blob: ef53f352b606a00502a87605f6acca40a317ba46 [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>
Rafael Aquinibf6bddf2012-12-11 16:02:42 -080017#include <linux/balloon_compaction.h>
Minchan Kim194159f2013-02-22 16:33:58 -080018#include <linux/page-isolation.h>
Mel Gorman748446b2010-05-24 14:32:27 -070019#include "internal.h"
20
Minchan Kim010fc292012-12-20 15:05:06 -080021#ifdef CONFIG_COMPACTION
22static inline void count_compact_event(enum vm_event_item item)
23{
24 count_vm_event(item);
25}
26
27static inline void count_compact_events(enum vm_event_item item, long delta)
28{
29 count_vm_events(item, delta);
30}
31#else
32#define count_compact_event(item) do { } while (0)
33#define count_compact_events(item, delta) do { } while (0)
34#endif
35
Michal Nazarewiczff9543f2011-12-29 13:09:50 +010036#if defined CONFIG_COMPACTION || defined CONFIG_CMA
37
Mel Gormanb7aba692011-01-13 15:45:54 -080038#define CREATE_TRACE_POINTS
39#include <trace/events/compaction.h>
40
Mel Gorman748446b2010-05-24 14:32:27 -070041static unsigned long release_freepages(struct list_head *freelist)
42{
43 struct page *page, *next;
44 unsigned long count = 0;
45
46 list_for_each_entry_safe(page, next, freelist, lru) {
47 list_del(&page->lru);
48 __free_page(page);
49 count++;
50 }
51
52 return count;
53}
54
Michal Nazarewiczff9543f2011-12-29 13:09:50 +010055static void map_pages(struct list_head *list)
56{
57 struct page *page;
58
59 list_for_each_entry(page, list, lru) {
60 arch_alloc_page(page, 0);
61 kernel_map_pages(page, 1, 1);
62 }
63}
64
Michal Nazarewicz47118af2011-12-29 13:09:50 +010065static inline bool migrate_async_suitable(int migratetype)
66{
67 return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE;
68}
69
Mel Gormanbb13ffe2012-10-08 16:32:41 -070070#ifdef CONFIG_COMPACTION
71/* Returns true if the pageblock should be scanned for pages to isolate. */
72static inline bool isolation_suitable(struct compact_control *cc,
73 struct page *page)
74{
75 if (cc->ignore_skip_hint)
76 return true;
77
78 return !get_pageblock_skip(page);
79}
80
81/*
82 * This function is called to clear all cached information on pageblocks that
83 * should be skipped for page isolation when the migrate and free page scanner
84 * meet.
85 */
Mel Gorman62997022012-10-08 16:32:47 -070086static void __reset_isolation_suitable(struct zone *zone)
Mel Gormanbb13ffe2012-10-08 16:32:41 -070087{
88 unsigned long start_pfn = zone->zone_start_pfn;
89 unsigned long end_pfn = zone->zone_start_pfn + zone->spanned_pages;
90 unsigned long pfn;
91
Mel Gormanc89511a2012-10-08 16:32:45 -070092 zone->compact_cached_migrate_pfn = start_pfn;
93 zone->compact_cached_free_pfn = end_pfn;
Mel Gorman62997022012-10-08 16:32:47 -070094 zone->compact_blockskip_flush = false;
Mel Gormanbb13ffe2012-10-08 16:32:41 -070095
96 /* Walk the zone and mark every pageblock as suitable for isolation */
97 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
98 struct page *page;
99
100 cond_resched();
101
102 if (!pfn_valid(pfn))
103 continue;
104
105 page = pfn_to_page(pfn);
106 if (zone != page_zone(page))
107 continue;
108
109 clear_pageblock_skip(page);
110 }
111}
112
Mel Gorman62997022012-10-08 16:32:47 -0700113void reset_isolation_suitable(pg_data_t *pgdat)
114{
115 int zoneid;
116
117 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
118 struct zone *zone = &pgdat->node_zones[zoneid];
119 if (!populated_zone(zone))
120 continue;
121
122 /* Only flush if a full compaction finished recently */
123 if (zone->compact_blockskip_flush)
124 __reset_isolation_suitable(zone);
125 }
126}
127
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700128/*
129 * If no pages were isolated then mark this pageblock to be skipped in the
Mel Gorman62997022012-10-08 16:32:47 -0700130 * future. The information is later cleared by __reset_isolation_suitable().
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700131 */
Mel Gormanc89511a2012-10-08 16:32:45 -0700132static void update_pageblock_skip(struct compact_control *cc,
133 struct page *page, unsigned long nr_isolated,
134 bool migrate_scanner)
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700135{
Mel Gormanc89511a2012-10-08 16:32:45 -0700136 struct zone *zone = cc->zone;
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700137 if (!page)
138 return;
139
Mel Gormanc89511a2012-10-08 16:32:45 -0700140 if (!nr_isolated) {
141 unsigned long pfn = page_to_pfn(page);
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700142 set_pageblock_skip(page);
Mel Gormanc89511a2012-10-08 16:32:45 -0700143
144 /* Update where compaction should restart */
145 if (migrate_scanner) {
146 if (!cc->finished_update_migrate &&
147 pfn > zone->compact_cached_migrate_pfn)
148 zone->compact_cached_migrate_pfn = pfn;
149 } else {
150 if (!cc->finished_update_free &&
151 pfn < zone->compact_cached_free_pfn)
152 zone->compact_cached_free_pfn = pfn;
153 }
154 }
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700155}
156#else
157static inline bool isolation_suitable(struct compact_control *cc,
158 struct page *page)
159{
160 return true;
161}
162
Mel Gormanc89511a2012-10-08 16:32:45 -0700163static void update_pageblock_skip(struct compact_control *cc,
164 struct page *page, unsigned long nr_isolated,
165 bool migrate_scanner)
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700166{
167}
168#endif /* CONFIG_COMPACTION */
169
Mel Gorman2a1402a2012-10-08 16:32:33 -0700170static inline bool should_release_lock(spinlock_t *lock)
171{
172 return need_resched() || spin_is_contended(lock);
173}
174
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100175/*
Mel Gormanc67fe372012-08-21 16:16:17 -0700176 * Compaction requires the taking of some coarse locks that are potentially
177 * very heavily contended. Check if the process needs to be scheduled or
178 * if the lock is contended. For async compaction, back out in the event
179 * if contention is severe. For sync compaction, schedule.
180 *
181 * Returns true if the lock is held.
