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 | ||
4f9bc69a KW |
40 | VM_BUG_ON(pageblock_start_pfn(start_pfn) != |
41 | pageblock_start_pfn(end_pfn - 1)); | |
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)) { | |
affd26b1 | 82 | struct folio *folio = page_folio(page); |
b48d8a8e ZY |
83 | unsigned int skip_pages; |
84 | ||
85 | if (PageHuge(page)) { | |
affd26b1 | 86 | if (!hugepage_migration_supported(folio_hstate(folio))) |
b48d8a8e | 87 | return page; |
affd26b1 | 88 | } else if (!folio_test_lru(folio) && !__folio_test_movable(folio)) { |
b48d8a8e ZY |
89 | return page; |
90 | } | |
91 | ||
affd26b1 | 92 | skip_pages = folio_nr_pages(folio) - folio_page_idx(folio, page); |
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 | 174 | check_unmovable_start = max(page_to_pfn(page), start_pfn); |
4f9bc69a | 175 | check_unmovable_end = min(pageblock_end_pfn(page_to_pfn(page)), |
844fbae6 ZY |
176 | end_pfn); |
177 | ||
178 | unmovable = has_unmovable_pages(check_unmovable_start, check_unmovable_end, | |
179 | migratetype, isol_flags); | |
4a55c047 | 180 | if (!unmovable) { |
c0cd6f55 | 181 | int nr_pages; |
4da2ce25 | 182 | int mt = get_pageblock_migratetype(page); |
2139cbe6 | 183 | |
c0cd6f55 JW |
184 | nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE); |
185 | /* Block spans zone boundaries? */ | |
186 | if (nr_pages == -1) { | |
187 | spin_unlock_irqrestore(&zone->lock, flags); | |
188 | return -EBUSY; | |
189 | } | |
190 | __mod_zone_freepage_state(zone, -nr_pages, mt); | |
ad53f92e | 191 | zone->nr_isolate_pageblock++; |
1c31cb49 | 192 | spin_unlock_irqrestore(&zone->lock, flags); |
1c31cb49 | 193 | return 0; |
ee6f509c MK |
194 | } |
195 | ||
196 | spin_unlock_irqrestore(&zone->lock, flags); | |
1c31cb49 | 197 | if (isol_flags & REPORT_FAILURE) { |
48381d7e DH |
198 | /* |
199 | * printk() with zone->lock held will likely trigger a | |
200 | * lockdep splat, so defer it here. | |
201 | */ | |
202 | dump_page(unmovable, "unmovable page"); | |
3d680bdf | 203 | } |
4a55c047 | 204 | |
1c31cb49 | 205 | return -EBUSY; |
ee6f509c MK |
206 | } |
207 | ||
b2c9e2fb | 208 | static void unset_migratetype_isolate(struct page *page, int migratetype) |
ee6f509c MK |
209 | { |
210 | struct zone *zone; | |
c0cd6f55 | 211 | unsigned long flags; |
e3a2713c | 212 | bool isolated_page = false; |
3c605096 | 213 | unsigned int order; |
3c605096 | 214 | struct page *buddy; |
2139cbe6 | 215 | |
ee6f509c MK |
216 | zone = page_zone(page); |
217 | spin_lock_irqsave(&zone->lock, flags); | |
bbf9ce97 | 218 | if (!is_migrate_isolate_page(page)) |
ee6f509c | 219 | goto out; |
3c605096 JK |
220 | |
221 | /* | |
222 | * Because freepage with more than pageblock_order on isolated | |
223 | * pageblock is restricted to merge due to freepage counting problem, | |
224 | * it is possible that there is free buddy page. | |
225 | * move_freepages_block() doesn't care of merge so we need other | |
226 | * approach in order to merge them. Isolation and free will make | |
227 | * these pages to be merged. | |
228 | */ | |
229 | if (PageBuddy(page)) { | |
ab130f91 | 230 | order = buddy_order(page); |
5e0a760b | 231 | if (order >= pageblock_order && order < MAX_PAGE_ORDER) { |
8170ac47 ZY |
232 | buddy = find_buddy_page_pfn(page, page_to_pfn(page), |
233 | order, NULL); | |
234 | if (buddy && !is_migrate_isolate_page(buddy)) { | |
a500cb34 ML |
