Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
a5d76b54 KH |
2 | /* |
3 | * linux/mm/page_isolation.c | |
4 | */ | |
5 | ||
a5d76b54 KH |
6 | #include <linux/mm.h> |
7 | #include <linux/page-isolation.h> | |
8 | #include <linux/pageblock-flags.h> | |
ee6f509c | 9 | #include <linux/memory.h> |
c8721bbb | 10 | #include <linux/hugetlb.h> |
83358ece | 11 | #include <linux/page_owner.h> |
8b913238 | 12 | #include <linux/migrate.h> |
a5d76b54 KH |
13 | #include "internal.h" |
14 | ||
0f0848e5 JK |
15 | #define CREATE_TRACE_POINTS |
16 | #include <trace/events/page_isolation.h> | |
17 | ||
d381c547 | 18 | static int set_migratetype_isolate(struct page *page, int migratetype, int isol_flags) |
ee6f509c MK |
19 | { |
20 | struct zone *zone; | |
21 | unsigned long flags, pfn; | |
22 | struct memory_isolate_notify arg; | |
23 | int notifier_ret; | |
24 | int ret = -EBUSY; | |
25 | ||
26 | zone = page_zone(page); | |
27 | ||
28 | spin_lock_irqsave(&zone->lock, flags); | |
29 | ||
2c7452a0 MK |
30 | /* |
31 | * We assume the caller intended to SET migrate type to isolate. | |
32 | * If it is already set, then someone else must have raced and | |
33 | * set it before us. Return -EBUSY | |
34 | */ | |
35 | if (is_migrate_isolate_page(page)) | |
36 | goto out; | |
37 | ||
ee6f509c MK |
38 | pfn = page_to_pfn(page); |
39 | arg.start_pfn = pfn; | |
40 | arg.nr_pages = pageblock_nr_pages; | |
41 | arg.pages_found = 0; | |
42 | ||
43 | /* | |
44 | * It may be possible to isolate a pageblock even if the | |
45 | * migratetype is not MIGRATE_MOVABLE. The memory isolation | |
46 | * notifier chain is used by balloon drivers to return the | |
47 | * number of pages in a range that are held by the balloon | |
48 | * driver to shrink memory. If all the pages are accounted for | |
49 | * by balloons, are free, or on the LRU, isolation can continue. | |
50 | * Later, for example, when memory hotplug notifier runs, these | |
51 | * pages reported as "can be isolated" should be isolated(freed) | |
52 | * by the balloon driver through the memory notifier chain. | |
53 | */ | |
54 | notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg); | |
55 | notifier_ret = notifier_to_errno(notifier_ret); | |
56 | if (notifier_ret) | |
57 | goto out; | |
58 | /* | |
59 | * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself. | |
60 | * We just check MOVABLE pages. | |
61 | */ | |
d381c547 | 62 | if (!has_unmovable_pages(zone, page, arg.pages_found, migratetype, flags)) |
ee6f509c MK |
63 | ret = 0; |
64 | ||
65 | /* | |
ac34dcd2 | 66 | * immobile means "not-on-lru" pages. If immobile is larger than |
ee6f509c MK |
67 | * removable-by-driver pages reported by notifier, we'll fail. |
68 | */ | |
69 | ||
70 | out: | |
71 | if (!ret) { | |
2139cbe6 | 72 | unsigned long nr_pages; |
4da2ce25 | 73 | int mt = get_pageblock_migratetype(page); |
2139cbe6 | 74 | |
a458431e | 75 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); |
ad53f92e | 76 | zone->nr_isolate_pageblock++; |
02aa0cdd VB |
77 | nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE, |
78 | NULL); | |
2139cbe6 | 79 | |
4da2ce25 | 80 | __mod_zone_freepage_state(zone, -nr_pages, mt); |
ee6f509c MK |
81 | } |
82 | ||
83 | spin_unlock_irqrestore(&zone->lock, flags); | |
84 | if (!ret) | |
ec25af84 | 85 | drain_all_pages(zone); |
ee6f509c MK |
86 | return ret; |
87 | } | |
88 | ||
c5b4e1b0 | 89 | static void unset_migratetype_isolate(struct page *page, unsigned migratetype) |
ee6f509c MK |
90 | { |
91 | struct zone *zone; | |
2139cbe6 | 92 | unsigned long flags, nr_pages; |
e3a2713c | 93 | bool isolated_page = false; |
3c605096 | 94 | unsigned int order; |
76741e77 | 95 | unsigned long pfn, buddy_pfn; |
3c605096 | 96 | struct page *buddy; |
2139cbe6 | 97 | |
ee6f509c MK |
98 | zone = page_zone(page); |
99 | spin_lock_irqsave(&zone->lock, flags); | |
bbf9ce97 | 100 | if (!is_migrate_isolate_page(page)) |
ee6f509c | 101 | goto out; |
3c605096 JK |
102 | |
103 | /* | |
104 | * Because freepage with more than pageblock_order on isolated | |
105 | * pageblock is restricted to merge due to freepage counting problem, | |
106 | * it is possible that there is free buddy page. | |
