1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 /* 2 /*
3 * linux/mm/page_isolation.c 3 * linux/mm/page_isolation.c
4 */ 4 */
5 5
6 #include <linux/mm.h> 6 #include <linux/mm.h>
7 #include <linux/page-isolation.h> 7 #include <linux/page-isolation.h>
8 #include <linux/pageblock-flags.h> 8 #include <linux/pageblock-flags.h>
9 #include <linux/memory.h> 9 #include <linux/memory.h>
10 #include <linux/hugetlb.h> 10 #include <linux/hugetlb.h>
11 #include <linux/page_owner.h> 11 #include <linux/page_owner.h>
12 #include <linux/migrate.h> 12 #include <linux/migrate.h>
13 #include "internal.h" 13 #include "internal.h"
14 14
15 #define CREATE_TRACE_POINTS 15 #define CREATE_TRACE_POINTS
16 #include <trace/events/page_isolation.h> 16 #include <trace/events/page_isolation.h>
17 17
18 /* !! 18 static int set_migratetype_isolate(struct page *page, int migratetype, int isol_flags)
19 * This function checks whether the range [sta <<
20 * unmovable pages or not. The range must fall <<
21 * consequently belong to a single zone. <<
22 * <<
23 * PageLRU check without isolation or lru_lock <<
24 * MIGRATE_MOVABLE block might include unmovab <<
25 * check without lock_page also may miss some <<
26 * race condition. So you can't expect this fu <<
27 * <<
28 * Returns a page without holding a reference. <<
29 * dereference that page (e.g., dumping), it h <<
30 * cannot get removed (e.g., via memory unplug <<
31 * <<
32 */ <<
33 static struct page *has_unmovable_pages(unsign <<
34 int migratetyp <<
35 { <<
36 struct page *page = pfn_to_page(start_ <<
37 struct zone *zone = page_zone(page); <<
38 unsigned long pfn; <<
39 <<
40 VM_BUG_ON(pageblock_start_pfn(start_pf <<
41 pageblock_start_pfn(end_pfn <<
42 <<
43 if (is_migrate_cma_page(page)) { <<
44 /* <<
45 * CMA allocations (alloc_cont <<
46 * isolate CMA pageblocks even <<
47 * so consider them movable he <<
48 */ <<
49 if (is_migrate_cma(migratetype <<
50 return NULL; <<
51 <<
52 return page; <<
53 } <<
54 <<
55 for (pfn = start_pfn; pfn < end_pfn; p <<
56 page = pfn_to_page(pfn); <<
57 <<
58 /* <<
59 * Both, bootmem allocations a <<
60 * PG_reserved and are unmovab <<
61 * allocations inside ZONE_MOV <<
62 * specifying "movablecore". <<
63 */ <<
64 if (PageReserved(page)) <<
65 return page; <<
66 <<
67 /* <<
68 * If the zone is movable and <<
69 * pages then it should be rea <<
70 * is movable. <<
71 */ <<
72 if (zone_idx(zone) == ZONE_MOV <<
73 continue; <<
74 <<
75 /* <<
76 * Hugepages are not in LRU li <<
77 * THPs are on the LRU, but ne <<
78 * We need not scan over tail <<
79 * handle each tail page indiv <<
80 */ <<
81 if (PageHuge(page) || PageTran <<
82 struct folio *folio = <<
83 unsigned int skip_page <<
84 <<
85 if (PageHuge(page)) { <<
86 if (!hugepage_ <<
87 return <<
88 } else if (!folio_test <<
89 return page; <<
90 } <<
91 <<
92 skip_pages = folio_nr_ <<
93 pfn += skip_pages - 1; <<
94 continue; <<
95 } <<
96 <<
97 /* <<
98 * We can't use page_count wit <<
99 * because another CPU can fre <<
100 * This check already skips co <<
101 * because their page->_refcou <<
102 */ <<
103 if (!page_ref_count(page)) { <<
104 if (PageBuddy(page)) <<
105 pfn += (1 << b <<
106 continue; <<
107 } <<
108 <<
109 /* <<
110 * The HWPoisoned page may be <<
111 * page_count() is not 0. <<
112 */ <<
113 if ((flags & MEMORY_OFFLINE) & <<
