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 #define CREATE_TRACE_POINTS
13 #include <trace/events/page_isolation.h>
15 static int set_migratetype_isolate(struct page *page,
16 bool skip_hwpoisoned_pages)
19 unsigned long flags, pfn;
20 struct memory_isolate_notify arg;
24 zone = page_zone(page);
26 spin_lock_irqsave(&zone->lock, flags);
28 pfn = page_to_pfn(page);
30 arg.nr_pages = pageblock_nr_pages;
34 * It may be possible to isolate a pageblock even if the
35 * migratetype is not MIGRATE_MOVABLE. The memory isolation
36 * notifier chain is used by balloon drivers to return the
37 * number of pages in a range that are held by the balloon
38 * driver to shrink memory. If all the pages are accounted for
39 * by balloons, are free, or on the LRU, isolation can continue.
40 * Later, for example, when memory hotplug notifier runs, these
41 * pages reported as "can be isolated" should be isolated(freed)
42 * by the balloon driver through the memory notifier chain.
44 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
45 notifier_ret = notifier_to_errno(notifier_ret);
49 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
50 * We just check MOVABLE pages.
52 if (!has_unmovable_pages(zone, page, arg.pages_found,
53 skip_hwpoisoned_pages))
57 * immobile means "not-on-lru" paes. If immobile is larger than
58 * removable-by-driver pages reported by notifier, we'll fail.
63 unsigned long nr_pages;
64 int migratetype = get_pageblock_migratetype(page);
66 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
67 zone->nr_isolate_pageblock++;
68 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE);
70 __mod_zone_freepage_state(zone, -nr_pages, migratetype);
73 spin_unlock_irqrestore(&zone->lock, flags);
75 drain_all_pages(zone);
79 static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
82 unsigned long flags, nr_pages;
83 struct page *isolated_page = NULL;
85 unsigned long page_idx, buddy_idx;
88 zone = page_zone(page);
89 spin_lock_irqsave(&zone->lock, flags);
90 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
94 * Because freepage with more than pageblock_order on isolated
95 * pageblock is restricted to merge due to freepage counting problem,
96 * it is possible that there is free buddy page.
97 * move_freepages_block() doesn't care of merge so we need other
98 * approach in order to merge them. Isolation and free will make
99 * these pages to be merged.
101 if (PageBuddy(page)) {
102 order = page_order(page);
103 if (order >= pageblock_order) {
104 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
105 buddy_idx = __find_buddy_index(page_idx, order);
106 buddy = page + (buddy_idx - page_idx);
108 if (pfn_valid_within(page_to_pfn(buddy)) &&
109 !is_migrate_isolate_page(buddy)) {
110 __isolate_free_page(page, order);
111 kernel_map_pages(page, (1 << order), 1);
112 set_page_refcounted(page);
113 isolated_page = page;
119 * If we isolate freepage with more than pageblock_order, there
120 * should be no freepage in the range, so we could avoid costly
121 * pageblock scanning for freepage moving.
123 if (!isolated_page) {
124 nr_pages = move_freepages_block(zone, page, migratetype);
125 __mod_zone_freepage_state(zone, nr_pages, migratetype);
127 set_pageblock_migratetype(page, migratetype);
128 zone->nr_isolate_pageblock--;
130 spin_unlock_irqrestore(&zone->lock, flags);
132 __free_pages(isolated_page, order);
135 static inline struct page *
136 __first_valid_page(unsigned long pfn, unsigned long nr_pages)
139 for (i = 0; i < nr_pages; i++)
140 if (pfn_valid_within(pfn + i))
142 if (unlikely(i == nr_pages))
144 return pfn_to_page(pfn + i);
148 * start_isolate_page_range() -- make page-allocation-type of range of pages
149 * to be MIGRATE_ISOLATE.
150 * @start_pfn: The lower PFN of the range to be isolated.
151 * @end_pfn: The upper PFN of the range to be isolated.
152 * @migratetype: migrate type to set in error recovery.
154 * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
155 * the range will never be allocated. Any free pages and pages freed in the
156 * future will not be allocated again.
158 * start_pfn/end_pfn must be aligned to pageblock_order.
159 * Returns 0 on success and -EBUSY if any part of range cannot be isolated.
161 int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
162 unsigned migratetype, bool skip_hwpoisoned_pages)
165 unsigned long undo_pfn;
168 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
169 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
171 for (pfn = start_pfn;
173 pfn += pageblock_nr_pages) {
174 page = __first_valid_page(pfn, pageblock_nr_pages);
176 set_migratetype_isolate(page, skip_hwpoisoned_pages)) {
183 for (pfn = start_pfn;
185 pfn += pageblock_nr_pages)
186 unset_migratetype_isolate(pfn_to_page(pfn), migratetype);
192 * Make isolated pages available again.
194 int undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
195 unsigned migratetype)
200 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
201 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
203 for (pfn = start_pfn;
205 pfn += pageblock_nr_pages) {
206 page = __first_valid_page(pfn, pageblock_nr_pages);
207 if (!page || get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
209 unset_migratetype_isolate(page, migratetype);
214 * Test all pages in the range is free(means isolated) or not.
215 * all pages in [start_pfn...end_pfn) must be in the same zone.
216 * zone->lock must be held before call this.
218 * Returns the last tested pfn.
221 __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
222 bool skip_hwpoisoned_pages)
226 while (pfn < end_pfn) {
227 if (!pfn_valid_within(pfn)) {
231 page = pfn_to_page(pfn);
234 * If the page is on a free list, it has to be on
235 * the correct MIGRATE_ISOLATE freelist. There is no
236 * simple way to verify that as VM_BUG_ON(), though.
238 pfn += 1 << page_order(page);
239 else if (skip_hwpoisoned_pages && PageHWPoison(page))
240 /* A HWPoisoned page cannot be also PageBuddy */
249 int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
250 bool skip_hwpoisoned_pages)
252 unsigned long pfn, flags;
257 * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
258 * are not aligned to pageblock_nr_pages.
259 * Then we just check migratetype first.
261 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
262 page = __first_valid_page(pfn, pageblock_nr_pages);
263 if (page && get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
266 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
267 if ((pfn < end_pfn) || !page)
269 /* Check all pages are free or marked as ISOLATED */
270 zone = page_zone(page);
271 spin_lock_irqsave(&zone->lock, flags);
272 pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn,
273 skip_hwpoisoned_pages);
274 spin_unlock_irqrestore(&zone->lock, flags);
276 trace_test_pages_isolated(start_pfn, end_pfn, pfn);
278 return pfn < end_pfn ? -EBUSY : 0;
281 struct page *alloc_migrate_target(struct page *page, unsigned long private,
284 gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
287 * TODO: allocate a destination hugepage from a nearest neighbor node,
288 * accordance with memory policy of the user process if possible. For
289 * now as a simple work-around, we use the next node for destination.
291 if (PageHuge(page)) {
292 int node = next_online_node(page_to_nid(page));
293 if (node == MAX_NUMNODES)
294 node = first_online_node;
295 return alloc_huge_page_node(page_hstate(compound_head(page)),
299 if (PageHighMem(page))
300 gfp_mask |= __GFP_HIGHMEM;
302 return alloc_page(gfp_mask);