Loading...
1/*
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>
15#include <linux/sysctl.h>
16#include <linux/sysfs.h>
17#include "internal.h"
18
19#define CREATE_TRACE_POINTS
20#include <trace/events/compaction.h>
21
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 */
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 */
36 bool sync; /* Synchronous migration */
37
38 /* Account for isolated anon and file pages */
39 unsigned long nr_anon;
40 unsigned long nr_file;
41
42 unsigned int order; /* order a direct compactor needs */
43 int migratetype; /* MOVABLE, RECLAIMABLE etc */
44 struct zone *zone;
45};
46
47static unsigned long release_freepages(struct list_head *freelist)
48{
49 struct page *page, *next;
50 unsigned long count = 0;
51
52 list_for_each_entry_safe(page, next, freelist, lru) {
53 list_del(&page->lru);
54 __free_page(page);
55 count++;
56 }
57
58 return count;
59}
60
61/* Isolate free pages onto a private freelist. Must hold zone->lock */
62static unsigned long isolate_freepages_block(struct zone *zone,
63 unsigned long blockpfn,
64 struct list_head *freelist)
65{
66 unsigned long zone_end_pfn, end_pfn;
67 int nr_scanned = 0, total_isolated = 0;
68 struct page *cursor;
69
70 /* Get the last PFN we should scan for free pages at */
71 zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
72 end_pfn = min(blockpfn + pageblock_nr_pages, zone_end_pfn);
73
74 /* Find the first usable PFN in the block to initialse page cursor */
75 for (; blockpfn < end_pfn; blockpfn++) {
76 if (pfn_valid_within(blockpfn))
77 break;
78 }
79 cursor = pfn_to_page(blockpfn);
80
81 /* Isolate free pages. This assumes the block is valid */
82 for (; blockpfn < end_pfn; blockpfn++, cursor++) {
83 int isolated, i;
84 struct page *page = cursor;
85
86 if (!pfn_valid_within(blockpfn))
87 continue;
88 nr_scanned++;
89
90 if (!PageBuddy(page))
91 continue;
92
93 /* Found a free page, break it into order-0 pages */
94 isolated = split_free_page(page);
95 total_isolated += isolated;
96 for (i = 0; i < isolated; i++) {
97 list_add(&page->lru, freelist);
98 page++;
99 }
100
101 /* If a page was split, advance to the end of it */
102 if (isolated) {
103 blockpfn += isolated - 1;
104 cursor += isolated - 1;
105 }
106 }
107
108 trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated);
109 return total_isolated;
110}
111
112/* Returns true if the page is within a block suitable for migration to */
113static bool suitable_migration_target(struct page *page)
114{
115
116 int migratetype = get_pageblock_migratetype(page);
117
118 /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */
119 if (migratetype == MIGRATE_ISOLATE || migratetype == MIGRATE_RESERVE)
120 return false;
121
122 /* If the page is a large free page, then allow migration */
123 if (PageBuddy(page) && page_order(page) >= pageblock_order)
124 return true;
125
126 /* If the block is MIGRATE_MOVABLE, allow migration */
127 if (migratetype == MIGRATE_MOVABLE)
128 return true;
129
130 /* Otherwise skip the block */
131 return false;
132}
133
134/*
135 * Based on information in the current compact_control, find blocks
136 * suitable for isolating free pages from and then isolate them.
137 */
138static void isolate_freepages(struct zone *zone,
139 struct compact_control *cc)
140{
141 struct page *page;
142 unsigned long high_pfn, low_pfn, pfn;
143 unsigned long flags;
144 int nr_freepages = cc->nr_freepages;
145 struct list_head *freelist = &cc->freepages;
146
147 /*
148 * Initialise the free scanner. The starting point is where we last
149 * scanned from (or the end of the zone if starting). The low point
150 * is the end of the pageblock the migration scanner is using.
151 */
152 pfn = cc->free_pfn;
153 low_pfn = cc->migrate_pfn + pageblock_nr_pages;
154
155 /*
156 * Take care that if the migration scanner is at the end of the zone
157 * that the free scanner does not accidentally move to the next zone
158 * in the next isolation cycle.
159 */
160 high_pfn = min(low_pfn, pfn);
161
162 /*
163 * Isolate free pages until enough are available to migrate the
164 * pages on cc->migratepages. We stop searching if the migrate
165 * and free page scanners meet or enough free pages are isolated.
166 */
167 for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages;
168 pfn -= pageblock_nr_pages) {
169 unsigned long isolated;
170
171 if (!pfn_valid(pfn))
172 continue;
173
174 /*
175 * Check for overlapping nodes/zones. It's possible on some
176 * configurations to have a setup like
177 * node0 node1 node0
178 * i.e. it's possible that all pages within a zones range of
179 * pages do not belong to a single zone.
180 */
181 page = pfn_to_page(pfn);
182 if (page_zone(page) != zone)
183 continue;
184
185 /* Check the block is suitable for migration */
186 if (!suitable_migration_target(page))
187 continue;
188
189 /*
190 * Found a block suitable for isolating free pages from. Now
191 * we disabled interrupts, double check things are ok and
192 * isolate the pages. This is to minimise the time IRQs
193 * are disabled
194 */
195 isolated = 0;
196 spin_lock_irqsave(&zone->lock, flags);
197 if (suitable_migration_target(page)) {
198 isolated = isolate_freepages_block(zone, pfn, freelist);
199 nr_freepages += isolated;
200 }
201 spin_unlock_irqrestore(&zone->lock, flags);
202
203 /*
204 * Record the highest PFN we isolated pages from. When next
205 * looking for free pages, the search will restart here as
206 * page migration may have returned some pages to the allocator
207 */
208 if (isolated)
209 high_pfn = max(high_pfn, pfn);
210 }
211
212 /* split_free_page does not map the pages */
213 list_for_each_entry(page, freelist, lru) {
214 arch_alloc_page(page, 0);
215 kernel_map_pages(page, 1, 1);
216 }
217
218 cc->free_pfn = high_pfn;
219 cc->nr_freepages = nr_freepages;
220}
221
222/* Update the number of anon and file isolated pages in the zone */
223static void acct_isolated(struct zone *zone, struct compact_control *cc)
224{
225 struct page *page;
226 unsigned int count[NR_LRU_LISTS] = { 0, };
227
228 list_for_each_entry(page, &cc->migratepages, lru) {
229 int lru = page_lru_base_type(page);
230 count[lru]++;
231 }
232
233 cc->nr_anon = count[LRU_ACTIVE_ANON] + count[LRU_INACTIVE_ANON];
234 cc->nr_file = count[LRU_ACTIVE_FILE] + count[LRU_INACTIVE_FILE];
235 __mod_zone_page_state(zone, NR_ISOLATED_ANON, cc->nr_anon);
236 __mod_zone_page_state(zone, NR_ISOLATED_FILE, cc->nr_file);
237}
238
239/* Similar to reclaim, but different enough that they don't share logic */
240static bool too_many_isolated(struct zone *zone)
241{
242 unsigned long active, inactive, isolated;
243
244 inactive = zone_page_state(zone, NR_INACTIVE_FILE) +
245 zone_page_state(zone, NR_INACTIVE_ANON);
246 active = zone_page_state(zone, NR_ACTIVE_FILE) +
247 zone_page_state(zone, NR_ACTIVE_ANON);
248 isolated = zone_page_state(zone, NR_ISOLATED_FILE) +
249 zone_page_state(zone, NR_ISOLATED_ANON);
250
251 return isolated > (inactive + active) / 2;
252}
253
254/* possible outcome of isolate_migratepages */
255typedef enum {
256 ISOLATE_ABORT, /* Abort compaction now */
257 ISOLATE_NONE, /* No pages isolated, continue scanning */
258 ISOLATE_SUCCESS, /* Pages isolated, migrate */
259} isolate_migrate_t;
260
261/*
262 * Isolate all pages that can be migrated from the block pointed to by
263 * the migrate scanner within compact_control.