182 * Returns false if the lock is released and compaction should abort
183 */
184static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags,
185 bool locked, struct compact_control *cc)
186{
Mel Gorman2a1402a2012-10-08 16:32:33 -0700187 if (should_release_lock(lock)) {
Mel Gormanc67fe372012-08-21 16:16:17 -0700188 if (locked) {
189 spin_unlock_irqrestore(lock, *flags);
190 locked = false;
191 }
192
193 /* async aborts if taking too long or contended */
194 if (!cc->sync) {
Shaohua Lie64c5232012-10-08 16:32:27 -0700195 cc->contended = true;
Mel Gormanc67fe372012-08-21 16:16:17 -0700196 return false;
197 }
198
199 cond_resched();
Mel Gormanc67fe372012-08-21 16:16:17 -0700200 }
201
202 if (!locked)
203 spin_lock_irqsave(lock, *flags);
204 return true;
205}
206
207static inline bool compact_trylock_irqsave(spinlock_t *lock,
208 unsigned long *flags, struct compact_control *cc)
209{
210 return compact_checklock_irqsave(lock, flags, false, cc);
211}
212
Mel Gormanf40d1e42012-10-08 16:32:36 -0700213/* Returns true if the page is within a block suitable for migration to */
214static bool suitable_migration_target(struct page *page)
215{
216 int migratetype = get_pageblock_migratetype(page);
217
218 /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */
Minchan Kim194159f2013-02-22 16:33:58 -0800219 if (migratetype == MIGRATE_RESERVE)
220 return false;
221
222 if (is_migrate_isolate(migratetype))
Mel Gormanf40d1e42012-10-08 16:32:36 -0700223 return false;
224
225 /* If the page is a large free page, then allow migration */
226 if (PageBuddy(page) && page_order(page) >= pageblock_order)
227 return true;
228
229 /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
230 if (migrate_async_suitable(migratetype))
231 return true;
232
233 /* Otherwise skip the block */
234 return false;
235}
236
Mel Gormanc67fe372012-08-21 16:16:17 -0700237/*
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100238 * Isolate free pages onto a private freelist. Caller must hold zone->lock.
239 * If @strict is true, will abort returning 0 on any invalid PFNs or non-free
240 * pages inside of the pageblock (even though it may still end up isolating
241 * some pages).
242 */
Mel Gormanf40d1e42012-10-08 16:32:36 -0700243static unsigned long isolate_freepages_block(struct compact_control *cc,
244 unsigned long blockpfn,
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100245 unsigned long end_pfn,
246 struct list_head *freelist,
247 bool strict)
Mel Gorman748446b2010-05-24 14:32:27 -0700248{
Mel Gormanb7aba692011-01-13 15:45:54 -0800249 int nr_scanned = 0, total_isolated = 0;
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700250 struct page *cursor, *valid_page = NULL;
Mel Gormanf40d1e42012-10-08 16:32:36 -0700251 unsigned long nr_strict_required = end_pfn - blockpfn;
252 unsigned long flags;
253 bool locked = false;
Mel Gorman748446b2010-05-24 14:32:27 -0700254
Mel Gorman748446b2010-05-24 14:32:27 -0700255 cursor = pfn_to_page(blockpfn);
256
Mel Gormanf40d1e42012-10-08 16:32:36 -0700257 /* Isolate free pages. */
Mel Gorman748446b2010-05-24 14:32:27 -0700258 for (; blockpfn < end_pfn; blockpfn++, cursor++) {
259 int isolated, i;
260 struct page *page = cursor;
261
Mel Gormanb7aba692011-01-13 15:45:54 -0800262 nr_scanned++;
Mel Gormanf40d1e42012-10-08 16:32:36 -0700263 if (!pfn_valid_within(blockpfn))
Mel Gorman748446b2010-05-24 14:32:27 -0700264 continue;
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700265 if (!valid_page)
266 valid_page = page;
Mel Gormanf40d1e42012-10-08 16:32:36 -0700267 if (!PageBuddy(page))
268 continue;
269
270 /*
271 * The zone lock must be held to isolate freepages.
272 * Unfortunately this is a very coarse lock and can be
273 * heavily contended if there are parallel allocations
274 * or parallel compactions. For async compaction do not
275 * spin on the lock and we acquire the lock as late as
276 * possible.
277 */
278 locked = compact_checklock_irqsave(&cc->zone->lock, &flags,
279 locked, cc);
280 if (!locked)
281 break;
282
283 /* Recheck this is a suitable migration target under lock */
284 if (!strict && !suitable_migration_target(page))
285 break;
286
287 /* Recheck this is a buddy page under lock */
288 if (!PageBuddy(page))
289 continue;
Mel Gorman748446b2010-05-24 14:32:27 -0700290
291 /* Found a free page, break it into order-0 pages */
292 isolated = split_free_page(page);
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100293 if (!isolated && strict)
Mel Gormanf40d1e42012-10-08 16:32:36 -0700294 break;
Mel Gorman748446b2010-05-24 14:32:27 -0700295 total_isolated += isolated;
296 for (i = 0; i < isolated; i++) {
297 list_add(&page->lru, freelist);
298 page++;
299 }
300
301 /* If a page was split, advance to the end of it */
302 if (isolated) {
303 blockpfn += isolated - 1;
304 cursor += isolated - 1;
305 }
306 }
307
Mel Gormanb7aba692011-01-13 15:45:54 -0800308 trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated);
Mel Gormanf40d1e42012-10-08 16:32:36 -0700309
310 /*
311 * If strict isolation is requested by CMA then check that all the
312 * pages requested were isolated. If there were any failures, 0 is
313 * returned and CMA will fail.
314 */
Mel Gorman0db63d72012-10-19 13:56:57 -0700315 if (strict && nr_strict_required > total_isolated)
Mel Gormanf40d1e42012-10-08 16:32:36 -0700316 total_isolated = 0;
317
318 if (locked)
319 spin_unlock_irqrestore(&cc->zone->lock, flags);
320
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700321 /* Update the pageblock-skip if the whole pageblock was scanned */
322 if (blockpfn == end_pfn)
Mel Gormanc89511a2012-10-08 16:32:45 -0700323 update_pageblock_skip(cc, valid_page, total_isolated, false);
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700324
Minchan Kim010fc292012-12-20 15:05:06 -0800325 count_compact_events(COMPACTFREE_SCANNED, nr_scanned);
Mel Gorman397487d2012-10-19 12:00:10 +0100326 if (total_isolated)
Minchan Kim010fc292012-12-20 15:05:06 -0800327 count_compact_events(COMPACTISOLATED, total_isolated);
Mel Gorman748446b2010-05-24 14:32:27 -0700328 return total_isolated;
329}
330
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100331/**
332 * isolate_freepages_range() - isolate free pages.
333 * @start_pfn: The first PFN to start isolating.
334 * @end_pfn: The one-past-last PFN.
335 *
336 * Non-free pages, invalid PFNs, or zone boundaries within the
337 * [start_pfn, end_pfn) range are considered errors, cause function to
338 * undo its actions and return zero.
339 *
340 * Otherwise, function returns one-past-the-last PFN of isolated page
341 * (which may be greater then end_pfn if end fell in a middle of
342 * a free page).
343 */
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100344unsigned long
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700345isolate_freepages_range(struct compact_control *cc,
346 unsigned long start_pfn, unsigned long end_pfn)
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100347{
Mel Gormanf40d1e42012-10-08 16:32:36 -0700348 unsigned long isolated, pfn, block_end_pfn;
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100349 LIST_HEAD(freelist);
350
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100351 for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) {
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700352 if (!pfn_valid(pfn) || cc->zone != page_zone(pfn_to_page(pfn)))
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100353 break;
354
355 /*
356 * On subsequent iterations ALIGN() is actually not needed,
357 * but we keep it that we not to complicate the code.
358 */
359 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
360 block_end_pfn = min(block_end_pfn, end_pfn);
361
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700362 isolated = isolate_freepages_block(cc, pfn, block_end_pfn,
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100363 &freelist, true);
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100364
365 /*
366 * In strict mode, isolate_freepages_block() returns 0 if
367 * there are any holes in the block (ie. invalid PFNs or
368 * non-free pages).
