diff options
Diffstat (limited to 'mm/compaction.c')
-rw-r--r-- | mm/compaction.c | 328 |
1 files changed, 157 insertions, 171 deletions
diff --git a/mm/compaction.c b/mm/compaction.c index 06b198fa9abe..7a92e418a187 100644 --- a/mm/compaction.c +++ b/mm/compaction.c | |||
@@ -16,30 +16,11 @@ | |||
16 | #include <linux/sysfs.h> | 16 | #include <linux/sysfs.h> |
17 | #include "internal.h" | 17 | #include "internal.h" |
18 | 18 | ||
19 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA | ||
20 | |||
19 | #define CREATE_TRACE_POINTS | 21 | #define CREATE_TRACE_POINTS |
20 | #include <trace/events/compaction.h> | 22 | #include <trace/events/compaction.h> |
21 | 23 | ||
22 | /* | ||
23 | * compact_control is used to track pages being migrated and the free pages | ||
24 | * they are being migrated to during memory compaction. The free_pfn starts | ||
25 | * at the end of a zone and migrate_pfn begins at the start. Movable pages | ||
26 | * are moved to the end of a zone during a compaction run and the run | ||
27 | * completes when free_pfn <= migrate_pfn | ||
28 | */ | ||
29 | struct compact_control { | ||
30 | struct list_head freepages; /* List of free pages to migrate to */ | ||
31 | struct list_head migratepages; /* List of pages being migrated */ | ||
32 | unsigned long nr_freepages; /* Number of isolated free pages */ | ||
33 | unsigned long nr_migratepages; /* Number of pages to migrate */ | ||
34 | unsigned long free_pfn; /* isolate_freepages search base */ | ||
35 | unsigned long migrate_pfn; /* isolate_migratepages search base */ | ||
36 | bool sync; /* Synchronous migration */ | ||
37 | |||
38 | int order; /* order a direct compactor needs */ | ||
39 | int migratetype; /* MOVABLE, RECLAIMABLE etc */ | ||
40 | struct zone *zone; | ||
41 | }; | ||
42 | |||
43 | static unsigned long release_freepages(struct list_head *freelist) | 24 | static unsigned long release_freepages(struct list_head *freelist) |
44 | { | 25 | { |
45 | struct page *page, *next; | 26 | struct page *page, *next; |
@@ -54,6 +35,16 @@ static unsigned long release_freepages(struct list_head *freelist) | |||
54 | return count; | 35 | return count; |
55 | } | 36 | } |
56 | 37 | ||
38 | static void map_pages(struct list_head *list) | ||
39 | { | ||
40 | struct page *page; | ||
41 | |||
42 | list_for_each_entry(page, list, lru) { | ||
43 | arch_alloc_page(page, 0); | ||
44 | kernel_map_pages(page, 1, 1); | ||
45 | } | ||
46 | } | ||
47 | |||
57 | /* | 48 | /* |
58 | * Isolate free pages onto a private freelist. Caller must hold zone->lock. | 49 | * Isolate free pages onto a private freelist. Caller must hold zone->lock. |
59 | * If @strict is true, will abort returning 0 on any invalid PFNs or non-free | 50 | * If @strict is true, will abort returning 0 on any invalid PFNs or non-free |
@@ -122,7 +113,7 @@ static unsigned long isolate_freepages_block(unsigned long blockpfn, | |||
122 | * (which may be greater then end_pfn if end fell in a middle of | 113 | * (which may be greater then end_pfn if end fell in a middle of |
123 | * a free page). | 114 | * a free page). |
124 | */ | 115 | */ |
125 | static unsigned long | 116 | unsigned long |
126 | isolate_freepages_range(unsigned long start_pfn, unsigned long end_pfn) | 117 | isolate_freepages_range(unsigned long start_pfn, unsigned long end_pfn) |
127 | { | 118 | { |
128 | unsigned long isolated, pfn, block_end_pfn, flags; | 119 | unsigned long isolated, pfn, block_end_pfn, flags; |
@@ -176,127 +167,6 @@ isolate_freepages_range(unsigned long start_pfn, unsigned long end_pfn) | |||