264 */
265static isolate_migrate_t isolate_migratepages(struct zone *zone,
266 struct compact_control *cc)
267{
268 unsigned long low_pfn, end_pfn;
269 unsigned long last_pageblock_nr = 0, pageblock_nr;
270 unsigned long nr_scanned = 0, nr_isolated = 0;
271 struct list_head *migratelist = &cc->migratepages;
272
273 /* Do not scan outside zone boundaries */
274 low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn);
275
276 /* Only scan within a pageblock boundary */
277 end_pfn = ALIGN(low_pfn + pageblock_nr_pages, pageblock_nr_pages);
278
279 /* Do not cross the free scanner or scan within a memory hole */
280 if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) {
281 cc->migrate_pfn = end_pfn;
282 return ISOLATE_NONE;
283 }
284
285 /*
286 * Ensure that there are not too many pages isolated from the LRU
287 * list by either parallel reclaimers or compaction. If there are,
288 * delay for some time until fewer pages are isolated
289 */
290 while (unlikely(too_many_isolated(zone))) {
291 /* async migration should just abort */
292 if (!cc->sync)
293 return ISOLATE_ABORT;
294
295 congestion_wait(BLK_RW_ASYNC, HZ/10);
296
297 if (fatal_signal_pending(current))
298 return ISOLATE_ABORT;
299 }
300
301 /* Time to isolate some pages for migration */
302 cond_resched();
303 spin_lock_irq(&zone->lru_lock);
304 for (; low_pfn < end_pfn; low_pfn++) {
305 struct page *page;
306 bool locked = true;
307
308 /* give a chance to irqs before checking need_resched() */
309 if (!((low_pfn+1) % SWAP_CLUSTER_MAX)) {
310 spin_unlock_irq(&zone->lru_lock);
311 locked = false;
312 }
313 if (need_resched() || spin_is_contended(&zone->lru_lock)) {
314 if (locked)
315 spin_unlock_irq(&zone->lru_lock);
316 cond_resched();
317 spin_lock_irq(&zone->lru_lock);
318 if (fatal_signal_pending(current))
319 break;
320 } else if (!locked)
321 spin_lock_irq(&zone->lru_lock);
322
323 if (!pfn_valid_within(low_pfn))
324 continue;
325 nr_scanned++;
326
327 /* Get the page and skip if free */
328 page = pfn_to_page(low_pfn);
329 if (PageBuddy(page))
330 continue;
331
332 /*
333 * For async migration, also only scan in MOVABLE blocks. Async
334 * migration is optimistic to see if the minimum amount of work
335 * satisfies the allocation
336 */
337 pageblock_nr = low_pfn >> pageblock_order;
338 if (!cc->sync && last_pageblock_nr != pageblock_nr &&
339 get_pageblock_migratetype(page) != MIGRATE_MOVABLE) {
340 low_pfn += pageblock_nr_pages;
341 low_pfn = ALIGN(low_pfn, pageblock_nr_pages) - 1;
342 last_pageblock_nr = pageblock_nr;
343 continue;
344 }
345
346 if (!PageLRU(page))
347 continue;
348
349 /*
350 * PageLRU is set, and lru_lock excludes isolation,
351 * splitting and collapsing (collapsing has already
352 * happened if PageLRU is set).
353 */
354 if (PageTransHuge(page)) {
355 low_pfn += (1 << compound_order(page)) - 1;
356 continue;
357 }
358
359 /* Try isolate the page */
360 if (__isolate_lru_page(page, ISOLATE_BOTH, 0) != 0)
361 continue;
362
363 VM_BUG_ON(PageTransCompound(page));
364
365 /* Successfully isolated */
366 del_page_from_lru_list(zone, page, page_lru(page));
367 list_add(&page->lru, migratelist);
368 cc->nr_migratepages++;
369 nr_isolated++;
370
371 /* Avoid isolating too much */
372 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX)
373 break;
374 }
375
376 acct_isolated(zone, cc);
377
378 spin_unlock_irq(&zone->lru_lock);
379 cc->migrate_pfn = low_pfn;
380
381 trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated);
382
383 return ISOLATE_SUCCESS;
384}
385
386/*
387 * This is a migrate-callback that "allocates" freepages by taking pages
388 * from the isolated freelists in the block we are migrating to.
389 */
390static struct page *compaction_alloc(struct page *migratepage,
391 unsigned long data,
392 int **result)
393{
394 struct compact_control *cc = (struct compact_control *)data;
395 struct page *freepage;
396
397 /* Isolate free pages if necessary */
398 if (list_empty(&cc->freepages)) {
399 isolate_freepages(cc->zone, cc);
400
401 if (list_empty(&cc->freepages))
402 return NULL;
403 }
404
405 freepage = list_entry(cc->freepages.next, struct page, lru);
406 list_del(&freepage->lru);
407 cc->nr_freepages--;
408
409 return freepage;
410}
411
412/*
413 * We cannot control nr_migratepages and nr_freepages fully when migration is
414 * running as migrate_pages() has no knowledge of compact_control. When
415 * migration is complete, we count the number of pages on the lists by hand.
416 */
417static void update_nr_listpages(struct compact_control *cc)
418{
419 int nr_migratepages = 0;
420 int nr_freepages = 0;
421 struct page *page;
422
423 list_for_each_entry(page, &cc->migratepages, lru)
424 nr_migratepages++;
425 list_for_each_entry(page, &cc->freepages, lru)
426 nr_freepages++;
427
428 cc->nr_migratepages = nr_migratepages;
429 cc->nr_freepages = nr_freepages;
430}
431
432static int compact_finished(struct zone *zone,
433 struct compact_control *cc)
434{
435 unsigned int order;
436 unsigned long watermark;
437
438 if (fatal_signal_pending(current))
439 return COMPACT_PARTIAL;
440
441 /* Compaction run completes if the migrate and free scanner meet */
442 if (cc->free_pfn <= cc->migrate_pfn)
443 return COMPACT_COMPLETE;
444
445 /*
446 * order == -1 is expected when compacting via
447 * /proc/sys/vm/compact_memory
448 */
449 if (cc->order == -1)
450 return COMPACT_CONTINUE;
451
452 /* Compaction run is not finished if the watermark is not met */
453 watermark = low_wmark_pages(zone);
454 watermark += (1 << cc->order);
455
456 if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0))
457 return COMPACT_CONTINUE;
458
459 /* Direct compactor: Is a suitable page free? */
460 for (order = cc->order; order < MAX_ORDER; order++) {
461 /* Job done if page is free of the right migratetype */
462 if (!list_empty(&zone->free_area[order].free_list[cc->migratetype]))
463 return COMPACT_PARTIAL;
464
465 /* Job done if allocation would set block type */
466 if (order >= pageblock_order && zone->free_area[order].nr_free)
467 return COMPACT_PARTIAL;
468 }
469
470 return COMPACT_CONTINUE;
471}
472
473/*
474 * compaction_suitable: Is this suitable to run compaction on this zone now?
