Loading...
1/*
2 * linux/mm/page_isolation.c
3 */
4
5#include <linux/mm.h>
6#include <linux/page-isolation.h>
7#include <linux/pageblock-flags.h>
8#include <linux/memory.h>
9#include <linux/hugetlb.h>
10#include "internal.h"
11
12#define CREATE_TRACE_POINTS
13#include <trace/events/page_isolation.h>
14
15static int set_migratetype_isolate(struct page *page,
16 bool skip_hwpoisoned_pages)
17{
18 struct zone *zone;
19 unsigned long flags, pfn;
20 struct memory_isolate_notify arg;
21 int notifier_ret;
22 int ret = -EBUSY;
23
24 zone = page_zone(page);
25
26 spin_lock_irqsave(&zone->lock, flags);
27
28 pfn = page_to_pfn(page);
29 arg.start_pfn = pfn;
30 arg.nr_pages = pageblock_nr_pages;
31 arg.pages_found = 0;
32
33 /*
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.
43 */
44 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
45 notifier_ret = notifier_to_errno(notifier_ret);
46 if (notifier_ret)
47 goto out;
48 /*
49 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
50 * We just check MOVABLE pages.
51 */
52 if (!has_unmovable_pages(zone, page, arg.pages_found,
53 skip_hwpoisoned_pages))
54 ret = 0;
55
56 /*
57 * immobile means "not-on-lru" paes. If immobile is larger than
58 * removable-by-driver pages reported by notifier, we'll fail.
59 */
60
61out:
62 if (!ret) {
63 unsigned long nr_pages;
64 int migratetype = get_pageblock_migratetype(page);
65
66 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
67 zone->nr_isolate_pageblock++;
68 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE);
69
70 __mod_zone_freepage_state(zone, -nr_pages, migratetype);
71 }
72
73 spin_unlock_irqrestore(&zone->lock, flags);
74 if (!ret)
75 drain_all_pages(zone);
76 return ret;
77}
78
79static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
80{
81 struct zone *zone;
82 unsigned long flags, nr_pages;
83 struct page *isolated_page = NULL;
84 unsigned int order;
85 unsigned long page_idx, buddy_idx;
86 struct page *buddy;
87
88 zone = page_zone(page);
89 spin_lock_irqsave(&zone->lock, flags);
90 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
91 goto out;
92
93 /*
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.
100 */
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);
107
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;
114 }
115 }
116 }
117
118 /*
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.
122 */
123 if (!isolated_page) {
124 nr_pages = move_freepages_block(zone, page, migratetype);
125 __mod_zone_freepage_state(zone, nr_pages, migratetype);
126 }
127 set_pageblock_migratetype(page, migratetype);
128 zone->nr_isolate_pageblock--;
129out:
130 spin_unlock_irqrestore(&zone->lock, flags);
131 if (isolated_page)
132 __free_pages(isolated_page, order);
133}
134
135static inline struct page *
136__first_valid_page(unsigned long pfn, unsigned long nr_pages)
137{
138 int i;
139 for (i = 0; i < nr_pages; i++)
140 if (pfn_valid_within(pfn + i))
141 break;
142 if (unlikely(i == nr_pages))
143 return NULL;
144 return pfn_to_page(pfn + i);
145}
146
147/*
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.
153 *
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.
157 *
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.
160 */
161int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
162 unsigned migratetype, bool skip_hwpoisoned_pages)
163{
164 unsigned long pfn;
165 unsigned long undo_pfn;
166 struct page *page;
167
168 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
169 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
170
171 for (pfn = start_pfn;
172 pfn < end_pfn;
173 pfn += pageblock_nr_pages) {
174 page = __first_valid_page(pfn, pageblock_nr_pages);
175 if (page &&
176 set_migratetype_isolate(page, skip_hwpoisoned_pages)) {
177 undo_pfn = pfn;
178 goto undo;
179 }
180 }
181 return 0;
182undo:
183 for (pfn = start_pfn;
184 pfn < undo_pfn;
185 pfn += pageblock_nr_pages)
186 unset_migratetype_isolate(pfn_to_page(pfn), migratetype);
187
188 return -EBUSY;
189}
190
191/*
192 * Make isolated pages available again.