176 | return pfn; | 167 | return pfn; |
177 | } | 168 | } |
178 | 169 | ||
179 | /* Returns true if the page is within a block suitable for migration to */ | ||
180 | static bool suitable_migration_target(struct page *page) | ||
181 | { | ||
182 | |||
183 | int migratetype = get_pageblock_migratetype(page); | ||
184 | |||
185 | /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */ | ||
186 | if (migratetype == MIGRATE_ISOLATE || migratetype == MIGRATE_RESERVE) | ||
187 | return false; | ||
188 | |||
189 | /* If the page is a large free page, then allow migration */ | ||
190 | if (PageBuddy(page) && page_order(page) >= pageblock_order) | ||
191 | return true; | ||
192 | |||
193 | /* If the block is MIGRATE_MOVABLE, allow migration */ | ||
194 | if (migratetype == MIGRATE_MOVABLE) | ||
195 | return true; | ||
196 | |||
197 | /* Otherwise skip the block */ | ||
198 | return false; | ||
199 | } | ||
200 | |||
201 | static void map_pages(struct list_head *list) | ||
202 | { | ||
203 | struct page *page; | ||
204 | |||
205 | list_for_each_entry(page, list, lru) { | ||
206 | arch_alloc_page(page, 0); | ||
207 | kernel_map_pages(page, 1, 1); | ||
208 | } | ||
209 | } | ||
210 | |||
211 | /* | ||
212 | * Based on information in the current compact_control, find blocks | ||
213 | * suitable for isolating free pages from and then isolate them. | ||
214 | */ | ||
215 | static void isolate_freepages(struct zone *zone, | ||
216 | struct compact_control *cc) | ||
217 | { | ||
218 | struct page *page; | ||
219 | unsigned long high_pfn, low_pfn, pfn, zone_end_pfn, end_pfn; | ||
220 | unsigned long flags; | ||
221 | int nr_freepages = cc->nr_freepages; | ||
222 | struct list_head *freelist = &cc->freepages; | ||
223 | |||
224 | /* | ||
225 | * Initialise the free scanner. The starting point is where we last | ||
226 | * scanned from (or the end of the zone if starting). The low point | ||
227 | * is the end of the pageblock the migration scanner is using. | ||
228 | */ | ||
229 | pfn = cc->free_pfn; | ||
230 | low_pfn = cc->migrate_pfn + pageblock_nr_pages; | ||
231 | |||
232 | /* | ||
233 | * Take care that if the migration scanner is at the end of the zone | ||
234 | * that the free scanner does not accidentally move to the next zone | ||
235 | * in the next isolation cycle. | ||
236 | */ | ||
237 | high_pfn = min(low_pfn, pfn); | ||
238 | |||
239 | zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; | ||
240 | |||
241 | /* | ||
242 | * Isolate free pages until enough are available to migrate the | ||
243 | * pages on cc->migratepages. We stop searching if the migrate | ||
244 | * and free page scanners meet or enough free pages are isolated. | ||
245 | */ | ||
246 | for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages; | ||
247 | pfn -= pageblock_nr_pages) { | ||
248 | unsigned long isolated; | ||
249 | |||
250 | if (!pfn_valid(pfn)) | ||
251 | continue; | ||
252 | |||
253 | /* | ||
254 | * Check for overlapping nodes/zones. It's possible on some | ||
255 | * configurations to have a setup like | ||
256 | * node0 node1 node0 | ||
257 | * i.e. it's possible that all pages within a zones range of | ||
258 | * pages do not belong to a single zone. | ||
259 | */ | ||
260 | page = pfn_to_page(pfn); | ||
261 | if (page_zone(page) != zone) | ||
262 | continue; | ||
263 | |||
264 | /* Check the block is suitable for migration */ | ||
265 | if (!suitable_migration_target(page)) | ||
266 | continue; | ||
267 | |||
268 | /* | ||
269 | * Found a block suitable for isolating free pages from. Now | ||