235 | isolated_page = !!__isolate_free_page(page, order); |
236 | /* | |
237 | * Isolating a free page in an isolated pageblock | |
238 | * is expected to always work as watermarks don't | |
239 | * apply here. | |
240 | */ | |
241 | VM_WARN_ON(!isolated_page); | |
3c605096 JK |
242 | } |
243 | } | |
244 | } | |
245 | ||
246 | /* | |
247 | * If we isolate freepage with more than pageblock_order, there | |
248 | * should be no freepage in the range, so we could avoid costly | |
249 | * pageblock scanning for freepage moving. | |
293ffa5e DH |
250 | * |
251 | * We didn't actually touch any of the isolated pages, so place them | |
252 | * to the tail of the freelist. This is an optimization for memory | |
253 | * onlining - just onlined memory won't immediately be considered for | |
254 | * allocation. | |
3c605096 | 255 | */ |
a85468b7 | 256 | if (!isolated_page) { |
c0cd6f55 JW |
257 | int nr_pages = move_freepages_block(zone, page, migratetype); |
258 | /* | |
259 | * Isolating this block already succeeded, so this | |
260 | * should not fail on zone boundaries. | |
261 | */ | |
262 | WARN_ON_ONCE(nr_pages == -1); | |
3c605096 | 263 | __mod_zone_freepage_state(zone, nr_pages, migratetype); |
f37c0f68 ZY |
264 | } else { |
265 | set_pageblock_migratetype(page, migratetype); | |
624f58d8 | 266 | __putback_isolated_page(page, order, migratetype); |
f37c0f68 | 267 | } |
ad53f92e | 268 | zone->nr_isolate_pageblock--; |
ee6f509c MK |
269 | out: |
270 | spin_unlock_irqrestore(&zone->lock, flags); | |
271 | } | |
272 | ||
a5d76b54 KH |
273 | static inline struct page * |
274 | __first_valid_page(unsigned long pfn, unsigned long nr_pages) | |
275 | { | |
276 | int i; | |
2ce13640 MH |
277 | |
278 | for (i = 0; i < nr_pages; i++) { | |
279 | struct page *page; | |
280 | ||
2ce13640 MH |
281 | page = pfn_to_online_page(pfn + i); |
282 | if (!page) | |
283 | continue; | |
284 | return page; | |
285 | } | |
286 | return NULL; | |
a5d76b54 KH |
287 | } |
288 | ||
b2c9e2fb ZY |
289 | /** |
290 | * isolate_single_pageblock() -- tries to isolate a pageblock that might be | |
291 | * within a free or in-use page. | |
292 | * @boundary_pfn: pageblock-aligned pfn that a page might cross | |
88ee1343 | 293 | * @flags: isolation flags |
b2c9e2fb ZY |
294 | * @gfp_flags: GFP flags used for migrating pages |
295 | * @isolate_before: isolate the pageblock before the boundary_pfn | |
04299938 YL |
296 | * @skip_isolation: the flag to skip the pageblock isolation in second |
297 | * isolate_single_pageblock() | |
80e2b584 | 298 | * @migratetype: migrate type to set in error recovery. |
b2c9e2fb | 299 | * |
5e0a760b | 300 | * Free and in-use pages can be as big as MAX_PAGE_ORDER and contain more than one |
b2c9e2fb ZY |
301 | * pageblock. When not all pageblocks within a page are isolated at the same |
302 | * time, free page accounting can go wrong. For example, in the case of | |
5e0a760b KS |
303 | * MAX_PAGE_ORDER = pageblock_order + 1, a MAX_PAGE_ORDER page has two |
304 | * pagelbocks. | |
305 | * [ MAX_PAGE_ORDER ] | |
b2c9e2fb ZY |
306 | * [ pageblock0 | pageblock1 ] |
307 | * When either pageblock is isolated, if it is a free page, the page is not | |
308 | * split into separate migratetype lists, which is supposed to; if it is an | |
309 | * in-use page and freed later, __free_one_page() does not split the free page | |
310 | * either. The function handles this by splitting the free page or migrating | |
