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 | 19 | { |
4a55c047 | 20 | struct page *unmovable = NULL; |
ee6f509c | 21 | struct zone *zone; |
3f9903b9 | 22 | unsigned long flags; |
ee6f509c MK |
23 | int ret = -EBUSY; |
24 | ||
25 | zone = page_zone(page); | |
26 | ||
27 | spin_lock_irqsave(&zone->lock, flags); | |
28 | ||
2c7452a0 MK |
29 | /* |
30 | * We assume the caller intended to SET migrate type to isolate. | |
31 | * If it is already set, then someone else must have raced and | |
32 | * set it before us. Return -EBUSY | |
33 | */ | |
34 | if (is_migrate_isolate_page(page)) | |
35 | goto out; | |
36 | ||
ee6f509c MK |
37 | /* |
38 | * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself. | |
39 | * We just check MOVABLE pages. | |
40 | */ | |
4a55c047 QC |
41 | unmovable = has_unmovable_pages(zone, page, migratetype, isol_flags); |
42 | if (!unmovable) { | |
2139cbe6 | 43 | unsigned long nr_pages; |
4da2ce25 | 44 | int mt = get_pageblock_migratetype(page); |
2139cbe6 | 45 | |
a458431e | 46 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); |
ad53f92e | 47 | zone->nr_isolate_pageblock++; |
02aa0cdd VB |
48 | nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE, |
49 | NULL); | |
2139cbe6 | 50 | |
4da2ce25 | 51 | __mod_zone_freepage_state(zone, -nr_pages, mt); |
3f9903b9 | 52 | ret = 0; |
ee6f509c MK |
53 | } |
54 | ||
3f9903b9 | 55 | out: |
ee6f509c | 56 | spin_unlock_irqrestore(&zone->lock, flags); |
3d680bdf | 57 | if (!ret) { |
ec25af84 | 58 | drain_all_pages(zone); |
3d680bdf QC |
59 | } else { |
60 | WARN_ON_ONCE(zone_idx(zone) == ZONE_MOVABLE); | |
61 | ||
62 | if ((isol_flags & REPORT_FAILURE) && unmovable) | |
63 | /* | |
64 | * printk() with zone->lock held will likely trigger a | |
65 | * lockdep splat, so defer it here. | |
66 | */ | |
67 | dump_page(unmovable, "unmovable page"); | |
68 | } | |
4a55c047 | 69 | |
ee6f509c MK |
70 | return ret; |
71 | } | |
72 | ||
c5b4e1b0 | 73 | static void unset_migratetype_isolate(struct page *page, unsigned migratetype) |
ee6f509c MK |
74 | { |
75 | struct zone *zone; | |
2139cbe6 | 76 | unsigned long flags, nr_pages; |
e3a2713c | 77 | bool isolated_page = false; |
3c605096 | 78 | unsigned int order; |
76741e77 | 79 | unsigned long pfn, buddy_pfn; |
3c605096 | 80 | struct page *buddy; |
2139cbe6 | 81 | |
ee6f509c MK |
82 | zone = page_zone(page); |
83 | spin_lock_irqsave(&zone->lock, flags); | |
bbf9ce97 | 84 | if (!is_migrate_isolate_page(page)) |
ee6f509c | 85 | goto out; |
3c605096 JK |
86 | |
87 | /* | |
88 | * Because freepage with more than pageblock_order on isolated | |
89 | * pageblock is restricted to merge due to freepage counting problem, | |
90 | * it is possible that there is free buddy page. | |
91 | * move_freepages_block() doesn't care of merge so we need other | |
92 | * approach in order to merge them. Isolation and free will make | |
93 | * these pages to be merged. | |
94 | */ | |
95 | if (PageBuddy(page)) { | |
96 | order = page_order(page); | |
97 | if (order >= pageblock_order) { | |
76741e77 VB |
98 | pfn = page_to_pfn(page); |
99 | buddy_pfn = __find_buddy_pfn(pfn, order); | |
100 | buddy = page + (buddy_pfn - pfn); | |
3c605096 | 101 | |
13ad59df | 102 | if (pfn_valid_within(buddy_pfn) && |
1ae7013d | 103 | !is_migrate_isolate_page(buddy)) { |
3c605096 | 104 | __isolate_free_page(page, order); |
e3a2713c | 105 | isolated_page = true; |
3c605096 JK |
106 | } |
107 | } | |
108 | } | |
109 | ||
110 | /* | |
111 | * If we isolate freepage with more than pageblock_order, there | |
112 | * should be no freepage in the range, so we could avoid costly | |
113 | * pageblock scanning for freepage moving. | |
114 | */ | |
115 | if (!isolated_page) { | |
02aa0cdd | 116 | nr_pages = move_freepages_block(zone, page, migratetype, NULL); |
3c605096 JK |
117 | __mod_zone_freepage_state(zone, nr_pages, migratetype); |
118 | } | |
a458431e | 119 | set_pageblock_migratetype(page, migratetype); |
