Commit | Line | Data |
---|---|---|
748446bb MG |
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> | |
76ab0f53 | 15 | #include <linux/sysctl.h> |
ed4a6d7f | 16 | #include <linux/sysfs.h> |
748446bb MG |
17 | #include "internal.h" |
18 | ||
b7aba698 MG |
19 | #define CREATE_TRACE_POINTS |
20 | #include <trace/events/compaction.h> | |
21 | ||
748446bb MG |
22 | /* |
23 | * compact_control is used to track pages being migrated and the free pages | |
24 | * they are being migrated to during memory compaction. The free_pfn starts | |
25 | * at the end of a zone and migrate_pfn begins at the start. Movable pages | |
26 | * are moved to the end of a zone during a compaction run and the run | |
27 | * completes when free_pfn <= migrate_pfn | |
28 | */ | |
29 | struct compact_control { | |
30 | struct list_head freepages; /* List of free pages to migrate to */ | |
31 | struct list_head migratepages; /* List of pages being migrated */ | |
32 | unsigned long nr_freepages; /* Number of isolated free pages */ | |
33 | unsigned long nr_migratepages; /* Number of pages to migrate */ | |
34 | unsigned long free_pfn; /* isolate_freepages search base */ | |
35 | unsigned long migrate_pfn; /* isolate_migratepages search base */ | |
77f1fe6b | 36 | bool sync; /* Synchronous migration */ |
748446bb | 37 | |
aad6ec37 | 38 | int order; /* order a direct compactor needs */ |
56de7263 | 39 | int migratetype; /* MOVABLE, RECLAIMABLE etc */ |
748446bb MG |
40 | struct zone *zone; |
41 | }; | |
42 | ||
43 | static unsigned long release_freepages(struct list_head *freelist) | |
44 | { | |
45 | struct page *page, *next; | |
46 | unsigned long count = 0; | |
47 | ||
48 | list_for_each_entry_safe(page, next, freelist, lru) { | |
49 | list_del(&page->lru); | |
50 | __free_page(page); | |
51 | count++; | |
52 | } | |
53 | ||
54 | return count; | |
55 | } | |
56 | ||
57 | /* Isolate free pages onto a private freelist. Must hold zone->lock */ | |
58 | static unsigned long isolate_freepages_block(struct zone *zone, | |
59 | unsigned long blockpfn, | |
60 | struct list_head *freelist) | |
61 | { | |
62 | unsigned long zone_end_pfn, end_pfn; | |
b7aba698 | 63 | int nr_scanned = 0, total_isolated = 0; |
748446bb MG |
64 | struct page *cursor; |
65 | ||
66 | /* Get the last PFN we should scan for free pages at */ | |
67 | zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
68 | end_pfn = min(blockpfn + pageblock_nr_pages, zone_end_pfn); | |
69 | ||
70 | /* Find the first usable PFN in the block to initialse page cursor */ | |
71 | for (; blockpfn < end_pfn; blockpfn++) { | |
72 | if (pfn_valid_within(blockpfn)) | |
73 | break; | |
74 | } | |
75 | cursor = pfn_to_page(blockpfn); | |
76 | ||
77 | /* Isolate free pages. This assumes the block is valid */ | |
78 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { | |
79 | int isolated, i; | |
80 | struct page *page = cursor; | |
81 | ||
82 | if (!pfn_valid_within(blockpfn)) | |
83 | continue; | |
b7aba698 | 84 | nr_scanned++; |
748446bb MG |
85 | |
86 | if (!PageBuddy(page)) | |
87 | continue; | |
88 | ||
89 | /* Found a free page, break it into order-0 pages */ | |
90 | isolated = split_free_page(page); | |
91 | total_isolated += isolated; | |
92 | for (i = 0; i < isolated; i++) { | |
93 | list_add(&page->lru, freelist); | |
94 | page++; | |
95 | } | |
96 | ||
97 | /* If a page was split, advance to the end of it */ | |
98 | if (isolated) { | |
99 | blockpfn += isolated - 1; | |
100 | cursor += isolated - 1; | |
101 | } | |
102 | } | |
103 | ||
b7aba698 | 104 | trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated); |
748446bb MG |
105 | return total_isolated; |
106 | } | |