311 | * the in-use page then splitting the free page. | |
312 | */ | |
88ee1343 | 313 | static int isolate_single_pageblock(unsigned long boundary_pfn, int flags, |
80e2b584 ZY |
314 | gfp_t gfp_flags, bool isolate_before, bool skip_isolation, |
315 | int migratetype) | |
b2c9e2fb | 316 | { |
b2c9e2fb ZY |
317 | unsigned long start_pfn; |
318 | unsigned long isolate_pageblock; | |
319 | unsigned long pfn; | |
320 | struct zone *zone; | |
88ee1343 | 321 | int ret; |
b2c9e2fb | 322 | |
ee0913c4 | 323 | VM_BUG_ON(!pageblock_aligned(boundary_pfn)); |
b2c9e2fb ZY |
324 | |
325 | if (isolate_before) | |
326 | isolate_pageblock = boundary_pfn - pageblock_nr_pages; | |
327 | else | |
328 | isolate_pageblock = boundary_pfn; | |
329 | ||
330 | /* | |
331 | * scan at the beginning of MAX_ORDER_NR_PAGES aligned range to avoid | |
332 | * only isolating a subset of pageblocks from a bigger than pageblock | |
333 | * free or in-use page. Also make sure all to-be-isolated pageblocks | |
334 | * are within the same zone. | |
335 | */ | |
336 | zone = page_zone(pfn_to_page(isolate_pageblock)); | |
337 | start_pfn = max(ALIGN_DOWN(isolate_pageblock, MAX_ORDER_NR_PAGES), | |
338 | zone->zone_start_pfn); | |
339 | ||
80e2b584 | 340 | if (skip_isolation) { |
fba4eaf9 | 341 | int mt __maybe_unused = get_pageblock_migratetype(pfn_to_page(isolate_pageblock)); |
88ee1343 | 342 | |
80e2b584 ZY |
343 | VM_BUG_ON(!is_migrate_isolate(mt)); |
344 | } else { | |
345 | ret = set_migratetype_isolate(pfn_to_page(isolate_pageblock), migratetype, | |
346 | flags, isolate_pageblock, isolate_pageblock + pageblock_nr_pages); | |
9b209e55 ZY |
347 | |
348 | if (ret) | |
349 | return ret; | |
350 | } | |
b2c9e2fb ZY |
351 | |
352 | /* | |
353 | * Bail out early when the to-be-isolated pageblock does not form | |
354 | * a free or in-use page across boundary_pfn: | |
355 | * | |
356 | * 1. isolate before boundary_pfn: the page after is not online | |
357 | * 2. isolate after boundary_pfn: the page before is not online | |
358 | * | |
359 | * This also ensures correctness. Without it, when isolate after | |
360 | * boundary_pfn and [start_pfn, boundary_pfn) are not online, | |
361 | * __first_valid_page() will return unexpected NULL in the for loop | |
362 | * below. | |
363 | */ | |
364 | if (isolate_before) { | |
365 | if (!pfn_to_online_page(boundary_pfn)) | |
366 | return 0; | |
367 | } else { | |
368 | if (!pfn_to_online_page(boundary_pfn - 1)) | |
369 | return 0; | |
370 | } | |
371 | ||
372 | for (pfn = start_pfn; pfn < boundary_pfn;) { | |
373 | struct page *page = __first_valid_page(pfn, boundary_pfn - pfn); | |
374 | ||
375 | VM_BUG_ON(!page); | |
376 | pfn = page_to_pfn(page); | |
377 | /* | |
378 | * start_pfn is MAX_ORDER_NR_PAGES aligned, if there is any | |
379 | * free pages in [start_pfn, boundary_pfn), its head page will | |
380 | * always be in the range. | |
381 | */ | |
382 | if (PageBuddy(page)) { | |
383 | int order = buddy_order(page); | |
384 | ||
86d28b07 ZY |
385 | if (pfn + (1UL << order) > boundary_pfn) { |
386 | /* free page changed before split, check it again */ | |
387 | if (split_free_page(page, order, boundary_pfn - pfn)) | |
388 | continue; | |
389 | } | |
390 | ||
391 | pfn += 1UL << order; | |
b2c9e2fb ZY |
392 | continue; |
393 | } | |
394 | /* | |
395 | * migrate compound pages then let the free page handling code | |
