blob: 8be430b812def9f32058ffb8c7e259ff1375d98e [file] [log] [blame]
Mel Gorman748446b2010-05-24 14:32:27 -07001/*
2 * linux/mm/compaction.c
3 *
4 * Memory compaction for the reduction of external fragmentation. Note that
5 * this heavily depends upon page migration to do all the real heavy
6 * lifting
7 *
8 * Copyright IBM Corp. 2007-2010 Mel Gorman <mel@csn.ul.ie>
9 */
10#include <linux/swap.h>
11#include <linux/migrate.h>
12#include <linux/compaction.h>
13#include <linux/mm_inline.h>
14#include <linux/backing-dev.h>
Mel Gorman76ab0f52010-05-24 14:32:28 -070015#include <linux/sysctl.h>
Mel Gormaned4a6d72010-05-24 14:32:29 -070016#include <linux/sysfs.h>
Mel Gorman748446b2010-05-24 14:32:27 -070017#include "internal.h"
18
Mel Gormanb7aba692011-01-13 15:45:54 -080019#define CREATE_TRACE_POINTS
20#include <trace/events/compaction.h>
21
Mel Gorman748446b2010-05-24 14:32:27 -070022/*
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 */
29struct 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 */
Mel Gorman77f1fe62011-01-13 15:45:57 -080036 bool sync; /* Synchronous migration */
Mel Gorman748446b2010-05-24 14:32:27 -070037
38 /* Account for isolated anon and file pages */
39 unsigned long nr_anon;
40 unsigned long nr_file;
41
Mel Gorman56de7262010-05-24 14:32:30 -070042 unsigned int order; /* order a direct compactor needs */
43 int migratetype; /* MOVABLE, RECLAIMABLE etc */
Mel Gorman748446b2010-05-24 14:32:27 -070044 struct zone *zone;
Andrea Arcangeli5a03b052011-01-13 15:47:11 -080045
46 int compact_mode;
Mel Gorman748446b2010-05-24 14:32:27 -070047};
48
49static unsigned long release_freepages(struct list_head *freelist)
50{
51 struct page *page, *next;
52 unsigned long count = 0;
53
54 list_for_each_entry_safe(page, next, freelist, lru) {
55 list_del(&page->lru);
56 __free_page(page);
57 count++;
58 }
59
60 return count;
61}
62
63/* Isolate free pages onto a private freelist. Must hold zone->lock */
64static unsigned long isolate_freepages_block(struct zone *zone,
65 unsigned long blockpfn,
66 struct list_head *freelist)
67{
68 unsigned long zone_end_pfn, end_pfn;
Mel Gormanb7aba692011-01-13 15:45:54 -080069 int nr_scanned = 0, total_isolated = 0;
Mel Gorman748446b2010-05-24 14:32:27 -070070 struct page *cursor;
71
72 /* Get the last PFN we should scan for free pages at */
73 zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
74 end_pfn = min(blockpfn + pageblock_nr_pages, zone_end_pfn);
75
76 /* Find the first usable PFN in the block to initialse page cursor */
77 for (; blockpfn < end_pfn; blockpfn++) {
78 if (pfn_valid_within(blockpfn))
79 break;
80 }
81 cursor = pfn_to_page(blockpfn);
82
83 /* Isolate free pages. This assumes the block is valid */
84 for (; blockpfn < end_pfn; blockpfn++, cursor++) {
85 int isolated, i;
86 struct page *page = cursor;
87
88 if (!pfn_valid_within(blockpfn))
89 continue;
Mel Gormanb7aba692011-01-13 15:45:54 -080090 nr_scanned++;
Mel Gorman748446b2010-05-24 14:32:27 -070091
92 if (!PageBuddy(page))
93 continue;
94
95 /* Found a free page, break it into order-0 pages */
96 isolated = split_free_page(page);
97 total_isolated += isolated;
98 for (i = 0; i < isolated; i++) {
99 list_add(&page->lru, freelist);
100 page++;
101 }
102
103 /* If a page was split, advance to the end of it */
104 if (isolated) {
105 blockpfn += isolated - 1;
106 cursor += isolated - 1;
107 }
108 }
109
Mel Gormanb7aba692011-01-13 15:45:54 -0800110 trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated);
Mel Gorman748446b2010-05-24 14:32:27 -0700111 return total_isolated;
112}
113
114/* Returns true if the page is within a block suitable for migration to */
115static bool suitable_migration_target(struct page *page)
116{
117
118 int migratetype = get_pageblock_migratetype(page);
119
120 /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */
121 if (migratetype == MIGRATE_ISOLATE || migratetype == MIGRATE_RESERVE)
122 return false;
123
124 /* If the page is a large free page, then allow migration */
125 if (PageBuddy(page) && page_order(page) >= pageblock_order)
126 return true;
127
128 /* If the block is MIGRATE_MOVABLE, allow migration */
129 if (migratetype == MIGRATE_MOVABLE)
130 return true;
131
132 /* Otherwise skip the block */
133 return false;
134}
135
136/*
137 * Based on information in the current compact_control, find blocks
138 * suitable for isolating free pages from and then isolate them.