107 | ||
108 | /* Returns true if the page is within a block suitable for migration to */ | |
109 | static bool suitable_migration_target(struct page *page) | |
110 | { | |
111 | ||
112 | int migratetype = get_pageblock_migratetype(page); | |
113 | ||
114 | /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */ | |
115 | if (migratetype == MIGRATE_ISOLATE || migratetype == MIGRATE_RESERVE) | |
116 | return false; | |
117 | ||
118 | /* If the page is a large free page, then allow migration */ | |
119 | if (PageBuddy(page) && page_order(page) >= pageblock_order) | |
120 | return true; | |
121 | ||
122 | /* If the block is MIGRATE_MOVABLE, allow migration */ | |
123 | if (migratetype == MIGRATE_MOVABLE) | |
124 | return true; | |
125 | ||
126 | /* Otherwise skip the block */ | |
127 | return false; | |
128 | } | |
129 | ||
130 | /* | |
131 | * Based on information in the current compact_control, find blocks | |
132 | * suitable for isolating free pages from and then isolate them. | |
133 | */ | |
134 | static void isolate_freepages(struct zone *zone, | |
135 | struct compact_control *cc) | |
136 | { | |
137 | struct page *page; | |
138 | unsigned long high_pfn, low_pfn, pfn; | |
139 | unsigned long flags; | |
140 | int nr_freepages = cc->nr_freepages; | |
141 | struct list_head *freelist = &cc->freepages; | |
142 | ||
7454f4ba MG |
143 | /* |
144 | * Initialise the free scanner. The starting point is where we last | |
145 | * scanned from (or the end of the zone if starting). The low point | |
146 | * is the end of the pageblock the migration scanner is using. | |
147 | */ | |
748446bb MG |
148 | pfn = cc->free_pfn; |
149 | low_pfn = cc->migrate_pfn + pageblock_nr_pages; | |
7454f4ba MG |
150 | |
151 | /* | |
152 | * Take care that if the migration scanner is at the end of the zone | |
153 | * that the free scanner does not accidentally move to the next zone | |
154 | * in the next isolation cycle. | |
155 | */ | |
156 | high_pfn = min(low_pfn, pfn); | |
748446bb MG |
157 | |
158 | /* | |
159 | * Isolate free pages until enough are available to migrate the | |
160 | * pages on cc->migratepages. We stop searching if the migrate | |
161 | * and free page scanners meet or enough free pages are isolated. | |
162 | */ | |
748446bb MG |
163 | for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages; |
164 | pfn -= pageblock_nr_pages) { | |
165 | unsigned long isolated; | |
166 | ||
167 | if (!pfn_valid(pfn)) | |
168 | continue; | |
169 | ||
170 | /* | |
171 | * Check for overlapping nodes/zones. It's possible on some | |
172 | * configurations to have a setup like | |
173 | * node0 node1 node0 | |
174 | * i.e. it's possible that all pages within a zones range of | |
175 | * pages do not belong to a single zone. | |
176 | */ | |
177 | page = pfn_to_page(pfn); | |
178 | if (page_zone(page) != zone) | |
179 | continue; | |
180 | ||
181 | /* Check the block is suitable for migration */ | |
182 | if (!suitable_migration_target(page)) | |
183 | continue; | |
184 | ||
602605a4 MG |
185 | /* |
186 | * Found a block suitable for isolating free pages from. Now | |
187 | * we disabled interrupts, double check things are ok and | |
188 | * isolate the pages. This is to minimise the time IRQs | |
189 | * are disabled | |
190 | */ | |
191 | isolated = 0; | |
192 | spin_lock_irqsave(&zone->lock, flags); | |
193 | if (suitable_migration_target(page)) { | |
194 | isolated = isolate_freepages_block(zone, pfn, freelist); | |
195 | nr_freepages += isolated; | |
196 | } | |
197 | spin_unlock_irqrestore(&zone->lock, flags); | |
748446bb MG |
198 | |
199 | /* | |
200 | * Record the highest PFN we isolated pages from. When next | |