396 | * above do the rest. If migration is not possible, just fail. | |
397 | */ | |
398 | if (PageCompound(page)) { | |
b2c9e2fb ZY |
399 | struct page *head = compound_head(page); |
400 | unsigned long head_pfn = page_to_pfn(head); | |
547be963 | 401 | unsigned long nr_pages = compound_nr(head); |
b2c9e2fb | 402 | |
88ee1343 | 403 | if (head_pfn + nr_pages <= boundary_pfn) { |
b2c9e2fb ZY |
404 | pfn = head_pfn + nr_pages; |
405 | continue; | |
406 | } | |
407 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA | |
408 | /* | |
409 | * hugetlb, lru compound (THP), and movable compound pages | |
410 | * can be migrated. Otherwise, fail the isolation. | |
411 | */ | |
412 | if (PageHuge(page) || PageLRU(page) || __PageMovable(page)) { | |
413 | int order; | |
414 | unsigned long outer_pfn; | |
88ee1343 ZY |
415 | int page_mt = get_pageblock_migratetype(page); |
416 | bool isolate_page = !is_migrate_isolate_page(page); | |
b2c9e2fb ZY |
417 | struct compact_control cc = { |
418 | .nr_migratepages = 0, | |
419 | .order = -1, | |
420 | .zone = page_zone(pfn_to_page(head_pfn)), | |
421 | .mode = MIGRATE_SYNC, | |
422 | .ignore_skip_hint = true, | |
423 | .no_set_skip_hint = true, | |
424 | .gfp_mask = gfp_flags, | |
425 | .alloc_contig = true, | |
426 | }; | |
427 | INIT_LIST_HEAD(&cc.migratepages); | |
428 | ||
88ee1343 ZY |
429 | /* |
430 | * XXX: mark the page as MIGRATE_ISOLATE so that | |
431 | * no one else can grab the freed page after migration. | |
432 | * Ideally, the page should be freed as two separate | |
433 | * pages to be added into separate migratetype free | |
434 | * lists. | |
435 | */ | |
436 | if (isolate_page) { | |
437 | ret = set_migratetype_isolate(page, page_mt, | |
438 | flags, head_pfn, head_pfn + nr_pages); | |
439 | if (ret) | |
440 | goto failed; | |
441 | } | |
442 | ||
b2c9e2fb | 443 | ret = __alloc_contig_migrate_range(&cc, head_pfn, |
c8b36003 | 444 | head_pfn + nr_pages, page_mt); |
b2c9e2fb | 445 | |
88ee1343 ZY |
446 | /* |
447 | * restore the page's migratetype so that it can | |
448 | * be split into separate migratetype free lists | |
449 | * later. | |
450 | */ | |
451 | if (isolate_page) | |
452 | unset_migratetype_isolate(page, page_mt); | |
453 | ||
b2c9e2fb ZY |
454 | if (ret) |
455 | goto failed; | |
456 | /* | |
457 | * reset pfn to the head of the free page, so | |
458 | * that the free page handling code above can split | |
459 | * the free page to the right migratetype list. | |
460 | * | |
461 | * head_pfn is not used here as a hugetlb page order | |
5e0a760b | 462 | * can be bigger than MAX_PAGE_ORDER, but after it is |
b2c9e2fb ZY |
463 | * freed, the free page order is not. Use pfn within |
464 | * the range to find the head of the free page. | |
465 | */ | |
466 | order = 0; | |
467 | outer_pfn = pfn; | |
468 | while (!PageBuddy(pfn_to_page(outer_pfn))) { | |
88ee1343 | 469 | /* stop if we cannot find the free page */ |
5e0a760b | 470 | if (++order > MAX_PAGE_ORDER) |
88ee1343 | 471 | goto failed; |
b2c9e2fb ZY |
472 | outer_pfn &= ~0UL << order; |
473 | } | |
474 | pfn = outer_pfn; | |
475 | continue; | |
476 | } else | |
477 | #endif | |
478 | goto failed; | |
479 | } | |
480 | ||
481 | pfn++; | |
482 | } | |
483 | return 0; | |
484 | failed: | |
485 | /* restore the original migratetype */ | |
9b209e55 | 486 | if (!skip_isolation) |