369 */
370 if (!isolated)
371 break;
372
373 /*
374 * If we managed to isolate pages, it is always (1 << n) *
375 * pageblock_nr_pages for some non-negative n. (Max order
376 * page may span two pageblocks).
377 */
378 }
379
380 /* split_free_page does not map the pages */
381 map_pages(&freelist);
382
383 if (pfn < end_pfn) {
384 /* Loop terminated early, cleanup. */
385 release_freepages(&freelist);
386 return 0;
387 }
388
389 /* We don't use freelists for anything. */
390 return pfn;
391}
392
Mel Gorman748446b2010-05-24 14:32:27 -0700393/* Update the number of anon and file isolated pages in the zone */
Mel Gormanc67fe372012-08-21 16:16:17 -0700394static void acct_isolated(struct zone *zone, bool locked, struct compact_control *cc)
Mel Gorman748446b2010-05-24 14:32:27 -0700395{
396 struct page *page;
Minchan Kimb9e84ac2011-10-31 17:06:44 -0700397 unsigned int count[2] = { 0, };
Mel Gorman748446b2010-05-24 14:32:27 -0700398
Minchan Kimb9e84ac2011-10-31 17:06:44 -0700399 list_for_each_entry(page, &cc->migratepages, lru)
400 count[!!page_is_file_cache(page)]++;
Mel Gorman748446b2010-05-24 14:32:27 -0700401
Mel Gormanc67fe372012-08-21 16:16:17 -0700402 /* If locked we can use the interrupt unsafe versions */
403 if (locked) {
404 __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
405 __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
406 } else {
407 mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
408 mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
409 }
Mel Gorman748446b2010-05-24 14:32:27 -0700410}
411
412/* Similar to reclaim, but different enough that they don't share logic */
413static bool too_many_isolated(struct zone *zone)
414{
Minchan Kimbc693042010-09-09 16:38:00 -0700415 unsigned long active, inactive, isolated;
Mel Gorman748446b2010-05-24 14:32:27 -0700416
417 inactive = zone_page_state(zone, NR_INACTIVE_FILE) +
418 zone_page_state(zone, NR_INACTIVE_ANON);
Minchan Kimbc693042010-09-09 16:38:00 -0700419 active = zone_page_state(zone, NR_ACTIVE_FILE) +
420 zone_page_state(zone, NR_ACTIVE_ANON);
Mel Gorman748446b2010-05-24 14:32:27 -0700421 isolated = zone_page_state(zone, NR_ISOLATED_FILE) +
422 zone_page_state(zone, NR_ISOLATED_ANON);
423
Minchan Kimbc693042010-09-09 16:38:00 -0700424 return isolated > (inactive + active) / 2;
Mel Gorman748446b2010-05-24 14:32:27 -0700425}
426
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100427/**
428 * isolate_migratepages_range() - isolate all migrate-able pages in range.
429 * @zone: Zone pages are in.
430 * @cc: Compaction control structure.
431 * @low_pfn: The first PFN of the range.
432 * @end_pfn: The one-past-the-last PFN of the range.
Minchan Kime46a2872012-10-08 16:33:48 -0700433 * @unevictable: true if it allows to isolate unevictable pages
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100434 *
435 * Isolate all pages that can be migrated from the range specified by
436 * [low_pfn, end_pfn). Returns zero if there is a fatal signal
437 * pending), otherwise PFN of the first page that was not scanned
438 * (which may be both less, equal to or more then end_pfn).
439 *
440 * Assumes that cc->migratepages is empty and cc->nr_migratepages is
441 * zero.
442 *
443 * Apart from cc->migratepages and cc->nr_migratetypes this function
444 * does not modify any cc's fields, in particular it does not modify
445 * (or read for that matter) cc->migrate_pfn.
Mel Gorman748446b2010-05-24 14:32:27 -0700446 */
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100447unsigned long
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100448isolate_migratepages_range(struct zone *zone, struct compact_control *cc,
Minchan Kime46a2872012-10-08 16:33:48 -0700449 unsigned long low_pfn, unsigned long end_pfn, bool unevictable)
Mel Gorman748446b2010-05-24 14:32:27 -0700450{
Mel Gorman9927af742011-01-13 15:45:59 -0800451 unsigned long last_pageblock_nr = 0, pageblock_nr;
Mel Gormanb7aba692011-01-13 15:45:54 -0800452 unsigned long nr_scanned = 0, nr_isolated = 0;
Mel Gorman748446b2010-05-24 14:32:27 -0700453 struct list_head *migratelist = &cc->migratepages;
Konstantin Khlebnikovf3fd4a62012-05-29 15:06:54 -0700454 isolate_mode_t mode = 0;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700455 struct lruvec *lruvec;
Mel Gormanc67fe372012-08-21 16:16:17 -0700456 unsigned long flags;
Mel Gorman2a1402a2012-10-08 16:32:33 -0700457 bool locked = false;
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700458 struct page *page = NULL, *valid_page = NULL;
Mel Gorman748446b2010-05-24 14:32:27 -0700459
Mel Gorman748446b2010-05-24 14:32:27 -0700460 /*
461 * Ensure that there are not too many pages isolated from the LRU
462 * list by either parallel reclaimers or compaction. If there are,
463 * delay for some time until fewer pages are isolated
464 */
465 while (unlikely(too_many_isolated(zone))) {
Mel Gormanf9e35b32011-06-15 15:08:52 -0700466 /* async migration should just abort */
Linus Torvalds68e3e922012-06-03 20:05:57 -0700467 if (!cc->sync)
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100468 return 0;
Mel Gormanf9e35b32011-06-15 15:08:52 -0700469
Mel Gorman748446b2010-05-24 14:32:27 -0700470 congestion_wait(BLK_RW_ASYNC, HZ/10);
471
472 if (fatal_signal_pending(current))
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100473 return 0;
Mel Gorman748446b2010-05-24 14:32:27 -0700474 }
475
476 /* Time to isolate some pages for migration */
Andrea Arcangelib2eef8c2011-03-22 16:33:10 -0700477 cond_resched();
Mel Gorman748446b2010-05-24 14:32:27 -0700478 for (; low_pfn < end_pfn; low_pfn++) {
Andrea Arcangelib2eef8c2011-03-22 16:33:10 -0700479 /* give a chance to irqs before checking need_resched() */
Mel Gorman2a1402a2012-10-08 16:32:33 -0700480 if (locked && !((low_pfn+1) % SWAP_CLUSTER_MAX)) {
481 if (should_release_lock(&zone->lru_lock)) {
482 spin_unlock_irqrestore(&zone->lru_lock, flags);
483 locked = false;
484 }
Andrea Arcangelib2eef8c2011-03-22 16:33:10 -0700485 }
Mel Gormanc67fe372012-08-21 16:16:17 -0700486
Mel Gorman0bf380b2012-02-03 15:37:18 -0800487 /*
488 * migrate_pfn does not necessarily start aligned to a
489 * pageblock. Ensure that pfn_valid is called when moving
490 * into a new MAX_ORDER_NR_PAGES range in case of large
491 * memory holes within the zone
492 */
493 if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
494 if (!pfn_valid(low_pfn)) {
495 low_pfn += MAX_ORDER_NR_PAGES - 1;
496 continue;
497 }
498 }
499
Mel Gorman748446b2010-05-24 14:32:27 -0700500 if (!pfn_valid_within(low_pfn))
501 continue;
Mel Gormanb7aba692011-01-13 15:45:54 -0800502 nr_scanned++;
Mel Gorman748446b2010-05-24 14:32:27 -0700503
Mel Gormandc908602012-02-08 17:13:38 -0800504 /*
505 * Get the page and ensure the page is within the same zone.