193 */
194int undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
195 unsigned migratetype)
196{
197 unsigned long pfn;
198 struct page *page;
199
200 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
201 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
202
203 for (pfn = start_pfn;
204 pfn < end_pfn;
205 pfn += pageblock_nr_pages) {
206 page = __first_valid_page(pfn, pageblock_nr_pages);
207 if (!page || get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
208 continue;
209 unset_migratetype_isolate(page, migratetype);
210 }
211 return 0;
212}
213/*
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.
217 *
218 * Returns the last tested pfn.
219 */
220static unsigned long
221__test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
222 bool skip_hwpoisoned_pages)
223{
224 struct page *page;
225
226 while (pfn < end_pfn) {
227 if (!pfn_valid_within(pfn)) {
228 pfn++;
229 continue;
230 }
231 page = pfn_to_page(pfn);
232 if (PageBuddy(page))
233 /*
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.
237 */
238 pfn += 1 << page_order(page);
239 else if (skip_hwpoisoned_pages && PageHWPoison(page))
240 /* A HWPoisoned page cannot be also PageBuddy */
241 pfn++;
242 else
243 break;
244 }
245
246 return pfn;
247}
248
249int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
250 bool skip_hwpoisoned_pages)
251{
252 unsigned long pfn, flags;
253 struct page *page;
254 struct zone *zone;
255
256 /*
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.
260 */
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)
264 break;
265 }
266 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
267 if ((pfn < end_pfn) || !page)
268 return -EBUSY;
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);
275
276 trace_test_pages_isolated(start_pfn, end_pfn, pfn);
277
278 return pfn < end_pfn ? -EBUSY : 0;
279}
280
281struct page *alloc_migrate_target(struct page *page, unsigned long private,
282 int **resultp)
283{
284 gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
285
286 /*
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.
290 */
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)),
296 node);
297 }
298
299 if (PageHighMem(page))
300 gfp_mask |= __GFP_HIGHMEM;
301
302 return alloc_page(gfp_mask);
303}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/mm/page_isolation.c
4 */
5
6#include <linux/mm.h>
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>
13#include "internal.h"
14
15#define CREATE_TRACE_POINTS
16#include <trace/events/page_isolation.h>
17
18/*
19 * This function checks whether the range [start_pfn, end_pfn) includes
20 * unmovable pages or not. The range must fall into a single pageblock and
21 * consequently belong to a single zone.
22 *
23 * PageLRU check without isolation or lru_lock could race so that
24 * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
25 * check without lock_page also may miss some movable non-lru pages at
26 * race condition. So you can't expect this function should be exact.
27 *
28 * Returns a page without holding a reference. If the caller wants to
29 * dereference that page (e.g., dumping), it has to make sure that it
30 * cannot get removed (e.g., via memory unplug) concurrently.
31 *
32 */
33static struct page *has_unmovable_pages(unsigned long start_pfn, unsigned long end_pfn,
34 int migratetype, int flags)
35{
36 struct page *page = pfn_to_page(start_pfn);
37 struct zone *zone = page_zone(page);
38 unsigned long pfn;
39
40 VM_BUG_ON(pageblock_start_pfn(start_pfn) !=
41 pageblock_start_pfn(end_pfn - 1));
42
43 if (is_migrate_cma_page(page)) {
44 /*
45 * CMA allocations (alloc_contig_range) really need to mark
46 * isolate CMA pageblocks even when they are not movable in fact
47 * so consider them movable here.
48 */
49 if (is_migrate_cma(migratetype))
50 return NULL;
51
52 return page;
53 }
54
55 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
56 page = pfn_to_page(pfn);
57
58 /*
59 * Both, bootmem allocations and memory holes are marked
60 * PG_reserved and are unmovable. We can even have unmovable
61 * allocations inside ZONE_MOVABLE, for example when
62 * specifying "movablecore".
63 */
64 if (PageReserved(page))
65 return page;
66
67 /*
68 * If the zone is movable and we have ruled out all reserved
69 * pages then it should be reasonably safe to assume the rest
70 * is movable.
71 */
72 if (zone_idx(zone) == ZONE_MOVABLE)
73 continue;
74
75 /*
76 * Hugepages are not in LRU lists, but they're movable.
77 * THPs are on the LRU, but need to be counted as #small pages.