114 continue; <<
115 <<
116 /* <<
117 * We treat all PageOffline() <<
118 * to give drivers a chance to <<
119 * in MEM_GOING_OFFLINE in ord <<
120 * can be offlined as there ar <<
121 * For actually unmovable Page <<
122 * not support this, we will f <<
123 * move these pages that still <<
124 * (false negatives in this fu <<
125 */ <<
126 if ((flags & MEMORY_OFFLINE) & <<
127 continue; <<
128 <<
129 if (__PageMovable(page) || Pag <<
130 continue; <<
131 <<
132 /* <<
133 * If there are RECLAIMABLE pa <<
134 * it. But now, memory offlin <<
135 * shrink_node_slabs() and it <<
136 */ <<
137 return page; <<
138 } <<
139 return NULL; <<
140 } <<
141 <<
142 /* <<
143 * This function set pageblock migratetype to <<
144 * present in [start_pfn, end_pfn). The pagebl <<
145 * [start_pfn, end_pfn). <<
146 */ <<
147 static int set_migratetype_isolate(struct page <<
148 unsigned long start_pf <<
149 { 19 {
150 struct zone *zone = page_zone(page); 20 struct zone *zone = page_zone(page);
151 struct page *unmovable; 21 struct page *unmovable;
152 unsigned long flags; 22 unsigned long flags;
153 unsigned long check_unmovable_start, c <<
154 <<
155 if (PageUnaccepted(page)) <<
156 accept_page(page); <<
157 23
158 spin_lock_irqsave(&zone->lock, flags); 24 spin_lock_irqsave(&zone->lock, flags);
159 25
160 /* 26 /*
161 * We assume the caller intended to SE 27 * We assume the caller intended to SET migrate type to isolate.
162 * If it is already set, then someone 28 * If it is already set, then someone else must have raced and
163 * set it before us. 29 * set it before us.
164 */ 30 */
165 if (is_migrate_isolate_page(page)) { 31 if (is_migrate_isolate_page(page)) {
166 spin_unlock_irqrestore(&zone-> 32 spin_unlock_irqrestore(&zone->lock, flags);
167 return -EBUSY; 33 return -EBUSY;
168 } 34 }
169 35
170 /* 36 /*
171 * FIXME: Now, memory hotplug doesn't 37 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
172 * We just check MOVABLE pages. 38 * We just check MOVABLE pages.
173 * <<
174 * Pass the intersection of [start_pfn <<
175 * to avoid redundant checks. <<
176 */ 39 */
177 check_unmovable_start = max(page_to_pf !! 40 unmovable = has_unmovable_pages(zone, page, migratetype, isol_flags);
178 check_unmovable_end = min(pageblock_en <<
179 end_pfn); <<
180 <<
181 unmovable = has_unmovable_pages(check_ <<
182 migratetype, isol_flag <<
183 if (!unmovable) { 41 if (!unmovable) {
184 if (!move_freepages_block_isol !! 42 unsigned long nr_pages;
185 spin_unlock_irqrestore !! 43 int mt = get_pageblock_migratetype(page);
186 return -EBUSY; !! 44
187 } !! 45 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
188 zone->nr_isolate_pageblock++; 46 zone->nr_isolate_pageblock++;
>> 47 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE,
>> 48 NULL);
>> 49
>> 50 __mod_zone_freepage_state(zone, -nr_pages, mt);
189 spin_unlock_irqrestore(&zone-> 51 spin_unlock_irqrestore(&zone->lock, flags);
190 return 0; 52 return 0;
191 } 53 }
192 54
193 spin_unlock_irqrestore(&zone->lock, fl 55 spin_unlock_irqrestore(&zone->lock, flags);
194 if (isol_flags & REPORT_FAILURE) { 56 if (isol_flags & REPORT_FAILURE) {
195 /* 57 /*
196 * printk() with zone->lock he 58 * printk() with zone->lock held will likely trigger a
197 * lockdep splat, so defer it 59 * lockdep splat, so defer it here.