506 * See the comment in isolate_freepages about overlapping
507 * nodes. It is deliberate that the new zone lock is not taken
508 * as memory compaction should not move pages between nodes.
509 */
Mel Gorman748446b2010-05-24 14:32:27 -0700510 page = pfn_to_page(low_pfn);
Mel Gormandc908602012-02-08 17:13:38 -0800511 if (page_zone(page) != zone)
512 continue;
513
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700514 if (!valid_page)
515 valid_page = page;
516
517 /* If isolation recently failed, do not retry */
518 pageblock_nr = low_pfn >> pageblock_order;
519 if (!isolation_suitable(cc, page))
520 goto next_pageblock;
521
Mel Gormandc908602012-02-08 17:13:38 -0800522 /* Skip if free */
Mel Gorman748446b2010-05-24 14:32:27 -0700523 if (PageBuddy(page))
524 continue;
525
Mel Gorman9927af742011-01-13 15:45:59 -0800526 /*
527 * For async migration, also only scan in MOVABLE blocks. Async
528 * migration is optimistic to see if the minimum amount of work
529 * satisfies the allocation
530 */
Linus Torvalds68e3e922012-06-03 20:05:57 -0700531 if (!cc->sync && last_pageblock_nr != pageblock_nr &&
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100532 !migrate_async_suitable(get_pageblock_migratetype(page))) {
Mel Gormanc89511a2012-10-08 16:32:45 -0700533 cc->finished_update_migrate = true;
Mel Gorman2a1402a2012-10-08 16:32:33 -0700534 goto next_pageblock;
Mel Gorman9927af742011-01-13 15:45:59 -0800535 }
536
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800537 /*
538 * Check may be lockless but that's ok as we recheck later.
539 * It's possible to migrate LRU pages and balloon pages
540 * Skip any other type of page
541 */
542 if (!PageLRU(page)) {
543 if (unlikely(balloon_page_movable(page))) {
544 if (locked && balloon_page_isolate(page)) {
545 /* Successfully isolated */
546 cc->finished_update_migrate = true;
547 list_add(&page->lru, migratelist);
548 cc->nr_migratepages++;
549 nr_isolated++;
550 goto check_compact_cluster;
551 }
552 }
Andrea Arcangelibc835012011-01-13 15:47:08 -0800553 continue;
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800554 }
Andrea Arcangelibc835012011-01-13 15:47:08 -0800555
556 /*
Mel Gorman2a1402a2012-10-08 16:32:33 -0700557 * PageLRU is set. lru_lock normally excludes isolation
558 * splitting and collapsing (collapsing has already happened
559 * if PageLRU is set) but the lock is not necessarily taken
560 * here and it is wasteful to take it just to check transhuge.
561 * Check TransHuge without lock and skip the whole pageblock if
562 * it's either a transhuge or hugetlbfs page, as calling
563 * compound_order() without preventing THP from splitting the
564 * page underneath us may return surprising results.
Andrea Arcangelibc835012011-01-13 15:47:08 -0800565 */
566 if (PageTransHuge(page)) {
Mel Gorman2a1402a2012-10-08 16:32:33 -0700567 if (!locked)
568 goto next_pageblock;
569 low_pfn += (1 << compound_order(page)) - 1;
570 continue;
571 }
572
573 /* Check if it is ok to still hold the lock */
574 locked = compact_checklock_irqsave(&zone->lru_lock, &flags,
575 locked, cc);
576 if (!locked || fatal_signal_pending(current))
577 break;
578
579 /* Recheck PageLRU and PageTransHuge under lock */
580 if (!PageLRU(page))
581 continue;
582 if (PageTransHuge(page)) {
Andrea Arcangelibc835012011-01-13 15:47:08 -0800583 low_pfn += (1 << compound_order(page)) - 1;
584 continue;
585 }
586
Linus Torvalds68e3e922012-06-03 20:05:57 -0700587 if (!cc->sync)
Mel Gormanc8244932012-01-12 17:19:38 -0800588 mode |= ISOLATE_ASYNC_MIGRATE;
589
Minchan Kime46a2872012-10-08 16:33:48 -0700590 if (unevictable)
591 mode |= ISOLATE_UNEVICTABLE;
592
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700593 lruvec = mem_cgroup_page_lruvec(page, zone);
594
Mel Gorman748446b2010-05-24 14:32:27 -0700595 /* Try isolate the page */
Konstantin Khlebnikovf3fd4a62012-05-29 15:06:54 -0700596 if (__isolate_lru_page(page, mode) != 0)
Mel Gorman748446b2010-05-24 14:32:27 -0700597 continue;
598
Andrea Arcangelibc835012011-01-13 15:47:08 -0800599 VM_BUG_ON(PageTransCompound(page));
600
Mel Gorman748446b2010-05-24 14:32:27 -0700601 /* Successfully isolated */
Mel Gormanc89511a2012-10-08 16:32:45 -0700602 cc->finished_update_migrate = true;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700603 del_page_from_lru_list(page, lruvec, page_lru(page));
Mel Gorman748446b2010-05-24 14:32:27 -0700604 list_add(&page->lru, migratelist);
Mel Gorman748446b2010-05-24 14:32:27 -0700605 cc->nr_migratepages++;
Mel Gormanb7aba692011-01-13 15:45:54 -0800606 nr_isolated++;
Mel Gorman748446b2010-05-24 14:32:27 -0700607
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800608check_compact_cluster:
Mel Gorman748446b2010-05-24 14:32:27 -0700609 /* Avoid isolating too much */
Hillf Danton31b83842012-01-10 15:07:59 -0800610 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) {
611 ++low_pfn;
Mel Gorman748446b2010-05-24 14:32:27 -0700612 break;
Hillf Danton31b83842012-01-10 15:07:59 -0800613 }
Mel Gorman2a1402a2012-10-08 16:32:33 -0700614
615 continue;
616
617next_pageblock:
Mel Gormana9aacbc2013-02-22 16:32:25 -0800618 low_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages) - 1;
Mel Gorman2a1402a2012-10-08 16:32:33 -0700619 last_pageblock_nr = pageblock_nr;
Mel Gorman748446b2010-05-24 14:32:27 -0700620 }
621
Mel Gormanc67fe372012-08-21 16:16:17 -0700622 acct_isolated(zone, locked, cc);
Mel Gorman748446b2010-05-24 14:32:27 -0700623
Mel Gormanc67fe372012-08-21 16:16:17 -0700624 if (locked)
625 spin_unlock_irqrestore(&zone->lru_lock, flags);
Mel Gorman748446b2010-05-24 14:32:27 -0700626
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700627 /* Update the pageblock-skip if the whole pageblock was scanned */
628 if (low_pfn == end_pfn)
Mel Gormanc89511a2012-10-08 16:32:45 -0700629 update_pageblock_skip(cc, valid_page, nr_isolated, true);
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700630
Mel Gormanb7aba692011-01-13 15:45:54 -0800631 trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated);
632
Minchan Kim010fc292012-12-20 15:05:06 -0800633 count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned);
Mel Gorman397487d2012-10-19 12:00:10 +0100634 if (nr_isolated)
Minchan Kim010fc292012-12-20 15:05:06 -0800635 count_compact_events(COMPACTISOLATED, nr_isolated);
Mel Gorman397487d2012-10-19 12:00:10 +0100636
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100637 return low_pfn;
638}
639
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100640#endif /* CONFIG_COMPACTION || CONFIG_CMA */
641#ifdef CONFIG_COMPACTION
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100642/*
643 * Based on information in the current compact_control, find blocks
644 * suitable for isolating free pages from and then isolate them.