270 | * we disabled interrupts, double check things are ok and | ||
271 | * isolate the pages. This is to minimise the time IRQs | ||
272 | * are disabled | ||
273 | */ | ||
274 | isolated = 0; | ||
275 | spin_lock_irqsave(&zone->lock, flags); | ||
276 | if (suitable_migration_target(page)) { | ||
277 | end_pfn = min(pfn + pageblock_nr_pages, zone_end_pfn); | ||
278 | isolated = isolate_freepages_block(pfn, end_pfn, | ||
279 | freelist, false); | ||
280 | nr_freepages += isolated; | ||
281 | } | ||
282 | spin_unlock_irqrestore(&zone->lock, flags); | ||
283 | |||
284 | /* | ||
285 | * Record the highest PFN we isolated pages from. When next | ||
286 | * looking for free pages, the search will restart here as | ||
287 | * page migration may have returned some pages to the allocator | ||
288 | */ | ||
289 | if (isolated) | ||
290 | high_pfn = max(high_pfn, pfn); | ||
291 | } | ||
292 | |||
293 | /* split_free_page does not map the pages */ | ||
294 | map_pages(freelist); | ||
295 | |||
296 | cc->free_pfn = high_pfn; | ||
297 | cc->nr_freepages = nr_freepages; | ||
298 | } | ||
299 | |||
300 | /* Update the number of anon and file isolated pages in the zone */ | 170 | /* Update the number of anon and file isolated pages in the zone */ |
301 | static void acct_isolated(struct zone *zone, struct compact_control *cc) | 171 | static void acct_isolated(struct zone *zone, struct compact_control *cc) |
302 | { | 172 | { |
@@ -325,13 +195,6 @@ static bool too_many_isolated(struct zone *zone) | |||
325 | return isolated > (inactive + active) / 2; | 195 | return isolated > (inactive + active) / 2; |
326 | } | 196 | } |
327 | 197 | ||
328 | /* possible outcome of isolate_migratepages */ | ||
329 | typedef enum { | ||
330 | ISOLATE_ABORT, /* Abort compaction now */ | ||
331 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | ||
332 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | ||
333 | } isolate_migrate_t; | ||
334 | |||
335 | /** | 198 | /** |
336 | * isolate_migratepages_range() - isolate all migrate-able pages in range. | 199 | * isolate_migratepages_range() - isolate all migrate-able pages in range. |
337 | * @zone: Zone pages are in. | 200 | * @zone: Zone pages are in. |
@@ -351,7 +214,7 @@ typedef enum { | |||
351 | * does not modify any cc's fields, in particular it does not modify | 214 | * does not modify any cc's fields, in particular it does not modify |
352 | * (or read for that matter) cc->migrate_pfn. | 215 | * (or read for that matter) cc->migrate_pfn. |
353 | */ | 216 | */ |
354 | static unsigned long | 217 | unsigned long |
355 | isolate_migratepages_range(struct zone *zone, struct compact_control *cc, | 218 | isolate_migratepages_range(struct zone *zone, struct compact_control *cc, |
356 | unsigned long low_pfn, unsigned long end_pfn) | 219 | unsigned long low_pfn, unsigned long end_pfn) |
357 | { | 220 | { |
@@ -487,35 +350,118 @@ isolate_migratepages_range(struct zone *zone, struct compact_control *cc, | |||
487 | return low_pfn; | 350 | return low_pfn; |
488 | } | 351 | } |
489 | 352 | ||
353 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ | ||
354 | #ifdef CONFIG_COMPACTION | ||
355 | |||
356 | /* Returns true if the page is within a block suitable for migration to */ | ||
357 | static bool suitable_migration_target(struct page *page) | ||
358 | { | ||
359 | |||
360 | int migratetype = get_pageblock_migratetype(page); | ||
361 | |||
362 | /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */ | ||
363 | if (migratetype == MIGRATE_ISOLATE || migratetype == MIGRATE_RESERVE) | ||
364 | return false; | ||