201 | * looking for free pages, the search will restart here as | |
202 | * page migration may have returned some pages to the allocator | |
203 | */ | |
204 | if (isolated) | |
205 | high_pfn = max(high_pfn, pfn); | |
206 | } | |
748446bb MG |
207 | |
208 | /* split_free_page does not map the pages */ | |
209 | list_for_each_entry(page, freelist, lru) { | |
210 | arch_alloc_page(page, 0); | |
211 | kernel_map_pages(page, 1, 1); | |
212 | } | |
213 | ||
214 | cc->free_pfn = high_pfn; | |
215 | cc->nr_freepages = nr_freepages; | |
216 | } | |
217 | ||
218 | /* Update the number of anon and file isolated pages in the zone */ | |
219 | static void acct_isolated(struct zone *zone, struct compact_control *cc) | |
220 | { | |
221 | struct page *page; | |
b9e84ac1 | 222 | unsigned int count[2] = { 0, }; |
748446bb | 223 | |
b9e84ac1 MK |
224 | list_for_each_entry(page, &cc->migratepages, lru) |
225 | count[!!page_is_file_cache(page)]++; | |
748446bb | 226 | |
b9e84ac1 MK |
227 | __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); |
228 | __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
748446bb MG |
229 | } |
230 | ||
231 | /* Similar to reclaim, but different enough that they don't share logic */ | |
232 | static bool too_many_isolated(struct zone *zone) | |
233 | { | |
bc693045 | 234 | unsigned long active, inactive, isolated; |
748446bb MG |
235 | |
236 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + | |
237 | zone_page_state(zone, NR_INACTIVE_ANON); | |
bc693045 MK |
238 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
239 | zone_page_state(zone, NR_ACTIVE_ANON); | |
748446bb MG |
240 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
241 | zone_page_state(zone, NR_ISOLATED_ANON); | |
242 | ||
bc693045 | 243 | return isolated > (inactive + active) / 2; |
748446bb MG |
244 | } |
245 | ||
f9e35b3b MG |
246 | /* possible outcome of isolate_migratepages */ |
247 | typedef enum { | |
248 | ISOLATE_ABORT, /* Abort compaction now */ | |
249 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
250 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
251 | } isolate_migrate_t; | |
252 | ||
748446bb MG |
253 | /* |
254 | * Isolate all pages that can be migrated from the block pointed to by | |
255 | * the migrate scanner within compact_control. | |
256 | */ | |
f9e35b3b | 257 | static isolate_migrate_t isolate_migratepages(struct zone *zone, |
748446bb MG |
258 | struct compact_control *cc) |
259 | { | |
260 | unsigned long low_pfn, end_pfn; | |
9927af74 | 261 | unsigned long last_pageblock_nr = 0, pageblock_nr; |
b7aba698 | 262 | unsigned long nr_scanned = 0, nr_isolated = 0; |
748446bb | 263 | struct list_head *migratelist = &cc->migratepages; |
39deaf85 | 264 | isolate_mode_t mode = ISOLATE_ACTIVE|ISOLATE_INACTIVE; |
748446bb MG |
265 | |
266 | /* Do not scan outside zone boundaries */ | |
267 | low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn); | |
268 | ||
269 | /* Only scan within a pageblock boundary */ | |
270 | end_pfn = ALIGN(low_pfn + pageblock_nr_pages, pageblock_nr_pages); | |
271 | ||
272 | /* Do not cross the free scanner or scan within a memory hole */ | |
273 | if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) { | |
274 | cc->migrate_pfn = end_pfn; | |
f9e35b3b | 275 | return ISOLATE_NONE; |
748446bb MG |
276 | } |
277 | ||
278 | /* | |
279 | * Ensure that there are not too many pages isolated from the LRU | |
280 | * list by either parallel reclaimers or compaction. If there are, | |
281 | * delay for some time until fewer pages are isolated | |
282 | */ | |
283 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b MG |