107 | * move_freepages_block() doesn't care of merge so we need other | |
108 | * approach in order to merge them. Isolation and free will make | |
109 | * these pages to be merged. | |
110 | */ | |
111 | if (PageBuddy(page)) { | |
112 | order = page_order(page); | |
113 | if (order >= pageblock_order) { | |
76741e77 VB |
114 | pfn = page_to_pfn(page); |
115 | buddy_pfn = __find_buddy_pfn(pfn, order); | |
116 | buddy = page + (buddy_pfn - pfn); | |
3c605096 | 117 | |
13ad59df | 118 | if (pfn_valid_within(buddy_pfn) && |
1ae7013d | 119 | !is_migrate_isolate_page(buddy)) { |
3c605096 | 120 | __isolate_free_page(page, order); |
e3a2713c | 121 | isolated_page = true; |
3c605096 JK |
122 | } |
123 | } | |
124 | } | |
125 | ||
126 | /* | |
127 | * If we isolate freepage with more than pageblock_order, there | |
128 | * should be no freepage in the range, so we could avoid costly | |
129 | * pageblock scanning for freepage moving. | |
130 | */ | |
131 | if (!isolated_page) { | |
02aa0cdd | 132 | nr_pages = move_freepages_block(zone, page, migratetype, NULL); |
3c605096 JK |
133 | __mod_zone_freepage_state(zone, nr_pages, migratetype); |
134 | } | |
a458431e | 135 | set_pageblock_migratetype(page, migratetype); |
ad53f92e | 136 | zone->nr_isolate_pageblock--; |
ee6f509c MK |
137 | out: |
138 | spin_unlock_irqrestore(&zone->lock, flags); | |
83358ece | 139 | if (isolated_page) { |
46f24fd8 | 140 | post_alloc_hook(page, order, __GFP_MOVABLE); |
e3a2713c | 141 | __free_pages(page, order); |
83358ece | 142 | } |
ee6f509c MK |
143 | } |
144 | ||
a5d76b54 KH |
145 | static inline struct page * |
146 | __first_valid_page(unsigned long pfn, unsigned long nr_pages) | |
147 | { | |
148 | int i; | |
2ce13640 MH |
149 | |
150 | for (i = 0; i < nr_pages; i++) { | |
151 | struct page *page; | |
152 | ||
153 | if (!pfn_valid_within(pfn + i)) | |
154 | continue; | |
155 | page = pfn_to_online_page(pfn + i); | |
156 | if (!page) | |
157 | continue; | |
158 | return page; | |
159 | } | |
160 | return NULL; | |
a5d76b54 KH |
161 | } |
162 | ||
163 | /* | |
164 | * start_isolate_page_range() -- make page-allocation-type of range of pages | |
165 | * to be MIGRATE_ISOLATE. | |
166 | * @start_pfn: The lower PFN of the range to be isolated. | |
167 | * @end_pfn: The upper PFN of the range to be isolated. | |
0815f3d8 | 168 | * @migratetype: migrate type to set in error recovery. |
a5d76b54 KH |
169 | * |
170 | * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in | |
171 | * the range will never be allocated. Any free pages and pages freed in the | |
172 | * future will not be allocated again. | |
173 | * | |
174 | * start_pfn/end_pfn must be aligned to pageblock_order. | |
2c7452a0 MK |
175 | * Return 0 on success and -EBUSY if any part of range cannot be isolated. |
176 | * | |
177 | * There is no high level synchronization mechanism that prevents two threads | |
178 | * from trying to isolate overlapping ranges. If this happens, one thread | |
179 | * will notice pageblocks in the overlapping range already set to isolate. | |
180 | * This happens in set_migratetype_isolate, and set_migratetype_isolate | |
181 | * returns an error. We then clean up by restoring the migration type on | |
182 | * pageblocks we may have modified and return -EBUSY to caller. This | |
183 | * prevents two threads from simultaneously working on overlapping ranges. | |
a5d76b54 | 184 | */ |
0815f3d8 | 185 | int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn, |
d381c547 | 186 | unsigned migratetype, int flags) |
a5d76b54 KH |
187 | { |
188 | unsigned long pfn; | |
189 | unsigned long undo_pfn; | |
190 | struct page *page; | |
191 | ||
fec174d6 NH |
192 | BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages)); |
193 | BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages)); | |
a5d76b54 KH |
194 | |
195 | for (pfn = start_pfn; | |
196 | pfn < end_pfn; | |
197 | pfn += pageblock_nr_pages) { | |
198 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
b023f468 | 199 | if (page && |
d381c547 | 200 | set_migratetype_isolate(page, migratetype, flags)) { |
a5d76b54 KH |
201 | undo_pfn = pfn; |
202 | goto undo; | |
203 | } | |
204 | } | |
205 | return 0; | |
206 | undo: | |
207 | for (pfn = start_pfn; | |