78 * We need not scan over tail pages because we don't
79 * handle each tail page individually in migration.
80 */
81 if (PageHuge(page) || PageTransCompound(page)) {
82 struct page *head = compound_head(page);
83 unsigned int skip_pages;
84
85 if (PageHuge(page)) {
86 if (!hugepage_migration_supported(page_hstate(head)))
87 return page;
88 } else if (!PageLRU(head) && !__PageMovable(head)) {
89 return page;
90 }
91
92 skip_pages = compound_nr(head) - (page - head);
93 pfn += skip_pages - 1;
94 continue;
95 }
96
97 /*
98 * We can't use page_count without pin a page
99 * because another CPU can free compound page.
100 * This check already skips compound tails of THP
101 * because their page->_refcount is zero at all time.
102 */
103 if (!page_ref_count(page)) {
104 if (PageBuddy(page))
105 pfn += (1 << buddy_order(page)) - 1;
106 continue;
107 }
108
109 /*
110 * The HWPoisoned page may be not in buddy system, and
111 * page_count() is not 0.
112 */
113 if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
114 continue;
115
116 /*
117 * We treat all PageOffline() pages as movable when offlining
118 * to give drivers a chance to decrement their reference count
119 * in MEM_GOING_OFFLINE in order to indicate that these pages
120 * can be offlined as there are no direct references anymore.
121 * For actually unmovable PageOffline() where the driver does
122 * not support this, we will fail later when trying to actually
123 * move these pages that still have a reference count > 0.
124 * (false negatives in this function only)
125 */
126 if ((flags & MEMORY_OFFLINE) && PageOffline(page))
127 continue;
128
129 if (__PageMovable(page) || PageLRU(page))
130 continue;
131
132 /*
133 * If there are RECLAIMABLE pages, we need to check
134 * it. But now, memory offline itself doesn't call
135 * shrink_node_slabs() and it still to be fixed.
136 */
137 return page;
138 }
139 return NULL;
140}
141
142/*
143 * This function set pageblock migratetype to isolate if no unmovable page is
144 * present in [start_pfn, end_pfn). The pageblock must intersect with
145 * [start_pfn, end_pfn).
146 */
147static int set_migratetype_isolate(struct page *page, int migratetype, int isol_flags,
148 unsigned long start_pfn, unsigned long end_pfn)
149{
150 struct zone *zone = page_zone(page);
151 struct page *unmovable;
152 unsigned long flags;
153 unsigned long check_unmovable_start, check_unmovable_end;
154
155 spin_lock_irqsave(&zone->lock, flags);
156
157 /*
158 * We assume the caller intended to SET migrate type to isolate.
159 * If it is already set, then someone else must have raced and
160 * set it before us.
161 */
162 if (is_migrate_isolate_page(page)) {
163 spin_unlock_irqrestore(&zone->lock, flags);
164 return -EBUSY;
165 }
166
167 /*
168 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
169 * We just check MOVABLE pages.
170 *
171 * Pass the intersection of [start_pfn, end_pfn) and the page's pageblock
172 * to avoid redundant checks.
173 */
174 check_unmovable_start = max(page_to_pfn(page), start_pfn);
175 check_unmovable_end = min(pageblock_end_pfn(page_to_pfn(page)),
176 end_pfn);
177
178 unmovable = has_unmovable_pages(check_unmovable_start, check_unmovable_end,
179 migratetype, isol_flags);
180 if (!unmovable) {
181 unsigned long nr_pages;
182 int mt = get_pageblock_migratetype(page);
183
184 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
185 zone->nr_isolate_pageblock++;
186 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE,
187 NULL);
188
189 __mod_zone_freepage_state(zone, -nr_pages, mt);
190 spin_unlock_irqrestore(&zone->lock, flags);
191 return 0;
192 }
193
194 spin_unlock_irqrestore(&zone->lock, flags);
195 if (isol_flags & REPORT_FAILURE) {
196 /*
197 * printk() with zone->lock held will likely trigger a
198 * lockdep splat, so defer it here.