80e2b584 | 487 | unset_migratetype_isolate(pfn_to_page(isolate_pageblock), migratetype); |
b2c9e2fb ZY |
488 | return -EBUSY; |
489 | } | |
490 | ||
9b7ea46a | 491 | /** |
ce5df776 JW |
492 | * start_isolate_page_range() - mark page range MIGRATE_ISOLATE |
493 | * @start_pfn: The first PFN of the range to be isolated. | |
494 | * @end_pfn: The last PFN of the range to be isolated. | |
9b7ea46a QC |
495 | * @migratetype: Migrate type to set in error recovery. |
496 | * @flags: The following flags are allowed (they can be combined in | |
497 | * a bit mask) | |
756d25be DH |
498 | * MEMORY_OFFLINE - isolate to offline (!allocate) memory |
499 | * e.g., skip over PageHWPoison() pages | |
aa218795 | 500 | * and PageOffline() pages. |
9b7ea46a QC |
501 | * REPORT_FAILURE - report details about the failure to |
502 | * isolate the range | |
b2c9e2fb ZY |
503 | * @gfp_flags: GFP flags used for migrating pages that sit across the |
504 | * range boundaries. | |
a5d76b54 KH |
505 | * |
506 | * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in | |
507 | * the range will never be allocated. Any free pages and pages freed in the | |
9b7ea46a QC |
508 | * future will not be allocated again. If specified range includes migrate types |
509 | * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all | |
510 | * pages in the range finally, the caller have to free all pages in the range. | |
511 | * test_page_isolated() can be used for test it. | |
2c7452a0 | 512 | * |
b2c9e2fb ZY |
513 | * The function first tries to isolate the pageblocks at the beginning and end |
514 | * of the range, since there might be pages across the range boundaries. | |
515 | * Afterwards, it isolates the rest of the range. | |
516 | * | |
2c7452a0 | 517 | * There is no high level synchronization mechanism that prevents two threads |
9b7ea46a | 518 | * from trying to isolate overlapping ranges. If this happens, one thread |
2c7452a0 MK |
519 | * will notice pageblocks in the overlapping range already set to isolate. |
520 | * This happens in set_migratetype_isolate, and set_migratetype_isolate | |
9b7ea46a QC |
521 | * returns an error. We then clean up by restoring the migration type on |
522 | * pageblocks we may have modified and return -EBUSY to caller. This | |
2c7452a0 | 523 | * prevents two threads from simultaneously working on overlapping ranges. |
9b7ea46a | 524 | * |
96831826 PT |
525 | * Please note that there is no strong synchronization with the page allocator |
526 | * either. Pages might be freed while their page blocks are marked ISOLATED. | |
7612921f VB |
527 | * A call to drain_all_pages() after isolation can flush most of them. However |
528 | * in some cases pages might still end up on pcp lists and that would allow | |
96831826 | 529 | * for their allocation even when they are in fact isolated already. Depending |
ec6e8c7e VB |
530 | * on how strong of a guarantee the caller needs, zone_pcp_disable/enable() |
531 | * might be used to flush and disable pcplist before isolation and enable after | |
532 | * unisolation. | |
96831826 | 533 | * |
3fa0c7c7 | 534 | * Return: 0 on success and -EBUSY if any part of range cannot be isolated. |
a5d76b54 | 535 | */ |
0815f3d8 | 536 | int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn, |
b2c9e2fb | 537 | int migratetype, int flags, gfp_t gfp_flags) |
a5d76b54 KH |
538 | { |
539 | unsigned long pfn; | |
a5d76b54 | 540 | struct page *page; |
