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 | ||
b48d8a8e | 18 | /* |
844fbae6 ZY |
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. | |
b48d8a8e ZY |
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 | */ | |
844fbae6 ZY |
33 | static struct page *has_unmovable_pages(unsigned long start_pfn, unsigned long end_pfn, |
34 | int migratetype, int flags) | |
b48d8a8e | 35 | { |
844fbae6 ZY |
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(ALIGN_DOWN(start_pfn, pageblock_nr_pages) != | |
41 | ALIGN_DOWN(end_pfn - 1, pageblock_nr_pages)); | |
b48d8a8e ZY |
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 | ||
844fbae6 ZY |
55 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
56 | page = pfn_to_page(pfn); | |
b48d8a8e ZY |
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); | |
844fbae6 | 93 | pfn += skip_pages - 1; |
b48d8a8e ZY |
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)) | |
844fbae6 | 105 | pfn += (1 << buddy_order(page)) - 1; |
b48d8a8e ZY |
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 | ||
844fbae6 ZY |
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 | */ | |
147 | static int set_migratetype_isolate(struct page *page, int migratetype, int isol_flags, | |
148 | unsigned long start_pfn, unsigned long end_pfn) | |
ee6f509c | 149 | { |
1c31cb49 DH |
150 | struct zone *zone = page_zone(page); |
151 | struct page *unmovable; | |
3f9903b9 | 152 | unsigned long flags; |
844fbae6 | 153 | unsigned long check_unmovable_start, check_unmovable_end; |
ee6f509c MK |
154 | |
155 | spin_lock_irqsave(&zone->lock, flags); | |
156 | ||
2c7452a0 MK |
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 | |
51030a53 | 160 | * set it before us. |
2c7452a0 | 161 | */ |
51030a53 DH |
162 | if (is_migrate_isolate_page(page)) { |
163 | spin_unlock_irqrestore(&zone->lock, flags); | |
164 | return -EBUSY; | |
165 | } | |
2c7452a0 | 166 | |
ee6f509c MK |
167 | /* |
168 | * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself. | |
169 | * We just check MOVABLE pages. | |
844fbae6 ZY |
170 | * |
171 | * Pass the intersection of [start_pfn, end_pfn) and the page's pageblock | |
172 | * to avoid redundant checks. | |
ee6f509c | 173 | */ |
844fbae6 ZY |
174 | check_unmovable_start = max(page_to_pfn(page), start_pfn); |
175 | check_unmovable_end = min(ALIGN(page_to_pfn(page) + 1, pageblock_nr_pages), | |
176 | end_pfn); | |
177 | ||
178 | unmovable = has_unmovable_pages(check_unmovable_start, check_unmovable_end, | |
179 | migratetype, isol_flags); | |
4a55c047 | 180 | if (!unmovable) { |
2139cbe6 | 181 | unsigned long nr_pages; |
4da2ce25 | 182 | int mt = get_pageblock_migratetype(page); |
2139cbe6 | 183 | |
a458431e | 184 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); |
ad53f92e | 185 | zone->nr_isolate_pageblock++; |
02aa0cdd VB |
186 | nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE, |
187 | NULL); | |
2139cbe6 | 188 | |
4da2ce25 | 189 | __mod_zone_freepage_state(zone, -nr_pages, mt); |
1c31cb49 | 190 | spin_unlock_irqrestore(&zone->lock, flags); |
1c31cb49 | 191 | return 0; |
ee6f509c MK |
192 | } |
193 | ||
194 | spin_unlock_irqrestore(&zone->lock, flags); | |
1c31cb49 | 195 | if (isol_flags & REPORT_FAILURE) { |
48381d7e DH |
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"); | |
3d680bdf | 201 | } |
4a55c047 | 202 | |
1c31cb49 | 203 | return -EBUSY; |
ee6f509c MK |
204 | } |
205 | ||
c5b4e1b0 | 206 | static void unset_migratetype_isolate(struct page *page, unsigned migratetype) |
ee6f509c MK |
207 | { |
208 | struct zone *zone; | |
2139cbe6 | 209 | unsigned long flags, nr_pages; |
e3a2713c | 210 | bool isolated_page = false; |
3c605096 | 211 | unsigned int order; |
3c605096 | 212 | struct page *buddy; |
2139cbe6 | 213 | |
ee6f509c MK |
214 | zone = page_zone(page); |
215 | spin_lock_irqsave(&zone->lock, flags); | |
bbf9ce97 | 216 | if (!is_migrate_isolate_page(page)) |
ee6f509c | 217 | goto out; |
3c605096 JK |
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)) { | |
ab130f91 | 228 | order = buddy_order(page); |
2484be0f | 229 | if (order >= pageblock_order && order < MAX_ORDER - 1) { |
8170ac47 ZY |
230 | buddy = find_buddy_page_pfn(page, page_to_pfn(page), |
231 | order, NULL); | |
232 | if (buddy && !is_migrate_isolate_page(buddy)) { | |
a500cb34 ML |
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); | |