139 */
140static void isolate_freepages(struct zone *zone,
141 struct compact_control *cc)
142{
143 struct page *page;
144 unsigned long high_pfn, low_pfn, pfn;
145 unsigned long flags;
146 int nr_freepages = cc->nr_freepages;
147 struct list_head *freelist = &cc->freepages;
148
149 pfn = cc->free_pfn;
150 low_pfn = cc->migrate_pfn + pageblock_nr_pages;
151 high_pfn = low_pfn;
152
153 /*
154 * Isolate free pages until enough are available to migrate the
155 * pages on cc->migratepages. We stop searching if the migrate
156 * and free page scanners meet or enough free pages are isolated.
157 */
158 spin_lock_irqsave(&zone->lock, flags);
159 for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages;
160 pfn -= pageblock_nr_pages) {
161 unsigned long isolated;
162
163 if (!pfn_valid(pfn))
164 continue;
165
166 /*
167 * Check for overlapping nodes/zones. It's possible on some
168 * configurations to have a setup like
169 * node0 node1 node0
170 * i.e. it's possible that all pages within a zones range of
171 * pages do not belong to a single zone.
172 */
173 page = pfn_to_page(pfn);
174 if (page_zone(page) != zone)
175 continue;
176
177 /* Check the block is suitable for migration */
178 if (!suitable_migration_target(page))
179 continue;
180
181 /* Found a block suitable for isolating free pages from */
182 isolated = isolate_freepages_block(zone, pfn, freelist);
183 nr_freepages += isolated;
184
185 /*
186 * Record the highest PFN we isolated pages from. When next
187 * looking for free pages, the search will restart here as
188 * page migration may have returned some pages to the allocator
189 */
190 if (isolated)
191 high_pfn = max(high_pfn, pfn);
192 }
193 spin_unlock_irqrestore(&zone->lock, flags);
194
195 /* split_free_page does not map the pages */
196 list_for_each_entry(page, freelist, lru) {
197 arch_alloc_page(page, 0);
198 kernel_map_pages(page, 1, 1);
199 }
200
201 cc->free_pfn = high_pfn;
202 cc->nr_freepages = nr_freepages;
203}
204
205/* Update the number of anon and file isolated pages in the zone */
206static void acct_isolated(struct zone *zone, struct compact_control *cc)
207{
208 struct page *page;
209 unsigned int count[NR_LRU_LISTS] = { 0, };
210
211 list_for_each_entry(page, &cc->migratepages, lru) {
212 int lru = page_lru_base_type(page);
213 count[lru]++;
214 }
215
216 cc->nr_anon = count[LRU_ACTIVE_ANON] + count[LRU_INACTIVE_ANON];
217 cc->nr_file = count[LRU_ACTIVE_FILE] + count[LRU_INACTIVE_FILE];
218 __mod_zone_page_state(zone, NR_ISOLATED_ANON, cc->nr_anon);
219 __mod_zone_page_state(zone, NR_ISOLATED_FILE, cc->nr_file);
220}
221
222/* Similar to reclaim, but different enough that they don't share logic */
223static bool too_many_isolated(struct zone *zone)
224{
Minchan Kimbc693042010-09-09 16:38:00 -0700225 unsigned long active, inactive, isolated;
Mel Gorman748446b2010-05-24 14:32:27 -0700226
227 inactive = zone_page_state(zone, NR_INACTIVE_FILE) +
228 zone_page_state(zone, NR_INACTIVE_ANON);
Minchan Kimbc693042010-09-09 16:38:00 -0700229 active = zone_page_state(zone, NR_ACTIVE_FILE) +
230 zone_page_state(zone, NR_ACTIVE_ANON);
Mel Gorman748446b2010-05-24 14:32:27 -0700231 isolated = zone_page_state(zone, NR_ISOLATED_FILE) +
232 zone_page_state(zone, NR_ISOLATED_ANON);
233
Minchan Kimbc693042010-09-09 16:38:00 -0700234 return isolated > (inactive + active) / 2;
Mel Gorman748446b2010-05-24 14:32:27 -0700235}
236
237/*
238 * Isolate all pages that can be migrated from the block pointed to by
239 * the migrate scanner within compact_control.