199 */
200 dump_page(unmovable, "unmovable page");
201 }
202
203 return -EBUSY;
204}
205
206static void unset_migratetype_isolate(struct page *page, int migratetype)
207{
208 struct zone *zone;
209 unsigned long flags, nr_pages;
210 bool isolated_page = false;
211 unsigned int order;
212 struct page *buddy;
213
214 zone = page_zone(page);
215 spin_lock_irqsave(&zone->lock, flags);
216 if (!is_migrate_isolate_page(page))
217 goto out;
218
219 /*
220 * Because freepage with more than pageblock_order on isolated
221 * pageblock is restricted to merge due to freepage counting problem,
222 * it is possible that there is free buddy page.
223 * move_freepages_block() doesn't care of merge so we need other
224 * approach in order to merge them. Isolation and free will make
225 * these pages to be merged.
226 */
227 if (PageBuddy(page)) {
228 order = buddy_order(page);
229 if (order >= pageblock_order && order < MAX_ORDER - 1) {
230 buddy = find_buddy_page_pfn(page, page_to_pfn(page),
231 order, NULL);
232 if (buddy && !is_migrate_isolate_page(buddy)) {
233 isolated_page = !!__isolate_free_page(page, order);
234 /*
235 * Isolating a free page in an isolated pageblock
236 * is expected to always work as watermarks don't
237 * apply here.
238 */
239 VM_WARN_ON(!isolated_page);
240 }
241 }
242 }
243
244 /*
245 * If we isolate freepage with more than pageblock_order, there
246 * should be no freepage in the range, so we could avoid costly
247 * pageblock scanning for freepage moving.
248 *
249 * We didn't actually touch any of the isolated pages, so place them
250 * to the tail of the freelist. This is an optimization for memory
251 * onlining - just onlined memory won't immediately be considered for
252 * allocation.
253 */
254 if (!isolated_page) {
255 nr_pages = move_freepages_block(zone, page, migratetype, NULL);
256 __mod_zone_freepage_state(zone, nr_pages, migratetype);
257 }
258 set_pageblock_migratetype(page, migratetype);
259 if (isolated_page)
260 __putback_isolated_page(page, order, migratetype);
261 zone->nr_isolate_pageblock--;
262out:
263 spin_unlock_irqrestore(&zone->lock, flags);
264}
265
266static inline struct page *
267__first_valid_page(unsigned long pfn, unsigned long nr_pages)
268{
269 int i;
270
271 for (i = 0; i < nr_pages; i++) {
272 struct page *page;
273
274 page = pfn_to_online_page(pfn + i);
275 if (!page)
276 continue;
277 return page;
278 }
279 return NULL;
280}
281
282/**
283 * isolate_single_pageblock() -- tries to isolate a pageblock that might be
284 * within a free or in-use page.
285 * @boundary_pfn: pageblock-aligned pfn that a page might cross
286 * @flags: isolation flags
287 * @gfp_flags: GFP flags used for migrating pages
288 * @isolate_before: isolate the pageblock before the boundary_pfn
289 * @skip_isolation: the flag to skip the pageblock isolation in second
290 * isolate_single_pageblock()
291 * @migratetype: migrate type to set in error recovery.
292 *
293 * Free and in-use pages can be as big as MAX_ORDER-1 and contain more than one
294 * pageblock. When not all pageblocks within a page are isolated at the same
295 * time, free page accounting can go wrong. For example, in the case of
296 * MAX_ORDER-1 = pageblock_order + 1, a MAX_ORDER-1 page has two pagelbocks.
297 * [ MAX_ORDER-1 ]
298 * [ pageblock0 | pageblock1 ]
299 * When either pageblock is isolated, if it is a free page, the page is not
300 * split into separate migratetype lists, which is supposed to; if it is an
301 * in-use page and freed later, __free_one_page() does not split the free page
302 * either. The function handles this by splitting the free page or migrating
303 * the in-use page then splitting the free page.
304 */
305static int isolate_single_pageblock(unsigned long boundary_pfn, int flags,
306 gfp_t gfp_flags, bool isolate_before, bool skip_isolation,
307 int migratetype)
308{
309 unsigned long start_pfn;
310 unsigned long isolate_pageblock;
311 unsigned long pfn;
312 struct zone *zone;
313 int ret;
314
315 VM_BUG_ON(!pageblock_aligned(boundary_pfn));
316
317 if (isolate_before)
318 isolate_pageblock = boundary_pfn - pageblock_nr_pages;
319 else
320 isolate_pageblock = boundary_pfn;
321
322 /*
323 * scan at the beginning of MAX_ORDER_NR_PAGES aligned range to avoid
324 * only isolating a subset of pageblocks from a bigger than pageblock
325 * free or in-use page. Also make sure all to-be-isolated pageblocks
326 * are within the same zone.