645 */
646static void isolate_freepages(struct zone *zone,
647 struct compact_control *cc)
648{
649 struct page *page;
650 unsigned long high_pfn, low_pfn, pfn, zone_end_pfn, end_pfn;
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100651 int nr_freepages = cc->nr_freepages;
652 struct list_head *freelist = &cc->freepages;
653
654 /*
655 * Initialise the free scanner. The starting point is where we last
656 * scanned from (or the end of the zone if starting). The low point
657 * is the end of the pageblock the migration scanner is using.
658 */
659 pfn = cc->free_pfn;
660 low_pfn = cc->migrate_pfn + pageblock_nr_pages;
661
662 /*
663 * Take care that if the migration scanner is at the end of the zone
664 * that the free scanner does not accidentally move to the next zone
665 * in the next isolation cycle.
666 */
667 high_pfn = min(low_pfn, pfn);
668
669 zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
670
671 /*
672 * Isolate free pages until enough are available to migrate the
673 * pages on cc->migratepages. We stop searching if the migrate
674 * and free page scanners meet or enough free pages are isolated.
675 */
676 for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages;
677 pfn -= pageblock_nr_pages) {
678 unsigned long isolated;
679
680 if (!pfn_valid(pfn))
681 continue;
682
683 /*
684 * Check for overlapping nodes/zones. It's possible on some
685 * configurations to have a setup like
686 * node0 node1 node0
687 * i.e. it's possible that all pages within a zones range of
688 * pages do not belong to a single zone.
689 */
690 page = pfn_to_page(pfn);
691 if (page_zone(page) != zone)
692 continue;
693
694 /* Check the block is suitable for migration */
Linus Torvalds68e3e922012-06-03 20:05:57 -0700695 if (!suitable_migration_target(page))
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100696 continue;
Linus Torvalds68e3e922012-06-03 20:05:57 -0700697
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700698 /* If isolation recently failed, do not retry */
699 if (!isolation_suitable(cc, page))
700 continue;
701
Mel Gormanf40d1e42012-10-08 16:32:36 -0700702 /* Found a block suitable for isolating free pages from */
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100703 isolated = 0;
Mel Gorman60177d32012-12-06 19:01:14 +0000704
705 /*
706 * As pfn may not start aligned, pfn+pageblock_nr_page
707 * may cross a MAX_ORDER_NR_PAGES boundary and miss
708 * a pfn_valid check. Ensure isolate_freepages_block()
709 * only scans within a pageblock
710 */
711 end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
712 end_pfn = min(end_pfn, zone_end_pfn);
Mel Gormanf40d1e42012-10-08 16:32:36 -0700713 isolated = isolate_freepages_block(cc, pfn, end_pfn,
714 freelist, false);
715 nr_freepages += isolated;
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100716
717 /*
718 * Record the highest PFN we isolated pages from. When next
719 * looking for free pages, the search will restart here as
720 * page migration may have returned some pages to the allocator
721 */
Mel Gormanc89511a2012-10-08 16:32:45 -0700722 if (isolated) {
723 cc->finished_update_free = true;
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100724 high_pfn = max(high_pfn, pfn);
Mel Gormanc89511a2012-10-08 16:32:45 -0700725 }
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100726 }
727
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100728 /* split_free_page does not map the pages */
729 map_pages(freelist);
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100730
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100731 cc->free_pfn = high_pfn;
732 cc->nr_freepages = nr_freepages;
Mel Gorman748446b2010-05-24 14:32:27 -0700733}
734
735/*
736 * This is a migrate-callback that "allocates" freepages by taking pages
737 * from the isolated freelists in the block we are migrating to.
738 */
739static struct page *compaction_alloc(struct page *migratepage,
740 unsigned long data,
741 int **result)
742{
743 struct compact_control *cc = (struct compact_control *)data;
744 struct page *freepage;
745
746 /* Isolate free pages if necessary */
747 if (list_empty(&cc->freepages)) {
748 isolate_freepages(cc->zone, cc);
749
750 if (list_empty(&cc->freepages))
751 return NULL;
752 }
753
754 freepage = list_entry(cc->freepages.next, struct page, lru);
755 list_del(&freepage->lru);
756 cc->nr_freepages--;
757
758 return freepage;
759}
760
761/*
762 * We cannot control nr_migratepages and nr_freepages fully when migration is
763 * running as migrate_pages() has no knowledge of compact_control. When
764 * migration is complete, we count the number of pages on the lists by hand.
765 */
766static void update_nr_listpages(struct compact_control *cc)
767{
768 int nr_migratepages = 0;
769 int nr_freepages = 0;
770 struct page *page;
771
772 list_for_each_entry(page, &cc->migratepages, lru)
773 nr_migratepages++;
774 list_for_each_entry(page, &cc->freepages, lru)
775 nr_freepages++;
776
777 cc->nr_migratepages = nr_migratepages;
778 cc->nr_freepages = nr_freepages;
779}
780
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100781/* possible outcome of isolate_migratepages */
782typedef enum {
783 ISOLATE_ABORT, /* Abort compaction now */
784 ISOLATE_NONE, /* No pages isolated, continue scanning */
785 ISOLATE_SUCCESS, /* Pages isolated, migrate */
786} isolate_migrate_t;
787
788/*
789 * Isolate all pages that can be migrated from the block pointed to by
790 * the migrate scanner within compact_control.