284 | /* async migration should just abort */ |
285 | if (!cc->sync) | |
286 | return ISOLATE_ABORT; | |
287 | ||
748446bb MG |
288 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
289 | ||
290 | if (fatal_signal_pending(current)) | |
f9e35b3b | 291 | return ISOLATE_ABORT; |
748446bb MG |
292 | } |
293 | ||
294 | /* Time to isolate some pages for migration */ | |
b2eef8c0 | 295 | cond_resched(); |
748446bb MG |
296 | spin_lock_irq(&zone->lru_lock); |
297 | for (; low_pfn < end_pfn; low_pfn++) { | |
298 | struct page *page; | |
b2eef8c0 AA |
299 | bool locked = true; |
300 | ||
301 | /* give a chance to irqs before checking need_resched() */ | |
302 | if (!((low_pfn+1) % SWAP_CLUSTER_MAX)) { | |
303 | spin_unlock_irq(&zone->lru_lock); | |
304 | locked = false; | |
305 | } | |
306 | if (need_resched() || spin_is_contended(&zone->lru_lock)) { | |
307 | if (locked) | |
308 | spin_unlock_irq(&zone->lru_lock); | |
309 | cond_resched(); | |
310 | spin_lock_irq(&zone->lru_lock); | |
311 | if (fatal_signal_pending(current)) | |
312 | break; | |
313 | } else if (!locked) | |
314 | spin_lock_irq(&zone->lru_lock); | |
315 | ||
0bf380bc MG |
316 | /* |
317 | * migrate_pfn does not necessarily start aligned to a | |
318 | * pageblock. Ensure that pfn_valid is called when moving | |
319 | * into a new MAX_ORDER_NR_PAGES range in case of large | |
320 | * memory holes within the zone | |
321 | */ | |
322 | if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { | |
323 | if (!pfn_valid(low_pfn)) { | |
324 | low_pfn += MAX_ORDER_NR_PAGES - 1; | |
325 | continue; | |
326 | } | |
327 | } | |
328 | ||
748446bb MG |
329 | if (!pfn_valid_within(low_pfn)) |
330 | continue; | |
b7aba698 | 331 | nr_scanned++; |
748446bb | 332 | |
dc908600 MG |
333 | /* |
334 | * Get the page and ensure the page is within the same zone. | |
335 | * See the comment in isolate_freepages about overlapping | |
336 | * nodes. It is deliberate that the new zone lock is not taken | |
337 | * as memory compaction should not move pages between nodes. | |
338 | */ | |
748446bb | 339 | page = pfn_to_page(low_pfn); |
dc908600 MG |
340 | if (page_zone(page) != zone) |
341 | continue; | |
342 | ||
343 | /* Skip if free */ | |
748446bb MG |
344 | if (PageBuddy(page)) |
345 | continue; | |
346 | ||
9927af74 MG |
347 | /* |
348 | * For async migration, also only scan in MOVABLE blocks. Async | |
349 | * migration is optimistic to see if the minimum amount of work | |
350 | * satisfies the allocation | |
351 | */ | |
352 | pageblock_nr = low_pfn >> pageblock_order; | |
353 | if (!cc->sync && last_pageblock_nr != pageblock_nr && | |
354 | get_pageblock_migratetype(page) != MIGRATE_MOVABLE) { | |
355 | low_pfn += pageblock_nr_pages; | |
356 | low_pfn = ALIGN(low_pfn, pageblock_nr_pages) - 1; | |
357 | last_pageblock_nr = pageblock_nr; | |
358 | continue; | |
359 | } | |
360 | ||
bc835011 AA |
361 | if (!PageLRU(page)) |
362 | continue; | |
363 | ||
364 | /* | |
365 | * PageLRU is set, and lru_lock excludes isolation, | |
366 | * splitting and collapsing (collapsing has already | |
367 | * happened if PageLRU is set). | |
368 | */ | |
369 | if (PageTransHuge(page)) { | |
370 | low_pfn += (1 << compound_order(page)) - 1; | |
371 | continue; | |
372 | } | |
373 | ||
c8244935 MG |
374 | if (!cc->sync) |
375 | mode |= ISOLATE_ASYNC_MIGRATE; | |
376 | ||
748446bb | 377 | /* Try isolate the page */ |
39deaf85 | 378 | if (__isolate_lru_page(page, mode, 0) != 0) |
748446bb MG |
379 | continue; |
380 | ||
bc835011 AA |
381 | VM_BUG_ON(PageTransCompound(page)); |
382 | ||
748446bb MG |
383 | /* Successfully isolated */ |
384 | del_page_from_lru_list(zone, page, page_lru(page)); | |
