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 | { |
1c31cb49 DH |
20 | struct zone *zone = page_zone(page); |
21 | struct page *unmovable; | |
3f9903b9 | 22 | unsigned long flags; |
ee6f509c MK |
23 | |
24 | spin_lock_irqsave(&zone->lock, flags); | |
25 | ||
2c7452a0 MK |
26 | /* |
27 | * We assume the caller intended to SET migrate type to isolate. | |
28 | * If it is already set, then someone else must have raced and | |
51030a53 | 29 | * set it before us. |
2c7452a0 | 30 | */ |
51030a53 DH |
31 | if (is_migrate_isolate_page(page)) { |
32 | spin_unlock_irqrestore(&zone->lock, flags); | |
33 | return -EBUSY; | |
34 | } | |
2c7452a0 | 35 | |
ee6f509c MK |
36 | /* |
37 | * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself. | |
38 | * We just check MOVABLE pages. | |
39 | */ | |
4a55c047 QC |
40 | unmovable = has_unmovable_pages(zone, page, migratetype, isol_flags); |
41 | if (!unmovable) { | |
2139cbe6 | 42 | unsigned long nr_pages; |
4da2ce25 | 43 | int mt = get_pageblock_migratetype(page); |
2139cbe6 | 44 | |
a458431e | 45 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); |
ad53f92e | 46 | zone->nr_isolate_pageblock++; |
02aa0cdd VB |
47 | nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE, |
48 | NULL); | |
2139cbe6 | 49 | |
4da2ce25 | 50 | __mod_zone_freepage_state(zone, -nr_pages, mt); |
1c31cb49 | 51 | spin_unlock_irqrestore(&zone->lock, flags); |
1c31cb49 | 52 | return 0; |
ee6f509c MK |
53 | } |
54 | ||
55 | spin_unlock_irqrestore(&zone->lock, flags); | |
1c31cb49 | 56 | if (isol_flags & REPORT_FAILURE) { |
48381d7e DH |
57 | /* |
58 | * printk() with zone->lock held will likely trigger a | |
59 | * lockdep splat, so defer it here. | |
60 | */ | |
61 | dump_page(unmovable, "unmovable page"); | |
3d680bdf | 62 | } |
4a55c047 | 63 | |
1c31cb49 | 64 | return -EBUSY; |
ee6f509c MK |
65 | } |
66 | ||
c5b4e1b0 | 67 | static void unset_migratetype_isolate(struct page *page, unsigned migratetype) |
ee6f509c MK |
68 | { |
69 | struct zone *zone; | |
2139cbe6 | 70 | unsigned long flags, nr_pages; |
e3a2713c | 71 | bool isolated_page = false; |
3c605096 | 72 | unsigned int order; |
76741e77 | 73 | unsigned long pfn, buddy_pfn; |
3c605096 | 74 | struct page *buddy; |
2139cbe6 | 75 | |
ee6f509c MK |
76 | zone = page_zone(page); |
77 | spin_lock_irqsave(&zone->lock, flags); | |
bbf9ce97 | 78 | if (!is_migrate_isolate_page(page)) |
ee6f509c | 79 | goto out; |
3c605096 JK |
80 | |
81 | /* | |
82 | * Because freepage with more than pageblock_order on isolated | |
83 | * pageblock is restricted to merge due to freepage counting problem, | |
84 | * it is possible that there is free buddy page. | |
85 | * move_freepages_block() doesn't care of merge so we need other | |
86 | * approach in order to merge them. Isolation and free will make | |
87 | * these pages to be merged. | |
88 | */ | |
89 | if (PageBuddy(page)) { | |
ab130f91 | 90 | order = buddy_order(page); |
2484be0f | 91 | if (order >= pageblock_order && order < MAX_ORDER - 1) { |
76741e77 VB |
92 | pfn = page_to_pfn(page); |
93 | buddy_pfn = __find_buddy_pfn(pfn, order); | |
94 | buddy = page + (buddy_pfn - pfn); | |
3c605096 | 95 | |
859a85dd | 96 | if (!is_migrate_isolate_page(buddy)) { |
a500cb34 ML |
97 | isolated_page = !!__isolate_free_page(page, order); |
98 | /* | |
99 | * Isolating a free page in an isolated pageblock | |
100 | * is expected to always work as watermarks don't | |
101 | * apply here. | |
102 | */ | |
103 | VM_WARN_ON(!isolated_page); | |
3c605096 JK |
104 | } |
105 | } | |
106 | } | |
107 | ||
108 | /* | |
109 | * If we isolate freepage with more than pageblock_order, there | |
110 | * should be no freepage in the range, so we could avoid costly | |
111 | * pageblock scanning for freepage moving. | |
293ffa5e DH |
112 | * |
113 | * We didn't actually touch any of the isolated pages, so place them | |
114 | * to the tail of the freelist. This is an optimization for memory | |
115 | * onlining - just onlined memory won't immediately be considered for | |
116 | * allocation. | |
3c605096 | 117 | */ |
a85468b7 | 118 | if (!isolated_page) { |
02aa0cdd | 119 | nr_pages = move_freepages_block(zone, page, migratetype, NULL); |