791 */
792static isolate_migrate_t isolate_migratepages(struct zone *zone,
793 struct compact_control *cc)
794{
795 unsigned long low_pfn, end_pfn;
796
797 /* Do not scan outside zone boundaries */
798 low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn);
799
800 /* Only scan within a pageblock boundary */
Mel Gormana9aacbc2013-02-22 16:32:25 -0800801 end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages);
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100802
803 /* Do not cross the free scanner or scan within a memory hole */
804 if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) {
805 cc->migrate_pfn = end_pfn;
806 return ISOLATE_NONE;
807 }
808
809 /* Perform the isolation */
Minchan Kime46a2872012-10-08 16:33:48 -0700810 low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn, false);
Shaohua Lie64c5232012-10-08 16:32:27 -0700811 if (!low_pfn || cc->contended)
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100812 return ISOLATE_ABORT;
813
814 cc->migrate_pfn = low_pfn;
815
816 return ISOLATE_SUCCESS;
817}
818
Mel Gorman748446b2010-05-24 14:32:27 -0700819static int compact_finished(struct zone *zone,
Andrea Arcangeli5a03b052011-01-13 15:47:11 -0800820 struct compact_control *cc)
Mel Gorman748446b2010-05-24 14:32:27 -0700821{
Mel Gorman8fb74b92013-01-11 14:32:16 -0800822 unsigned int order;
Andrea Arcangeli5a03b052011-01-13 15:47:11 -0800823 unsigned long watermark;
Mel Gorman56de7262010-05-24 14:32:30 -0700824
Mel Gorman748446b2010-05-24 14:32:27 -0700825 if (fatal_signal_pending(current))
826 return COMPACT_PARTIAL;
827
Mel Gorman753341a2012-10-08 16:32:40 -0700828 /* Compaction run completes if the migrate and free scanner meet */
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700829 if (cc->free_pfn <= cc->migrate_pfn) {
Mel Gorman62997022012-10-08 16:32:47 -0700830 /*
831 * Mark that the PG_migrate_skip information should be cleared
832 * by kswapd when it goes to sleep. kswapd does not set the
833 * flag itself as the decision to be clear should be directly
834 * based on an allocation request.
835 */
836 if (!current_is_kswapd())
837 zone->compact_blockskip_flush = true;
838
Mel Gorman748446b2010-05-24 14:32:27 -0700839 return COMPACT_COMPLETE;
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700840 }
Mel Gorman748446b2010-05-24 14:32:27 -0700841
Johannes Weiner82478fb2011-01-20 14:44:21 -0800842 /*
843 * order == -1 is expected when compacting via
844 * /proc/sys/vm/compact_memory
845 */
Mel Gorman56de7262010-05-24 14:32:30 -0700846 if (cc->order == -1)
847 return COMPACT_CONTINUE;
848
Michal Hocko3957c772011-06-15 15:08:25 -0700849 /* Compaction run is not finished if the watermark is not met */
850 watermark = low_wmark_pages(zone);
851 watermark += (1 << cc->order);
852
853 if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0))
854 return COMPACT_CONTINUE;
855
Mel Gorman56de7262010-05-24 14:32:30 -0700856 /* Direct compactor: Is a suitable page free? */
Mel Gorman8fb74b92013-01-11 14:32:16 -0800857 for (order = cc->order; order < MAX_ORDER; order++) {
858 struct free_area *area = &zone->free_area[order];
Mel Gorman56de7262010-05-24 14:32:30 -0700859
Mel Gorman8fb74b92013-01-11 14:32:16 -0800860 /* Job done if page is free of the right migratetype */
861 if (!list_empty(&area->free_list[cc->migratetype]))
862 return COMPACT_PARTIAL;
863
864 /* Job done if allocation would set block type */
865 if (cc->order >= pageblock_order && area->nr_free)
866 return COMPACT_PARTIAL;
Mel Gorman56de7262010-05-24 14:32:30 -0700867 }
868
Mel Gorman748446b2010-05-24 14:32:27 -0700869 return COMPACT_CONTINUE;
870}
871
Mel Gorman3e7d3442011-01-13 15:45:56 -0800872/*
873 * compaction_suitable: Is this suitable to run compaction on this zone now?
874 * Returns
875 * COMPACT_SKIPPED - If there are too few free pages for compaction
876 * COMPACT_PARTIAL - If the allocation would succeed without compaction
877 * COMPACT_CONTINUE - If compaction should run now
878 */
879unsigned long compaction_suitable(struct zone *zone, int order)
880{
881 int fragindex;
882 unsigned long watermark;
883
884 /*
Michal Hocko3957c772011-06-15 15:08:25 -0700885 * order == -1 is expected when compacting via
886 * /proc/sys/vm/compact_memory
887 */
888 if (order == -1)
889 return COMPACT_CONTINUE;
890
891 /*
Mel Gorman3e7d3442011-01-13 15:45:56 -0800892 * Watermarks for order-0 must be met for compaction. Note the 2UL.
893 * This is because during migration, copies of pages need to be
894 * allocated and for a short time, the footprint is higher
895 */
896 watermark = low_wmark_pages(zone) + (2UL << order);
897 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
898 return COMPACT_SKIPPED;
899
900 /*
901 * fragmentation index determines if allocation failures are due to
902 * low memory or external fragmentation
903 *
Shaohua Lia582a732011-06-15 15:08:49 -0700904 * index of -1000 implies allocations might succeed depending on
905 * watermarks
Mel Gorman3e7d3442011-01-13 15:45:56 -0800906 * index towards 0 implies failure is due to lack of memory
907 * index towards 1000 implies failure is due to fragmentation
908 *
909 * Only compact if a failure would be due to fragmentation.
910 */
911 fragindex = fragmentation_index(zone, order);
912 if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
913 return COMPACT_SKIPPED;
914
Shaohua Lia582a732011-06-15 15:08:49 -0700915 if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark,
916 0, 0))
Mel Gorman3e7d3442011-01-13 15:45:56 -0800917 return COMPACT_PARTIAL;
918
919 return COMPACT_CONTINUE;
920}
921
Mel Gorman748446b2010-05-24 14:32:27 -0700922static int compact_zone(struct zone *zone, struct compact_control *cc)
923{
924 int ret;
Mel Gormanc89511a2012-10-08 16:32:45 -0700925 unsigned long start_pfn = zone->zone_start_pfn;
926 unsigned long end_pfn = zone->zone_start_pfn + zone->spanned_pages;
Mel Gorman748446b2010-05-24 14:32:27 -0700927
Mel Gorman3e7d3442011-01-13 15:45:56 -0800928 ret = compaction_suitable(zone, cc->order);
929 switch (ret) {
930 case COMPACT_PARTIAL:
931 case COMPACT_SKIPPED:
932 /* Compaction is likely to fail */
933 return ret;
934 case COMPACT_CONTINUE:
935 /* Fall through to compaction */
936 ;
937 }
938
Mel Gormanc89511a2012-10-08 16:32:45 -0700939 /*
940 * Setup to move all movable pages to the end of the zone. Used cached
941 * information on where the scanners should start but check that it
942 * is initialised by ensuring the values are within zone boundaries.
943 */
944 cc->migrate_pfn = zone->compact_cached_migrate_pfn;
945 cc->free_pfn = zone->compact_cached_free_pfn;
946 if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) {
947 cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1);
948 zone->compact_cached_free_pfn = cc->free_pfn;
949 }
950 if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) {
951 cc->migrate_pfn = start_pfn;
952 zone->compact_cached_migrate_pfn = cc->migrate_pfn;
953 }
Mel Gorman748446b2010-05-24 14:32:27 -0700954
Mel Gorman62997022012-10-08 16:32:47 -0700955 /*
956 * Clear pageblock skip if there were failures recently and compaction
957 * is about to be retried after being deferred. kswapd does not do
958 * this reset as it'll reset the cached information when going to sleep.