240 */
241static unsigned long isolate_migratepages(struct zone *zone,
242 struct compact_control *cc)
243{
244 unsigned long low_pfn, end_pfn;
Mel Gorman9927af742011-01-13 15:45:59 -0800245 unsigned long last_pageblock_nr = 0, pageblock_nr;
Mel Gormanb7aba692011-01-13 15:45:54 -0800246 unsigned long nr_scanned = 0, nr_isolated = 0;
Mel Gorman748446b2010-05-24 14:32:27 -0700247 struct list_head *migratelist = &cc->migratepages;
248
249 /* Do not scan outside zone boundaries */
250 low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn);
251
252 /* Only scan within a pageblock boundary */
253 end_pfn = ALIGN(low_pfn + pageblock_nr_pages, pageblock_nr_pages);
254
255 /* Do not cross the free scanner or scan within a memory hole */
256 if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) {
257 cc->migrate_pfn = end_pfn;
258 return 0;
259 }
260
261 /*
262 * Ensure that there are not too many pages isolated from the LRU
263 * list by either parallel reclaimers or compaction. If there are,
264 * delay for some time until fewer pages are isolated
265 */
266 while (unlikely(too_many_isolated(zone))) {
267 congestion_wait(BLK_RW_ASYNC, HZ/10);
268
269 if (fatal_signal_pending(current))
270 return 0;
271 }
272
273 /* Time to isolate some pages for migration */
274 spin_lock_irq(&zone->lru_lock);
275 for (; low_pfn < end_pfn; low_pfn++) {
276 struct page *page;
277 if (!pfn_valid_within(low_pfn))
278 continue;
Mel Gormanb7aba692011-01-13 15:45:54 -0800279 nr_scanned++;
Mel Gorman748446b2010-05-24 14:32:27 -0700280
281 /* Get the page and skip if free */
282 page = pfn_to_page(low_pfn);
283 if (PageBuddy(page))
284 continue;
285
Mel Gorman9927af742011-01-13 15:45:59 -0800286 /*
287 * For async migration, also only scan in MOVABLE blocks. Async
288 * migration is optimistic to see if the minimum amount of work
289 * satisfies the allocation
290 */
291 pageblock_nr = low_pfn >> pageblock_order;
292 if (!cc->sync && last_pageblock_nr != pageblock_nr &&
293 get_pageblock_migratetype(page) != MIGRATE_MOVABLE) {
294 low_pfn += pageblock_nr_pages;
295 low_pfn = ALIGN(low_pfn, pageblock_nr_pages) - 1;
296 last_pageblock_nr = pageblock_nr;
297 continue;
298 }
299
Andrea Arcangelibc835012011-01-13 15:47:08 -0800300 if (!PageLRU(page))
301 continue;
302
303 /*
304 * PageLRU is set, and lru_lock excludes isolation,
305 * splitting and collapsing (collapsing has already
306 * happened if PageLRU is set).