3c605096 JK |
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. | |
293ffa5e DH |
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. | |
3c605096 | 253 | */ |
a85468b7 | 254 | if (!isolated_page) { |
02aa0cdd | 255 | nr_pages = move_freepages_block(zone, page, migratetype, NULL); |
3c605096 JK |
256 | __mod_zone_freepage_state(zone, nr_pages, migratetype); |
257 | } | |
a458431e | 258 | set_pageblock_migratetype(page, migratetype); |
624f58d8 AD |
259 | if (isolated_page) |
260 | __putback_isolated_page(page, order, migratetype); | |
ad53f92e | 261 | zone->nr_isolate_pageblock--; |
ee6f509c MK |
262 | out: |
263 | spin_unlock_irqrestore(&zone->lock, flags); | |
264 | } | |
265 | ||
a5d76b54 KH |
266 | static inline struct page * |
267 | __first_valid_page(unsigned long pfn, unsigned long nr_pages) | |
268 | { | |
269 | int i; | |
2ce13640 MH |
270 | |
271 | for (i = 0; i < nr_pages; i++) { | |
272 | struct page *page; | |
273 | ||
2ce13640 MH |
274 | page = pfn_to_online_page(pfn + i); |
275 | if (!page) | |
276 | continue; | |
277 | return page; | |
278 | } | |
279 | return NULL; | |
a5d76b54 KH |
280 | } |
281 | ||
9b7ea46a QC |
282 | /** |
283 | * start_isolate_page_range() - make page-allocation-type of range of pages to | |
284 | * be MIGRATE_ISOLATE. | |
285 | * @start_pfn: The lower PFN of the range to be isolated. | |
286 | * @end_pfn: The upper PFN of the range to be isolated. | |
287 | * start_pfn/end_pfn must be aligned to pageblock_order. | |
288 | * @migratetype: Migrate type to set in error recovery. | |
289 | * @flags: The following flags are allowed (they can be combined in | |
290 | * a bit mask) | |
756d25be DH |
291 | * MEMORY_OFFLINE - isolate to offline (!allocate) memory |
292 | * e.g., skip over PageHWPoison() pages | |
aa218795 | 293 | * and PageOffline() pages. |
9b7ea46a QC |
294 | * REPORT_FAILURE - report details about the failure to |
295 | * isolate the range | |
a5d76b54 KH |
296 | * |
297 | * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in | |
298 | * the range will never be allocated. Any free pages and pages freed in the | |
9b7ea46a QC |
299 | * future will not be allocated again. If specified range includes migrate types |
300 | * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all | |
301 | * pages in the range finally, the caller have to free all pages in the range. | |
302 | * test_page_isolated() can be used for test it. | |
2c7452a0 MK |
303 | * |
304 | * There is no high level synchronization mechanism that prevents two threads | |
9b7ea46a | 305 | * from trying to isolate overlapping ranges. If this happens, one thread |
2c7452a0 MK |
306 | * will notice pageblocks in the overlapping range already set to isolate. |
307 | * This happens in set_migratetype_isolate, and set_migratetype_isolate | |
9b7ea46a QC |
308 | * returns an error. We then clean up by restoring the migration type on |
309 | * pageblocks we may have modified and return -EBUSY to caller. This | |
2c7452a0 | 310 | * prevents two threads from simultaneously working on overlapping ranges. |
9b7ea46a | 311 | * |
96831826 PT |
312 | * Please note that there is no strong synchronization with the page allocator |
313 | * either. Pages might be freed while their page blocks are marked ISOLATED. | |
7612921f VB |
314 | * A call to drain_all_pages() after isolation can flush most of them. However |
315 | * in some cases pages might still end up on pcp lists and that would allow | |
96831826 | 316 | * for their allocation even when they are in fact isolated already. Depending |
ec6e8c7e VB |
317 | * on how strong of a guarantee the caller needs, zone_pcp_disable/enable() |
318 | * might be used to flush and disable pcplist before isolation and enable after | |
319 | * unisolation. | |
96831826 | 320 | * |
3fa0c7c7 | 321 | * Return: 0 on success and -EBUSY if any part of range cannot be isolated. |
a5d76b54 | 322 | */ |
0815f3d8 | 323 | int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn, |
d381c547 | 324 | unsigned migratetype, int flags) |
a5d76b54 KH |
325 | { |
326 | unsigned long pfn; | |
a5d76b54 KH |
327 | struct page *page; |
328 | ||
fec174d6 NH |
329 | BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages)); |
330 | BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages)); | |
a5d76b54 KH |
331 | |
332 | for (pfn = start_pfn; | |
333 | pfn < end_pfn; | |