6e263fff | 541 | /* isolation is done at page block granularity */ |
4f9bc69a | 542 | unsigned long isolate_start = pageblock_start_pfn(start_pfn); |
5f7fa13f | 543 | unsigned long isolate_end = pageblock_align(end_pfn); |
b2c9e2fb | 544 | int ret; |
9b209e55 | 545 | bool skip_isolation = false; |
a5d76b54 | 546 | |
6e263fff | 547 | /* isolate [isolate_start, isolate_start + pageblock_nr_pages) pageblock */ |
80e2b584 ZY |
548 | ret = isolate_single_pageblock(isolate_start, flags, gfp_flags, false, |
549 | skip_isolation, migratetype); | |
b2c9e2fb ZY |
550 | if (ret) |
551 | return ret; | |
552 | ||
9b209e55 ZY |
553 | if (isolate_start == isolate_end - pageblock_nr_pages) |
554 | skip_isolation = true; | |
555 | ||
6e263fff | 556 | /* isolate [isolate_end - pageblock_nr_pages, isolate_end) pageblock */ |
80e2b584 ZY |
557 | ret = isolate_single_pageblock(isolate_end, flags, gfp_flags, true, |
558 | skip_isolation, migratetype); | |
b2c9e2fb | 559 | if (ret) { |
6e263fff | 560 | unset_migratetype_isolate(pfn_to_page(isolate_start), migratetype); |
b2c9e2fb ZY |
561 | return ret; |
562 | } | |
563 | ||
564 | /* skip isolated pageblocks at the beginning and end */ | |
6e263fff ZY |
565 | for (pfn = isolate_start + pageblock_nr_pages; |
566 | pfn < isolate_end - pageblock_nr_pages; | |
a5d76b54 KH |
567 | pfn += pageblock_nr_pages) { |
568 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
844fbae6 ZY |
569 | if (page && set_migratetype_isolate(page, migratetype, flags, |
570 | start_pfn, end_pfn)) { | |
6e263fff | 571 | undo_isolate_page_range(isolate_start, pfn, migratetype); |
b2c9e2fb | 572 | unset_migratetype_isolate( |
6e263fff | 573 | pfn_to_page(isolate_end - pageblock_nr_pages), |
b2c9e2fb | 574 | migratetype); |
e1d8c966 | 575 | return -EBUSY; |
a5d76b54 KH |
576 | } |
577 | } | |
3fa0c7c7 | 578 | return 0; |
a5d76b54 KH |
579 | } |
580 | ||
ce5df776 JW |
581 | /** |
582 | * undo_isolate_page_range - undo effects of start_isolate_page_range() | |
583 | * @start_pfn: The first PFN of the isolated range | |
584 | * @end_pfn: The last PFN of the isolated range | |
585 | * @migratetype: New migrate type to set on the range | |
586 | * | |
587 | * This finds every MIGRATE_ISOLATE page block in the given range | |
588 | * and switches it to @migratetype. | |
a5d76b54 | 589 | */ |
1fcf0a56 | 590 | void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn, |
b2c9e2fb | 591 | int migratetype) |
a5d76b54 KH |
592 | { |
593 | unsigned long pfn; | |
594 | struct page *page; | |
4f9bc69a | 595 | unsigned long isolate_start = pageblock_start_pfn(start_pfn); |
5f7fa13f | 596 | unsigned long isolate_end = pageblock_align(end_pfn); |
6f8d2b8a | 597 | |
6e263fff ZY |
598 | for (pfn = isolate_start; |
599 | pfn < isolate_end; | |
a5d76b54 KH |
600 | pfn += pageblock_nr_pages) { |
601 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
bbf9ce97 | 602 | if (!page || !is_migrate_isolate_page(page)) |
a5d76b54 | 603 | continue; |
0815f3d8 | 604 | unset_migratetype_isolate(page, migratetype); |
a5d76b54 | 605 | } |
a5d76b54 KH |
606 | } |
607 | /* | |
608 | * Test all pages in the range is free(means isolated) or not. | |
609 | * all pages in [start_pfn...end_pfn) must be in the same zone. | |
610 | * zone->lock must be held before call this. | |
611 | * | |
ec3b6882 | 612 | * Returns the last tested pfn. |
a5d76b54 | 613 | */ |
fea85cff | 614 | static unsigned long |