307 */
308 if (PageTransHuge(page)) {
309 low_pfn += (1 << compound_order(page)) - 1;
310 continue;
311 }
312
Mel Gorman748446b2010-05-24 14:32:27 -0700313 /* Try isolate the page */
314 if (__isolate_lru_page(page, ISOLATE_BOTH, 0) != 0)
315 continue;
316
Andrea Arcangelibc835012011-01-13 15:47:08 -0800317 VM_BUG_ON(PageTransCompound(page));
318
Mel Gorman748446b2010-05-24 14:32:27 -0700319 /* Successfully isolated */
320 del_page_from_lru_list(zone, page, page_lru(page));
321 list_add(&page->lru, migratelist);
Mel Gorman748446b2010-05-24 14:32:27 -0700322 cc->nr_migratepages++;
Mel Gormanb7aba692011-01-13 15:45:54 -0800323 nr_isolated++;
Mel Gorman748446b2010-05-24 14:32:27 -0700324
325 /* Avoid isolating too much */
326 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX)
327 break;
328 }
329
330 acct_isolated(zone, cc);
331
332 spin_unlock_irq(&zone->lru_lock);
333 cc->migrate_pfn = low_pfn;
334
Mel Gormanb7aba692011-01-13 15:45:54 -0800335 trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated);
336
Mel Gorman748446b2010-05-24 14:32:27 -0700337 return cc->nr_migratepages;
338}
339
340/*
341 * This is a migrate-callback that "allocates" freepages by taking pages
342 * from the isolated freelists in the block we are migrating to.
343 */
344static struct page *compaction_alloc(struct page *migratepage,
345 unsigned long data,
346 int **result)
347{
348 struct compact_control *cc = (struct compact_control *)data;
349 struct page *freepage;
350
351 /* Isolate free pages if necessary */
352 if (list_empty(&cc->freepages)) {
353 isolate_freepages(cc->zone, cc);
354
355 if (list_empty(&cc->freepages))
356 return NULL;
357 }
358
359 freepage = list_entry(cc->freepages.next, struct page, lru);
360 list_del(&freepage->lru);
361 cc->nr_freepages--;
362
363 return freepage;
364}
365
366/*
367 * We cannot control nr_migratepages and nr_freepages fully when migration is
368 * running as migrate_pages() has no knowledge of compact_control. When
369 * migration is complete, we count the number of pages on the lists by hand.
370 */
371static void update_nr_listpages(struct compact_control *cc)
372{
373 int nr_migratepages = 0;
374 int nr_freepages = 0;
375 struct page *page;
376
377 list_for_each_entry(page, &cc->migratepages, lru)
378 nr_migratepages++;
379 list_for_each_entry(page, &cc->freepages, lru)
380 nr_freepages++;
381
382 cc->nr_migratepages = nr_migratepages;
383 cc->nr_freepages = nr_freepages;
384}
385
386static int compact_finished(struct zone *zone,
Andrea Arcangeli5a03b052011-01-13 15:47:11 -0800387 struct compact_control *cc)
Mel Gorman748446b2010-05-24 14:32:27 -0700388{
Mel Gorman56de7262010-05-24 14:32:30 -0700389 unsigned int order;
Andrea Arcangeli5a03b052011-01-13 15:47:11 -0800390 unsigned long watermark;
Mel Gorman56de7262010-05-24 14:32:30 -0700391
Mel Gorman748446b2010-05-24 14:32:27 -0700392 if (fatal_signal_pending(current))
393 return COMPACT_PARTIAL;
394
395 /* Compaction run completes if the migrate and free scanner meet */
396 if (cc->free_pfn <= cc->migrate_pfn)
397 return COMPACT_COMPLETE;
398
Mel Gorman56de7262010-05-24 14:32:30 -0700399 /* Compaction run is not finished if the watermark is not met */
Andrea Arcangeli5a03b052011-01-13 15:47:11 -0800400 if (cc->compact_mode != COMPACT_MODE_KSWAPD)
401 watermark = low_wmark_pages(zone);
402 else
403 watermark = high_wmark_pages(zone);
404 watermark += (1 << cc->order);
405
Mel Gorman56de7262010-05-24 14:32:30 -0700406 if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0))
407 return COMPACT_CONTINUE;
408
Johannes Weiner82478fb2011-01-20 14:44:21 -0800409 /*
410 * order == -1 is expected when compacting via
411 * /proc/sys/vm/compact_memory
412 */
Mel Gorman56de7262010-05-24 14:32:30 -0700413 if (cc->order == -1)
414 return COMPACT_CONTINUE;
415
Andrea Arcangeli5a03b052011-01-13 15:47:11 -0800416 /*
417 * Generating only one page of the right order is not enough
418 * for kswapd, we must continue until we're above the high
419 * watermark as a pool for high order GFP_ATOMIC allocations
420 * too.