385 | list_add(&page->lru, migratelist); | |
748446bb | 386 | cc->nr_migratepages++; |
b7aba698 | 387 | nr_isolated++; |
748446bb MG |
388 | |
389 | /* Avoid isolating too much */ | |
31b8384a HD |
390 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
391 | ++low_pfn; | |
748446bb | 392 | break; |
31b8384a | 393 | } |
748446bb MG |
394 | } |
395 | ||
396 | acct_isolated(zone, cc); | |
397 | ||
398 | spin_unlock_irq(&zone->lru_lock); | |
399 | cc->migrate_pfn = low_pfn; | |
400 | ||
b7aba698 MG |
401 | trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated); |
402 | ||
f9e35b3b | 403 | return ISOLATE_SUCCESS; |
748446bb MG |
404 | } |
405 | ||
406 | /* | |
407 | * This is a migrate-callback that "allocates" freepages by taking pages | |
408 | * from the isolated freelists in the block we are migrating to. | |
409 | */ | |
410 | static struct page *compaction_alloc(struct page *migratepage, | |
411 | unsigned long data, | |
412 | int **result) | |
413 | { | |
414 | struct compact_control *cc = (struct compact_control *)data; | |
415 | struct page *freepage; | |
416 | ||
417 | /* Isolate free pages if necessary */ | |
418 | if (list_empty(&cc->freepages)) { | |
419 | isolate_freepages(cc->zone, cc); | |
420 | ||
421 | if (list_empty(&cc->freepages)) | |
422 | return NULL; | |
423 | } | |
424 | ||
425 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
426 | list_del(&freepage->lru); | |
427 | cc->nr_freepages--; | |
428 | ||
429 | return freepage; | |
430 | } | |
431 | ||
432 | /* | |
433 | * We cannot control nr_migratepages and nr_freepages fully when migration is | |
434 | * running as migrate_pages() has no knowledge of compact_control. When | |
435 | * migration is complete, we count the number of pages on the lists by hand. | |
436 | */ | |
437 | static void update_nr_listpages(struct compact_control *cc) | |
438 | { | |
439 | int nr_migratepages = 0; | |
440 | int nr_freepages = 0; | |
441 | struct page *page; | |
442 | ||
443 | list_for_each_entry(page, &cc->migratepages, lru) | |
444 | nr_migratepages++; | |
445 | list_for_each_entry(page, &cc->freepages, lru) | |
446 | nr_freepages++; | |
447 | ||
448 | cc->nr_migratepages = nr_migratepages; | |
449 | cc->nr_freepages = nr_freepages; | |
450 | } | |
451 | ||
452 | static int compact_finished(struct zone *zone, | |
5a03b051 | 453 | struct compact_control *cc) |
748446bb | 454 | { |
56de7263 | 455 | unsigned int order; |
5a03b051 | 456 | unsigned long watermark; |
56de7263 | 457 | |
748446bb MG |
458 | if (fatal_signal_pending(current)) |
459 | return COMPACT_PARTIAL; | |
460 | ||
461 | /* Compaction run completes if the migrate and free scanner meet */ | |
462 | if (cc->free_pfn <= cc->migrate_pfn) | |
463 | return COMPACT_COMPLETE; | |
464 | ||
82478fb7 JW |
465 | /* |
466 | * order == -1 is expected when compacting via | |
467 | * /proc/sys/vm/compact_memory | |
468 | */ | |
56de7263 MG |
469 | if (cc->order == -1) |
470 | return COMPACT_CONTINUE; | |
471 | ||
3957c776 MH |
472 | /* Compaction run is not finished if the watermark is not met */ |
473 | watermark = low_wmark_pages(zone); | |
474 | watermark += (1 << cc->order); | |
475 | ||
476 | if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0)) | |
477 | return COMPACT_CONTINUE; | |
478 | ||
56de7263 MG |
479 | /* Direct compactor: Is a suitable page free? */ |
480 | for (order = cc->order; order < MAX_ORDER; order++) { | |
481 | /* Job done if page is free of the right migratetype */ | |
482 | if (!list_empty(&zone->free_area[order].free_list[cc->migratetype])) | |
483 | return COMPACT_PARTIAL; | |
484 | ||
485 | /* Job done if allocation would set block type */ | |