327 */
328 zone = page_zone(pfn_to_page(isolate_pageblock));
329 start_pfn = max(ALIGN_DOWN(isolate_pageblock, MAX_ORDER_NR_PAGES),
330 zone->zone_start_pfn);
331
332 if (skip_isolation) {
333 int mt __maybe_unused = get_pageblock_migratetype(pfn_to_page(isolate_pageblock));
334
335 VM_BUG_ON(!is_migrate_isolate(mt));
336 } else {
337 ret = set_migratetype_isolate(pfn_to_page(isolate_pageblock), migratetype,
338 flags, isolate_pageblock, isolate_pageblock + pageblock_nr_pages);
339
340 if (ret)
341 return ret;
342 }
343
344 /*
345 * Bail out early when the to-be-isolated pageblock does not form
346 * a free or in-use page across boundary_pfn:
347 *
348 * 1. isolate before boundary_pfn: the page after is not online
349 * 2. isolate after boundary_pfn: the page before is not online
350 *
351 * This also ensures correctness. Without it, when isolate after
352 * boundary_pfn and [start_pfn, boundary_pfn) are not online,
353 * __first_valid_page() will return unexpected NULL in the for loop
354 * below.
355 */
356 if (isolate_before) {
357 if (!pfn_to_online_page(boundary_pfn))
358 return 0;
359 } else {
360 if (!pfn_to_online_page(boundary_pfn - 1))
361 return 0;
362 }
363
364 for (pfn = start_pfn; pfn < boundary_pfn;) {
365 struct page *page = __first_valid_page(pfn, boundary_pfn - pfn);
366
367 VM_BUG_ON(!page);
368 pfn = page_to_pfn(page);
369 /*
370 * start_pfn is MAX_ORDER_NR_PAGES aligned, if there is any
371 * free pages in [start_pfn, boundary_pfn), its head page will
372 * always be in the range.
373 */
374 if (PageBuddy(page)) {
375 int order = buddy_order(page);
376
377 if (pfn + (1UL << order) > boundary_pfn) {
378 /* free page changed before split, check it again */
379 if (split_free_page(page, order, boundary_pfn - pfn))
380 continue;
381 }
382
383 pfn += 1UL << order;
384 continue;
385 }
386 /*
387 * migrate compound pages then let the free page handling code
388 * above do the rest. If migration is not possible, just fail.
389 */
390 if (PageCompound(page)) {
391 struct page *head = compound_head(page);
392 unsigned long head_pfn = page_to_pfn(head);
393 unsigned long nr_pages = compound_nr(head);
394
395 if (head_pfn + nr_pages <= boundary_pfn) {
396 pfn = head_pfn + nr_pages;
397 continue;
398 }
399#if defined CONFIG_COMPACTION || defined CONFIG_CMA
400 /*
401 * hugetlb, lru compound (THP), and movable compound pages
402 * can be migrated. Otherwise, fail the isolation.
403 */
404 if (PageHuge(page) || PageLRU(page) || __PageMovable(page)) {
405 int order;
406 unsigned long outer_pfn;
407 int page_mt = get_pageblock_migratetype(page);
408 bool isolate_page = !is_migrate_isolate_page(page);
409 struct compact_control cc = {
410 .nr_migratepages = 0,
411 .order = -1,
412 .zone = page_zone(pfn_to_page(head_pfn)),
413 .mode = MIGRATE_SYNC,
414 .ignore_skip_hint = true,
415 .no_set_skip_hint = true,
416 .gfp_mask = gfp_flags,
417 .alloc_contig = true,
418 };
419 INIT_LIST_HEAD(&cc.migratepages);
420
421 /*
422 * XXX: mark the page as MIGRATE_ISOLATE so that
423 * no one else can grab the freed page after migration.
424 * Ideally, the page should be freed as two separate
425 * pages to be added into separate migratetype free
426 * lists.