198 */ 60 */
199 dump_page(unmovable, "unmovabl 61 dump_page(unmovable, "unmovable page");
200 } 62 }
201 63
202 return -EBUSY; 64 return -EBUSY;
203 } 65 }
204 66
205 static void unset_migratetype_isolate(struct p !! 67 static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
206 { 68 {
207 struct zone *zone; 69 struct zone *zone;
208 unsigned long flags; !! 70 unsigned long flags, nr_pages;
209 bool isolated_page = false; 71 bool isolated_page = false;
210 unsigned int order; 72 unsigned int order;
>> 73 unsigned long pfn, buddy_pfn;
211 struct page *buddy; 74 struct page *buddy;
212 75
213 zone = page_zone(page); 76 zone = page_zone(page);
214 spin_lock_irqsave(&zone->lock, flags); 77 spin_lock_irqsave(&zone->lock, flags);
215 if (!is_migrate_isolate_page(page)) 78 if (!is_migrate_isolate_page(page))
216 goto out; 79 goto out;
217 80
218 /* 81 /*
219 * Because freepage with more than pag 82 * Because freepage with more than pageblock_order on isolated
220 * pageblock is restricted to merge du 83 * pageblock is restricted to merge due to freepage counting problem,
221 * it is possible that there is free b 84 * it is possible that there is free buddy page.
222 * move_freepages_block() doesn't care 85 * move_freepages_block() doesn't care of merge so we need other
223 * approach in order to merge them. Is 86 * approach in order to merge them. Isolation and free will make
224 * these pages to be merged. 87 * these pages to be merged.
225 */ 88 */
226 if (PageBuddy(page)) { 89 if (PageBuddy(page)) {
227 order = buddy_order(page); 90 order = buddy_order(page);
228 if (order >= pageblock_order & !! 91 if (order >= pageblock_order && order < MAX_ORDER - 1) {
229 buddy = find_buddy_pag !! 92 pfn = page_to_pfn(page);
230 !! 93 buddy_pfn = __find_buddy_pfn(pfn, order);
231 if (buddy && !is_migra !! 94 buddy = page + (buddy_pfn - pfn);
232 isolated_page !! 95
233 /* !! 96 if (pfn_valid_within(buddy_pfn) &&
234 * Isolating a !! 97 !is_migrate_isolate_page(buddy)) {
235 * is expected !! 98 __isolate_free_page(page, order);
236 * apply here. !! 99 isolated_page = true;
237 */ <<
238 VM_WARN_ON(!is <<
239 } 100 }
240 } 101 }
241 } 102 }
242 103
243 /* 104 /*
244 * If we isolate freepage with more th 105 * If we isolate freepage with more than pageblock_order, there
245 * should be no freepage in the range, 106 * should be no freepage in the range, so we could avoid costly
246 * pageblock scanning for freepage mov 107 * pageblock scanning for freepage moving.
247 * 108 *
248 * We didn't actually touch any of the 109 * We didn't actually touch any of the isolated pages, so place them
249 * to the tail of the freelist. This i 110 * to the tail of the freelist. This is an optimization for memory
250 * onlining - just onlined memory won' 111 * onlining - just onlined memory won't immediately be considered for
251 * allocation. 112 * allocation.
