Commit | Line | Data |
---|---|---|
a5d76b54 KH |
1 | /* |
2 | * linux/mm/page_isolation.c | |
3 | */ | |
4 | ||
a5d76b54 KH |
5 | #include <linux/mm.h> |
6 | #include <linux/page-isolation.h> | |
7 | #include <linux/pageblock-flags.h> | |
ee6f509c | 8 | #include <linux/memory.h> |
c8721bbb | 9 | #include <linux/hugetlb.h> |
83358ece | 10 | #include <linux/page_owner.h> |
a5d76b54 KH |
11 | #include "internal.h" |
12 | ||
0f0848e5 JK |
13 | #define CREATE_TRACE_POINTS |
14 | #include <trace/events/page_isolation.h> | |
15 | ||
c5b4e1b0 NH |
16 | static int set_migratetype_isolate(struct page *page, |
17 | bool skip_hwpoisoned_pages) | |
ee6f509c MK |
18 | { |
19 | struct zone *zone; | |
20 | unsigned long flags, pfn; | |
21 | struct memory_isolate_notify arg; | |
22 | int notifier_ret; | |
23 | int ret = -EBUSY; | |
24 | ||
25 | zone = page_zone(page); | |
26 | ||
27 | spin_lock_irqsave(&zone->lock, flags); | |
28 | ||
29 | pfn = page_to_pfn(page); | |
30 | arg.start_pfn = pfn; | |
31 | arg.nr_pages = pageblock_nr_pages; | |
32 | arg.pages_found = 0; | |
33 | ||
34 | /* | |
35 | * It may be possible to isolate a pageblock even if the | |
36 | * migratetype is not MIGRATE_MOVABLE. The memory isolation | |
37 | * notifier chain is used by balloon drivers to return the | |
38 | * number of pages in a range that are held by the balloon | |
39 | * driver to shrink memory. If all the pages are accounted for | |
40 | * by balloons, are free, or on the LRU, isolation can continue. | |
41 | * Later, for example, when memory hotplug notifier runs, these | |
42 | * pages reported as "can be isolated" should be isolated(freed) | |
43 | * by the balloon driver through the memory notifier chain. | |
44 | */ | |
45 | notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg); | |
46 | notifier_ret = notifier_to_errno(notifier_ret); | |
47 | if (notifier_ret) | |
48 | goto out; | |
49 | /* | |
50 | * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself. | |
51 | * We just check MOVABLE pages. | |
52 | */ | |
b023f468 WC |
53 | if (!has_unmovable_pages(zone, page, arg.pages_found, |
54 | skip_hwpoisoned_pages)) | |
ee6f509c MK |
55 | ret = 0; |
56 | ||
57 | /* | |
ac34dcd2 | 58 | * immobile means "not-on-lru" pages. If immobile is larger than |
ee6f509c MK |
59 | * removable-by-driver pages reported by notifier, we'll fail. |
60 | */ | |
61 | ||
62 | out: | |
63 | if (!ret) { | |
2139cbe6 | 64 | unsigned long nr_pages; |
d1ce749a | 65 | int migratetype = get_pageblock_migratetype(page); |
2139cbe6 | 66 | |
a458431e | 67 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); |
ad53f92e | 68 | zone->nr_isolate_pageblock++; |
02aa0cdd VB |
69 | nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE, |
70 | NULL); | |
2139cbe6 | 71 | |
d1ce749a | 72 | __mod_zone_freepage_state(zone, -nr_pages, migratetype); |
ee6f509c MK |
73 | } |
74 | ||
75 | spin_unlock_irqrestore(&zone->lock, flags); | |
76 | if (!ret) | |
ec25af84 | 77 | drain_all_pages(zone); |
ee6f509c MK |
78 | return ret; |
79 | } | |
80 | ||
c5b4e1b0 | 81 | static void unset_migratetype_isolate(struct page *page, unsigned migratetype) |
ee6f509c MK |
82 | { |
83 | struct zone *zone; | |
2139cbe6 | 84 | unsigned long flags, nr_pages; |
e3a2713c | 85 | bool isolated_page = false; |
3c605096 | 86 | unsigned int order; |
76741e77 | 87 | unsigned long pfn, buddy_pfn; |
3c605096 | 88 | struct page *buddy; |
2139cbe6 | 89 | |
ee6f509c MK |
90 | zone = page_zone(page); |
91 | spin_lock_irqsave(&zone->lock, flags); | |
bbf9ce97 | 92 | if (!is_migrate_isolate_page(page)) |
ee6f509c | 93 | goto out; |
3c605096 JK |
94 | |
95 | /* | |
96 | * Because freepage with more than pageblock_order on isolated | |
97 | * pageblock is restricted to merge due to freepage counting problem, | |
98 | * it is possible that there is free buddy page. | |
99 | * move_freepages_block() doesn't care of merge so we need other | |
