]>
Commit | Line | Data |
---|---|---|
b20a3503 CL |
1 | /* |
2 | * Memory Migration functionality - linux/mm/migration.c | |
3 | * | |
4 | * Copyright (C) 2006 Silicon Graphics, Inc., Christoph Lameter | |
5 | * | |
6 | * Page migration was first developed in the context of the memory hotplug | |
7 | * project. The main authors of the migration code are: | |
8 | * | |
9 | * IWAMOTO Toshihiro <iwamoto@valinux.co.jp> | |
10 | * Hirokazu Takahashi <taka@valinux.co.jp> | |
11 | * Dave Hansen <haveblue@us.ibm.com> | |
12 | * Christoph Lameter <clameter@sgi.com> | |
13 | */ | |
14 | ||
15 | #include <linux/migrate.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/swap.h> | |
0697212a | 18 | #include <linux/swapops.h> |
b20a3503 | 19 | #include <linux/pagemap.h> |
e23ca00b | 20 | #include <linux/buffer_head.h> |
b20a3503 CL |
21 | #include <linux/mm_inline.h> |
22 | #include <linux/pagevec.h> | |
23 | #include <linux/rmap.h> | |
24 | #include <linux/topology.h> | |
25 | #include <linux/cpu.h> | |
26 | #include <linux/cpuset.h> | |
b20a3503 CL |
27 | |
28 | #include "internal.h" | |
29 | ||
b20a3503 CL |
30 | /* The maximum number of pages to take off the LRU for migration */ |
31 | #define MIGRATE_CHUNK_SIZE 256 | |
32 | ||
33 | #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru)) | |
34 | ||
35 | /* | |
36 | * Isolate one page from the LRU lists. If successful put it onto | |
37 | * the indicated list with elevated page count. | |
38 | * | |
39 | * Result: | |
40 | * -EBUSY: page not on LRU list | |
41 | * 0: page removed from LRU list and added to the specified list. | |
42 | */ | |
43 | int isolate_lru_page(struct page *page, struct list_head *pagelist) | |
44 | { | |
45 | int ret = -EBUSY; | |
46 | ||
47 | if (PageLRU(page)) { | |
48 | struct zone *zone = page_zone(page); | |
49 | ||
50 | spin_lock_irq(&zone->lru_lock); | |
51 | if (PageLRU(page)) { | |
52 | ret = 0; | |
53 | get_page(page); | |
54 | ClearPageLRU(page); | |
55 | if (PageActive(page)) | |
56 | del_page_from_active_list(zone, page); | |
57 | else | |
58 | del_page_from_inactive_list(zone, page); | |
59 | list_add_tail(&page->lru, pagelist); | |
60 | } | |
61 | spin_unlock_irq(&zone->lru_lock); | |
62 | } | |
63 | return ret; | |
64 | } | |
65 | ||
66 | /* | |
67 | * migrate_prep() needs to be called after we have compiled the list of pages | |
68 | * to be migrated using isolate_lru_page() but before we begin a series of calls | |
69 | * to migrate_pages(). | |
70 | */ | |
71 | int migrate_prep(void) | |
72 | { | |
b20a3503 CL |
73 | /* |
74 | * Clear the LRU lists so pages can be isolated. | |
75 | * Note that pages may be moved off the LRU after we have | |
76 | * drained them. Those pages will fail to migrate like other | |
77 | * pages that may be busy. | |
78 | */ | |
79 | lru_add_drain_all(); | |
80 | ||
81 | return 0; | |
82 | } | |
83 | ||
84 | static inline void move_to_lru(struct page *page) | |
85 | { | |
86 | list_del(&page->lru); | |
87 | if (PageActive(page)) { | |
88 | /* | |
89 | * lru_cache_add_active checks that | |
90 | * the PG_active bit is off. | |
91 | */ | |