ad53f92e | 120 | zone->nr_isolate_pageblock--; |
ee6f509c MK |
121 | out: |
122 | spin_unlock_irqrestore(&zone->lock, flags); | |
83358ece | 123 | if (isolated_page) { |
46f24fd8 | 124 | post_alloc_hook(page, order, __GFP_MOVABLE); |
e3a2713c | 125 | __free_pages(page, order); |
83358ece | 126 | } |
ee6f509c MK |
127 | } |
128 | ||
a5d76b54 KH |
129 | static inline struct page * |
130 | __first_valid_page(unsigned long pfn, unsigned long nr_pages) | |
131 | { | |
132 | int i; | |
2ce13640 MH |
133 | |
134 | for (i = 0; i < nr_pages; i++) { | |
135 | struct page *page; | |
136 | ||
2ce13640 MH |
137 | page = pfn_to_online_page(pfn + i); |
138 | if (!page) | |
139 | continue; | |
140 | return page; | |
141 | } | |
142 | return NULL; | |
a5d76b54 KH |
143 | } |
144 | ||
9b7ea46a QC |
145 | /** |
146 | * start_isolate_page_range() - make page-allocation-type of range of pages to | |
147 | * be MIGRATE_ISOLATE. | |
148 | * @start_pfn: The lower PFN of the range to be isolated. | |
149 | * @end_pfn: The upper PFN of the range to be isolated. | |
150 | * start_pfn/end_pfn must be aligned to pageblock_order. | |
151 | * @migratetype: Migrate type to set in error recovery. | |
152 | * @flags: The following flags are allowed (they can be combined in | |
153 | * a bit mask) | |
756d25be DH |
154 | * MEMORY_OFFLINE - isolate to offline (!allocate) memory |
155 | * e.g., skip over PageHWPoison() pages | |
9b7ea46a QC |
156 | * REPORT_FAILURE - report details about the failure to |
157 | * isolate the range | |
a5d76b54 KH |
158 | * |
159 | * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in | |
160 | * the range will never be allocated. Any free pages and pages freed in the | |
9b7ea46a QC |
161 | * future will not be allocated again. If specified range includes migrate types |
162 | * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all | |
163 | * pages in the range finally, the caller have to free all pages in the range. | |
164 | * test_page_isolated() can be used for test it. | |
2c7452a0 MK |
165 | * |
166 | * There is no high level synchronization mechanism that prevents two threads | |
9b7ea46a | 167 | * from trying to isolate overlapping ranges. If this happens, one thread |
2c7452a0 MK |
168 | * will notice pageblocks in the overlapping range already set to isolate. |
169 | * This happens in set_migratetype_isolate, and set_migratetype_isolate | |
9b7ea46a QC |
170 | * returns an error. We then clean up by restoring the migration type on |
171 | * pageblocks we may have modified and return -EBUSY to caller. This | |
2c7452a0 | 172 | * prevents two threads from simultaneously working on overlapping ranges. |
9b7ea46a QC |
173 | * |
174 | * Return: the number of isolated pageblocks on success and -EBUSY if any part | |
175 | * of range cannot be isolated. | |
a5d76b54 | 176 | */ |
0815f3d8 | 177 | int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn, |
d381c547 | 178 | unsigned migratetype, int flags) |
a5d76b54 KH |
179 | { |
180 | unsigned long pfn; | |
181 | unsigned long undo_pfn; | |
182 | struct page *page; | |
9b7ea46a | 183 | int nr_isolate_pageblock = 0; |
a5d76b54 | 184 | |
fec174d6 NH |
185 | BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages)); |
186 | BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages)); | |
a5d76b54 KH |
187 | |
188 | for (pfn = start_pfn; | |
189 | pfn < end_pfn; | |
190 | pfn += pageblock_nr_pages) { | |
191 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
9b7ea46a QC |
192 | if (page) { |
193 | if (set_migratetype_isolate(page, migratetype, flags)) { | |
194 | undo_pfn = pfn; | |
195 | goto undo; | |
196 | } | |
197 | nr_isolate_pageblock++; | |
a5d76b54 KH |
198 | } |
199 | } | |
9b7ea46a | 200 | return nr_isolate_pageblock; |
a5d76b54 KH |
201 | undo: |
202 | for (pfn = start_pfn; | |
dbc0e4ce | 203 | pfn < undo_pfn; |
2ce13640 MH |
204 | pfn += pageblock_nr_pages) { |
205 | struct page *page = pfn_to_online_page(pfn); | |
206 | if (!page) | |
207 | continue; | |