252 */ 113 */
253 if (!isolated_page) { 114 if (!isolated_page) {
254 /* !! 115 nr_pages = move_freepages_block(zone, page, migratetype, NULL);
255 * Isolating this block alread !! 116 __mod_zone_freepage_state(zone, nr_pages, migratetype);
256 * should not fail on zone bou <<
257 */ <<
258 WARN_ON_ONCE(!move_freepages_b <<
259 } else { <<
260 set_pageblock_migratetype(page <<
261 __putback_isolated_page(page, <<
262 } 117 }
>> 118 set_pageblock_migratetype(page, migratetype);
>> 119 if (isolated_page)
>> 120 __putback_isolated_page(page, order, migratetype);
263 zone->nr_isolate_pageblock--; 121 zone->nr_isolate_pageblock--;
264 out: 122 out:
265 spin_unlock_irqrestore(&zone->lock, fl 123 spin_unlock_irqrestore(&zone->lock, flags);
266 } 124 }
267 125
268 static inline struct page * 126 static inline struct page *
269 __first_valid_page(unsigned long pfn, unsigned 127 __first_valid_page(unsigned long pfn, unsigned long nr_pages)
270 { 128 {
271 int i; 129 int i;
272 130
273 for (i = 0; i < nr_pages; i++) { 131 for (i = 0; i < nr_pages; i++) {
274 struct page *page; 132 struct page *page;
275 133
276 page = pfn_to_online_page(pfn 134 page = pfn_to_online_page(pfn + i);
277 if (!page) 135 if (!page)
278 continue; 136 continue;
279 return page; 137 return page;
280 } 138 }
281 return NULL; 139 return NULL;
282 } 140 }
283 141
284 /** 142 /**
285 * isolate_single_pageblock() -- tries to isol !! 143 * start_isolate_page_range() - make page-allocation-type of range of pages to
286 * within a free or in-use page. !! 144 * be MIGRATE_ISOLATE.
287 * @boundary_pfn: pageblock-alig !! 145 * @start_pfn: The lower PFN of the range to be isolated.
288 * @flags: isolation flag !! 146 * @end_pfn: The upper PFN of the range to be isolated.
289 * @gfp_flags: GFP flags used !! 147 * start_pfn/end_pfn must be aligned to pageblock_order.
290 * @isolate_before: isolate the pageblock <<
291 * @skip_isolation: the flag to skip the p <<
292 * isolate_single_pageblo <<
293 * @migratetype: migrate type to set in <<
294 * <<
295 * Free and in-use pages can be as big as MAX_ <<
296 * pageblock. When not all pageblocks within a <<
297 * time, free page accounting can go wrong. Fo <<
298 * MAX_PAGE_ORDER = pageblock_order + 1, a MAX <<
299 * pagelbocks. <<
300 * [ MAX_PAGE_ORDER ] <<
301 * [ pageblock0 | pageblock1 ] <<
302 * When either pageblock is isolated, if it is <<
303 * split into separate migratetype lists, whic <<
304 * in-use page and freed later, __free_one_pag <<
305 * either. The function handles this by splitt <<
306 * the in-use page then splitting the free pag <<
307 */ <<
308 static int isolate_single_pageblock(unsigned l <<
309 gfp_t gfp_flags, bool <<
310 int migratetype) <<
311 { <<
312 unsigned long start_pfn; <<
313 unsigned long isolate_pageblock; <<
314 unsigned long pfn; <<
315 struct zone *zone; <<
316 int ret; <<
317 <<
318 VM_BUG_ON(!pageblock_aligned(boundary_ <<
319 <<
320 if (isolate_before) <<
321 isolate_pageblock = boundary_p <<
322 else <<
323 isolate_pageblock = boundary_p <<
324 <<
325 /* <<
326 * scan at the beginning of MAX_ORDER_ <<
327 * only isolating a subset of pagebloc <<
328 * free or in-use page. Also make sure <<
329 * are within the same zone. <<
330 */ <<
331 zone = page_zone(pfn_to_page(isolate_ <<
332 start_pfn = max(ALIGN_DOWN(isolate_pa <<
333 zone->zo <<
334 <<
335 if (skip_isolation) { <<
336 int mt __maybe_unused = get_pa <<
337 <<
338 VM_BUG_ON(!is_migrate_isolate( <<
339 } else { <<
340 ret = set_migratetype_isolate( <<
341 flags, isolate <<
342 <<