92 | ClearPageActive(page); | |
93 | lru_cache_add_active(page); | |
94 | } else { | |
95 | lru_cache_add(page); | |
96 | } | |
97 | put_page(page); | |
98 | } | |
99 | ||
100 | /* | |
101 | * Add isolated pages on the list back to the LRU. | |
102 | * | |
103 | * returns the number of pages put back. | |
104 | */ | |
105 | int putback_lru_pages(struct list_head *l) | |
106 | { | |
107 | struct page *page; | |
108 | struct page *page2; | |
109 | int count = 0; | |
110 | ||
111 | list_for_each_entry_safe(page, page2, l, lru) { | |
112 | move_to_lru(page); | |
113 | count++; | |
114 | } | |
115 | return count; | |
116 | } | |
117 | ||
0697212a CL |
118 | static inline int is_swap_pte(pte_t pte) |
119 | { | |
120 | return !pte_none(pte) && !pte_present(pte) && !pte_file(pte); | |
121 | } | |
122 | ||
123 | /* | |
124 | * Restore a potential migration pte to a working pte entry | |
125 | */ | |
126 | static void remove_migration_pte(struct vm_area_struct *vma, unsigned long addr, | |
127 | struct page *old, struct page *new) | |
128 | { | |
129 | struct mm_struct *mm = vma->vm_mm; | |
130 | swp_entry_t entry; | |
131 | pgd_t *pgd; | |
132 | pud_t *pud; | |
133 | pmd_t *pmd; | |
134 | pte_t *ptep, pte; | |
135 | spinlock_t *ptl; | |
136 | ||
137 | pgd = pgd_offset(mm, addr); | |
138 | if (!pgd_present(*pgd)) | |
139 | return; | |
140 | ||
141 | pud = pud_offset(pgd, addr); | |
142 | if (!pud_present(*pud)) | |
143 | return; | |
144 | ||
145 | pmd = pmd_offset(pud, addr); | |
146 | if (!pmd_present(*pmd)) | |
147 | return; | |
148 | ||
149 | ptep = pte_offset_map(pmd, addr); | |
150 | ||
151 | if (!is_swap_pte(*ptep)) { | |
152 | pte_unmap(ptep); | |
153 | return; | |
154 | } | |
155 | ||
156 | ptl = pte_lockptr(mm, pmd); | |
157 | spin_lock(ptl); | |
158 | pte = *ptep; | |
159 | if (!is_swap_pte(pte)) | |
160 | goto out; | |
161 | ||
162 | entry = pte_to_swp_entry(pte); | |
163 | ||
164 | if (!is_migration_entry(entry) || migration_entry_to_page(entry) != old) | |
165 | goto out; | |
166 | ||
0697212a CL |
167 | get_page(new); |
168 | pte = pte_mkold(mk_pte(new, vma->vm_page_prot)); | |
169 | if (is_write_migration_entry(entry)) | |
170 | pte = pte_mkwrite(pte); | |
171 | set_pte_at(mm, addr, ptep, pte); | |
172 | page_add_anon_rmap(new, vma, addr); | |
173 | out: | |
174 | pte_unmap_unlock(ptep, ptl); | |
175 | } | |
176 | ||
177 | /* | |
178 | * Get rid of all migration entries and replace them by | |
179 | * references to the indicated page. | |
180 | * | |
181 | * Must hold mmap_sem lock on at least one of the vmas containing | |
182 | * the page so that the anon_vma cannot vanish. | |
183 | */ | |
184 | static void remove_migration_ptes(struct page *old, struct page *new) | |
185 | { | |
186 | struct anon_vma *anon_vma; | |
187 | struct vm_area_struct *vma; | |
188 | unsigned long mapping; | |
189 | ||
190 | mapping = (unsigned long)new->mapping; | |
191 | ||
192 | if (!mapping || (mapping & PAGE_MAPPING_ANON) == 0) | |
193 | return; | |
194 | ||
195 | /* | |
196 | * We hold the mmap_sem lock. So no need to call page_lock_anon_vma. | |
197 | */ | |