486 | if (order >= pageblock_order && zone->free_area[order].nr_free) | |
487 | return COMPACT_PARTIAL; | |
488 | } | |
489 | ||
748446bb MG |
490 | return COMPACT_CONTINUE; |
491 | } | |
492 | ||
3e7d3449 MG |
493 | /* |
494 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
495 | * Returns | |
496 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
497 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
498 | * COMPACT_CONTINUE - If compaction should run now | |
499 | */ | |
500 | unsigned long compaction_suitable(struct zone *zone, int order) | |
501 | { | |
502 | int fragindex; | |
503 | unsigned long watermark; | |
504 | ||
3957c776 MH |
505 | /* |
506 | * order == -1 is expected when compacting via | |
507 | * /proc/sys/vm/compact_memory | |
508 | */ | |
509 | if (order == -1) | |
510 | return COMPACT_CONTINUE; | |
511 | ||
3e7d3449 MG |
512 | /* |
513 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
514 | * This is because during migration, copies of pages need to be | |
515 | * allocated and for a short time, the footprint is higher | |
516 | */ | |
517 | watermark = low_wmark_pages(zone) + (2UL << order); | |
518 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
519 | return COMPACT_SKIPPED; | |
520 | ||
521 | /* | |
522 | * fragmentation index determines if allocation failures are due to | |
523 | * low memory or external fragmentation | |
524 | * | |
a582a738 SL |
525 | * index of -1000 implies allocations might succeed depending on |
526 | * watermarks | |
3e7d3449 MG |
527 | * index towards 0 implies failure is due to lack of memory |
528 | * index towards 1000 implies failure is due to fragmentation | |
529 | * | |
530 | * Only compact if a failure would be due to fragmentation. | |
531 | */ | |
532 | fragindex = fragmentation_index(zone, order); | |
533 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
534 | return COMPACT_SKIPPED; | |
535 | ||
a582a738 SL |
536 | if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark, |
537 | 0, 0)) | |
3e7d3449 MG |
538 | return COMPACT_PARTIAL; |
539 | ||
540 | return COMPACT_CONTINUE; | |
541 | } | |
542 | ||
748446bb MG |
543 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
544 | { | |
545 | int ret; | |
546 | ||
3e7d3449 MG |
547 | ret = compaction_suitable(zone, cc->order); |
548 | switch (ret) { | |
549 | case COMPACT_PARTIAL: | |
550 | case COMPACT_SKIPPED: | |
551 | /* Compaction is likely to fail */ | |
552 | return ret; | |
553 | case COMPACT_CONTINUE: | |
554 | /* Fall through to compaction */ | |
555 | ; | |
556 | } | |
557 | ||
748446bb MG |
558 | /* Setup to move all movable pages to the end of the zone */ |
559 | cc->migrate_pfn = zone->zone_start_pfn; | |
560 | cc->free_pfn = cc->migrate_pfn + zone->spanned_pages; | |
561 | cc->free_pfn &= ~(pageblock_nr_pages-1); | |
562 | ||
563 | migrate_prep_local(); | |
564 | ||
565 | while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) { | |
566 | unsigned long nr_migrate, nr_remaining; | |
9d502c1c | 567 | int err; |
748446bb | 568 | |
f9e35b3b MG |
569 | switch (isolate_migratepages(zone, cc)) { |
570 | case ISOLATE_ABORT: | |
571 | ret = COMPACT_PARTIAL; | |
572 | goto out; | |
573 | case ISOLATE_NONE: | |
748446bb | 574 | continue; |
f9e35b3b MG |
575 | case ISOLATE_SUCCESS: |
576 | ; | |
577 | } | |
748446bb MG |
578 | |
579 | nr_migrate = cc->nr_migratepages; | |
9d502c1c | 580 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
7f0f2496 | 581 | (unsigned long)cc, false, |
a6bc32b8 | 582 | cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC); |
748446bb MG |
583 | update_nr_listpages(cc); |
584 | nr_remaining = cc->nr_migratepages; | |
585 | ||