343 if (ret) <<
344 return ret; <<
345 } <<
346 <<
347 /* <<
348 * Bail out early when the to-be-isola <<
349 * a free or in-use page across bounda <<
350 * <<
351 * 1. isolate before boundary_pfn: the <<
352 * 2. isolate after boundary_pfn: the <<
353 * <<
354 * This also ensures correctness. With <<
355 * boundary_pfn and [start_pfn, bounda <<
356 * __first_valid_page() will return un <<
357 * below. <<
358 */ <<
359 if (isolate_before) { <<
360 if (!pfn_to_online_page(bounda <<
361 return 0; <<
362 } else { <<
363 if (!pfn_to_online_page(bounda <<
364 return 0; <<
365 } <<
366 <<
367 for (pfn = start_pfn; pfn < boundary_p <<
368 struct page *page = __first_va <<
369 <<
370 VM_BUG_ON(!page); <<
371 pfn = page_to_pfn(page); <<
372 <<
373 if (PageUnaccepted(page)) { <<
374 pfn += MAX_ORDER_NR_PA <<
375 continue; <<
376 } <<
377 <<
378 if (PageBuddy(page)) { <<
379 int order = buddy_orde <<
380 <<
381 /* move_freepages_bloc <<
382 VM_WARN_ON_ONCE(pfn + <<
383 <<
384 pfn += 1UL << order; <<
385 continue; <<
386 } <<
387 <<
388 /* <<
389 * If a compound page is strad <<
390 * to migrate it out of the wa <<
391 * <<
392 * We don't have to worry abou <<
393 * free page that straddles in <<
394 * pages are freed as order-0 <<
395 * (currently) do not exceed p <<
396 * <<
397 * The block of interest has a <<
398 * MIGRATE_ISOLATE above, so w <<
399 * will free its pages onto th <<
400 */ <<
401 if (PageCompound(page)) { <<
402 struct page *head = co <<
403 unsigned long head_pfn <<
404 unsigned long nr_pages <<
405 <<
406 if (head_pfn + nr_page <<
407 PageHuge(page)) { <<
408 pfn = head_pfn <<
409 continue; <<
410 } <<
411 <<
412 /* <<
413 * These pages are mov <<
414 * not expected to exc <<
415 * <<
416 * Let us know when th <<
417 * proper free and spl <<
418 */ <<
419 VM_WARN_ON_ONCE_PAGE(P <<
420 VM_WARN_ON_ONCE_PAGE(_ <<
421 <<
422 goto failed; <<
423 } <<
424 <<
425 pfn++; <<
426 } <<
427 return 0; <<
428 failed: <<
429 /* restore the original migratetype */ <<
430 if (!skip_isolation) <<
431 unset_migratetype_isolate(pfn_ <<
432 return -EBUSY; <<
433 } <<
434 <<
435 /** <<
436 * start_isolate_page_range() - mark page rang <<
437 * @start_pfn: The first PFN of the r <<
438 * @end_pfn: The last PFN of the ra <<
439 * @migratetype: Migrate type to set in 148 * @migratetype: Migrate type to set in error recovery.
440 * @flags: The following flags ar 149 * @flags: The following flags are allowed (they can be combined in
441 * a bit mask) 150 * a bit mask)
442 * MEMORY_OFFLINE - isola 151 * MEMORY_OFFLINE - isolate to offline (!allocate) memory
443 * e.g., 152 * e.g., skip over PageHWPoison() pages
444 * and P 153 * and PageOffline() pages.
445 * REPORT_FAILURE - repor 154 * REPORT_FAILURE - report details about the failure to
446 * isolate the range 155 * isolate the range
447 * @gfp_flags: GFP flags used for mig <<
448 * range boundaries. <<
449 * 156 *
450 * Making page-allocation-type to be MIGRATE_I 157 * Making page-allocation-type to be MIGRATE_ISOLATE means free pages in
451 * the range will never be allocated. Any free 158 * the range will never be allocated. Any free pages and pages freed in the
452 * future will not be allocated again. If spec 159 * future will not be allocated again. If specified range includes migrate types
453 * other than MOVABLE or CMA, this will fail w 160 * other than MOVABLE or CMA, this will fail with -EBUSY. For isolating all
454 * pages in the range finally, the caller have 161 * pages in the range finally, the caller have to free all pages in the range.