198 | anon_vma = (struct anon_vma *) (mapping - PAGE_MAPPING_ANON); | |
199 | spin_lock(&anon_vma->lock); | |
200 | ||
201 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) | |
202 | remove_migration_pte(vma, page_address_in_vma(new, vma), | |
203 | old, new); | |
204 | ||
205 | spin_unlock(&anon_vma->lock); | |
206 | } | |
207 | ||
208 | /* | |
209 | * Something used the pte of a page under migration. We need to | |
210 | * get to the page and wait until migration is finished. | |
211 | * When we return from this function the fault will be retried. | |
212 | * | |
213 | * This function is called from do_swap_page(). | |
214 | */ | |
215 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, | |
216 | unsigned long address) | |
217 | { | |
218 | pte_t *ptep, pte; | |
219 | spinlock_t *ptl; | |
220 | swp_entry_t entry; | |
221 | struct page *page; | |
222 | ||
223 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
224 | pte = *ptep; | |
225 | if (!is_swap_pte(pte)) | |
226 | goto out; | |
227 | ||
228 | entry = pte_to_swp_entry(pte); | |
229 | if (!is_migration_entry(entry)) | |
230 | goto out; | |
231 | ||
232 | page = migration_entry_to_page(entry); | |
233 | ||
234 | get_page(page); | |
235 | pte_unmap_unlock(ptep, ptl); | |
236 | wait_on_page_locked(page); | |
237 | put_page(page); | |
238 | return; | |
239 | out: | |
240 | pte_unmap_unlock(ptep, ptl); | |
241 | } | |
242 | ||
b20a3503 | 243 | /* |
c3fcf8a5 | 244 | * Replace the page in the mapping. |
5b5c7120 CL |
245 | * |
246 | * The number of remaining references must be: | |
247 | * 1 for anonymous pages without a mapping | |
248 | * 2 for pages with a mapping | |
249 | * 3 for pages with a mapping and PagePrivate set. | |
b20a3503 | 250 | */ |
2d1db3b1 CL |
251 | static int migrate_page_move_mapping(struct address_space *mapping, |
252 | struct page *newpage, struct page *page) | |
b20a3503 | 253 | { |
b20a3503 CL |
254 | struct page **radix_pointer; |
255 | ||
6c5240ae CL |
256 | if (!mapping) { |
257 | /* Anonymous page */ | |
258 | if (page_count(page) != 1) | |
259 | return -EAGAIN; | |
260 | return 0; | |
261 | } | |
262 | ||
b20a3503 CL |
263 | write_lock_irq(&mapping->tree_lock); |
264 | ||
265 | radix_pointer = (struct page **)radix_tree_lookup_slot( | |
266 | &mapping->page_tree, | |
267 | page_index(page)); | |
268 | ||
6c5240ae | 269 | if (page_count(page) != 2 + !!PagePrivate(page) || |
b20a3503 CL |
270 | *radix_pointer != page) { |
271 | write_unlock_irq(&mapping->tree_lock); | |
e23ca00b | 272 | return -EAGAIN; |
b20a3503 CL |
273 | } |
274 | ||
275 | /* | |
276 | * Now we know that no one else is looking at the page. | |
b20a3503 CL |
277 | */ |
278 | get_page(newpage); | |
6c5240ae | 279 | #ifdef CONFIG_SWAP |
b20a3503 CL |
280 | if (PageSwapCache(page)) { |
281 | SetPageSwapCache(newpage); | |
282 | set_page_private(newpage, page_private(page)); | |
283 | } | |
6c5240ae | 284 | #endif |
b20a3503 CL |
285 | |
286 | *radix_pointer = newpage; | |
287 | __put_page(page); | |
288 | write_unlock_irq(&mapping->tree_lock); | |
289 | ||
290 | return 0; | |
291 | } | |
b20a3503 CL |
292 | |
293 | /* | |