100 | * approach in order to merge them. Isolation and free will make | |
101 | * these pages to be merged. | |
102 | */ | |
103 | if (PageBuddy(page)) { | |
104 | order = page_order(page); | |
105 | if (order >= pageblock_order) { | |
76741e77 VB |
106 | pfn = page_to_pfn(page); |
107 | buddy_pfn = __find_buddy_pfn(pfn, order); | |
108 | buddy = page + (buddy_pfn - pfn); | |
3c605096 | 109 | |
13ad59df | 110 | if (pfn_valid_within(buddy_pfn) && |
1ae7013d | 111 | !is_migrate_isolate_page(buddy)) { |
3c605096 | 112 | __isolate_free_page(page, order); |
e3a2713c | 113 | isolated_page = true; |
3c605096 JK |
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) { | |
02aa0cdd | 124 | nr_pages = move_freepages_block(zone, page, migratetype, NULL); |
3c605096 JK |
125 | __mod_zone_freepage_state(zone, nr_pages, migratetype); |
126 | } | |
a458431e | 127 | set_pageblock_migratetype(page, migratetype); |
ad53f92e | 128 | zone->nr_isolate_pageblock--; |
ee6f509c MK |
129 | out: |
130 | spin_unlock_irqrestore(&zone->lock, flags); | |
83358ece | 131 | if (isolated_page) { |
46f24fd8 | 132 | post_alloc_hook(page, order, __GFP_MOVABLE); |
e3a2713c | 133 | __free_pages(page, order); |
83358ece | 134 | } |
ee6f509c MK |
135 | } |
136 | ||
a5d76b54 KH |
137 | static inline struct page * |
138 | __first_valid_page(unsigned long pfn, unsigned long nr_pages) | |
139 | { | |
140 | int i; | |
141 | for (i = 0; i < nr_pages; i++) | |
142 | if (pfn_valid_within(pfn + i)) | |
143 | break; | |
144 | if (unlikely(i == nr_pages)) | |
145 | return NULL; | |
146 | return pfn_to_page(pfn + i); | |
147 | } | |
148 | ||
149 | /* | |
150 | * start_isolate_page_range() -- make page-allocation-type of range of pages | |
151 | * to be MIGRATE_ISOLATE. | |
152 | * @start_pfn: The lower PFN of the range to be isolated. | |
153 | * @end_pfn: The upper PFN of the range to be isolated. | |
0815f3d8 | 154 | * @migratetype: migrate type to set in error recovery. |
a5d76b54 KH |
155 | * |
156 | * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in | |
157 | * the range will never be allocated. Any free pages and pages freed in the | |
158 | * future will not be allocated again. | |
159 | * | |
160 | * start_pfn/end_pfn must be aligned to pageblock_order. | |
161 | * Returns 0 on success and -EBUSY if any part of range cannot be isolated. | |
162 | */ | |
0815f3d8 | 163 | int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn, |
b023f468 | 164 | unsigned migratetype, bool skip_hwpoisoned_pages) |
a5d76b54 KH |
165 | { |
166 | unsigned long pfn; | |
167 | unsigned long undo_pfn; | |
168 | struct page *page; | |
169 | ||
fec174d6 NH |
170 | BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages)); |
171 | BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages)); | |
a5d76b54 KH |
172 | |
173 | for (pfn = start_pfn; | |
174 | pfn < end_pfn; | |
175 | pfn += pageblock_nr_pages) { | |
176 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
b023f468 WC |
177 | if (page && |
178 | set_migratetype_isolate(page, skip_hwpoisoned_pages)) { | |
a5d76b54 KH |
179 | undo_pfn = pfn; |
180 | goto undo; | |
181 | } | |
182 | } | |
183 | return 0; | |
184 | undo: | |
185 | for (pfn = start_pfn; | |
dbc0e4ce | 186 | pfn < undo_pfn; |
a5d76b54 | 187 | pfn += pageblock_nr_pages) |
0815f3d8 | 188 | unset_migratetype_isolate(pfn_to_page(pfn), migratetype); |
a5d76b54 KH |
189 | |
190 | return -EBUSY; | |
191 | } | |
192 | ||
193 | /* | |
194 | * Make isolated pages available again. | |
195 | */ | |
0815f3d8 MN |
196 | int undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn, |
197 | unsigned migratetype) | |
a5d76b54 KH |
198 | { |
199 | unsigned long pfn; | |
200 | struct page *page; | |
6f8d2b8a WX |
201 | |
202 | BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages)); | |
203 | BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages)); | |
204 | ||
a5d76b54 KH |
205 | for (pfn = start_pfn; |
206 | pfn < end_pfn; | |
207 | pfn += pageblock_nr_pages) { | |
