn) must be in the same zone.
211 * zone->lock must be held before call this.
213 * Returns 1 if all pages in the range are isolated.
216 __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
217 bool skip_hwpoisoned_pages)
221 while (pfn < end_pfn) {
222 if (!pfn_valid_within(pfn)) {
226 page = pfn_to_page(pfn);
229 * If the page is on a free list, it has to be on
230 * the correct MIGRATE_ISOLATE freelist. There is no
231 * simple way to verify that as VM_BUG_ON(), though.
233 pfn += 1 << page_order(page);
234 else if (skip_hwpoisoned_pages && PageHWPoison(page))
235 /* A HWPoisoned page cannot be also PageBuddy */
245 int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
246 bool skip_hwpoisoned_pages)
248 unsigned long pfn, flags;
254 * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
255 * are not aligned to pageblock_nr_pages.
256 * Then we just check migratetype first.
258 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
259 page = __first_valid_page(pfn, pageblock_nr_pages);
260 if (page && get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
263 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
264 if ((pfn < end_pfn) || !page)
266 /* Check all pages are free or marked as ISOLATED */
267 zone = page_zone(page);
268 spin_lock_irqsave(&zone->lock, flags);
269 ret = __test_page_isolated_in_pageblock(start_pfn, end_pfn,
270 skip_hwpoisoned_pages);
271 spin_unlock_irqrestore(&zone->lock, flags);
272 return ret ? 0 : -EBUSY;
275 struct page *alloc_migrate_target(struct page *page, unsigned long private,
278 gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
281 * TODO: allocate a destination hugepage from a nearest neighbor node,
282 * accordance with memory policy of the user process if possible. For
283 * now as a simple work-around, we use the next node for destination.
285 if (PageHuge(page)) {
286 int node = next_online_node(page_to_nid(page));
287 if (node == MAX_NUMNODES)
288 node = first_online_node;
289 return alloc_huge_page_node(page_hstate(compound_head(page)),
293 if (PageHighMem(page))
294 gfp_mask |= __GFP_HIGHMEM;
296 return alloc_page(gfp_mask);