294 | * Copy the page to its new location | |
295 | */ | |
e7340f73 | 296 | static void migrate_page_copy(struct page *newpage, struct page *page) |
b20a3503 CL |
297 | { |
298 | copy_highpage(newpage, page); | |
299 | ||
300 | if (PageError(page)) | |
301 | SetPageError(newpage); | |
302 | if (PageReferenced(page)) | |
303 | SetPageReferenced(newpage); | |
304 | if (PageUptodate(page)) | |
305 | SetPageUptodate(newpage); | |
306 | if (PageActive(page)) | |
307 | SetPageActive(newpage); | |
308 | if (PageChecked(page)) | |
309 | SetPageChecked(newpage); | |
310 | if (PageMappedToDisk(page)) | |
311 | SetPageMappedToDisk(newpage); | |
312 | ||
313 | if (PageDirty(page)) { | |
314 | clear_page_dirty_for_io(page); | |
315 | set_page_dirty(newpage); | |
316 | } | |
317 | ||
6c5240ae | 318 | #ifdef CONFIG_SWAP |
b20a3503 | 319 | ClearPageSwapCache(page); |
6c5240ae | 320 | #endif |
b20a3503 CL |
321 | ClearPageActive(page); |
322 | ClearPagePrivate(page); | |
323 | set_page_private(page, 0); | |
324 | page->mapping = NULL; | |
325 | ||
326 | /* | |
327 | * If any waiters have accumulated on the new page then | |
328 | * wake them up. | |
329 | */ | |
330 | if (PageWriteback(newpage)) | |
331 | end_page_writeback(newpage); | |
332 | } | |
b20a3503 | 333 | |
1d8b85cc CL |
334 | /************************************************************ |
335 | * Migration functions | |
336 | ***********************************************************/ | |
337 | ||
338 | /* Always fail migration. Used for mappings that are not movable */ | |
2d1db3b1 CL |
339 | int fail_migrate_page(struct address_space *mapping, |
340 | struct page *newpage, struct page *page) | |
1d8b85cc CL |
341 | { |
342 | return -EIO; | |
343 | } | |
344 | EXPORT_SYMBOL(fail_migrate_page); | |
345 | ||
b20a3503 CL |
346 | /* |
347 | * Common logic to directly migrate a single page suitable for | |
348 | * pages that do not use PagePrivate. | |
349 | * | |
350 | * Pages are locked upon entry and exit. | |
351 | */ | |
2d1db3b1 CL |
352 | int migrate_page(struct address_space *mapping, |
353 | struct page *newpage, struct page *page) | |
b20a3503 CL |
354 | { |
355 | int rc; | |
356 | ||
357 | BUG_ON(PageWriteback(page)); /* Writeback must be complete */ | |
358 | ||
2d1db3b1 | 359 | rc = migrate_page_move_mapping(mapping, newpage, page); |
b20a3503 CL |
360 | |
361 | if (rc) | |
362 | return rc; | |
363 | ||
364 | migrate_page_copy(newpage, page); | |
b20a3503 CL |
365 | return 0; |
366 | } | |
367 | EXPORT_SYMBOL(migrate_page); | |
368 | ||
1d8b85cc CL |
369 | /* |
370 | * Migration function for pages with buffers. This function can only be used | |
371 | * if the underlying filesystem guarantees that no other references to "page" | |
372 | * exist. | |
373 | */ | |
2d1db3b1 CL |
374 | int buffer_migrate_page(struct address_space *mapping, |
375 | struct page *newpage, struct page *page) | |
1d8b85cc | 376 | { |
1d8b85cc CL |
377 | struct buffer_head *bh, *head; |
378 | int rc; | |
379 | ||
1d8b85cc | 380 | if (!page_has_buffers(page)) |
2d1db3b1 | 381 | return migrate_page(mapping, newpage, page); |
1d8b85cc CL |
382 | |
383 | head = page_buffers(page); | |
384 | ||
2d1db3b1 | 385 | rc = migrate_page_move_mapping(mapping, newpage, page); |
1d8b85cc CL |
386 | |
387 | if (rc) | |
388 | return rc; | |
389 | ||
390 | bh = head; | |
391 | do { | |
392 | get_bh(bh); | |
393 | lock_buffer(bh); | |
394 | bh = bh->b_this_page; | |
395 | ||
396 | } while (bh != head); | |
397 | ||
398 | ClearPagePrivate(page); | |
399 | set_page_private(newpage, page_private(page)); | |
400 | set_page_private(page, 0); | |
401 | put_page(page); | |
402 | get_page(newpage); | |
403 | ||
404 | bh = head; | |
405 | do { | |
406 | set_bh_page(bh, newpage, bh_offset(bh)); | |
407 | bh = bh->b_this_page; | |
408 | ||
409 | } while (bh != head); | |
410 | ||
411 | SetPagePrivate(newpage); | |
412 | ||
413 | migrate_page_copy(newpage, page); | |
414 | ||
415 | bh = head; | |
416 | do { | |
417 | unlock_buffer(bh); | |
418 | put_bh(bh); | |
419 | bh = bh->b_this_page; | |
420 | ||
421 | } while (bh != head); | |
422 | ||
423 | return 0; | |
424 | } | |
425 | EXPORT_SYMBOL(buffer_migrate_page); | |
426 | ||
8351a6e4 CL |
427 | static int fallback_migrate_page(struct address_space *mapping, |
428 | struct page *newpage, struct page *page) | |
429 | { | |
430 | /* | |
431 | * Default handling if a filesystem does not provide | |
432 | * a migration function. We can only migrate clean | |
433 | * pages so try to write out any dirty pages first. | |
434 | */ | |
435 | if (PageDirty(page)) { | |
436 | switch (pageout(page, mapping)) { | |
437 | case PAGE_KEEP: | |
438 | case PAGE_ACTIVATE: | |
439 | return -EAGAIN; | |
440 | ||
441 | case PAGE_SUCCESS: | |
442 | /* Relock since we lost the lock */ | |
443 | lock_page(page); | |
444 | /* Must retry since page state may have changed */ | |
445 | return -EAGAIN; | |
446 | ||
447 | case PAGE_CLEAN: | |
448 | ; /* try to migrate the page below */ | |
449 | } | |
450 | } | |
451 | ||
452 | /* | |
453 | * Buffers may be managed in a filesystem specific way. | |
454 | * We must have no buffers or drop them. | |
455 | */ | |
456 | if (page_has_buffers(page) && | |
457 | !try_to_release_page(page, GFP_KERNEL)) | |
458 | return -EAGAIN; | |
459 | ||
460 | return migrate_page(mapping, newpage, page); | |
461 | } | |
462 | ||
b20a3503 CL |
463 | /* |
464 | * migrate_pages | |
465 | * | |
466 | * Two lists are passed to this function. The first list | |
467 | * contains the pages isolated from the LRU to be migrated. | |
468 | * The second list contains new pages that the pages isolated | |
d75a0fcd | 469 | * can be moved to. |
b20a3503 CL |
470 | * |
471 | * The function returns after 10 attempts or if no pages | |
472 | * are movable anymore because to has become empty | |
473 | * or no retryable pages exist anymore. | |
474 | * | |
475 | * Return: Number of pages not migrated when "to" ran empty. | |
476 | */ | |
477 | int migrate_pages(struct list_head *from, struct list_head *to, | |
478 | struct list_head *moved, struct list_head *failed) | |
479 | { | |
480 | int retry; | |
481 | int nr_failed = 0; | |
482 | int pass = 0; | |
483 | struct page *page; | |