365 | |||
366 | /* If the page is a large free page, then allow migration */ | ||
367 | if (PageBuddy(page) && page_order(page) >= pageblock_order) | ||
368 | return true; | ||
369 | |||
370 | /* If the block is MIGRATE_MOVABLE, allow migration */ | ||
371 | if (migratetype == MIGRATE_MOVABLE) | ||
372 | return true; | ||
373 | |||
374 | /* Otherwise skip the block */ | ||
375 | return false; | ||
376 | } | ||
377 | |||
490 | /* | 378 | /* |
491 | * Isolate all pages that can be migrated from the block pointed to by | 379 | * Based on information in the current compact_control, find blocks |
492 | * the migrate scanner within compact_control. | 380 | * suitable for isolating free pages from and then isolate them. |
493 | */ | 381 | */ |
494 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | 382 | static void isolate_freepages(struct zone *zone, |
495 | struct compact_control *cc) | 383 | struct compact_control *cc) |
496 | { | 384 | { |
497 | unsigned long low_pfn, end_pfn; | 385 | struct page *page; |
386 | unsigned long high_pfn, low_pfn, pfn, zone_end_pfn, end_pfn; | ||
387 | unsigned long flags; | ||
388 | int nr_freepages = cc->nr_freepages; | ||
389 | struct list_head *freelist = &cc->freepages; | ||
498 | 390 | ||
499 | /* Do not scan outside zone boundaries */ | 391 | /* |
500 | low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn); | 392 | * Initialise the free scanner. The starting point is where we last |
393 | * scanned from (or the end of the zone if starting). The low point | ||
394 | * is the end of the pageblock the migration scanner is using. | ||
395 | */ | ||
396 | pfn = cc->free_pfn; | ||
397 | low_pfn = cc->migrate_pfn + pageblock_nr_pages; | ||
501 | 398 | ||
502 | /* Only scan within a pageblock boundary */ | 399 | /* |
503 | end_pfn = ALIGN(low_pfn + pageblock_nr_pages, pageblock_nr_pages); | 400 | * Take care that if the migration scanner is at the end of the zone |
401 | * that the free scanner does not accidentally move to the next zone | ||
402 | * in the next isolation cycle. | ||
403 | */ | ||
404 | high_pfn = min(low_pfn, pfn); | ||
504 | 405 | ||
505 | /* Do not cross the free scanner or scan within a memory hole */ | 406 | zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; |
506 | if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) { | ||
507 | cc->migrate_pfn = end_pfn; | ||
508 | return ISOLATE_NONE; | ||
509 | } | ||
510 | 407 | ||
511 | /* Perform the isolation */ | 408 | /* |
512 | low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn); | 409 | * Isolate free pages until enough are available to migrate the |
513 | if (!low_pfn) | 410 | * pages on cc->migratepages. We stop searching if the migrate |
514 | return ISOLATE_ABORT; | 411 | * and free page scanners meet or enough free pages are isolated. |
412 | */ | ||
413 | for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages; | ||
414 | pfn -= pageblock_nr_pages) { | ||
415 | unsigned long isolated; | ||
515 | 416 | ||
516 | cc->migrate_pfn = low_pfn; | 417 | if (!pfn_valid(pfn)) |
418 | continue; | ||
517 | 419 | ||
518 | return ISOLATE_SUCCESS; | 420 | /* |
421 | * Check for overlapping nodes/zones. It's possible on some | ||
422 | * configurations to have a setup like | ||
423 | * node0 node1 node0 | ||
424 | * i.e. it's possible that all pages within a zones range of | ||
425 | * pages do not belong to a single zone. | ||
426 | */ | ||
427 | page = pfn_to_page(pfn); | ||
428 | if (page_zone(page) != zone) | ||
429 | continue; | ||
430 | |||
431 | /* Check the block is suitable for migration */ | ||