421 */
422 if (cc->compact_mode == COMPACT_MODE_KSWAPD)
423 return COMPACT_CONTINUE;
424
Mel Gorman56de7262010-05-24 14:32:30 -0700425 /* Direct compactor: Is a suitable page free? */
426 for (order = cc->order; order < MAX_ORDER; order++) {
427 /* Job done if page is free of the right migratetype */
428 if (!list_empty(&zone->free_area[order].free_list[cc->migratetype]))
429 return COMPACT_PARTIAL;
430
431 /* Job done if allocation would set block type */
432 if (order >= pageblock_order && zone->free_area[order].nr_free)
433 return COMPACT_PARTIAL;
434 }
435
Mel Gorman748446b2010-05-24 14:32:27 -0700436 return COMPACT_CONTINUE;
437}
438
Mel Gorman3e7d3442011-01-13 15:45:56 -0800439/*
440 * compaction_suitable: Is this suitable to run compaction on this zone now?
441 * Returns
442 * COMPACT_SKIPPED - If there are too few free pages for compaction
443 * COMPACT_PARTIAL - If the allocation would succeed without compaction
444 * COMPACT_CONTINUE - If compaction should run now
445 */
446unsigned long compaction_suitable(struct zone *zone, int order)
447{
448 int fragindex;
449 unsigned long watermark;
450
451 /*
452 * Watermarks for order-0 must be met for compaction. Note the 2UL.
453 * This is because during migration, copies of pages need to be
454 * allocated and for a short time, the footprint is higher
455 */
456 watermark = low_wmark_pages(zone) + (2UL << order);
457 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
458 return COMPACT_SKIPPED;
459
460 /*
Johannes Weiner82478fb2011-01-20 14:44:21 -0800461 * order == -1 is expected when compacting via
462 * /proc/sys/vm/compact_memory
463 */
464 if (order == -1)
465 return COMPACT_CONTINUE;
466
467 /*
Mel Gorman3e7d3442011-01-13 15:45:56 -0800468 * fragmentation index determines if allocation failures are due to
469 * low memory or external fragmentation
470 *
471 * index of -1 implies allocations might succeed dependingon watermarks
472 * index towards 0 implies failure is due to lack of memory
473 * index towards 1000 implies failure is due to fragmentation
474 *
475 * Only compact if a failure would be due to fragmentation.
476 */
477 fragindex = fragmentation_index(zone, order);
478 if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
479 return COMPACT_SKIPPED;
480
481 if (fragindex == -1 && zone_watermark_ok(zone, order, watermark, 0, 0))
482 return COMPACT_PARTIAL;
483
484 return COMPACT_CONTINUE;
485}
486
Mel Gorman748446b2010-05-24 14:32:27 -0700487static int compact_zone(struct zone *zone, struct compact_control *cc)
488{
489 int ret;
490
Mel Gorman3e7d3442011-01-13 15:45:56 -0800491 ret = compaction_suitable(zone, cc->order);
492 switch (ret) {
493 case COMPACT_PARTIAL:
494 case COMPACT_SKIPPED:
495 /* Compaction is likely to fail */
496 return ret;
497 case COMPACT_CONTINUE:
498 /* Fall through to compaction */
499 ;
500 }
501