455 * test_page_isolated() can be used for test i 162 * test_page_isolated() can be used for test it.
456 * 163 *
457 * The function first tries to isolate the pag <<
458 * of the range, since there might be pages ac <<
459 * Afterwards, it isolates the rest of the ran <<
460 * <<
461 * There is no high level synchronization mech 164 * There is no high level synchronization mechanism that prevents two threads
462 * from trying to isolate overlapping ranges. 165 * from trying to isolate overlapping ranges. If this happens, one thread
463 * will notice pageblocks in the overlapping r 166 * will notice pageblocks in the overlapping range already set to isolate.
464 * This happens in set_migratetype_isolate, an 167 * This happens in set_migratetype_isolate, and set_migratetype_isolate
465 * returns an error. We then clean up by resto 168 * returns an error. We then clean up by restoring the migration type on
466 * pageblocks we may have modified and return 169 * pageblocks we may have modified and return -EBUSY to caller. This
467 * prevents two threads from simultaneously wo 170 * prevents two threads from simultaneously working on overlapping ranges.
468 * 171 *
469 * Please note that there is no strong synchro 172 * Please note that there is no strong synchronization with the page allocator
470 * either. Pages might be freed while their pa 173 * either. Pages might be freed while their page blocks are marked ISOLATED.
471 * A call to drain_all_pages() after isolation 174 * A call to drain_all_pages() after isolation can flush most of them. However
472 * in some cases pages might still end up on p 175 * in some cases pages might still end up on pcp lists and that would allow
473 * for their allocation even when they are in 176 * for their allocation even when they are in fact isolated already. Depending
474 * on how strong of a guarantee the caller nee 177 * on how strong of a guarantee the caller needs, zone_pcp_disable/enable()
475 * might be used to flush and disable pcplist 178 * might be used to flush and disable pcplist before isolation and enable after
476 * unisolation. 179 * unisolation.
477 * 180 *
478 * Return: 0 on success and -EBUSY if any part 181 * Return: 0 on success and -EBUSY if any part of range cannot be isolated.
479 */ 182 */
480 int start_isolate_page_range(unsigned long sta 183 int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
481 int migratetype, !! 184 unsigned migratetype, int flags)
482 { 185 {
483 unsigned long pfn; 186 unsigned long pfn;
>> 187 unsigned long undo_pfn;
484 struct page *page; 188 struct page *page;
485 /* isolation is done at page block gra !! 189
486 unsigned long isolate_start = pagebloc !! 190 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
487 unsigned long isolate_end = pageblock_ !! 191 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
488 int ret; !! 192
489 bool skip_isolation = false; !! 193 for (pfn = start_pfn;
490 !! 194 pfn < end_pfn;
491 /* isolate [isolate_start, isolate_sta <<
492 ret = isolate_single_pageblock(isolate <<
493 skip_isolation, migrat <<
494 if (ret) <<
495 return ret; <<
496 <<
497 if (isolate_start == isolate_end - pag <<
498 skip_isolation = true; <<
499 <<
500 /* isolate [isolate_end - pageblock_nr <<
501 ret = isolate_single_pageblock(isolate <<
502 skip_isolation, migrat <<
503 if (ret) { <<
504 unset_migratetype_isolate(pfn_ <<
505 return ret; <<
506 } <<
507 <<
508 /* skip isolated pageblocks at the beg <<
509 for (pfn = isolate_start + pageblock_n <<
510 pfn < isolate_end - pageblock_nr_ <<
511 pfn += pageblock_nr_pages) { 195 pfn += pageblock_nr_pages) {
512 page = __first_valid_page(pfn, 196 page = __first_valid_page(pfn, pageblock_nr_pages);
513 if (page && set_migratetype_is !! 197 if (page) {
514 start_ !! 198 if (set_migratetype_isolate(page, migratetype, flags)) {
515 undo_isolate_page_rang !! 199 undo_pfn = pfn;
516 unset_migratetype_isol !! 200 goto undo;
517 pfn_to_page(is !! 201 }
518 migratetype); <<
519 return -EBUSY; <<
520 } 202 }
521 } 203 }
522 return 0; 204 return 0;
>> 205 undo:
>> 206 for (pfn = start_pfn;
>> 207 pfn < undo_pfn;
>> 208 pfn += pageblock_nr_pages) {
>> 209 struct page *page = pfn_to_online_page(pfn);
>> 210 if (!page)
>> 211 continue;
>> 212 unset_migratetype_isolate(page, migratetype);
>> 213 }
>> 214
>> 215 return -EBUSY;
523 } 216 }
524 217
525 /** !! 218 /*
526 * undo_isolate_page_range - undo effects of s !! 219 * Make isolated pages available again.