475 * Returns
476 * COMPACT_SKIPPED - If there are too few free pages for compaction
477 * COMPACT_PARTIAL - If the allocation would succeed without compaction
478 * COMPACT_CONTINUE - If compaction should run now
479 */
480unsigned long compaction_suitable(struct zone *zone, int order)
481{
482 int fragindex;
483 unsigned long watermark;
484
485 /*
486 * order == -1 is expected when compacting via
487 * /proc/sys/vm/compact_memory
488 */
489 if (order == -1)
490 return COMPACT_CONTINUE;
491
492 /*
493 * Watermarks for order-0 must be met for compaction. Note the 2UL.
494 * This is because during migration, copies of pages need to be
495 * allocated and for a short time, the footprint is higher
496 */
497 watermark = low_wmark_pages(zone) + (2UL << order);
498 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
499 return COMPACT_SKIPPED;
500
501 /*
502 * fragmentation index determines if allocation failures are due to
503 * low memory or external fragmentation
504 *
505 * index of -1000 implies allocations might succeed depending on
506 * watermarks
507 * index towards 0 implies failure is due to lack of memory
508 * index towards 1000 implies failure is due to fragmentation
509 *
510 * Only compact if a failure would be due to fragmentation.
511 */
512 fragindex = fragmentation_index(zone, order);
513 if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
514 return COMPACT_SKIPPED;
515
516 if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark,
517 0, 0))
518 return COMPACT_PARTIAL;
519
520 return COMPACT_CONTINUE;
521}
522
523static int compact_zone(struct zone *zone, struct compact_control *cc)
524{
525 int ret;
526
527 ret = compaction_suitable(zone, cc->order);
528 switch (ret) {
529 case COMPACT_PARTIAL:
530 case COMPACT_SKIPPED:
531 /* Compaction is likely to fail */
532 return ret;
533 case COMPACT_CONTINUE:
534 /* Fall through to compaction */
535 ;
536 }
537
538 /* Setup to move all movable pages to the end of the zone */
539 cc->migrate_pfn = zone->zone_start_pfn;
540 cc->free_pfn = cc->migrate_pfn + zone->spanned_pages;
541 cc->free_pfn &= ~(pageblock_nr_pages-1);
542
543 migrate_prep_local();
544
545 while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) {
546 unsigned long nr_migrate, nr_remaining;
547 int err;
548
549 switch (isolate_migratepages(zone, cc)) {
550 case ISOLATE_ABORT:
551 ret = COMPACT_PARTIAL;
552 goto out;
553 case ISOLATE_NONE:
554 continue;
555 case ISOLATE_SUCCESS:
556 ;
557 }
558
559 nr_migrate = cc->nr_migratepages;
560 err = migrate_pages(&cc->migratepages, compaction_alloc,
561 (unsigned long)cc, false,
562 cc->sync);
563 update_nr_listpages(cc);
564 nr_remaining = cc->nr_migratepages;
565
566 count_vm_event(COMPACTBLOCKS);
567 count_vm_events(COMPACTPAGES, nr_migrate - nr_remaining);
568 if (nr_remaining)
569 count_vm_events(COMPACTPAGEFAILED, nr_remaining);
570 trace_mm_compaction_migratepages(nr_migrate - nr_remaining,
571 nr_remaining);
572
573 /* Release LRU pages not migrated */
574 if (err) {
575 putback_lru_pages(&cc->migratepages);
576 cc->nr_migratepages = 0;
577 }
578
579 }
580
581out:
582 /* Release free pages and check accounting */
583 cc->nr_freepages -= release_freepages(&cc->freepages);
584 VM_BUG_ON(cc->nr_freepages != 0);
585
586 return ret;
587}
588
589unsigned long compact_zone_order(struct zone *zone,
590 int order, gfp_t gfp_mask,
591 bool sync)
592{
593 struct compact_control cc = {
594 .nr_freepages = 0,
595 .nr_migratepages = 0,
596 .order = order,
597 .migratetype = allocflags_to_migratetype(gfp_mask),
598 .zone = zone,
599 .sync = sync,
600 };
601 INIT_LIST_HEAD(&cc.freepages);
602 INIT_LIST_HEAD(&cc.migratepages);
603
604 return compact_zone(zone, &cc);
605}
606
607int sysctl_extfrag_threshold = 500;
608
609/**
610 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
611 * @zonelist: The zonelist used for the current allocation
612 * @order: The order of the current allocation
613 * @gfp_mask: The GFP mask of the current allocation
614 * @nodemask: The allowed nodes to allocate from
615 * @sync: Whether migration is synchronous or not
616 *
617 * This is the main entry point for direct page compaction.
618 */
619unsigned long try_to_compact_pages(struct zonelist *zonelist,
620 int order, gfp_t gfp_mask, nodemask_t *nodemask,
621 bool sync)
622{
623 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
624 int may_enter_fs = gfp_mask & __GFP_FS;
625 int may_perform_io = gfp_mask & __GFP_IO;
626 struct zoneref *z;
627 struct zone *zone;
628 int rc = COMPACT_SKIPPED;
629
630 /*
631 * Check whether it is worth even starting compaction. The order check is
632 * made because an assumption is made that the page allocator can satisfy
633 * the "cheaper" orders without taking special steps
634 */
635 if (!order || !may_enter_fs || !may_perform_io)
636 return rc;
637
638 count_vm_event(COMPACTSTALL);
639
640 /* Compact each zone in the list */
641 for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx,
642 nodemask) {
643 int status;
644
645 status = compact_zone_order(zone, order, gfp_mask, sync);
646 rc = max(status, rc);
647
648 /* If a normal allocation would succeed, stop compacting */
649 if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, 0))
650 break;
651 }
652
653 return rc;
654}
655
656
657/* Compact all zones within a node */
658static int compact_node(int nid)
659{
660 int zoneid;
661 pg_data_t *pgdat;
662 struct zone *zone;
663
664 if (nid < 0 || nid >= nr_node_ids || !node_online(nid))
665 return -EINVAL;
666 pgdat = NODE_DATA(nid);
667
668 /* Flush pending updates to the LRU lists */
669 lru_add_drain_all();
670
671 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
672 struct compact_control cc = {
673 .nr_freepages = 0,
674 .nr_migratepages = 0,
675 .order = -1,
676 };
677
678 zone = &pgdat->node_zones[zoneid];
679 if (!populated_zone(zone))
680 continue;
681
682 cc.zone = zone;
683 INIT_LIST_HEAD(&cc.freepages);
684 INIT_LIST_HEAD(&cc.migratepages);
685
686 compact_zone(zone, &cc);
687
688 VM_BUG_ON(!list_empty(&cc.freepages));
689 VM_BUG_ON(!list_empty(&cc.migratepages));
690 }
691
692 return 0;
693}
694
695/* Compact all nodes in the system */
696static int compact_nodes(void)
697{
698 int nid;
699
700 for_each_online_node(nid)
701 compact_node(nid);
702
703 return COMPACT_COMPLETE;
704}
705
706/* The written value is actually unused, all memory is compacted */
707int sysctl_compact_memory;
708
709/* This is the entry point for compacting all nodes via /proc/sys/vm */
710int sysctl_compaction_handler(struct ctl_table *table, int write,
711 void __user *buffer, size_t *length, loff_t *ppos)
712{
713 if (write)
714 return compact_nodes();
715
716 return 0;
717}
718
719int sysctl_extfrag_handler(struct ctl_table *table, int write,
720 void __user *buffer, size_t *length, loff_t *ppos)
721{
722 proc_dointvec_minmax(table, write, buffer, length, ppos);
723
724 return 0;
725}
726
727#if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
728ssize_t sysfs_compact_node(struct sys_device *dev,
729 struct sysdev_attribute *attr,
730 const char *buf, size_t count)
731{
732 compact_node(dev->id);
733
734 return count;
735}
736static SYSDEV_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node);
737
738int compaction_register_node(struct node *node)
739{
740 return sysdev_create_file(&node->sysdev, &attr_compact);
741}
742
743void compaction_unregister_node(struct node *node)
744{
745 return sysdev_remove_file(&node->sysdev, &attr_compact);
746}
747#endif /* CONFIG_SYSFS && CONFIG_NUMA */
1/*
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>
15#include <linux/sysctl.h>
16#include <linux/sysfs.h>
17#include <linux/balloon_compaction.h>
18#include <linux/page-isolation.h>
19#include "internal.h"
20
21#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
36#if defined CONFIG_COMPACTION || defined CONFIG_CMA
37
38#define CREATE_TRACE_POINTS
39#include <trace/events/compaction.h>
40
41static 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
55static 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
65static inline bool migrate_async_suitable(int migratetype)
66{
67 return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE;
68}
69
70#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 */
86static void __reset_isolation_suitable(struct zone *zone)
87{
88 unsigned long start_pfn = zone->zone_start_pfn;
89 unsigned long end_pfn = zone_end_pfn(zone);
90 unsigned long pfn;
91
92 zone->compact_cached_migrate_pfn = start_pfn;
93 zone->compact_cached_free_pfn = end_pfn;
94 zone->compact_blockskip_flush = false;
95
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
113void 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
128/*
129 * If no pages were isolated then mark this pageblock to be skipped in the
130 * future. The information is later cleared by __reset_isolation_suitable().
131 */
132static void update_pageblock_skip(struct compact_control *cc,
133 struct page *page, unsigned long nr_isolated,
134 bool migrate_scanner)
135{
136 struct zone *zone = cc->zone;
137
138 if (cc->ignore_skip_hint)
139 return;
140
141 if (!page)
142 return;
143
144 if (!nr_isolated) {
145 unsigned long pfn = page_to_pfn(page);
146 set_pageblock_skip(page);
147
148 /* Update where compaction should restart */
149 if (migrate_scanner) {
150 if (!cc->finished_update_migrate &&
151 pfn > zone->compact_cached_migrate_pfn)
152 zone->compact_cached_migrate_pfn = pfn;
153 } else {
154 if (!cc->finished_update_free &&
155 pfn < zone->compact_cached_free_pfn)
156 zone->compact_cached_free_pfn = pfn;
157 }
158 }
159}
160#else
161static inline bool isolation_suitable(struct compact_control *cc,
162 struct page *page)
163{
164 return true;
165}
166
167static void update_pageblock_skip(struct compact_control *cc,
168 struct page *page, unsigned long nr_isolated,
169 bool migrate_scanner)
170{
171}
172#endif /* CONFIG_COMPACTION */
173
174static inline bool should_release_lock(spinlock_t *lock)
175{
176 return need_resched() || spin_is_contended(lock);
177}
178
179/*
180 * Compaction requires the taking of some coarse locks that are potentially
181 * very heavily contended. Check if the process needs to be scheduled or
182 * if the lock is contended. For async compaction, back out in the event
183 * if contention is severe. For sync compaction, schedule.
184 *
185 * Returns true if the lock is held.
186 * Returns false if the lock is released and compaction should abort
187 */
188static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags,
189 bool locked, struct compact_control *cc)
190{
191 if (should_release_lock(lock)) {
192 if (locked) {
193 spin_unlock_irqrestore(lock, *flags);
194 locked = false;
195 }
196
197 /* async aborts if taking too long or contended */
198 if (!cc->sync) {
199 cc->contended = true;
200 return false;
201 }
202
203 cond_resched();
204 }
205
206 if (!locked)
207 spin_lock_irqsave(lock, *flags);
208 return true;
209}
210
211static inline bool compact_trylock_irqsave(spinlock_t *lock,
212 unsigned long *flags, struct compact_control *cc)
213{
214 return compact_checklock_irqsave(lock, flags, false, cc);
215}
216
217/* Returns true if the page is within a block suitable for migration to */
218static bool suitable_migration_target(struct page *page)
219{
220 /* If the page is a large free page, then disallow migration */
221 if (PageBuddy(page) && page_order(page) >= pageblock_order)
222 return false;
223
224 /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
225 if (migrate_async_suitable(get_pageblock_migratetype(page)))
226 return true;
227
228 /* Otherwise skip the block */
229 return false;
230}
231
232/*
233 * Isolate free pages onto a private freelist. If @strict is true, will abort
234 * returning 0 on any invalid PFNs or non-free pages inside of the pageblock
235 * (even though it may still end up isolating some pages).
236 */
237static unsigned long isolate_freepages_block(struct compact_control *cc,
238 unsigned long blockpfn,
239 unsigned long end_pfn,
240 struct list_head *freelist,
241 bool strict)
242{
243 int nr_scanned = 0, total_isolated = 0;
244 struct page *cursor, *valid_page = NULL;
245 unsigned long flags;
246 bool locked = false;
247 bool checked_pageblock = false;
248
249 cursor = pfn_to_page(blockpfn);
250
251 /* Isolate free pages. */
252 for (; blockpfn < end_pfn; blockpfn++, cursor++) {
253 int isolated, i;
254 struct page *page = cursor;
255
256 nr_scanned++;
257 if (!pfn_valid_within(blockpfn))
258 goto isolate_fail;
259
260 if (!valid_page)
261 valid_page = page;
262 if (!PageBuddy(page))
263 goto isolate_fail;
264
265 /*
266 * The zone lock must be held to isolate freepages.
267 * Unfortunately this is a very coarse lock and can be
268 * heavily contended if there are parallel allocations
269 * or parallel compactions. For async compaction do not
270 * spin on the lock and we acquire the lock as late as
271 * possible.
272 */
273 locked = compact_checklock_irqsave(&cc->zone->lock, &flags,
274 locked, cc);
275 if (!locked)
276 break;
277
278 /* Recheck this is a suitable migration target under lock */
279 if (!strict && !checked_pageblock) {
280 /*
281 * We need to check suitability of pageblock only once
282 * and this isolate_freepages_block() is called with
283 * pageblock range, so just check once is sufficient.
284 */
285 checked_pageblock = true;
286 if (!suitable_migration_target(page))
287 break;
288 }
289
290 /* Recheck this is a buddy page under lock */
291 if (!PageBuddy(page))
292 goto isolate_fail;
293
294 /* Found a free page, break it into order-0 pages */
295 isolated = split_free_page(page);
296 total_isolated += isolated;
297 for (i = 0; i < isolated; i++) {
298 list_add(&page->lru, freelist);
299 page++;
300 }
301
302 /* If a page was split, advance to the end of it */
303 if (isolated) {
304 blockpfn += isolated - 1;
305 cursor += isolated - 1;
306 continue;
307 }
308
309isolate_fail:
310 if (strict)
311 break;
312 else
313 continue;
314
315 }
316
317 trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated);
318
319 /*
320 * If strict isolation is requested by CMA then check that all the
321 * pages requested were isolated. If there were any failures, 0 is
322 * returned and CMA will fail.
323 */
324 if (strict && blockpfn < end_pfn)
325 total_isolated = 0;
326
327 if (locked)
328 spin_unlock_irqrestore(&cc->zone->lock, flags);
329
330 /* Update the pageblock-skip if the whole pageblock was scanned */
331 if (blockpfn == end_pfn)
332 update_pageblock_skip(cc, valid_page, total_isolated, false);
333
334 count_compact_events(COMPACTFREE_SCANNED, nr_scanned);
335 if (total_isolated)
336 count_compact_events(COMPACTISOLATED, total_isolated);
337 return total_isolated;
338}
339
340/**
341 * isolate_freepages_range() - isolate free pages.