427 */
428 if (isolate_page) {
429 ret = set_migratetype_isolate(page, page_mt,
430 flags, head_pfn, head_pfn + nr_pages);
431 if (ret)
432 goto failed;
433 }
434
435 ret = __alloc_contig_migrate_range(&cc, head_pfn,
436 head_pfn + nr_pages);
437
438 /*
439 * restore the page's migratetype so that it can
440 * be split into separate migratetype free lists
441 * later.
442 */
443 if (isolate_page)
444 unset_migratetype_isolate(page, page_mt);
445
446 if (ret)
447 goto failed;
448 /*
449 * reset pfn to the head of the free page, so
450 * that the free page handling code above can split
451 * the free page to the right migratetype list.
452 *
453 * head_pfn is not used here as a hugetlb page order
454 * can be bigger than MAX_ORDER-1, but after it is
455 * freed, the free page order is not. Use pfn within
456 * the range to find the head of the free page.
457 */
458 order = 0;
459 outer_pfn = pfn;
460 while (!PageBuddy(pfn_to_page(outer_pfn))) {
461 /* stop if we cannot find the free page */
462 if (++order >= MAX_ORDER)
463 goto failed;
464 outer_pfn &= ~0UL << order;
465 }
466 pfn = outer_pfn;
467 continue;
468 } else
469#endif
470 goto failed;
471 }
472
473 pfn++;
474 }
475 return 0;
476failed:
477 /* restore the original migratetype */
478 if (!skip_isolation)
479 unset_migratetype_isolate(pfn_to_page(isolate_pageblock), migratetype);
480 return -EBUSY;
481}
482
483/**
484 * start_isolate_page_range() - make page-allocation-type of range of pages to
485 * be MIGRATE_ISOLATE.
486 * @start_pfn: The lower PFN of the range to be isolated.
487 * @end_pfn: The upper PFN of the range to be isolated.
488 * @migratetype: Migrate type to set in error recovery.
489 * @flags: The following flags are allowed (they can be combined in
490 * a bit mask)
491 * MEMORY_OFFLINE - isolate to offline (!allocate) memory
492 * e.g., skip over PageHWPoison() pages
493 * and PageOffline() pages.
494 * REPORT_FAILURE - report details about the failure to
495 * isolate the range
496 * @gfp_flags: GFP flags used for migrating pages that sit across the
497 * range boundaries.
498 *
499 * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
500 * the range will never be allocated. Any free pages and pages freed in the
501 * future will not be allocated again. If specified range includes migrate types
502 * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all
503 * pages in the range finally, the caller have to free all pages in the range.
504 * test_page_isolated() can be used for test it.
505 *
506 * The function first tries to isolate the pageblocks at the beginning and end
507 * of the range, since there might be pages across the range boundaries.
508 * Afterwards, it isolates the rest of the range.
509 *
510 * There is no high level synchronization mechanism that prevents two threads
511 * from trying to isolate overlapping ranges. If this happens, one thread
512 * will notice pageblocks in the overlapping range already set to isolate.
513 * This happens in set_migratetype_isolate, and set_migratetype_isolate
514 * returns an error. We then clean up by restoring the migration type on
515 * pageblocks we may have modified and return -EBUSY to caller. This
516 * prevents two threads from simultaneously working on overlapping ranges.
517 *
518 * Please note that there is no strong synchronization with the page allocator
519 * either. Pages might be freed while their page blocks are marked ISOLATED.
520 * A call to drain_all_pages() after isolation can flush most of them. However
521 * in some cases pages might still end up on pcp lists and that would allow
522 * for their allocation even when they are in fact isolated already. Depending
523 * on how strong of a guarantee the caller needs, zone_pcp_disable/enable()
524 * might be used to flush and disable pcplist before isolation and enable after
525 * unisolation.
526 *
527 * Return: 0 on success and -EBUSY if any part of range cannot be isolated.