3c605096 JK |
120 | __mod_zone_freepage_state(zone, nr_pages, migratetype); |
121 | } | |
a458431e | 122 | set_pageblock_migratetype(page, migratetype); |
624f58d8 AD |
123 | if (isolated_page) |
124 | __putback_isolated_page(page, order, migratetype); | |
ad53f92e | 125 | zone->nr_isolate_pageblock--; |
ee6f509c MK |
126 | out: |
127 | spin_unlock_irqrestore(&zone->lock, flags); | |
128 | } | |
129 | ||
a5d76b54 KH |
130 | static inline struct page * |
131 | __first_valid_page(unsigned long pfn, unsigned long nr_pages) | |
132 | { | |
133 | int i; | |
2ce13640 MH |
134 | |
135 | for (i = 0; i < nr_pages; i++) { | |
136 | struct page *page; | |
137 | ||
2ce13640 MH |
138 | page = pfn_to_online_page(pfn + i); |
139 | if (!page) | |
140 | continue; | |
141 | return page; | |
142 | } | |
143 | return NULL; | |
a5d76b54 KH |
144 | } |
145 | ||
9b7ea46a QC |
146 | /** |
147 | * start_isolate_page_range() - make page-allocation-type of range of pages to | |
148 | * be MIGRATE_ISOLATE. | |
149 | * @start_pfn: The lower PFN of the range to be isolated. | |
150 | * @end_pfn: The upper PFN of the range to be isolated. | |
151 | * start_pfn/end_pfn must be aligned to pageblock_order. | |
152 | * @migratetype: Migrate type to set in error recovery. | |
153 | * @flags: The following flags are allowed (they can be combined in | |
154 | * a bit mask) | |
756d25be DH |
155 | * MEMORY_OFFLINE - isolate to offline (!allocate) memory |
156 | * e.g., skip over PageHWPoison() pages | |
aa218795 | 157 | * and PageOffline() pages. |
9b7ea46a QC |
158 | * REPORT_FAILURE - report details about the failure to |
159 | * isolate the range | |
a5d76b54 KH |
160 | * |
161 | * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in | |
162 | * the range will never be allocated. Any free pages and pages freed in the | |
9b7ea46a QC |
163 | * future will not be allocated again. If specified range includes migrate types |
164 | * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all | |
165 | * pages in the range finally, the caller have to free all pages in the range. | |
166 | * test_page_isolated() can be used for test it. | |
2c7452a0 MK |
167 | * |
168 | * There is no high level synchronization mechanism that prevents two threads | |
9b7ea46a | 169 | * from trying to isolate overlapping ranges. If this happens, one thread |
2c7452a0 MK |
170 | * will notice pageblocks in the overlapping range already set to isolate. |
171 | * This happens in set_migratetype_isolate, and set_migratetype_isolate | |
9b7ea46a QC |
172 | * returns an error. We then clean up by restoring the migration type on |
173 | * pageblocks we may have modified and return -EBUSY to caller. This | |
2c7452a0 | 174 | * prevents two threads from simultaneously working on overlapping ranges. |
9b7ea46a | 175 | * |
96831826 PT |
176 | * Please note that there is no strong synchronization with the page allocator |
177 | * either. Pages might be freed while their page blocks are marked ISOLATED. | |
7612921f VB |
178 | * A call to drain_all_pages() after isolation can flush most of them. However |
179 | * in some cases pages might still end up on pcp lists and that would allow | |
96831826 | 180 | * for their allocation even when they are in fact isolated already. Depending |
ec6e8c7e VB |
181 | * on how strong of a guarantee the caller needs, zone_pcp_disable/enable() |
182 | * might be used to flush and disable pcplist before isolation and enable after | |
183 | * unisolation. | |
96831826 | 184 | * |
3fa0c7c7 | 185 | * Return: 0 on success and -EBUSY if any part of range cannot be isolated. |
a5d76b54 | 186 | */ |
0815f3d8 | 187 | int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn, |
d381c547 | 188 | unsigned migratetype, int flags) |
a5d76b54 KH |
189 | { |
190 | unsigned long pfn; | |
a5d76b54 KH |
191 | struct page *page; |
192 | ||
fec174d6 NH |
193 | BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages)); |
194 | BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages)); | |
a5d76b54 KH |
195 | |
196 | for (pfn = start_pfn; | |
197 | pfn < end_pfn; | |
198 | pfn += pageblock_nr_pages) { | |
199 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
e1d8c966 ML |