342 * @start_pfn: The first PFN to start isolating.
343 * @end_pfn: The one-past-last PFN.
344 *
345 * Non-free pages, invalid PFNs, or zone boundaries within the
346 * [start_pfn, end_pfn) range are considered errors, cause function to
347 * undo its actions and return zero.
348 *
349 * Otherwise, function returns one-past-the-last PFN of isolated page
350 * (which may be greater then end_pfn if end fell in a middle of
351 * a free page).
352 */
353unsigned long
354isolate_freepages_range(struct compact_control *cc,
355 unsigned long start_pfn, unsigned long end_pfn)
356{
357 unsigned long isolated, pfn, block_end_pfn;
358 LIST_HEAD(freelist);
359
360 for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) {
361 if (!pfn_valid(pfn) || cc->zone != page_zone(pfn_to_page(pfn)))
362 break;
363
364 /*
365 * On subsequent iterations ALIGN() is actually not needed,
366 * but we keep it that we not to complicate the code.
367 */
368 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
369 block_end_pfn = min(block_end_pfn, end_pfn);
370
371 isolated = isolate_freepages_block(cc, pfn, block_end_pfn,
372 &freelist, true);
373
374 /*
375 * In strict mode, isolate_freepages_block() returns 0 if
376 * there are any holes in the block (ie. invalid PFNs or
377 * non-free pages).
378 */
379 if (!isolated)
380 break;
381
382 /*
383 * If we managed to isolate pages, it is always (1 << n) *
384 * pageblock_nr_pages for some non-negative n. (Max order
385 * page may span two pageblocks).
386 */
387 }
388
389 /* split_free_page does not map the pages */
390 map_pages(&freelist);
391
392 if (pfn < end_pfn) {
393 /* Loop terminated early, cleanup. */
394 release_freepages(&freelist);
395 return 0;
396 }
397
398 /* We don't use freelists for anything. */
399 return pfn;
400}
401
402/* Update the number of anon and file isolated pages in the zone */
403static void acct_isolated(struct zone *zone, bool locked, struct compact_control *cc)
404{
405 struct page *page;
406 unsigned int count[2] = { 0, };
407
408 list_for_each_entry(page, &cc->migratepages, lru)
409 count[!!page_is_file_cache(page)]++;
410
411 /* If locked we can use the interrupt unsafe versions */
412 if (locked) {
413 __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
414 __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
415 } else {
416 mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
417 mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
418 }
419}
420
421/* Similar to reclaim, but different enough that they don't share logic */
422static bool too_many_isolated(struct zone *zone)
423{
424 unsigned long active, inactive, isolated;
425
426 inactive = zone_page_state(zone, NR_INACTIVE_FILE) +
427 zone_page_state(zone, NR_INACTIVE_ANON);
428 active = zone_page_state(zone, NR_ACTIVE_FILE) +
429 zone_page_state(zone, NR_ACTIVE_ANON);
430 isolated = zone_page_state(zone, NR_ISOLATED_FILE) +
431 zone_page_state(zone, NR_ISOLATED_ANON);
432
433 return isolated > (inactive + active) / 2;
434}
435
436/**
437 * isolate_migratepages_range() - isolate all migrate-able pages in range.
438 * @zone: Zone pages are in.
439 * @cc: Compaction control structure.
440 * @low_pfn: The first PFN of the range.
441 * @end_pfn: The one-past-the-last PFN of the range.
442 * @unevictable: true if it allows to isolate unevictable pages
443 *
444 * Isolate all pages that can be migrated from the range specified by
445 * [low_pfn, end_pfn). Returns zero if there is a fatal signal
446 * pending), otherwise PFN of the first page that was not scanned
447 * (which may be both less, equal to or more then end_pfn).
448 *
449 * Assumes that cc->migratepages is empty and cc->nr_migratepages is
450 * zero.
451 *
452 * Apart from cc->migratepages and cc->nr_migratetypes this function
453 * does not modify any cc's fields, in particular it does not modify
454 * (or read for that matter) cc->migrate_pfn.
455 */
456unsigned long
457isolate_migratepages_range(struct zone *zone, struct compact_control *cc,
458 unsigned long low_pfn, unsigned long end_pfn, bool unevictable)
459{
460 unsigned long last_pageblock_nr = 0, pageblock_nr;
461 unsigned long nr_scanned = 0, nr_isolated = 0;
462 struct list_head *migratelist = &cc->migratepages;
463 struct lruvec *lruvec;
464 unsigned long flags;
465 bool locked = false;
466 struct page *page = NULL, *valid_page = NULL;
467 bool skipped_async_unsuitable = false;
468 const isolate_mode_t mode = (!cc->sync ? ISOLATE_ASYNC_MIGRATE : 0) |
469 (unevictable ? ISOLATE_UNEVICTABLE : 0);
470
471 /*
472 * Ensure that there are not too many pages isolated from the LRU
473 * list by either parallel reclaimers or compaction. If there are,
474 * delay for some time until fewer pages are isolated
475 */
476 while (unlikely(too_many_isolated(zone))) {
477 /* async migration should just abort */
478 if (!cc->sync)
479 return 0;
480
481 congestion_wait(BLK_RW_ASYNC, HZ/10);
482
483 if (fatal_signal_pending(current))
484 return 0;
485 }
486
487 /* Time to isolate some pages for migration */
488 cond_resched();
489 for (; low_pfn < end_pfn; low_pfn++) {
490 /* give a chance to irqs before checking need_resched() */
491 if (locked && !(low_pfn % SWAP_CLUSTER_MAX)) {
492 if (should_release_lock(&zone->lru_lock)) {
493 spin_unlock_irqrestore(&zone->lru_lock, flags);
494 locked = false;
495 }
496 }
497
498 /*
499 * migrate_pfn does not necessarily start aligned to a
500 * pageblock. Ensure that pfn_valid is called when moving
501 * into a new MAX_ORDER_NR_PAGES range in case of large
502 * memory holes within the zone
503 */
504 if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
505 if (!pfn_valid(low_pfn)) {
506 low_pfn += MAX_ORDER_NR_PAGES - 1;
507 continue;
508 }
509 }
510
511 if (!pfn_valid_within(low_pfn))
512 continue;
513 nr_scanned++;
514
515 /*
516 * Get the page and ensure the page is within the same zone.
517 * See the comment in isolate_freepages about overlapping
518 * nodes. It is deliberate that the new zone lock is not taken
519 * as memory compaction should not move pages between nodes.
520 */
521 page = pfn_to_page(low_pfn);
522 if (page_zone(page) != zone)
523 continue;
524
525 if (!valid_page)
526 valid_page = page;
527
528 /* If isolation recently failed, do not retry */
529 pageblock_nr = low_pfn >> pageblock_order;
530 if (last_pageblock_nr != pageblock_nr) {
531 int mt;
532
533 last_pageblock_nr = pageblock_nr;
534 if (!isolation_suitable(cc, page))
535 goto next_pageblock;
536
537 /*
538 * For async migration, also only scan in MOVABLE
539 * blocks. Async migration is optimistic to see if
540 * the minimum amount of work satisfies the allocation
541 */
542 mt = get_pageblock_migratetype(page);
543 if (!cc->sync && !migrate_async_suitable(mt)) {
544 cc->finished_update_migrate = true;
545 skipped_async_unsuitable = true;
546 goto next_pageblock;
547 }
548 }
549
550 /*
551 * Skip if free. page_order cannot be used without zone->lock
552 * as nothing prevents parallel allocations or buddy merging.