208 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
bbf9ce97 | 209 | if (!page || !is_migrate_isolate_page(page)) |
a5d76b54 | 210 | continue; |
0815f3d8 | 211 | unset_migratetype_isolate(page, migratetype); |
a5d76b54 KH |
212 | } |
213 | return 0; | |
214 | } | |
215 | /* | |
216 | * Test all pages in the range is free(means isolated) or not. | |
217 | * all pages in [start_pfn...end_pfn) must be in the same zone. | |
218 | * zone->lock must be held before call this. | |
219 | * | |
ec3b6882 | 220 | * Returns the last tested pfn. |
a5d76b54 | 221 | */ |
fea85cff | 222 | static unsigned long |
b023f468 WC |
223 | __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn, |
224 | bool skip_hwpoisoned_pages) | |
a5d76b54 KH |
225 | { |
226 | struct page *page; | |
227 | ||
228 | while (pfn < end_pfn) { | |
229 | if (!pfn_valid_within(pfn)) { | |
230 | pfn++; | |
231 | continue; | |
232 | } | |
233 | page = pfn_to_page(pfn); | |
aa016d14 | 234 | if (PageBuddy(page)) |
435b405c | 235 | /* |
aa016d14 VB |
236 | * If the page is on a free list, it has to be on |
237 | * the correct MIGRATE_ISOLATE freelist. There is no | |
238 | * simple way to verify that as VM_BUG_ON(), though. | |
435b405c | 239 | */ |
a5d76b54 | 240 | pfn += 1 << page_order(page); |
aa016d14 VB |
241 | else if (skip_hwpoisoned_pages && PageHWPoison(page)) |
242 | /* A HWPoisoned page cannot be also PageBuddy */ | |
b023f468 | 243 | pfn++; |
a5d76b54 KH |
244 | else |
245 | break; | |
246 | } | |
fea85cff JK |
247 | |
248 | return pfn; | |
a5d76b54 KH |
249 | } |
250 | ||
b9eb6319 | 251 | /* Caller should ensure that requested range is in a single zone */ |
b023f468 WC |
252 | int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn, |
253 | bool skip_hwpoisoned_pages) | |
a5d76b54 | 254 | { |
6c1b7f68 | 255 | unsigned long pfn, flags; |
a5d76b54 | 256 | struct page *page; |
6c1b7f68 | 257 | struct zone *zone; |
a5d76b54 | 258 | |
a5d76b54 | 259 | /* |
85dbe706 TC |
260 | * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages |
261 | * are not aligned to pageblock_nr_pages. | |
262 | * Then we just check migratetype first. | |
a5d76b54 KH |
263 | */ |
264 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
265 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
bbf9ce97 | 266 | if (page && !is_migrate_isolate_page(page)) |
a5d76b54 KH |
267 | break; |
268 | } | |
a70dcb96 GS |
269 | page = __first_valid_page(start_pfn, end_pfn - start_pfn); |
270 | if ((pfn < end_pfn) || !page) | |
a5d76b54 | 271 | return -EBUSY; |
85dbe706 | 272 | /* Check all pages are free or marked as ISOLATED */ |
a70dcb96 | 273 | zone = page_zone(page); |
6c1b7f68 | 274 | spin_lock_irqsave(&zone->lock, flags); |
fea85cff | 275 | pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, |
b023f468 | 276 | skip_hwpoisoned_pages); |
6c1b7f68 | 277 | spin_unlock_irqrestore(&zone->lock, flags); |
fea85cff | 278 | |
0f0848e5 JK |
279 | trace_test_pages_isolated(start_pfn, end_pfn, pfn); |
280 | ||
fea85cff | 281 | return pfn < end_pfn ? -EBUSY : 0; |
a5d76b54 | 282 | } |
723a0644 MK |
283 | |
284 | struct page *alloc_migrate_target(struct page *page, unsigned long private, | |
285 | int **resultp) | |
286 | { | |
287 | gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE; | |
288 | ||
c8721bbb NH |
289 | /* |
290 | * TODO: allocate a destination hugepage from a nearest neighbor node, | |
291 | * accordance with memory policy of the user process if possible. For | |
292 | * now as a simple work-around, we use the next node for destination. | |
293 | */ | |
0edaf86c | 294 | if (PageHuge(page)) |
c8721bbb | 295 | return alloc_huge_page_node(page_hstate(compound_head(page)), |
0edaf86c AM |
296 | next_node_in(page_to_nid(page), |
297 | node_online_map)); | |
c8721bbb | 298 | |
723a0644 MK |
299 | if (PageHighMem(page)) |
300 | gfp_mask |= __GFP_HIGHMEM; | |
301 | ||
302 | return alloc_page(gfp_mask); | |
303 | } |