b023f468 | 615 | __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn, |
756d25be | 616 | int flags) |
a5d76b54 KH |
617 | { |
618 | struct page *page; | |
619 | ||
620 | while (pfn < end_pfn) { | |
a5d76b54 | 621 | page = pfn_to_page(pfn); |
aa016d14 | 622 | if (PageBuddy(page)) |
435b405c | 623 | /* |
aa016d14 VB |
624 | * If the page is on a free list, it has to be on |
625 | * the correct MIGRATE_ISOLATE freelist. There is no | |
626 | * simple way to verify that as VM_BUG_ON(), though. | |
435b405c | 627 | */ |
ab130f91 | 628 | pfn += 1 << buddy_order(page); |
756d25be | 629 | else if ((flags & MEMORY_OFFLINE) && PageHWPoison(page)) |
aa016d14 | 630 | /* A HWPoisoned page cannot be also PageBuddy */ |
b023f468 | 631 | pfn++; |
aa218795 DH |
632 | else if ((flags & MEMORY_OFFLINE) && PageOffline(page) && |
633 | !page_count(page)) | |
634 | /* | |
635 | * The responsible driver agreed to skip PageOffline() | |
636 | * pages when offlining memory by dropping its | |
637 | * reference in MEM_GOING_OFFLINE. | |
638 | */ | |
639 | pfn++; | |
a5d76b54 KH |
640 | else |
641 | break; | |
642 | } | |
fea85cff JK |
643 | |
644 | return pfn; | |
a5d76b54 KH |
645 | } |
646 | ||
ce5df776 JW |
647 | /** |
648 | * test_pages_isolated - check if pageblocks in range are isolated | |
649 | * @start_pfn: The first PFN of the isolated range | |
650 | * @end_pfn: The first PFN *after* the isolated range | |
651 | * @isol_flags: Testing mode flags | |
652 | * | |
653 | * This tests if all in the specified range are free. | |
654 | * | |
655 | * If %MEMORY_OFFLINE is specified in @flags, it will consider | |
656 | * poisoned and offlined pages free as well. | |
657 | * | |
658 | * Caller must ensure the requested range doesn't span zones. | |
659 | * | |
660 | * Returns 0 if true, -EBUSY if one or more pages are in use. | |
661 | */ | |
b023f468 | 662 | int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn, |
756d25be | 663 | int isol_flags) |
a5d76b54 | 664 | { |
6c1b7f68 | 665 | unsigned long pfn, flags; |
a5d76b54 | 666 | struct page *page; |
6c1b7f68 | 667 | struct zone *zone; |
1d09510b | 668 | int ret; |
a5d76b54 | 669 | |
a5d76b54 | 670 | /* |
5e0a760b KS |
671 | * Note: pageblock_nr_pages != MAX_PAGE_ORDER. Then, chunks of free |
672 | * pages are not aligned to pageblock_nr_pages. | |
85dbe706 | 673 | * Then we just check migratetype first. |
a5d76b54 KH |
674 | */ |
675 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
676 | page = __first_valid_page(pfn, pageblock_nr_pages); | |
bbf9ce97 | 677 | if (page && !is_migrate_isolate_page(page)) |
a5d76b54 KH |
678 | break; |
679 | } | |
a70dcb96 | 680 | page = __first_valid_page(start_pfn, end_pfn - start_pfn); |
1d09510b GD |
681 | if ((pfn < end_pfn) || !page) { |
682 | ret = -EBUSY; | |
683 | goto out; | |
684 | } | |
685 | ||
85dbe706 | 686 | /* Check all pages are free or marked as ISOLATED */ |
a70dcb96 | 687 | zone = page_zone(page); |
6c1b7f68 | 688 | spin_lock_irqsave(&zone->lock, flags); |
756d25be | 689 | pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, isol_flags); |
6c1b7f68 | 690 | spin_unlock_irqrestore(&zone->lock, flags); |
fea85cff | 691 | |
1d09510b GD |
692 | ret = pfn < end_pfn ? -EBUSY : 0; |
693 | ||
694 | out: | |
0f0848e5 JK |
695 | trace_test_pages_isolated(start_pfn, end_pfn, pfn); |
696 | ||
1d09510b | 697 | return ret; |
a5d76b54 | 698 | } |