432 | if (!suitable_migration_target(page)) | ||
433 | continue; | ||
434 | |||
435 | /* | ||
436 | * Found a block suitable for isolating free pages from. Now | ||
437 | * we disabled interrupts, double check things are ok and | ||
438 | * isolate the pages. This is to minimise the time IRQs | ||
439 | * are disabled | ||
440 | */ | ||
441 | isolated = 0; | ||
442 | spin_lock_irqsave(&zone->lock, flags); | ||
443 | if (suitable_migration_target(page)) { | ||
444 | end_pfn = min(pfn + pageblock_nr_pages, zone_end_pfn); | ||
445 | isolated = isolate_freepages_block(pfn, end_pfn, | ||
446 | freelist, false); | ||
447 | nr_freepages += isolated; | ||
448 | } | ||
449 | spin_unlock_irqrestore(&zone->lock, flags); | ||
450 | |||
451 | /* | ||
452 | * Record the highest PFN we isolated pages from. When next | ||
453 | * looking for free pages, the search will restart here as | ||
454 | * page migration may have returned some pages to the allocator | ||
455 | */ | ||
456 | if (isolated) | ||
457 | high_pfn = max(high_pfn, pfn); | ||
458 | } | ||
459 | |||
460 | /* split_free_page does not map the pages */ | ||
461 | map_pages(freelist); | ||
462 | |||
463 | cc->free_pfn = high_pfn; | ||
464 | cc->nr_freepages = nr_freepages; | ||
519 | } | 465 | } |
520 | 466 | ||
521 | /* | 467 | /* |
@@ -564,6 +510,44 @@ static void update_nr_listpages(struct compact_control *cc) | |||
564 | cc->nr_freepages = nr_freepages; | 510 | cc->nr_freepages = nr_freepages; |
565 | } | 511 | } |
566 | 512 | ||
513 | /* possible outcome of isolate_migratepages */ | ||
514 | typedef enum { | ||
515 | ISOLATE_ABORT, /* Abort compaction now */ | ||
516 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | ||
517 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | ||
518 | } isolate_migrate_t; | ||
519 | |||
520 | /* | ||
521 | * Isolate all pages that can be migrated from the block pointed to by | ||
522 | * the migrate scanner within compact_control. | ||
523 | */ | ||
524 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | ||
525 | struct compact_control *cc) | ||
526 | { | ||
527 | unsigned long low_pfn, end_pfn; | ||
528 | |||
529 | /* Do not scan outside zone boundaries */ | ||
530 | low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn); | ||
531 | |||
532 | /* Only scan within a pageblock boundary */ | ||
533 | end_pfn = ALIGN(low_pfn + pageblock_nr_pages, pageblock_nr_pages); | ||
534 | |||
535 | /* Do not cross the free scanner or scan within a memory hole */ | ||
536 | if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) { | ||
537 | cc->migrate_pfn = end_pfn; | ||
538 | return ISOLATE_NONE; | ||
539 | } | ||
540 | |||
541 | /* Perform the isolation */ | ||
542 | low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn); | ||
543 | if (!low_pfn) | ||
544 | return ISOLATE_ABORT; | ||
545 | |||
546 | cc->migrate_pfn = low_pfn; | ||
547 | |||
548 | return ISOLATE_SUCCESS; | ||
549 | } | ||
550 | |||
567 | static int compact_finished(struct zone *zone, | 551 | static int compact_finished(struct zone *zone, |
568 | struct compact_control *cc) | 552 | struct compact_control *cc) |
569 | { | 553 | { |
@@ -910,3 +894,5 @@ void compaction_unregister_node(struct node *node) | |||
910 | return device_remove_file(&node->dev, &dev_attr_compact); | 894 | return device_remove_file(&node->dev, &dev_attr_compact); |
911 | } | 895 | } |
912 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | 896 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ |
897 | |||
898 | #endif /* CONFIG_COMPACTION */ | ||