Mel Gorman748446b2010-05-24 14:32:27 -0700502 /* Setup to move all movable pages to the end of the zone */
503 cc->migrate_pfn = zone->zone_start_pfn;
504 cc->free_pfn = cc->migrate_pfn + zone->spanned_pages;
505 cc->free_pfn &= ~(pageblock_nr_pages-1);
506
507 migrate_prep_local();
508
509 while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) {
510 unsigned long nr_migrate, nr_remaining;
511
512 if (!isolate_migratepages(zone, cc))
513 continue;
514
515 nr_migrate = cc->nr_migratepages;
516 migrate_pages(&cc->migratepages, compaction_alloc,
Mel Gorman7f0f2492011-01-13 15:45:58 -0800517 (unsigned long)cc, false,
Mel Gorman77f1fe62011-01-13 15:45:57 -0800518 cc->sync);
Mel Gorman748446b2010-05-24 14:32:27 -0700519 update_nr_listpages(cc);
520 nr_remaining = cc->nr_migratepages;
521
522 count_vm_event(COMPACTBLOCKS);
523 count_vm_events(COMPACTPAGES, nr_migrate - nr_remaining);
524 if (nr_remaining)
525 count_vm_events(COMPACTPAGEFAILED, nr_remaining);
Mel Gormanb7aba692011-01-13 15:45:54 -0800526 trace_mm_compaction_migratepages(nr_migrate - nr_remaining,
527 nr_remaining);
Mel Gorman748446b2010-05-24 14:32:27 -0700528
529 /* Release LRU pages not migrated */
530 if (!list_empty(&cc->migratepages)) {
531 putback_lru_pages(&cc->migratepages);
532 cc->nr_migratepages = 0;
533 }
534
535 }
536
537 /* Release free pages and check accounting */
538 cc->nr_freepages -= release_freepages(&cc->freepages);
539 VM_BUG_ON(cc->nr_freepages != 0);
540
541 return ret;
542}
Mel Gorman76ab0f52010-05-24 14:32:28 -0700543
Mel Gorman3e7d3442011-01-13 15:45:56 -0800544unsigned long compact_zone_order(struct zone *zone,
Andrea Arcangeli5a03b052011-01-13 15:47:11 -0800545 int order, gfp_t gfp_mask,
546 bool sync,
547 int compact_mode)
Mel Gorman56de7262010-05-24 14:32:30 -0700548{
549 struct compact_control cc = {
550 .nr_freepages = 0,
551 .nr_migratepages = 0,
552 .order = order,
553 .migratetype = allocflags_to_migratetype(gfp_mask),
554 .zone = zone,
Mel Gorman77f1fe62011-01-13 15:45:57 -0800555 .sync = sync,
Andrea Arcangeli5a03b052011-01-13 15:47:11 -0800556 .compact_mode = compact_mode,
Mel Gorman56de7262010-05-24 14:32:30 -0700557 };
558 INIT_LIST_HEAD(&cc.freepages);
559 INIT_LIST_HEAD(&cc.migratepages);
560
561 return compact_zone(zone, &cc);
562}
563
Mel Gorman5e771902010-05-24 14:32:31 -0700564int sysctl_extfrag_threshold = 500;
565
Mel Gorman56de7262010-05-24 14:32:30 -0700566/**
567 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
568 * @zonelist: The zonelist used for the current allocation
569 * @order: The order of the current allocation
570 * @gfp_mask: The GFP mask of the current allocation
571 * @nodemask: The allowed nodes to allocate from
Mel Gorman77f1fe62011-01-13 15:45:57 -0800572 * @sync: Whether migration is synchronous or not
Mel Gorman56de7262010-05-24 14:32:30 -0700573 *
574 * This is the main entry point for direct page compaction.