dbc0e4ce | 208 | pfn < undo_pfn; |
2ce13640 MH |
209 | pfn += pageblock_nr_pages) { |
210 | struct page *page = pfn_to_online_page(pfn); | |
211 | if (!page) | |
212 | continue; | |
213 | unset_migratetype_isolate(page, migratetype); | |
214 | } | |
a5d76b54 KH |
215 | |
216 | return -EBUSY; | |
217 | } | |
218 | ||
219 | /* | |
220 | * Make isolated pages available again. | |
221 | */ | |
0815f3d8 MN |
222 | int undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn, |
223 | unsigned migratetype) | |
a5d76b54 KH |
224 | { |
225 | unsigned long pfn; | |
226 | struct page *page; | |
6f8d2b8a WX |
227 | |
228 | BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages)); | |
229 | BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages)); | |
230 | ||
a5d76b54 KH |
231 | for (pfn = start_pfn; |
232 | pfn < end_pfn; | |
233 | pfn += pageblock_nr_pages) { | |
234 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
bbf9ce97 | 235 | if (!page || !is_migrate_isolate_page(page)) |
a5d76b54 | 236 | continue; |
0815f3d8 | 237 | unset_migratetype_isolate(page, migratetype); |
a5d76b54 KH |
238 | } |
239 | return 0; | |
240 | } | |
241 | /* | |
242 | * Test all pages in the range is free(means isolated) or not. | |
243 | * all pages in [start_pfn...end_pfn) must be in the same zone. | |
244 | * zone->lock must be held before call this. | |
245 | * | |
ec3b6882 | 246 | * Returns the last tested pfn. |
a5d76b54 | 247 | */ |
fea85cff | 248 | static unsigned long |
b023f468 WC |
249 | __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn, |
250 | bool skip_hwpoisoned_pages) | |
a5d76b54 KH |
251 | { |
252 | struct page *page; | |
253 | ||
254 | while (pfn < end_pfn) { | |
255 | if (!pfn_valid_within(pfn)) { | |
256 | pfn++; | |
257 | continue; | |
258 | } | |
259 | page = pfn_to_page(pfn); | |
aa016d14 | 260 | if (PageBuddy(page)) |
435b405c | 261 | /* |
aa016d14 VB |
262 | * If the page is on a free list, it has to be on |
263 | * the correct MIGRATE_ISOLATE freelist. There is no | |
264 | * simple way to verify that as VM_BUG_ON(), though. | |
435b405c | 265 | */ |
a5d76b54 | 266 | pfn += 1 << page_order(page); |
aa016d14 VB |
267 | else if (skip_hwpoisoned_pages && PageHWPoison(page)) |
268 | /* A HWPoisoned page cannot be also PageBuddy */ | |
b023f468 | 269 | pfn++; |
a5d76b54 KH |
270 | else |
271 | break; | |
272 | } | |
fea85cff JK |
273 | |
274 | return pfn; | |
a5d76b54 KH |
275 | } |
276 | ||
b9eb6319 | 277 | /* Caller should ensure that requested range is in a single zone */ |
b023f468 WC |
278 | int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn, |
279 | bool skip_hwpoisoned_pages) | |
a5d76b54 | 280 | { |
6c1b7f68 | 281 | unsigned long pfn, flags; |
a5d76b54 | 282 | struct page *page; |
6c1b7f68 | 283 | struct zone *zone; |
a5d76b54 | 284 | |
a5d76b54 | 285 | /* |
85dbe706 TC |
286 | * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages |
287 | * are not aligned to pageblock_nr_pages. | |
288 | * Then we just check migratetype first. | |
a5d76b54 KH |
289 | */ |
290 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
291 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
bbf9ce97 | 292 | if (page && !is_migrate_isolate_page(page)) |
a5d76b54 KH |
293 | break; |
294 | } | |
a70dcb96 GS |
295 | page = __first_valid_page(start_pfn, end_pfn - start_pfn); |
296 | if ((pfn < end_pfn) || !page) | |
a5d76b54 | 297 | return -EBUSY; |
85dbe706 | 298 | /* Check all pages are free or marked as ISOLATED */ |
a70dcb96 | 299 | zone = page_zone(page); |
6c1b7f68 | 300 | spin_lock_irqsave(&zone->lock, flags); |
fea85cff | 301 | pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, |
b023f468 | 302 | skip_hwpoisoned_pages); |
6c1b7f68 | 303 | spin_unlock_irqrestore(&zone->lock, flags); |
fea85cff | 304 | |
0f0848e5 JK |
305 | trace_test_pages_isolated(start_pfn, end_pfn, pfn); |
306 | ||
fea85cff | 307 | return pfn < end_pfn ? -EBUSY : 0; |
a5d76b54 | 308 | } |
723a0644 | 309 | |
666feb21 | 310 | struct page *alloc_migrate_target(struct page *page, unsigned long private) |
723a0644 | 311 | { |
8b913238 | 312 | return new_page_nodemask(page, numa_node_id(), &node_states[N_MEMORY]); |
723a0644 | 313 | } |