959 */
960 if (compaction_restarting(zone, cc->order) && !current_is_kswapd())
961 __reset_isolation_suitable(zone);
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700962
Mel Gorman748446b2010-05-24 14:32:27 -0700963 migrate_prep_local();
964
965 while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) {
966 unsigned long nr_migrate, nr_remaining;
Minchan Kim9d502c12011-03-22 16:30:39 -0700967 int err;
Mel Gorman748446b2010-05-24 14:32:27 -0700968
Mel Gormanf9e35b32011-06-15 15:08:52 -0700969 switch (isolate_migratepages(zone, cc)) {
970 case ISOLATE_ABORT:
971 ret = COMPACT_PARTIAL;
Rafael Aquini5733c7d2012-12-11 16:02:47 -0800972 putback_movable_pages(&cc->migratepages);
Shaohua Lie64c5232012-10-08 16:32:27 -0700973 cc->nr_migratepages = 0;
Mel Gormanf9e35b32011-06-15 15:08:52 -0700974 goto out;
975 case ISOLATE_NONE:
Mel Gorman748446b2010-05-24 14:32:27 -0700976 continue;
Mel Gormanf9e35b32011-06-15 15:08:52 -0700977 case ISOLATE_SUCCESS:
978 ;
979 }
Mel Gorman748446b2010-05-24 14:32:27 -0700980
981 nr_migrate = cc->nr_migratepages;
Minchan Kim9d502c12011-03-22 16:30:39 -0700982 err = migrate_pages(&cc->migratepages, compaction_alloc,
Linus Torvalds68e3e922012-06-03 20:05:57 -0700983 (unsigned long)cc, false,
Mel Gorman7b2a2d42012-10-19 14:07:31 +0100984 cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC,
985 MR_COMPACTION);
Mel Gorman748446b2010-05-24 14:32:27 -0700986 update_nr_listpages(cc);
987 nr_remaining = cc->nr_migratepages;
988
Mel Gormanb7aba692011-01-13 15:45:54 -0800989 trace_mm_compaction_migratepages(nr_migrate - nr_remaining,
990 nr_remaining);
Mel Gorman748446b2010-05-24 14:32:27 -0700991
Rafael Aquini5733c7d2012-12-11 16:02:47 -0800992 /* Release isolated pages not migrated */
Minchan Kim9d502c12011-03-22 16:30:39 -0700993 if (err) {
Rafael Aquini5733c7d2012-12-11 16:02:47 -0800994 putback_movable_pages(&cc->migratepages);
Mel Gorman748446b2010-05-24 14:32:27 -0700995 cc->nr_migratepages = 0;
David Rientjes4bf2bba2012-07-11 14:02:13 -0700996 if (err == -ENOMEM) {
997 ret = COMPACT_PARTIAL;
998 goto out;
999 }
Mel Gorman748446b2010-05-24 14:32:27 -07001000 }
Mel Gorman748446b2010-05-24 14:32:27 -07001001 }
1002
Mel Gormanf9e35b32011-06-15 15:08:52 -07001003out:
Mel Gorman748446b2010-05-24 14:32:27 -07001004 /* Release free pages and check accounting */
1005 cc->nr_freepages -= release_freepages(&cc->freepages);
1006 VM_BUG_ON(cc->nr_freepages != 0);
1007
1008 return ret;
1009}
Mel Gorman76ab0f52010-05-24 14:32:28 -07001010
Kyungmin Parkd43a87e2011-10-31 17:09:08 -07001011static unsigned long compact_zone_order(struct zone *zone,
Andrea Arcangeli5a03b052011-01-13 15:47:11 -08001012 int order, gfp_t gfp_mask,
Mel Gorman8fb74b92013-01-11 14:32:16 -08001013 bool sync, bool *contended)
Mel Gorman56de7262010-05-24 14:32:30 -07001014{
Shaohua Lie64c5232012-10-08 16:32:27 -07001015 unsigned long ret;
Mel Gorman56de7262010-05-24 14:32:30 -07001016 struct compact_control cc = {
1017 .nr_freepages = 0,
1018 .nr_migratepages = 0,
1019 .order = order,
1020 .migratetype = allocflags_to_migratetype(gfp_mask),
1021 .zone = zone,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001022 .sync = sync,
Mel Gorman56de7262010-05-24 14:32:30 -07001023 };
1024 INIT_LIST_HEAD(&cc.freepages);
1025 INIT_LIST_HEAD(&cc.migratepages);
1026
Shaohua Lie64c5232012-10-08 16:32:27 -07001027 ret = compact_zone(zone, &cc);
1028
1029 VM_BUG_ON(!list_empty(&cc.freepages));
1030 VM_BUG_ON(!list_empty(&cc.migratepages));
1031
1032 *contended = cc.contended;
1033 return ret;
Mel Gorman56de7262010-05-24 14:32:30 -07001034}
1035
Mel Gorman5e771902010-05-24 14:32:31 -07001036int sysctl_extfrag_threshold = 500;
1037
Mel Gorman56de7262010-05-24 14:32:30 -07001038/**
1039 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
1040 * @zonelist: The zonelist used for the current allocation
1041 * @order: The order of the current allocation
1042 * @gfp_mask: The GFP mask of the current allocation
1043 * @nodemask: The allowed nodes to allocate from
Mel Gorman77f1fe62011-01-13 15:45:57 -08001044 * @sync: Whether migration is synchronous or not
Mel Gorman661c4cb2012-10-08 16:32:31 -07001045 * @contended: Return value that is true if compaction was aborted due to lock contention
1046 * @page: Optionally capture a free page of the requested order during compaction
Mel Gorman56de7262010-05-24 14:32:30 -07001047 *
1048 * This is the main entry point for direct page compaction.