528 */
529int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
530 int migratetype, int flags, gfp_t gfp_flags)
531{
532 unsigned long pfn;
533 struct page *page;
534 /* isolation is done at page block granularity */
535 unsigned long isolate_start = pageblock_start_pfn(start_pfn);
536 unsigned long isolate_end = pageblock_align(end_pfn);
537 int ret;
538 bool skip_isolation = false;
539
540 /* isolate [isolate_start, isolate_start + pageblock_nr_pages) pageblock */
541 ret = isolate_single_pageblock(isolate_start, flags, gfp_flags, false,
542 skip_isolation, migratetype);
543 if (ret)
544 return ret;
545
546 if (isolate_start == isolate_end - pageblock_nr_pages)
547 skip_isolation = true;
548
549 /* isolate [isolate_end - pageblock_nr_pages, isolate_end) pageblock */
550 ret = isolate_single_pageblock(isolate_end, flags, gfp_flags, true,
551 skip_isolation, migratetype);
552 if (ret) {
553 unset_migratetype_isolate(pfn_to_page(isolate_start), migratetype);
554 return ret;
555 }
556
557 /* skip isolated pageblocks at the beginning and end */
558 for (pfn = isolate_start + pageblock_nr_pages;
559 pfn < isolate_end - pageblock_nr_pages;
560 pfn += pageblock_nr_pages) {
561 page = __first_valid_page(pfn, pageblock_nr_pages);
562 if (page && set_migratetype_isolate(page, migratetype, flags,
563 start_pfn, end_pfn)) {
564 undo_isolate_page_range(isolate_start, pfn, migratetype);
565 unset_migratetype_isolate(
566 pfn_to_page(isolate_end - pageblock_nr_pages),
567 migratetype);
568 return -EBUSY;
569 }
570 }
571 return 0;
572}
573
574/*
575 * Make isolated pages available again.
576 */
577void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
578 int migratetype)
579{
580 unsigned long pfn;
581 struct page *page;
582 unsigned long isolate_start = pageblock_start_pfn(start_pfn);
583 unsigned long isolate_end = pageblock_align(end_pfn);
584
585 for (pfn = isolate_start;
586 pfn < isolate_end;
587 pfn += pageblock_nr_pages) {
588 page = __first_valid_page(pfn, pageblock_nr_pages);
589 if (!page || !is_migrate_isolate_page(page))
590 continue;
591 unset_migratetype_isolate(page, migratetype);
592 }
593}
594/*
595 * Test all pages in the range is free(means isolated) or not.
596 * all pages in [start_pfn...end_pfn) must be in the same zone.
597 * zone->lock must be held before call this.
598 *
599 * Returns the last tested pfn.
600 */
601static unsigned long
602__test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
603 int flags)
604{
605 struct page *page;
606
607 while (pfn < end_pfn) {
608 page = pfn_to_page(pfn);
609 if (PageBuddy(page))
610 /*
611 * If the page is on a free list, it has to be on
612 * the correct MIGRATE_ISOLATE freelist. There is no
613 * simple way to verify that as VM_BUG_ON(), though.
614 */
615 pfn += 1 << buddy_order(page);
616 else if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
617 /* A HWPoisoned page cannot be also PageBuddy */
618 pfn++;
619 else if ((flags & MEMORY_OFFLINE) && PageOffline(page) &&
620 !page_count(page))
621 /*
622 * The responsible driver agreed to skip PageOffline()
623 * pages when offlining memory by dropping its
624 * reference in MEM_GOING_OFFLINE.
625 */
626 pfn++;
627 else
628 break;
629 }
630
631 return pfn;
632}
633
634/* Caller should ensure that requested range is in a single zone */
635int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
636 int isol_flags)
637{
638 unsigned long pfn, flags;
639 struct page *page;
640 struct zone *zone;
641 int ret;
642
643 /*
644 * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
645 * are not aligned to pageblock_nr_pages.
646 * Then we just check migratetype first.
647 */
648 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
649 page = __first_valid_page(pfn, pageblock_nr_pages);
650 if (page && !is_migrate_isolate_page(page))
651 break;
652 }
653 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
654 if ((pfn < end_pfn) || !page) {
655 ret = -EBUSY;
656 goto out;
657 }
658
659 /* Check all pages are free or marked as ISOLATED */
660 zone = page_zone(page);
661 spin_lock_irqsave(&zone->lock, flags);
662 pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, isol_flags);
663 spin_unlock_irqrestore(&zone->lock, flags);
664
665 ret = pfn < end_pfn ? -EBUSY : 0;
666
667out:
668 trace_test_pages_isolated(start_pfn, end_pfn, pfn);
669
670 return ret;
671}