334 | pfn += pageblock_nr_pages) { | |
335 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
844fbae6 ZY |
336 | if (page && set_migratetype_isolate(page, migratetype, flags, |
337 | start_pfn, end_pfn)) { | |
e1d8c966 ML |
338 | undo_isolate_page_range(start_pfn, pfn, migratetype); |
339 | return -EBUSY; | |
a5d76b54 KH |
340 | } |
341 | } | |
3fa0c7c7 | 342 | return 0; |
a5d76b54 KH |
343 | } |
344 | ||
345 | /* | |
346 | * Make isolated pages available again. | |
347 | */ | |
1fcf0a56 | 348 | void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn, |
0815f3d8 | 349 | unsigned migratetype) |
a5d76b54 KH |
350 | { |
351 | unsigned long pfn; | |
352 | struct page *page; | |
6f8d2b8a WX |
353 | |
354 | BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages)); | |
355 | BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages)); | |
356 | ||
a5d76b54 KH |
357 | for (pfn = start_pfn; |
358 | pfn < end_pfn; | |
359 | pfn += pageblock_nr_pages) { | |
360 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
bbf9ce97 | 361 | if (!page || !is_migrate_isolate_page(page)) |
a5d76b54 | 362 | continue; |
0815f3d8 | 363 | unset_migratetype_isolate(page, migratetype); |
a5d76b54 | 364 | } |
a5d76b54 KH |
365 | } |
366 | /* | |
367 | * Test all pages in the range is free(means isolated) or not. | |
368 | * all pages in [start_pfn...end_pfn) must be in the same zone. | |
369 | * zone->lock must be held before call this. | |
370 | * | |
ec3b6882 | 371 | * Returns the last tested pfn. |
a5d76b54 | 372 | */ |
fea85cff | 373 | static unsigned long |
b023f468 | 374 | __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn, |
756d25be | 375 | int flags) |
a5d76b54 KH |
376 | { |
377 | struct page *page; | |
378 | ||
379 | while (pfn < end_pfn) { | |
a5d76b54 | 380 | page = pfn_to_page(pfn); |
aa016d14 | 381 | if (PageBuddy(page)) |
435b405c | 382 | /* |
aa016d14 VB |
383 | * If the page is on a free list, it has to be on |
384 | * the correct MIGRATE_ISOLATE freelist. There is no | |
385 | * simple way to verify that as VM_BUG_ON(), though. | |
435b405c | 386 | */ |
ab130f91 | 387 | pfn += 1 << buddy_order(page); |
756d25be | 388 | else if ((flags & MEMORY_OFFLINE) && PageHWPoison(page)) |
aa016d14 | 389 | /* A HWPoisoned page cannot be also PageBuddy */ |
b023f468 | 390 | pfn++; |
aa218795 DH |
391 | else if ((flags & MEMORY_OFFLINE) && PageOffline(page) && |
392 | !page_count(page)) | |
393 | /* | |
394 | * The responsible driver agreed to skip PageOffline() | |
395 | * pages when offlining memory by dropping its | |
396 | * reference in MEM_GOING_OFFLINE. | |
397 | */ | |
398 | pfn++; | |
a5d76b54 KH |
399 | else |
400 | break; | |
401 | } | |
fea85cff JK |
402 | |
403 | return pfn; | |
a5d76b54 KH |
404 | } |
405 | ||
b9eb6319 | 406 | /* Caller should ensure that requested range is in a single zone */ |
b023f468 | 407 | int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn, |
756d25be | 408 | int isol_flags) |
a5d76b54 | 409 | { |
6c1b7f68 | 410 | unsigned long pfn, flags; |
a5d76b54 | 411 | struct page *page; |
6c1b7f68 | 412 | struct zone *zone; |
1d09510b | 413 | int ret; |
a5d76b54 | 414 | |
a5d76b54 | 415 | /* |
85dbe706 TC |
416 | * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages |
417 | * are not aligned to pageblock_nr_pages. | |
418 | * Then we just check migratetype first. | |
a5d76b54 KH |
419 | */ |
420 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
421 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
bbf9ce97 | 422 | if (page && !is_migrate_isolate_page(page)) |
a5d76b54 KH |
423 | break; |
424 | } | |
a70dcb96 | 425 | page = __first_valid_page(start_pfn, end_pfn - start_pfn); |
1d09510b GD |
426 | if ((pfn < end_pfn) || !page) { |
427 | ret = -EBUSY; | |
428 | goto out; | |
429 | } | |
430 | ||
85dbe706 | 431 | /* Check all pages are free or marked as ISOLATED */ |
a70dcb96 | 432 | zone = page_zone(page); |
6c1b7f68 | 433 | spin_lock_irqsave(&zone->lock, flags); |
756d25be | 434 | pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, isol_flags); |
6c1b7f68 | 435 | spin_unlock_irqrestore(&zone->lock, flags); |
fea85cff | 436 | |
1d09510b GD |
437 | ret = pfn < end_pfn ? -EBUSY : 0; |
438 | ||
439 | out: | |
0f0848e5 JK |
440 | trace_test_pages_isolated(start_pfn, end_pfn, pfn); |
441 | ||
1d09510b | 442 | return ret; |
a5d76b54 | 443 | } |