553 */
554 if (PageBuddy(page))
555 continue;
556
557 /*
558 * Check may be lockless but that's ok as we recheck later.
559 * It's possible to migrate LRU pages and balloon pages
560 * Skip any other type of page
561 */
562 if (!PageLRU(page)) {
563 if (unlikely(balloon_page_movable(page))) {
564 if (locked && balloon_page_isolate(page)) {
565 /* Successfully isolated */
566 goto isolate_success;
567 }
568 }
569 continue;
570 }
571
572 /*
573 * PageLRU is set. lru_lock normally excludes isolation
574 * splitting and collapsing (collapsing has already happened
575 * if PageLRU is set) but the lock is not necessarily taken
576 * here and it is wasteful to take it just to check transhuge.
577 * Check TransHuge without lock and skip the whole pageblock if
578 * it's either a transhuge or hugetlbfs page, as calling
579 * compound_order() without preventing THP from splitting the
580 * page underneath us may return surprising results.
581 */
582 if (PageTransHuge(page)) {
583 if (!locked)
584 goto next_pageblock;
585 low_pfn += (1 << compound_order(page)) - 1;
586 continue;
587 }
588
589 /*
590 * Migration will fail if an anonymous page is pinned in memory,
591 * so avoid taking lru_lock and isolating it unnecessarily in an
592 * admittedly racy check.
593 */
594 if (!page_mapping(page) &&
595 page_count(page) > page_mapcount(page))
596 continue;
597
598 /* Check if it is ok to still hold the lock */
599 locked = compact_checklock_irqsave(&zone->lru_lock, &flags,
600 locked, cc);
601 if (!locked || fatal_signal_pending(current))
602 break;
603
604 /* Recheck PageLRU and PageTransHuge under lock */
605 if (!PageLRU(page))
606 continue;
607 if (PageTransHuge(page)) {
608 low_pfn += (1 << compound_order(page)) - 1;
609 continue;
610 }
611
612 lruvec = mem_cgroup_page_lruvec(page, zone);
613
614 /* Try isolate the page */
615 if (__isolate_lru_page(page, mode) != 0)
616 continue;
617
618 VM_BUG_ON_PAGE(PageTransCompound(page), page);
619
620 /* Successfully isolated */
621 del_page_from_lru_list(page, lruvec, page_lru(page));
622
623isolate_success:
624 cc->finished_update_migrate = true;
625 list_add(&page->lru, migratelist);
626 cc->nr_migratepages++;
627 nr_isolated++;
628
629 /* Avoid isolating too much */
630 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) {
631 ++low_pfn;
632 break;
633 }
634
635 continue;
636
637next_pageblock:
638 low_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages) - 1;
639 }
640
641 acct_isolated(zone, locked, cc);
642
643 if (locked)
644 spin_unlock_irqrestore(&zone->lru_lock, flags);
645
646 /*
647 * Update the pageblock-skip information and cached scanner pfn,
648 * if the whole pageblock was scanned without isolating any page.
649 * This is not done when pageblock was skipped due to being unsuitable
650 * for async compaction, so that eventual sync compaction can try.
651 */
652 if (low_pfn == end_pfn && !skipped_async_unsuitable)
653 update_pageblock_skip(cc, valid_page, nr_isolated, true);
654
655 trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated);
656
657 count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned);
658 if (nr_isolated)
659 count_compact_events(COMPACTISOLATED, nr_isolated);
660
661 return low_pfn;
662}
663
664#endif /* CONFIG_COMPACTION || CONFIG_CMA */
665#ifdef CONFIG_COMPACTION
666/*
667 * Based on information in the current compact_control, find blocks
668 * suitable for isolating free pages from and then isolate them.
669 */
670static void isolate_freepages(struct zone *zone,
671 struct compact_control *cc)
672{
673 struct page *page;
674 unsigned long high_pfn, low_pfn, pfn, z_end_pfn;
675 int nr_freepages = cc->nr_freepages;
676 struct list_head *freelist = &cc->freepages;
677
678 /*
679 * Initialise the free scanner. The starting point is where we last
680 * successfully isolated from, zone-cached value, or the end of the
681 * zone when isolating for the first time. We need this aligned to
682 * the pageblock boundary, because we do pfn -= pageblock_nr_pages
683 * in the for loop.
684 * The low boundary is the end of the pageblock the migration scanner
685 * is using.
686 */
687 pfn = cc->free_pfn & ~(pageblock_nr_pages-1);
688 low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages);
689
690 /*
691 * Take care that if the migration scanner is at the end of the zone
692 * that the free scanner does not accidentally move to the next zone
693 * in the next isolation cycle.
694 */
695 high_pfn = min(low_pfn, pfn);
696
697 z_end_pfn = zone_end_pfn(zone);
698
699 /*
700 * Isolate free pages until enough are available to migrate the
701 * pages on cc->migratepages. We stop searching if the migrate
702 * and free page scanners meet or enough free pages are isolated.
703 */
704 for (; pfn >= low_pfn && cc->nr_migratepages > nr_freepages;
705 pfn -= pageblock_nr_pages) {
706 unsigned long isolated;
707 unsigned long end_pfn;
708
709 /*
710 * This can iterate a massively long zone without finding any
711 * suitable migration targets, so periodically check if we need
712 * to schedule.
713 */
714 cond_resched();
715
716 if (!pfn_valid(pfn))
717 continue;
718
719 /*
720 * Check for overlapping nodes/zones. It's possible on some
721 * configurations to have a setup like
722 * node0 node1 node0
723 * i.e. it's possible that all pages within a zones range of
724 * pages do not belong to a single zone.
725 */
726 page = pfn_to_page(pfn);
727 if (page_zone(page) != zone)
728 continue;
729
730 /* Check the block is suitable for migration */
731 if (!suitable_migration_target(page))
732 continue;
733
734 /* If isolation recently failed, do not retry */
735 if (!isolation_suitable(cc, page))
736 continue;
737
738 /* Found a block suitable for isolating free pages from */
739 isolated = 0;
740
741 /*
742 * Take care when isolating in last pageblock of a zone which
743 * ends in the middle of a pageblock.
744 */
745 end_pfn = min(pfn + pageblock_nr_pages, z_end_pfn);
746 isolated = isolate_freepages_block(cc, pfn, end_pfn,
747 freelist, false);
748 nr_freepages += isolated;
749
750 /*
751 * Record the highest PFN we isolated pages from. When next
752 * looking for free pages, the search will restart here as
753 * page migration may have returned some pages to the allocator
754 */
755 if (isolated) {
756 cc->finished_update_free = true;
757 high_pfn = max(high_pfn, pfn);
758 }
759 }
760
761 /* split_free_page does not map the pages */
762 map_pages(freelist);
763
764 /*
765 * If we crossed the migrate scanner, we want to keep it that way
766 * so that compact_finished() may detect this
767 */
768 if (pfn < low_pfn)
769 cc->free_pfn = max(pfn, zone->zone_start_pfn);
770 else
771 cc->free_pfn = high_pfn;
772 cc->nr_freepages = nr_freepages;
773}
774
775/*
776 * This is a migrate-callback that "allocates" freepages by taking pages
777 * from the isolated freelists in the block we are migrating to.
778 */
779static struct page *compaction_alloc(struct page *migratepage,
780 unsigned long data,
781 int **result)
782{
783 struct compact_control *cc = (struct compact_control *)data;
784 struct page *freepage;
785
786 /* Isolate free pages if necessary */
787 if (list_empty(&cc->freepages)) {
788 isolate_freepages(cc->zone, cc);
789
790 if (list_empty(&cc->freepages))
791 return NULL;
792 }
793
794 freepage = list_entry(cc->freepages.next, struct page, lru);
795 list_del(&freepage->lru);
796 cc->nr_freepages--;
797
798 return freepage;
799}
800
801/*
802 * We cannot control nr_migratepages and nr_freepages fully when migration is
803 * running as migrate_pages() has no knowledge of compact_control. When
804 * migration is complete, we count the number of pages on the lists by hand.