575 */
576unsigned long try_to_compact_pages(struct zonelist *zonelist,
Mel Gorman77f1fe62011-01-13 15:45:57 -0800577 int order, gfp_t gfp_mask, nodemask_t *nodemask,
578 bool sync)
Mel Gorman56de7262010-05-24 14:32:30 -0700579{
580 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
581 int may_enter_fs = gfp_mask & __GFP_FS;
582 int may_perform_io = gfp_mask & __GFP_IO;
Mel Gorman56de7262010-05-24 14:32:30 -0700583 struct zoneref *z;
584 struct zone *zone;
585 int rc = COMPACT_SKIPPED;
586
587 /*
588 * Check whether it is worth even starting compaction. The order check is
589 * made because an assumption is made that the page allocator can satisfy
590 * the "cheaper" orders without taking special steps
591 */
Andrea Arcangelic5a73c32011-01-13 15:47:11 -0800592 if (!order || !may_enter_fs || !may_perform_io)
Mel Gorman56de7262010-05-24 14:32:30 -0700593 return rc;
594
595 count_vm_event(COMPACTSTALL);
596
597 /* Compact each zone in the list */
598 for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx,
599 nodemask) {
Mel Gorman56de7262010-05-24 14:32:30 -0700600 int status;
601
Andrea Arcangeli5a03b052011-01-13 15:47:11 -0800602 status = compact_zone_order(zone, order, gfp_mask, sync,
603 COMPACT_MODE_DIRECT_RECLAIM);
Mel Gorman56de7262010-05-24 14:32:30 -0700604 rc = max(status, rc);
605
Mel Gorman3e7d3442011-01-13 15:45:56 -0800606 /* If a normal allocation would succeed, stop compacting */
607 if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, 0))
Mel Gorman56de7262010-05-24 14:32:30 -0700608 break;
609 }
610
611 return rc;
612}
613
614
Mel Gorman76ab0f52010-05-24 14:32:28 -0700615/* Compact all zones within a node */
616static int compact_node(int nid)
617{
618 int zoneid;
619 pg_data_t *pgdat;
620 struct zone *zone;
621
622 if (nid < 0 || nid >= nr_node_ids || !node_online(nid))
623 return -EINVAL;
624 pgdat = NODE_DATA(nid);
625
626 /* Flush pending updates to the LRU lists */
627 lru_add_drain_all();
628
629 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
630 struct compact_control cc = {
631 .nr_freepages = 0,
632 .nr_migratepages = 0,
Mel Gorman56de7262010-05-24 14:32:30 -0700633 .order = -1,
Andrea Arcangeli5a03b052011-01-13 15:47:11 -0800634 .compact_mode = COMPACT_MODE_DIRECT_RECLAIM,
Mel Gorman76ab0f52010-05-24 14:32:28 -0700635 };
636
637 zone = &pgdat->node_zones[zoneid];
638 if (!populated_zone(zone))
639 continue;
640
641 cc.zone = zone;
642 INIT_LIST_HEAD(&cc.freepages);
643 INIT_LIST_HEAD(&cc.migratepages);
644
645 compact_zone(zone, &cc);
646
647 VM_BUG_ON(!list_empty(&cc.freepages));
648 VM_BUG_ON(!list_empty(&cc.migratepages));
649 }
650
651 return 0;
652}
653
654/* Compact all nodes in the system */
655static int compact_nodes(void)
656{
657 int nid;
658
659 for_each_online_node(nid)
660 compact_node(nid);
661
662 return COMPACT_COMPLETE;
663}
664
665/* The written value is actually unused, all memory is compacted */
666int sysctl_compact_memory;
667
668/* This is the entry point for compacting all nodes via /proc/sys/vm */
669int sysctl_compaction_handler(struct ctl_table *table, int write,
670 void __user *buffer, size_t *length, loff_t *ppos)
671{
672 if (write)
673 return compact_nodes();
674
675 return 0;
676}
Mel Gormaned4a6d72010-05-24 14:32:29 -0700677
Mel Gorman5e771902010-05-24 14:32:31 -0700678int sysctl_extfrag_handler(struct ctl_table *table, int write,
679 void __user *buffer, size_t *length, loff_t *ppos)
680{
681 proc_dointvec_minmax(table, write, buffer, length, ppos);
682
683 return 0;
684}
685
Mel Gormaned4a6d72010-05-24 14:32:29 -0700686#if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
687ssize_t sysfs_compact_node(struct sys_device *dev,
688 struct sysdev_attribute *attr,
689 const char *buf, size_t count)
690{
691 compact_node(dev->id);
692
693 return count;
694}
695static SYSDEV_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node);
696
697int compaction_register_node(struct node *node)
698{
699 return sysdev_create_file(&node->sysdev, &attr_compact);
700}
701
702void compaction_unregister_node(struct node *node)
703{
704 return sysdev_remove_file(&node->sysdev, &attr_compact);
705}
706#endif /* CONFIG_SYSFS && CONFIG_NUMA */