208 | unset_migratetype_isolate(page, migratetype); | |
209 | } | |
a5d76b54 KH |
210 | |
211 | return -EBUSY; | |
212 | } | |
213 | ||
214 | /* | |
215 | * Make isolated pages available again. | |
216 | */ | |
1fcf0a56 | 217 | void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn, |
0815f3d8 | 218 | unsigned migratetype) |
a5d76b54 KH |
219 | { |
220 | unsigned long pfn; | |
221 | struct page *page; | |
6f8d2b8a WX |
222 | |
223 | BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages)); | |
224 | BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages)); | |
225 | ||
a5d76b54 KH |
226 | for (pfn = start_pfn; |
227 | pfn < end_pfn; | |
228 | pfn += pageblock_nr_pages) { | |
229 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
bbf9ce97 | 230 | if (!page || !is_migrate_isolate_page(page)) |
a5d76b54 | 231 | continue; |
0815f3d8 | 232 | unset_migratetype_isolate(page, migratetype); |
a5d76b54 | 233 | } |
a5d76b54 KH |
234 | } |
235 | /* | |
236 | * Test all pages in the range is free(means isolated) or not. | |
237 | * all pages in [start_pfn...end_pfn) must be in the same zone. | |
238 | * zone->lock must be held before call this. | |
239 | * | |
ec3b6882 | 240 | * Returns the last tested pfn. |
a5d76b54 | 241 | */ |
fea85cff | 242 | static unsigned long |
b023f468 | 243 | __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn, |
756d25be | 244 | int flags) |
a5d76b54 KH |
245 | { |
246 | struct page *page; | |
247 | ||
248 | while (pfn < end_pfn) { | |
249 | if (!pfn_valid_within(pfn)) { | |
250 | pfn++; | |
251 | continue; | |
252 | } | |
253 | page = pfn_to_page(pfn); | |
aa016d14 | 254 | if (PageBuddy(page)) |
435b405c | 255 | /* |
aa016d14 VB |
256 | * If the page is on a free list, it has to be on |
257 | * the correct MIGRATE_ISOLATE freelist. There is no | |
258 | * simple way to verify that as VM_BUG_ON(), though. | |
435b405c | 259 | */ |
a5d76b54 | 260 | pfn += 1 << page_order(page); |
756d25be | 261 | else if ((flags & MEMORY_OFFLINE) && PageHWPoison(page)) |
aa016d14 | 262 | /* A HWPoisoned page cannot be also PageBuddy */ |
b023f468 | 263 | pfn++; |
a5d76b54 KH |
264 | else |
265 | break; | |
266 | } | |
fea85cff JK |
267 | |
268 | return pfn; | |
a5d76b54 KH |
269 | } |
270 | ||
b9eb6319 | 271 | /* Caller should ensure that requested range is in a single zone */ |
b023f468 | 272 | int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn, |
756d25be | 273 | int isol_flags) |
a5d76b54 | 274 | { |
6c1b7f68 | 275 | unsigned long pfn, flags; |
a5d76b54 | 276 | struct page *page; |
6c1b7f68 | 277 | struct zone *zone; |
a5d76b54 | 278 | |
a5d76b54 | 279 | /* |
85dbe706 TC |
280 | * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages |
281 | * are not aligned to pageblock_nr_pages. | |
282 | * Then we just check migratetype first. | |
a5d76b54 KH |
283 | */ |
284 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
285 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
bbf9ce97 | 286 | if (page && !is_migrate_isolate_page(page)) |
a5d76b54 KH |
287 | break; |
288 | } | |
a70dcb96 GS |
289 | page = __first_valid_page(start_pfn, end_pfn - start_pfn); |
290 | if ((pfn < end_pfn) || !page) | |
a5d76b54 | 291 | return -EBUSY; |
85dbe706 | 292 | /* Check all pages are free or marked as ISOLATED */ |
a70dcb96 | 293 | zone = page_zone(page); |
6c1b7f68 | 294 | spin_lock_irqsave(&zone->lock, flags); |
756d25be | 295 | pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, isol_flags); |
6c1b7f68 | 296 | spin_unlock_irqrestore(&zone->lock, flags); |
fea85cff | 297 | |
0f0848e5 JK |
298 | trace_test_pages_isolated(start_pfn, end_pfn, pfn); |
299 | ||
fea85cff | 300 | return pfn < end_pfn ? -EBUSY : 0; |
a5d76b54 | 301 | } |
723a0644 | 302 | |
666feb21 | 303 | struct page *alloc_migrate_target(struct page *page, unsigned long private) |
723a0644 | 304 | { |
8b913238 | 305 | return new_page_nodemask(page, numa_node_id(), &node_states[N_MEMORY]); |
723a0644 | 306 | } |