484 | struct page *page2; | |
485 | int swapwrite = current->flags & PF_SWAPWRITE; | |
486 | int rc; | |
487 | ||
488 | if (!swapwrite) | |
489 | current->flags |= PF_SWAPWRITE; | |
490 | ||
491 | redo: | |
492 | retry = 0; | |
493 | ||
494 | list_for_each_entry_safe(page, page2, from, lru) { | |
495 | struct page *newpage = NULL; | |
496 | struct address_space *mapping; | |
497 | ||
498 | cond_resched(); | |
499 | ||
500 | rc = 0; | |
501 | if (page_count(page) == 1) | |
502 | /* page was freed from under us. So we are done. */ | |
503 | goto next; | |
504 | ||
505 | if (to && list_empty(to)) | |
506 | break; | |
507 | ||
508 | /* | |
509 | * Skip locked pages during the first two passes to give the | |
510 | * functions holding the lock time to release the page. Later we | |
511 | * use lock_page() to have a higher chance of acquiring the | |
512 | * lock. | |
513 | */ | |
514 | rc = -EAGAIN; | |
515 | if (pass > 2) | |
516 | lock_page(page); | |
517 | else | |
518 | if (TestSetPageLocked(page)) | |
519 | goto next; | |
520 | ||
521 | /* | |
522 | * Only wait on writeback if we have already done a pass where | |
523 | * we we may have triggered writeouts for lots of pages. | |
524 | */ | |
d75a0fcd | 525 | if (pass > 0) |
b20a3503 | 526 | wait_on_page_writeback(page); |
d75a0fcd | 527 | else |
b20a3503 CL |
528 | if (PageWriteback(page)) |
529 | goto unlock_page; | |
b20a3503 | 530 | |
c3fcf8a5 | 531 | /* |
6c5240ae | 532 | * Establish migration ptes or remove ptes |
c3fcf8a5 CL |
533 | */ |
534 | rc = -EPERM; | |
535 | if (try_to_unmap(page, 1) == SWAP_FAIL) | |
536 | /* A vma has VM_LOCKED set -> permanent failure */ | |
2d1db3b1 | 537 | goto unlock_page; |
c3fcf8a5 CL |
538 | |
539 | rc = -EAGAIN; | |
540 | if (page_mapped(page)) | |
2d1db3b1 CL |
541 | goto unlock_page; |
542 | ||
543 | newpage = lru_to_page(to); | |
544 | lock_page(newpage); | |
545 | /* Prepare mapping for the new page.*/ | |
546 | newpage->index = page->index; | |
547 | newpage->mapping = page->mapping; | |
548 | ||
b20a3503 CL |
549 | /* |
550 | * Pages are properly locked and writeback is complete. | |
551 | * Try to migrate the page. | |
552 | */ | |
553 | mapping = page_mapping(page); | |
554 | if (!mapping) | |
6c5240ae | 555 | rc = migrate_page(mapping, newpage, page); |
b20a3503 | 556 | |
6c5240ae | 557 | else if (mapping->a_ops->migratepage) |
b20a3503 CL |
558 | /* |
559 | * Most pages have a mapping and most filesystems | |
560 | * should provide a migration function. Anonymous | |
561 | * pages are part of swap space which also has its | |
562 | * own migration function. This is the most common | |
563 | * path for page migration. | |
564 | */ | |
2d1db3b1 CL |
565 | rc = mapping->a_ops->migratepage(mapping, |
566 | newpage, page); | |
8351a6e4 CL |
567 | else |
568 | rc = fallback_migrate_page(mapping, newpage, page); | |
b20a3503 | 569 | |
6c5240ae CL |
570 | if (!rc) |
571 | remove_migration_ptes(page, newpage); | |
572 | ||
b20a3503 CL |
573 | unlock_page(newpage); |
574 | ||
575 | unlock_page: | |
6c5240ae CL |
576 | if (rc) |
577 | remove_migration_ptes(page, page); | |