805 */
806static void update_nr_listpages(struct compact_control *cc)
807{
808 int nr_migratepages = 0;
809 int nr_freepages = 0;
810 struct page *page;
811
812 list_for_each_entry(page, &cc->migratepages, lru)
813 nr_migratepages++;
814 list_for_each_entry(page, &cc->freepages, lru)
815 nr_freepages++;
816
817 cc->nr_migratepages = nr_migratepages;
818 cc->nr_freepages = nr_freepages;
819}
820
821/* possible outcome of isolate_migratepages */
822typedef enum {
823 ISOLATE_ABORT, /* Abort compaction now */
824 ISOLATE_NONE, /* No pages isolated, continue scanning */
825 ISOLATE_SUCCESS, /* Pages isolated, migrate */
826} isolate_migrate_t;
827
828/*
829 * Isolate all pages that can be migrated from the block pointed to by
830 * the migrate scanner within compact_control.
831 */
832static isolate_migrate_t isolate_migratepages(struct zone *zone,
833 struct compact_control *cc)
834{
835 unsigned long low_pfn, end_pfn;
836
837 /* Do not scan outside zone boundaries */
838 low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn);
839
840 /* Only scan within a pageblock boundary */
841 end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages);
842
843 /* Do not cross the free scanner or scan within a memory hole */
844 if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) {
845 cc->migrate_pfn = end_pfn;
846 return ISOLATE_NONE;
847 }
848
849 /* Perform the isolation */
850 low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn, false);
851 if (!low_pfn || cc->contended)
852 return ISOLATE_ABORT;
853
854 cc->migrate_pfn = low_pfn;
855
856 return ISOLATE_SUCCESS;
857}
858
859static int compact_finished(struct zone *zone,
860 struct compact_control *cc)
861{
862 unsigned int order;
863 unsigned long watermark;
864
865 if (fatal_signal_pending(current))
866 return COMPACT_PARTIAL;
867
868 /* Compaction run completes if the migrate and free scanner meet */
869 if (cc->free_pfn <= cc->migrate_pfn) {
870 /* Let the next compaction start anew. */
871 zone->compact_cached_migrate_pfn = zone->zone_start_pfn;
872 zone->compact_cached_free_pfn = zone_end_pfn(zone);
873
874 /*
875 * Mark that the PG_migrate_skip information should be cleared
876 * by kswapd when it goes to sleep. kswapd does not set the
877 * flag itself as the decision to be clear should be directly
878 * based on an allocation request.
879 */
880 if (!current_is_kswapd())
881 zone->compact_blockskip_flush = true;
882
883 return COMPACT_COMPLETE;
884 }
885
886 /*
887 * order == -1 is expected when compacting via
888 * /proc/sys/vm/compact_memory
889 */
890 if (cc->order == -1)
891 return COMPACT_CONTINUE;
892
893 /* Compaction run is not finished if the watermark is not met */
894 watermark = low_wmark_pages(zone);
895 watermark += (1 << cc->order);
896
897 if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0))
898 return COMPACT_CONTINUE;
899
900 /* Direct compactor: Is a suitable page free? */
901 for (order = cc->order; order < MAX_ORDER; order++) {
902 struct free_area *area = &zone->free_area[order];
903
904 /* Job done if page is free of the right migratetype */
905 if (!list_empty(&area->free_list[cc->migratetype]))
906 return COMPACT_PARTIAL;
907
908 /* Job done if allocation would set block type */
909 if (cc->order >= pageblock_order && area->nr_free)
910 return COMPACT_PARTIAL;
911 }
912
913 return COMPACT_CONTINUE;
914}
915
916/*
917 * compaction_suitable: Is this suitable to run compaction on this zone now?
918 * Returns
919 * COMPACT_SKIPPED - If there are too few free pages for compaction
920 * COMPACT_PARTIAL - If the allocation would succeed without compaction
921 * COMPACT_CONTINUE - If compaction should run now
922 */
923unsigned long compaction_suitable(struct zone *zone, int order)
924{
925 int fragindex;
926 unsigned long watermark;
927
928 /*
929 * order == -1 is expected when compacting via
930 * /proc/sys/vm/compact_memory
931 */
932 if (order == -1)
933 return COMPACT_CONTINUE;
934
935 /*
936 * Watermarks for order-0 must be met for compaction. Note the 2UL.
937 * This is because during migration, copies of pages need to be
938 * allocated and for a short time, the footprint is higher
939 */
940 watermark = low_wmark_pages(zone) + (2UL << order);
941 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
942 return COMPACT_SKIPPED;
943
944 /*
945 * fragmentation index determines if allocation failures are due to
946 * low memory or external fragmentation
947 *
948 * index of -1000 implies allocations might succeed depending on
949 * watermarks
950 * index towards 0 implies failure is due to lack of memory
951 * index towards 1000 implies failure is due to fragmentation
952 *
953 * Only compact if a failure would be due to fragmentation.
954 */
955 fragindex = fragmentation_index(zone, order);
956 if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
957 return COMPACT_SKIPPED;
958
959 if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark,
960 0, 0))
961 return COMPACT_PARTIAL;
962
963 return COMPACT_CONTINUE;
964}
965
966static int compact_zone(struct zone *zone, struct compact_control *cc)
967{
968 int ret;
969 unsigned long start_pfn = zone->zone_start_pfn;
970 unsigned long end_pfn = zone_end_pfn(zone);
971
972 ret = compaction_suitable(zone, cc->order);
973 switch (ret) {
974 case COMPACT_PARTIAL:
975 case COMPACT_SKIPPED:
976 /* Compaction is likely to fail */
977 return ret;
978 case COMPACT_CONTINUE:
979 /* Fall through to compaction */
980 ;
981 }
982
983 /*
984 * Clear pageblock skip if there were failures recently and compaction
985 * is about to be retried after being deferred. kswapd does not do
986 * this reset as it'll reset the cached information when going to sleep.
987 */
988 if (compaction_restarting(zone, cc->order) && !current_is_kswapd())
989 __reset_isolation_suitable(zone);
990
991 /*
992 * Setup to move all movable pages to the end of the zone. Used cached
993 * information on where the scanners should start but check that it
994 * is initialised by ensuring the values are within zone boundaries.