200 | if (page && set_migratetype_isolate(page, migratetype, flags)) { |
201 | undo_isolate_page_range(start_pfn, pfn, migratetype); | |
202 | return -EBUSY; | |
a5d76b54 KH |
203 | } |
204 | } | |
3fa0c7c7 | 205 | return 0; |
a5d76b54 KH |
206 | } |
207 | ||
208 | /* | |
209 | * Make isolated pages available again. | |
210 | */ | |
1fcf0a56 | 211 | void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn, |
0815f3d8 | 212 | unsigned migratetype) |
a5d76b54 KH |
213 | { |
214 | unsigned long pfn; | |
215 | struct page *page; | |
6f8d2b8a WX |
216 | |
217 | BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages)); | |
218 | BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages)); | |
219 | ||
a5d76b54 KH |
220 | for (pfn = start_pfn; |
221 | pfn < end_pfn; | |
222 | pfn += pageblock_nr_pages) { | |
223 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
bbf9ce97 | 224 | if (!page || !is_migrate_isolate_page(page)) |
a5d76b54 | 225 | continue; |
0815f3d8 | 226 | unset_migratetype_isolate(page, migratetype); |
a5d76b54 | 227 | } |
a5d76b54 KH |
228 | } |
229 | /* | |
230 | * Test all pages in the range is free(means isolated) or not. | |
231 | * all pages in [start_pfn...end_pfn) must be in the same zone. | |
232 | * zone->lock must be held before call this. | |
233 | * | |
ec3b6882 | 234 | * Returns the last tested pfn. |
a5d76b54 | 235 | */ |
fea85cff | 236 | static unsigned long |
b023f468 | 237 | __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn, |
756d25be | 238 | int flags) |
a5d76b54 KH |
239 | { |
240 | struct page *page; | |
241 | ||
242 | while (pfn < end_pfn) { | |
a5d76b54 | 243 | page = pfn_to_page(pfn); |
aa016d14 | 244 | if (PageBuddy(page)) |
435b405c | 245 | /* |
aa016d14 VB |
246 | * If the page is on a free list, it has to be on |
247 | * the correct MIGRATE_ISOLATE freelist. There is no | |
248 | * simple way to verify that as VM_BUG_ON(), though. | |
435b405c | 249 | */ |
ab130f91 | 250 | pfn += 1 << buddy_order(page); |
756d25be | 251 | else if ((flags & MEMORY_OFFLINE) && PageHWPoison(page)) |
aa016d14 | 252 | /* A HWPoisoned page cannot be also PageBuddy */ |
b023f468 | 253 | pfn++; |
aa218795 DH |
254 | else if ((flags & MEMORY_OFFLINE) && PageOffline(page) && |
255 | !page_count(page)) | |
256 | /* | |
257 | * The responsible driver agreed to skip PageOffline() | |
258 | * pages when offlining memory by dropping its | |
259 | * reference in MEM_GOING_OFFLINE. | |
260 | */ | |
261 | pfn++; | |
a5d76b54 KH |
262 | else |
263 | break; | |
264 | } | |
fea85cff JK |
265 | |
266 | return pfn; | |
a5d76b54 KH |
267 | } |
268 | ||
b9eb6319 | 269 | /* Caller should ensure that requested range is in a single zone */ |
b023f468 | 270 | int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn, |
756d25be | 271 | int isol_flags) |
a5d76b54 | 272 | { |
6c1b7f68 | 273 | unsigned long pfn, flags; |
a5d76b54 | 274 | struct page *page; |
6c1b7f68 | 275 | struct zone *zone; |
1d09510b | 276 | int ret; |
a5d76b54 | 277 | |
a5d76b54 | 278 | /* |
85dbe706 TC |
279 | * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages |
280 | * are not aligned to pageblock_nr_pages. | |
281 | * Then we just check migratetype first. | |
a5d76b54 KH |
282 | */ |
283 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
284 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
bbf9ce97 | 285 | if (page && !is_migrate_isolate_page(page)) |
a5d76b54 KH |
286 | break; |
287 | } | |
a70dcb96 | 288 | page = __first_valid_page(start_pfn, end_pfn - start_pfn); |
1d09510b GD |
289 | if ((pfn < end_pfn) || !page) { |
290 | ret = -EBUSY; | |
291 | goto out; | |
292 | } | |
293 | ||
85dbe706 | 294 | /* Check all pages are free or marked as ISOLATED */ |
a70dcb96 | 295 | zone = page_zone(page); |
6c1b7f68 | 296 | spin_lock_irqsave(&zone->lock, flags); |
756d25be | 297 | pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, isol_flags); |
6c1b7f68 | 298 | spin_unlock_irqrestore(&zone->lock, flags); |
fea85cff | 299 | |
1d09510b GD |
300 | ret = pfn < end_pfn ? -EBUSY : 0; |
301 | ||
302 | out: | |
0f0848e5 JK |
303 | trace_test_pages_isolated(start_pfn, end_pfn, pfn); |
304 | ||
1d09510b | 305 | return ret; |
a5d76b54 | 306 | } |