578 | ||
b20a3503 CL |
579 | unlock_page(page); |
580 | ||
581 | next: | |
2d1db3b1 CL |
582 | if (rc) { |
583 | if (newpage) | |
584 | newpage->mapping = NULL; | |
585 | ||
586 | if (rc == -EAGAIN) | |
587 | retry++; | |
588 | else { | |
589 | /* Permanent failure */ | |
590 | list_move(&page->lru, failed); | |
591 | nr_failed++; | |
592 | } | |
b20a3503 CL |
593 | } else { |
594 | if (newpage) { | |
595 | /* Successful migration. Return page to LRU */ | |
596 | move_to_lru(newpage); | |
597 | } | |
598 | list_move(&page->lru, moved); | |
599 | } | |
600 | } | |
601 | if (retry && pass++ < 10) | |
602 | goto redo; | |
603 | ||
604 | if (!swapwrite) | |
605 | current->flags &= ~PF_SWAPWRITE; | |
606 | ||
607 | return nr_failed + retry; | |
608 | } | |
609 | ||
b20a3503 CL |
610 | /* |
611 | * Migrate the list 'pagelist' of pages to a certain destination. | |
612 | * | |
613 | * Specify destination with either non-NULL vma or dest_node >= 0 | |
614 | * Return the number of pages not migrated or error code | |
615 | */ | |
616 | int migrate_pages_to(struct list_head *pagelist, | |
617 | struct vm_area_struct *vma, int dest) | |
618 | { | |
619 | LIST_HEAD(newlist); | |
620 | LIST_HEAD(moved); | |
621 | LIST_HEAD(failed); | |
622 | int err = 0; | |
623 | unsigned long offset = 0; | |
624 | int nr_pages; | |
625 | struct page *page; | |
626 | struct list_head *p; | |
627 | ||
628 | redo: | |
629 | nr_pages = 0; | |
630 | list_for_each(p, pagelist) { | |
631 | if (vma) { | |
632 | /* | |
633 | * The address passed to alloc_page_vma is used to | |
634 | * generate the proper interleave behavior. We fake | |
635 | * the address here by an increasing offset in order | |
636 | * to get the proper distribution of pages. | |
637 | * | |
638 | * No decision has been made as to which page | |
639 | * a certain old page is moved to so we cannot | |
640 | * specify the correct address. | |
641 | */ | |
642 | page = alloc_page_vma(GFP_HIGHUSER, vma, | |
643 | offset + vma->vm_start); | |
644 | offset += PAGE_SIZE; | |
645 | } | |
646 | else | |
647 | page = alloc_pages_node(dest, GFP_HIGHUSER, 0); | |
648 | ||
649 | if (!page) { | |
650 | err = -ENOMEM; | |
651 | goto out; | |
652 | } | |
653 | list_add_tail(&page->lru, &newlist); | |
654 | nr_pages++; | |
655 | if (nr_pages > MIGRATE_CHUNK_SIZE) | |
656 | break; | |
657 | } | |
658 | err = migrate_pages(pagelist, &newlist, &moved, &failed); | |
659 | ||
660 | putback_lru_pages(&moved); /* Call release pages instead ?? */ | |
661 | ||
662 | if (err >= 0 && list_empty(&newlist) && !list_empty(pagelist)) | |
663 | goto redo; | |
664 | out: | |
665 | /* Return leftover allocated pages */ | |
666 | while (!list_empty(&newlist)) { | |
667 | page = list_entry(newlist.next, struct page, lru); | |
668 | list_del(&page->lru); | |
669 | __free_page(page); | |
670 | } | |
671 | list_splice(&failed, pagelist); | |
672 | if (err < 0) | |
673 | return err; | |
674 | ||
675 | /* Calculate number of leftover pages */ | |
676 | nr_pages = 0; | |
677 | list_for_each(p, pagelist) | |
678 | nr_pages++; | |
679 | return nr_pages; | |
680 | } |