527 * @start_pfn: The first PFN of the i <<
528 * @end_pfn: The last PFN of the is <<
529 * @migratetype: New migrate type to se <<
530 * <<
531 * This finds every MIGRATE_ISOLATE page block <<
532 * and switches it to @migratetype. <<
533 */ 220 */
534 void undo_isolate_page_range(unsigned long sta 221 void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
535 int migratetype) !! 222 unsigned migratetype)
536 { 223 {
537 unsigned long pfn; 224 unsigned long pfn;
538 struct page *page; 225 struct page *page;
539 unsigned long isolate_start = pagebloc <<
540 unsigned long isolate_end = pageblock_ <<
541 226
542 for (pfn = isolate_start; !! 227 BUG_ON(!IS_ALIGNED(start_pfn, pageblock_nr_pages));
543 pfn < isolate_end; !! 228 BUG_ON(!IS_ALIGNED(end_pfn, pageblock_nr_pages));
>> 229
>> 230 for (pfn = start_pfn;
>> 231 pfn < end_pfn;
544 pfn += pageblock_nr_pages) { 232 pfn += pageblock_nr_pages) {
545 page = __first_valid_page(pfn, 233 page = __first_valid_page(pfn, pageblock_nr_pages);
546 if (!page || !is_migrate_isola 234 if (!page || !is_migrate_isolate_page(page))
547 continue; 235 continue;
548 unset_migratetype_isolate(page 236 unset_migratetype_isolate(page, migratetype);
549 } 237 }
550 } 238 }
551 /* 239 /*
552 * Test all pages in the range is free(means i 240 * Test all pages in the range is free(means isolated) or not.
553 * all pages in [start_pfn...end_pfn) must be 241 * all pages in [start_pfn...end_pfn) must be in the same zone.
554 * zone->lock must be held before call this. 242 * zone->lock must be held before call this.
555 * 243 *
556 * Returns the last tested pfn. 244 * Returns the last tested pfn.
557 */ 245 */
558 static unsigned long 246 static unsigned long
559 __test_page_isolated_in_pageblock(unsigned lon 247 __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
560 int flags) 248 int flags)
561 { 249 {
562 struct page *page; 250 struct page *page;
563 251
564 while (pfn < end_pfn) { 252 while (pfn < end_pfn) {
>> 253 if (!pfn_valid_within(pfn)) {
>> 254 pfn++;
>> 255 continue;
>> 256 }
565 page = pfn_to_page(pfn); 257 page = pfn_to_page(pfn);
566 if (PageBuddy(page)) 258 if (PageBuddy(page))
567 /* 259 /*
568 * If the page is on a 260 * If the page is on a free list, it has to be on
569 * the correct MIGRATE 261 * the correct MIGRATE_ISOLATE freelist. There is no
570 * simple way to verif 262 * simple way to verify that as VM_BUG_ON(), though.