586 | count_vm_event(COMPACTBLOCKS); | |
587 | count_vm_events(COMPACTPAGES, nr_migrate - nr_remaining); | |
588 | if (nr_remaining) | |
589 | count_vm_events(COMPACTPAGEFAILED, nr_remaining); | |
b7aba698 MG |
590 | trace_mm_compaction_migratepages(nr_migrate - nr_remaining, |
591 | nr_remaining); | |
748446bb MG |
592 | |
593 | /* Release LRU pages not migrated */ | |
9d502c1c | 594 | if (err) { |
748446bb MG |
595 | putback_lru_pages(&cc->migratepages); |
596 | cc->nr_migratepages = 0; | |
597 | } | |
598 | ||
599 | } | |
600 | ||
f9e35b3b | 601 | out: |
748446bb MG |
602 | /* Release free pages and check accounting */ |
603 | cc->nr_freepages -= release_freepages(&cc->freepages); | |
604 | VM_BUG_ON(cc->nr_freepages != 0); | |
605 | ||
606 | return ret; | |
607 | } | |
76ab0f53 | 608 | |
d43a87e6 | 609 | static unsigned long compact_zone_order(struct zone *zone, |
5a03b051 | 610 | int order, gfp_t gfp_mask, |
d527caf2 | 611 | bool sync) |
56de7263 MG |
612 | { |
613 | struct compact_control cc = { | |
614 | .nr_freepages = 0, | |
615 | .nr_migratepages = 0, | |
616 | .order = order, | |
617 | .migratetype = allocflags_to_migratetype(gfp_mask), | |
618 | .zone = zone, | |
77f1fe6b | 619 | .sync = sync, |
56de7263 MG |
620 | }; |
621 | INIT_LIST_HEAD(&cc.freepages); | |
622 | INIT_LIST_HEAD(&cc.migratepages); | |
623 | ||
624 | return compact_zone(zone, &cc); | |
625 | } | |
626 | ||
5e771905 MG |
627 | int sysctl_extfrag_threshold = 500; |
628 | ||
56de7263 MG |
629 | /** |
630 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
631 | * @zonelist: The zonelist used for the current allocation | |
632 | * @order: The order of the current allocation | |
633 | * @gfp_mask: The GFP mask of the current allocation | |
634 | * @nodemask: The allowed nodes to allocate from | |
77f1fe6b | 635 | * @sync: Whether migration is synchronous or not |
56de7263 MG |
636 | * |
637 | * This is the main entry point for direct page compaction. | |
638 | */ | |
639 | unsigned long try_to_compact_pages(struct zonelist *zonelist, | |
77f1fe6b MG |
640 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
641 | bool sync) | |
56de7263 MG |
642 | { |
643 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
644 | int may_enter_fs = gfp_mask & __GFP_FS; | |
645 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
646 | struct zoneref *z; |
647 | struct zone *zone; | |
648 | int rc = COMPACT_SKIPPED; | |
649 | ||
650 | /* | |
651 | * Check whether it is worth even starting compaction. The order check is | |
652 | * made because an assumption is made that the page allocator can satisfy | |
653 | * the "cheaper" orders without taking special steps | |
654 | */ | |
c5a73c3d | 655 | if (!order || !may_enter_fs || !may_perform_io) |
56de7263 MG |
656 | return rc; |
657 | ||
658 | count_vm_event(COMPACTSTALL); | |
659 | ||
660 | /* Compact each zone in the list */ | |
661 | for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx, | |
662 | nodemask) { | |
56de7263 MG |
663 | int status; |
664 | ||
d527caf2 | 665 | status = compact_zone_order(zone, order, gfp_mask, sync); |
56de7263 MG |
666 | rc = max(status, rc); |
667 | ||
3e7d3449 MG |
668 | /* If a normal allocation would succeed, stop compacting */ |
669 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, 0)) | |
56de7263 MG |
670 | break; |
671 | } | |
672 | ||
673 | return rc; | |
674 | } | |
675 | ||
676 | ||
76ab0f53 | 677 | /* Compact all zones within a node */ |
7be62de9 | 678 | static int __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
679 | { |
680 | int zoneid; | |
76ab0f53 MG |
681 | struct zone *zone; |
682 | ||
76ab0f53 | 683 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