1049 */
1050unsigned long try_to_compact_pages(struct zonelist *zonelist,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001051 int order, gfp_t gfp_mask, nodemask_t *nodemask,
Mel Gorman8fb74b92013-01-11 14:32:16 -08001052 bool sync, bool *contended)
Mel Gorman56de7262010-05-24 14:32:30 -07001053{
1054 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
1055 int may_enter_fs = gfp_mask & __GFP_FS;
1056 int may_perform_io = gfp_mask & __GFP_IO;
Mel Gorman56de7262010-05-24 14:32:30 -07001057 struct zoneref *z;
1058 struct zone *zone;
1059 int rc = COMPACT_SKIPPED;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001060 int alloc_flags = 0;
Mel Gorman56de7262010-05-24 14:32:30 -07001061
Mel Gorman4ffb6332012-10-08 16:29:09 -07001062 /* Check if the GFP flags allow compaction */
Andrea Arcangelic5a73c32011-01-13 15:47:11 -08001063 if (!order || !may_enter_fs || !may_perform_io)
Mel Gorman56de7262010-05-24 14:32:30 -07001064 return rc;
1065
Minchan Kim010fc292012-12-20 15:05:06 -08001066 count_compact_event(COMPACTSTALL);
Mel Gorman56de7262010-05-24 14:32:30 -07001067
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001068#ifdef CONFIG_CMA
1069 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
1070 alloc_flags |= ALLOC_CMA;
1071#endif
Mel Gorman56de7262010-05-24 14:32:30 -07001072 /* Compact each zone in the list */
1073 for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx,
1074 nodemask) {
Mel Gorman56de7262010-05-24 14:32:30 -07001075 int status;
1076
Mel Gormanc67fe372012-08-21 16:16:17 -07001077 status = compact_zone_order(zone, order, gfp_mask, sync,
Mel Gorman8fb74b92013-01-11 14:32:16 -08001078 contended);
Mel Gorman56de7262010-05-24 14:32:30 -07001079 rc = max(status, rc);
1080
Mel Gorman3e7d3442011-01-13 15:45:56 -08001081 /* If a normal allocation would succeed, stop compacting */
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07001082 if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0,
1083 alloc_flags))
Mel Gorman56de7262010-05-24 14:32:30 -07001084 break;
1085 }
1086
1087 return rc;
1088}
1089
1090
Mel Gorman76ab0f52010-05-24 14:32:28 -07001091/* Compact all zones within a node */
Andrew Morton7103f162013-02-22 16:32:33 -08001092static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc)
Mel Gorman76ab0f52010-05-24 14:32:28 -07001093{
1094 int zoneid;
Mel Gorman76ab0f52010-05-24 14:32:28 -07001095 struct zone *zone;
1096
Mel Gorman76ab0f52010-05-24 14:32:28 -07001097 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
Mel Gorman76ab0f52010-05-24 14:32:28 -07001098
1099 zone = &pgdat->node_zones[zoneid];
1100 if (!populated_zone(zone))
1101 continue;
1102
Rik van Riel7be62de2012-03-21 16:33:52 -07001103 cc->nr_freepages = 0;
1104 cc->nr_migratepages = 0;
1105 cc->zone = zone;
1106 INIT_LIST_HEAD(&cc->freepages);
1107 INIT_LIST_HEAD(&cc->migratepages);
Mel Gorman76ab0f52010-05-24 14:32:28 -07001108
Dan Carpenteraad6ec32012-03-21 16:33:54 -07001109 if (cc->order == -1 || !compaction_deferred(zone, cc->order))
Rik van Riel7be62de2012-03-21 16:33:52 -07001110 compact_zone(zone, cc);
Mel Gorman76ab0f52010-05-24 14:32:28 -07001111
Rik van Rielaff62242012-03-21 16:33:52 -07001112 if (cc->order > 0) {
1113 int ok = zone_watermark_ok(zone, cc->order,
1114 low_wmark_pages(zone), 0, 0);
Minchan Kimc81758f2012-08-21 16:16:03 -07001115 if (ok && cc->order >= zone->compact_order_failed)
Rik van Rielaff62242012-03-21 16:33:52 -07001116 zone->compact_order_failed = cc->order + 1;
1117 /* Currently async compaction is never deferred. */
Linus Torvalds68e3e922012-06-03 20:05:57 -07001118 else if (!ok && cc->sync)
Rik van Rielaff62242012-03-21 16:33:52 -07001119 defer_compaction(zone, cc->order);
1120 }
1121
Rik van Riel7be62de2012-03-21 16:33:52 -07001122 VM_BUG_ON(!list_empty(&cc->freepages));
1123 VM_BUG_ON(!list_empty(&cc->migratepages));
Mel Gorman76ab0f52010-05-24 14:32:28 -07001124 }
Mel Gorman76ab0f52010-05-24 14:32:28 -07001125}
1126
Andrew Morton7103f162013-02-22 16:32:33 -08001127void compact_pgdat(pg_data_t *pgdat, int order)
Rik van Riel7be62de2012-03-21 16:33:52 -07001128{
1129 struct compact_control cc = {
1130 .order = order,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001131 .sync = false,
Rik van Riel7be62de2012-03-21 16:33:52 -07001132 };
1133
Andrew Morton7103f162013-02-22 16:32:33 -08001134 __compact_pgdat(pgdat, &cc);
Rik van Riel7be62de2012-03-21 16:33:52 -07001135}
1136
Andrew Morton7103f162013-02-22 16:32:33 -08001137static void compact_node(int nid)
Rik van Riel7be62de2012-03-21 16:33:52 -07001138{
Rik van Riel7be62de2012-03-21 16:33:52 -07001139 struct compact_control cc = {
1140 .order = -1,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001141 .sync = true,
Rik van Riel7be62de2012-03-21 16:33:52 -07001142 };
1143
Andrew Morton7103f162013-02-22 16:32:33 -08001144 __compact_pgdat(NODE_DATA(nid), &cc);
Rik van Riel7be62de2012-03-21 16:33:52 -07001145}
1146
Mel Gorman76ab0f52010-05-24 14:32:28 -07001147/* Compact all nodes in the system */
Jason Liu7964c062013-01-11 14:31:47 -08001148static void compact_nodes(void)
Mel Gorman76ab0f52010-05-24 14:32:28 -07001149{
1150 int nid;
1151
Hugh Dickins8575ec22012-03-21 16:33:53 -07001152 /* Flush pending updates to the LRU lists */
1153 lru_add_drain_all();
1154
Mel Gorman76ab0f52010-05-24 14:32:28 -07001155 for_each_online_node(nid)
1156 compact_node(nid);
Mel Gorman76ab0f52010-05-24 14:32:28 -07001157}
1158
1159/* The written value is actually unused, all memory is compacted */
1160int sysctl_compact_memory;
1161
1162/* This is the entry point for compacting all nodes via /proc/sys/vm */
1163int sysctl_compaction_handler(struct ctl_table *table, int write,
1164 void __user *buffer, size_t *length, loff_t *ppos)
1165{
1166 if (write)
Jason Liu7964c062013-01-11 14:31:47 -08001167 compact_nodes();
Mel Gorman76ab0f52010-05-24 14:32:28 -07001168
1169 return 0;
1170}
Mel Gormaned4a6d72010-05-24 14:32:29 -07001171
Mel Gorman5e771902010-05-24 14:32:31 -07001172int sysctl_extfrag_handler(struct ctl_table *table, int write,
1173 void __user *buffer, size_t *length, loff_t *ppos)
1174{
1175 proc_dointvec_minmax(table, write, buffer, length, ppos);
1176
1177 return 0;
1178}
1179
Mel Gormaned4a6d72010-05-24 14:32:29 -07001180#if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
Kay Sievers10fbcf42011-12-21 14:48:43 -08001181ssize_t sysfs_compact_node(struct device *dev,
1182 struct device_attribute *attr,
Mel Gormaned4a6d72010-05-24 14:32:29 -07001183 const char *buf, size_t count)
1184{
Hugh Dickins8575ec22012-03-21 16:33:53 -07001185 int nid = dev->id;
1186
1187 if (nid >= 0 && nid < nr_node_ids && node_online(nid)) {
1188 /* Flush pending updates to the LRU lists */
1189 lru_add_drain_all();
1190
1191 compact_node(nid);
1192 }
Mel Gormaned4a6d72010-05-24 14:32:29 -07001193
1194 return count;
1195}
Kay Sievers10fbcf42011-12-21 14:48:43 -08001196static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node);
Mel Gormaned4a6d72010-05-24 14:32:29 -07001197
1198int compaction_register_node(struct node *node)
1199{
Kay Sievers10fbcf42011-12-21 14:48:43 -08001200 return device_create_file(&node->dev, &dev_attr_compact);
Mel Gormaned4a6d72010-05-24 14:32:29 -07001201}
1202
1203void compaction_unregister_node(struct node *node)
1204{
Kay Sievers10fbcf42011-12-21 14:48:43 -08001205 return device_remove_file(&node->dev, &dev_attr_compact);
Mel Gormaned4a6d72010-05-24 14:32:29 -07001206}
1207#endif /* CONFIG_SYSFS && CONFIG_NUMA */
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001208
1209#endif /* CONFIG_COMPACTION */