995 */
996 cc->migrate_pfn = zone->compact_cached_migrate_pfn;
997 cc->free_pfn = zone->compact_cached_free_pfn;
998 if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) {
999 cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1);
1000 zone->compact_cached_free_pfn = cc->free_pfn;
1001 }
1002 if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) {
1003 cc->migrate_pfn = start_pfn;
1004 zone->compact_cached_migrate_pfn = cc->migrate_pfn;
1005 }
1006
1007 trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, cc->free_pfn, end_pfn);
1008
1009 migrate_prep_local();
1010
1011 while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) {
1012 unsigned long nr_migrate, nr_remaining;
1013 int err;
1014
1015 switch (isolate_migratepages(zone, cc)) {
1016 case ISOLATE_ABORT:
1017 ret = COMPACT_PARTIAL;
1018 putback_movable_pages(&cc->migratepages);
1019 cc->nr_migratepages = 0;
1020 goto out;
1021 case ISOLATE_NONE:
1022 continue;
1023 case ISOLATE_SUCCESS:
1024 ;
1025 }
1026
1027 nr_migrate = cc->nr_migratepages;
1028 err = migrate_pages(&cc->migratepages, compaction_alloc,
1029 (unsigned long)cc,
1030 cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC,
1031 MR_COMPACTION);
1032 update_nr_listpages(cc);
1033 nr_remaining = cc->nr_migratepages;
1034
1035 trace_mm_compaction_migratepages(nr_migrate - nr_remaining,
1036 nr_remaining);
1037
1038 /* Release isolated pages not migrated */
1039 if (err) {
1040 putback_movable_pages(&cc->migratepages);
1041 cc->nr_migratepages = 0;
1042 /*
1043 * migrate_pages() may return -ENOMEM when scanners meet
1044 * and we want compact_finished() to detect it
1045 */
1046 if (err == -ENOMEM && cc->free_pfn > cc->migrate_pfn) {
1047 ret = COMPACT_PARTIAL;
1048 goto out;
1049 }
1050 }
1051 }
1052
1053out:
1054 /* Release free pages and check accounting */
1055 cc->nr_freepages -= release_freepages(&cc->freepages);
1056 VM_BUG_ON(cc->nr_freepages != 0);
1057
1058 trace_mm_compaction_end(ret);
1059
1060 return ret;
1061}
1062
1063static unsigned long compact_zone_order(struct zone *zone,
1064 int order, gfp_t gfp_mask,
1065 bool sync, bool *contended)
1066{
1067 unsigned long ret;
1068 struct compact_control cc = {
1069 .nr_freepages = 0,
1070 .nr_migratepages = 0,
1071 .order = order,
1072 .migratetype = allocflags_to_migratetype(gfp_mask),
1073 .zone = zone,
1074 .sync = sync,
1075 };
1076 INIT_LIST_HEAD(&cc.freepages);
1077 INIT_LIST_HEAD(&cc.migratepages);
1078
1079 ret = compact_zone(zone, &cc);
1080
1081 VM_BUG_ON(!list_empty(&cc.freepages));
1082 VM_BUG_ON(!list_empty(&cc.migratepages));
1083
1084 *contended = cc.contended;
1085 return ret;
1086}
1087
1088int sysctl_extfrag_threshold = 500;
1089
1090/**
1091 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
1092 * @zonelist: The zonelist used for the current allocation
1093 * @order: The order of the current allocation
1094 * @gfp_mask: The GFP mask of the current allocation
1095 * @nodemask: The allowed nodes to allocate from
1096 * @sync: Whether migration is synchronous or not
1097 * @contended: Return value that is true if compaction was aborted due to lock contention
1098 * @page: Optionally capture a free page of the requested order during compaction
1099 *
1100 * This is the main entry point for direct page compaction.
1101 */
1102unsigned long try_to_compact_pages(struct zonelist *zonelist,
1103 int order, gfp_t gfp_mask, nodemask_t *nodemask,
1104 bool sync, bool *contended)
1105{
1106 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
1107 int may_enter_fs = gfp_mask & __GFP_FS;
1108 int may_perform_io = gfp_mask & __GFP_IO;
1109 struct zoneref *z;
1110 struct zone *zone;
1111 int rc = COMPACT_SKIPPED;
1112 int alloc_flags = 0;
1113
1114 /* Check if the GFP flags allow compaction */
1115 if (!order || !may_enter_fs || !may_perform_io)
1116 return rc;
1117
1118 count_compact_event(COMPACTSTALL);
1119
1120#ifdef CONFIG_CMA
1121 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
1122 alloc_flags |= ALLOC_CMA;
1123#endif
1124 /* Compact each zone in the list */
1125 for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx,
1126 nodemask) {
1127 int status;
1128
1129 status = compact_zone_order(zone, order, gfp_mask, sync,
1130 contended);
1131 rc = max(status, rc);
1132
1133 /* If a normal allocation would succeed, stop compacting */
1134 if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0,
1135 alloc_flags))
1136 break;
1137 }
1138
1139 return rc;
1140}
1141
1142
1143/* Compact all zones within a node */
1144static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc)
1145{
1146 int zoneid;
1147 struct zone *zone;
1148
1149 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
1150
1151 zone = &pgdat->node_zones[zoneid];
1152 if (!populated_zone(zone))
1153 continue;
1154
1155 cc->nr_freepages = 0;
1156 cc->nr_migratepages = 0;
1157 cc->zone = zone;
1158 INIT_LIST_HEAD(&cc->freepages);
1159 INIT_LIST_HEAD(&cc->migratepages);
1160
1161 if (cc->order == -1 || !compaction_deferred(zone, cc->order))
1162 compact_zone(zone, cc);
1163
1164 if (cc->order > 0) {
1165 if (zone_watermark_ok(zone, cc->order,
1166 low_wmark_pages(zone), 0, 0))
1167 compaction_defer_reset(zone, cc->order, false);
1168 /* Currently async compaction is never deferred. */
1169 else if (cc->sync)
1170 defer_compaction(zone, cc->order);
1171 }
1172
1173 VM_BUG_ON(!list_empty(&cc->freepages));
1174 VM_BUG_ON(!list_empty(&cc->migratepages));
1175 }
1176}
1177
1178void compact_pgdat(pg_data_t *pgdat, int order)
1179{
1180 struct compact_control cc = {
1181 .order = order,
1182 .sync = false,
1183 };
1184
1185 if (!order)
1186 return;
1187
1188 __compact_pgdat(pgdat, &cc);
1189}
1190
1191static void compact_node(int nid)
1192{
1193 struct compact_control cc = {
1194 .order = -1,
1195 .sync = true,
1196 .ignore_skip_hint = true,
1197 };
1198
1199 __compact_pgdat(NODE_DATA(nid), &cc);
1200}
1201
1202/* Compact all nodes in the system */
1203static void compact_nodes(void)
1204{
1205 int nid;
1206
1207 /* Flush pending updates to the LRU lists */
1208 lru_add_drain_all();
1209
1210 for_each_online_node(nid)
1211 compact_node(nid);
1212}
1213
1214/* The written value is actually unused, all memory is compacted */
1215int sysctl_compact_memory;
1216
1217/* This is the entry point for compacting all nodes via /proc/sys/vm */
1218int sysctl_compaction_handler(struct ctl_table *table, int write,
1219 void __user *buffer, size_t *length, loff_t *ppos)
1220{
1221 if (write)
1222 compact_nodes();
1223
1224 return 0;
1225}
1226
1227int sysctl_extfrag_handler(struct ctl_table *table, int write,
1228 void __user *buffer, size_t *length, loff_t *ppos)
1229{
1230 proc_dointvec_minmax(table, write, buffer, length, ppos);
1231
1232 return 0;
1233}
1234
1235#if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
1236static ssize_t sysfs_compact_node(struct device *dev,
1237 struct device_attribute *attr,
1238 const char *buf, size_t count)
1239{
1240 int nid = dev->id;
1241
1242 if (nid >= 0 && nid < nr_node_ids && node_online(nid)) {
1243 /* Flush pending updates to the LRU lists */
1244 lru_add_drain_all();
1245
1246 compact_node(nid);
1247 }
1248
1249 return count;
1250}
1251static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node);
1252
1253int compaction_register_node(struct node *node)
1254{
1255 return device_create_file(&node->dev, &dev_attr_compact);
1256}
1257
1258void compaction_unregister_node(struct node *node)
1259{
1260 return device_remove_file(&node->dev, &dev_attr_compact);
1261}
1262#endif /* CONFIG_SYSFS && CONFIG_NUMA */
1263
1264#endif /* CONFIG_COMPACTION */