571 */ 263 */
572 pfn += 1 << buddy_orde 264 pfn += 1 << buddy_order(page);
573 else if ((flags & MEMORY_OFFLI 265 else if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
574 /* A HWPoisoned page c 266 /* A HWPoisoned page cannot be also PageBuddy */
575 pfn++; 267 pfn++;
576 else if ((flags & MEMORY_OFFLI 268 else if ((flags & MEMORY_OFFLINE) && PageOffline(page) &&
577 !page_count(page)) 269 !page_count(page))
578 /* 270 /*
579 * The responsible dri 271 * The responsible driver agreed to skip PageOffline()
580 * pages when offlinin 272 * pages when offlining memory by dropping its
581 * reference in MEM_GO 273 * reference in MEM_GOING_OFFLINE.
582 */ 274 */
583 pfn++; 275 pfn++;
584 else 276 else
585 break; 277 break;
586 } 278 }
587 279
588 return pfn; 280 return pfn;
589 } 281 }
590 282
591 /** !! 283 /* Caller should ensure that requested range is in a single zone */
592 * test_pages_isolated - check if pageblocks i <<
593 * @start_pfn: The first PFN of the i <<
594 * @end_pfn: The first PFN *after* <<
595 * @isol_flags: Testing mode flags <<
596 * <<
597 * This tests if all in the specified range ar <<
598 * <<
599 * If %MEMORY_OFFLINE is specified in @flags, <<
600 * poisoned and offlined pages free as well. <<
601 * <<
602 * Caller must ensure the requested range does <<
603 * <<
604 * Returns 0 if true, -EBUSY if one or more pa <<
605 */ <<
606 int test_pages_isolated(unsigned long start_pf 284 int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
607 int isol_flags) 285 int isol_flags)
608 { 286 {
609 unsigned long pfn, flags; 287 unsigned long pfn, flags;
610 struct page *page; 288 struct page *page;
611 struct zone *zone; 289 struct zone *zone;
612 int ret; <<
613 290
614 /* 291 /*
615 * Note: pageblock_nr_pages != MAX_PAG !! 292 * Note: pageblock_nr_pages != MAX_ORDER. Then, chunks of free pages
616 * pages are not aligned to pageblock_ !! 293 * are not aligned to pageblock_nr_pages.
617 * Then we just check migratetype firs 294 * Then we just check migratetype first.
618 */ 295 */
619 for (pfn = start_pfn; pfn < end_pfn; p 296 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
620 page = __first_valid_page(pfn, 297 page = __first_valid_page(pfn, pageblock_nr_pages);
621 if (page && !is_migrate_isolat 298 if (page && !is_migrate_isolate_page(page))
622 break; 299 break;
623 } 300 }
624 page = __first_valid_page(start_pfn, e 301 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
625 if ((pfn < end_pfn) || !page) { !! 302 if ((pfn < end_pfn) || !page)
626 ret = -EBUSY; !! 303 return -EBUSY;
627 goto out; <<
628 } <<
629 <<
630 /* Check all pages are free or marked 304 /* Check all pages are free or marked as ISOLATED */
631 zone = page_zone(page); 305 zone = page_zone(page);
632 spin_lock_irqsave(&zone->lock, flags); 306 spin_lock_irqsave(&zone->lock, flags);
633 pfn = __test_page_isolated_in_pagebloc 307 pfn = __test_page_isolated_in_pageblock(start_pfn, end_pfn, isol_flags);
634 spin_unlock_irqrestore(&zone->lock, fl 308 spin_unlock_irqrestore(&zone->lock, flags);
635 309
636 ret = pfn < end_pfn ? -EBUSY : 0; <<
637 <<
638 out: <<
639 trace_test_pages_isolated(start_pfn, e 310 trace_test_pages_isolated(start_pfn, end_pfn, pfn);
640 311
641 return ret; !! 312 return pfn < end_pfn ? -EBUSY : 0;
642 } 313 }
643 314
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.