684 | |
685 | zone = &pgdat->node_zones[zoneid]; | |
686 | if (!populated_zone(zone)) | |
687 | continue; | |
688 | ||
7be62de9 RR |
689 | cc->nr_freepages = 0; |
690 | cc->nr_migratepages = 0; | |
691 | cc->zone = zone; | |
692 | INIT_LIST_HEAD(&cc->freepages); | |
693 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 694 | |
aad6ec37 | 695 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 696 | compact_zone(zone, cc); |
76ab0f53 | 697 | |
aff62249 RR |
698 | if (cc->order > 0) { |
699 | int ok = zone_watermark_ok(zone, cc->order, | |
700 | low_wmark_pages(zone), 0, 0); | |
701 | if (ok && cc->order > zone->compact_order_failed) | |
702 | zone->compact_order_failed = cc->order + 1; | |
703 | /* Currently async compaction is never deferred. */ | |
704 | else if (!ok && cc->sync) | |
705 | defer_compaction(zone, cc->order); | |
706 | } | |
707 | ||
7be62de9 RR |
708 | VM_BUG_ON(!list_empty(&cc->freepages)); |
709 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 MG |
710 | } |
711 | ||
712 | return 0; | |
713 | } | |
714 | ||
7be62de9 RR |
715 | int compact_pgdat(pg_data_t *pgdat, int order) |
716 | { | |
717 | struct compact_control cc = { | |
718 | .order = order, | |
719 | .sync = false, | |
720 | }; | |
721 | ||
722 | return __compact_pgdat(pgdat, &cc); | |
723 | } | |
724 | ||
725 | static int compact_node(int nid) | |
726 | { | |
7be62de9 RR |
727 | struct compact_control cc = { |
728 | .order = -1, | |
729 | .sync = true, | |
730 | }; | |
731 | ||
8575ec29 | 732 | return __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
733 | } |
734 | ||
76ab0f53 MG |
735 | /* Compact all nodes in the system */ |
736 | static int compact_nodes(void) | |
737 | { | |
738 | int nid; | |
739 | ||
8575ec29 HD |
740 | /* Flush pending updates to the LRU lists */ |
741 | lru_add_drain_all(); | |
742 | ||
76ab0f53 MG |
743 | for_each_online_node(nid) |
744 | compact_node(nid); | |
745 | ||
746 | return COMPACT_COMPLETE; | |
747 | } | |
748 | ||
749 | /* The written value is actually unused, all memory is compacted */ | |
750 | int sysctl_compact_memory; | |
751 | ||
752 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
753 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
754 | void __user *buffer, size_t *length, loff_t *ppos) | |
755 | { | |
756 | if (write) | |
757 | return compact_nodes(); | |
758 | ||
759 | return 0; | |
760 | } | |
ed4a6d7f | 761 | |
5e771905 MG |
762 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
763 | void __user *buffer, size_t *length, loff_t *ppos) | |
764 | { | |
765 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
766 | ||
767 | return 0; | |
768 | } | |
769 | ||
ed4a6d7f | 770 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
10fbcf4c KS |
771 | ssize_t sysfs_compact_node(struct device *dev, |
772 | struct device_attribute *attr, | |
ed4a6d7f MG |
773 | const char *buf, size_t count) |
774 | { | |
8575ec29 HD |
775 | int nid = dev->id; |
776 | ||
777 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
778 | /* Flush pending updates to the LRU lists */ | |
779 | lru_add_drain_all(); | |
780 | ||
781 | compact_node(nid); | |
782 | } | |
ed4a6d7f MG |
783 | |
784 | return count; | |
785 | } | |
10fbcf4c | 786 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
787 | |
788 | int compaction_register_node(struct node *node) | |
789 | { | |
10fbcf4c | 790 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
791 | } |
792 | ||
793 | void compaction_unregister_node(struct node *node) | |
794 | { | |
10fbcf4c | 795 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
796 | } |
797 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ |