Branch data Line data Source code
1 : : /*
2 : : * linux/mm/vmscan.c
3 : : *
4 : : * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 : : *
6 : : * Swap reorganised 29.12.95, Stephen Tweedie.
7 : : * kswapd added: 7.1.96 sct
8 : : * Removed kswapd_ctl limits, and swap out as many pages as needed
9 : : * to bring the system back to freepages.high: 2.4.97, Rik van Riel.
10 : : * Zone aware kswapd started 02/00, Kanoj Sarcar (kanoj@sgi.com).
11 : : * Multiqueue VM started 5.8.00, Rik van Riel.
12 : : */
13 : :
14 : : #include <linux/mm.h>
15 : : #include <linux/module.h>
16 : : #include <linux/gfp.h>
17 : : #include <linux/kernel_stat.h>
18 : : #include <linux/swap.h>
19 : : #include <linux/pagemap.h>
20 : : #include <linux/init.h>
21 : : #include <linux/highmem.h>
22 : : #include <linux/vmpressure.h>
23 : : #include <linux/vmstat.h>
24 : : #include <linux/file.h>
25 : : #include <linux/writeback.h>
26 : : #include <linux/blkdev.h>
27 : : #include <linux/buffer_head.h> /* for try_to_release_page(),
28 : : buffer_heads_over_limit */
29 : : #include <linux/mm_inline.h>
30 : : #include <linux/backing-dev.h>
31 : : #include <linux/rmap.h>
32 : : #include <linux/topology.h>
33 : : #include <linux/cpu.h>
34 : : #include <linux/cpuset.h>
35 : : #include <linux/compaction.h>
36 : : #include <linux/notifier.h>
37 : : #include <linux/rwsem.h>
38 : : #include <linux/delay.h>
39 : : #include <linux/kthread.h>
40 : : #include <linux/freezer.h>
41 : : #include <linux/memcontrol.h>
42 : : #include <linux/delayacct.h>
43 : : #include <linux/sysctl.h>
44 : : #include <linux/oom.h>
45 : : #include <linux/prefetch.h>
46 : : #include <linux/debugfs.h>
47 : :
48 : : #include <asm/tlbflush.h>
49 : : #include <asm/div64.h>
50 : :
51 : : #include <linux/swapops.h>
52 : : #include <linux/balloon_compaction.h>
53 : :
54 : : #include "internal.h"
55 : :
56 : : #define CREATE_TRACE_POINTS
57 : : #include <trace/events/vmscan.h>
58 : :
59 : : struct scan_control {
60 : : /* Incremented by the number of inactive pages that were scanned */
61 : : unsigned long nr_scanned;
62 : :
63 : : /* Number of pages freed so far during a call to shrink_zones() */
64 : : unsigned long nr_reclaimed;
65 : :
66 : : /* How many pages shrink_list() should reclaim */
67 : : unsigned long nr_to_reclaim;
68 : :
69 : : unsigned long hibernation_mode;
70 : :
71 : : /* This context's GFP mask */
72 : : gfp_t gfp_mask;
73 : :
74 : : int may_writepage;
75 : :
76 : : /* Can mapped pages be reclaimed? */
77 : : int may_unmap;
78 : :
79 : : /* Can pages be swapped as part of reclaim? */
80 : : int may_swap;
81 : :
82 : : int order;
83 : :
84 : : /* Scan (total_size >> priority) pages at once */
85 : : int priority;
86 : :
87 : : /*
88 : : * The memory cgroup that hit its limit and as a result is the
89 : : * primary target of this reclaim invocation.
90 : : */
91 : : struct mem_cgroup *target_mem_cgroup;
92 : :
93 : : /*
94 : : * Nodemask of nodes allowed by the caller. If NULL, all nodes
95 : : * are scanned.
96 : : */
97 : : nodemask_t *nodemask;
98 : : };
99 : :
100 : : #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
101 : :
102 : : #ifdef ARCH_HAS_PREFETCH
103 : : #define prefetch_prev_lru_page(_page, _base, _field) \
104 : : do { \
105 : : if ((_page)->lru.prev != _base) { \
106 : : struct page *prev; \
107 : : \
108 : : prev = lru_to_page(&(_page->lru)); \
109 : : prefetch(&prev->_field); \
110 : : } \
111 : : } while (0)
112 : : #else
113 : : #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0)
114 : : #endif
115 : :
116 : : #ifdef ARCH_HAS_PREFETCHW
117 : : #define prefetchw_prev_lru_page(_page, _base, _field) \
118 : : do { \
119 : : if ((_page)->lru.prev != _base) { \
120 : : struct page *prev; \
121 : : \
122 : : prev = lru_to_page(&(_page->lru)); \
123 : : prefetchw(&prev->_field); \
124 : : } \
125 : : } while (0)
126 : : #else
127 : : #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0)
128 : : #endif
129 : :
130 : : /*
131 : : * From 0 .. 100. Higher means more swappy.
132 : : */
133 : : int vm_swappiness = 60;
134 : : unsigned long vm_total_pages; /* The total number of pages which the VM controls */
135 : :
136 : : static LIST_HEAD(shrinker_list);
137 : : static DECLARE_RWSEM(shrinker_rwsem);
138 : :
139 : : #ifdef CONFIG_MEMCG
140 : : static bool global_reclaim(struct scan_control *sc)
141 : : {
142 : : return !sc->target_mem_cgroup;
143 : : }
144 : : #else
145 : : static bool global_reclaim(struct scan_control *sc)
146 : : {
147 : : return true;
148 : : }
149 : : #endif
150 : :
151 : 0 : unsigned long zone_reclaimable_pages(struct zone *zone)
152 : : {
153 : : int nr;
154 : :
155 : 2673939 : nr = zone_page_state(zone, NR_ACTIVE_FILE) +
156 : : zone_page_state(zone, NR_INACTIVE_FILE);
157 : :
158 [ + ]: 2673939 : if (get_nr_swap_pages() > 0)
159 : 1278 : nr += zone_page_state(zone, NR_ACTIVE_ANON) +
160 : : zone_page_state(zone, NR_INACTIVE_ANON);
161 : :
162 : 0 : return nr;
163 : : }
164 : :
165 : 0 : bool zone_reclaimable(struct zone *zone)
166 : : {
167 : 464618 : return zone->pages_scanned < zone_reclaimable_pages(zone) * 6;
168 : : }
169 : :
170 : : static unsigned long get_lru_size(struct lruvec *lruvec, enum lru_list lru)
171 : : {
172 : : if (!mem_cgroup_disabled())
173 : : return mem_cgroup_get_lru_size(lruvec, lru);
174 : :
175 : 811779 : return zone_page_state(lruvec_zone(lruvec), NR_LRU_BASE + lru);
176 : : }
177 : :
178 : : struct dentry *debug_file;
179 : :
180 : 0 : static int debug_shrinker_show(struct seq_file *s, void *unused)
181 : : {
182 : : struct shrinker *shrinker;
183 : : struct shrink_control sc;
184 : :
185 : 0 : sc.gfp_mask = -1;
186 : 0 : sc.nr_to_scan = 0;
187 : :
188 : 0 : down_read(&shrinker_rwsem);
189 [ # # ]: 0 : list_for_each_entry(shrinker, &shrinker_list, list) {
190 : : int num_objs;
191 : :
192 : 0 : num_objs = shrinker->count_objects(shrinker, &sc);
193 : 0 : seq_printf(s, "%pf %d\n", shrinker->count_objects, num_objs);
194 : : }
195 : 0 : up_read(&shrinker_rwsem);
196 : 0 : return 0;
197 : : }
198 : :
199 : 0 : static int debug_shrinker_open(struct inode *inode, struct file *file)
200 : : {
201 : 0 : return single_open(file, debug_shrinker_show, inode->i_private);
202 : : }
203 : :
204 : : static const struct file_operations debug_shrinker_fops = {
205 : : .open = debug_shrinker_open,
206 : : .read = seq_read,
207 : : .llseek = seq_lseek,
208 : : .release = single_release,
209 : : };
210 : :
211 : : /*
212 : : * Add a shrinker callback to be called from the vm.
213 : : */
214 : 0 : int register_shrinker(struct shrinker *shrinker)
215 : : {
216 : : size_t size = sizeof(*shrinker->nr_deferred);
217 : :
218 : : /*
219 : : * If we only have one possible node in the system anyway, save
220 : : * ourselves the trouble and disable NUMA aware behavior. This way we
221 : : * will save memory and some small loop time later.
222 : : */
223 : : if (nr_node_ids == 1)
224 : 3 : shrinker->flags &= ~SHRINKER_NUMA_AWARE;
225 : :
226 : : if (shrinker->flags & SHRINKER_NUMA_AWARE)
227 : : size *= nr_node_ids;
228 : :
229 : 3 : shrinker->nr_deferred = kzalloc(size, GFP_KERNEL);
230 [ + - ]: 3 : if (!shrinker->nr_deferred)
231 : : return -ENOMEM;
232 : :
233 : 3 : down_write(&shrinker_rwsem);
234 : 3 : list_add_tail(&shrinker->list, &shrinker_list);
235 : 3 : up_write(&shrinker_rwsem);
236 : 3 : return 0;
237 : : }
238 : : EXPORT_SYMBOL(register_shrinker);
239 : :
240 : 0 : static int __init add_shrinker_debug(void)
241 : : {
242 : 0 : debugfs_create_file("shrinker", 0644, NULL, NULL,
243 : : &debug_shrinker_fops);
244 : 0 : return 0;
245 : : }
246 : :
247 : : late_initcall(add_shrinker_debug);
248 : :
249 : : /*
250 : : * Remove one
251 : : */
252 : 0 : void unregister_shrinker(struct shrinker *shrinker)
253 : : {
254 : 3 : down_write(&shrinker_rwsem);
255 : : list_del(&shrinker->list);
256 : 3 : up_write(&shrinker_rwsem);
257 : 3 : kfree(shrinker->nr_deferred);
258 : 3 : }
259 : : EXPORT_SYMBOL(unregister_shrinker);
260 : :
261 : : #define SHRINK_BATCH 128
262 : :
263 : : static unsigned long
264 : 0 : shrink_slab_node(struct shrink_control *shrinkctl, struct shrinker *shrinker,
265 : : unsigned long nr_pages_scanned, unsigned long lru_pages)
266 : : {
267 : : unsigned long freed = 0;
268 : : unsigned long long delta;
269 : : long total_scan;
270 : : long max_pass;
271 : : long nr;
272 : : long new_nr;
273 : 4604685 : int nid = shrinkctl->nid;
274 : 4604685 : long batch_size = shrinker->batch ? shrinker->batch
275 [ + + ]: 4604685 : : SHRINK_BATCH;
276 : :
277 : 4604685 : max_pass = shrinker->count_objects(shrinker, shrinkctl);
278 [ + + ]: 4604843 : if (max_pass == 0)
279 : : return 0;
280 : :
281 : : /*
282 : : * copy the current shrinker scan count into a local variable
283 : : * and zero it so that other concurrent shrinker invocations
284 : : * don't also do this scanning work.
285 : : */
286 : 5965396 : nr = atomic_long_xchg(&shrinker->nr_deferred[nid], 0);
287 : :
288 : : total_scan = nr;
289 : 5284764 : delta = (4 * nr_pages_scanned) / shrinker->seeks;
290 : 5284764 : delta *= max_pass;
291 [ - + ][ # # ]: 5284764 : do_div(delta, lru_pages + 1);
[ - + ][ - + ]
[ - + ][ - + ]
[ - + ][ - + ]
[ - + ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ]
292 : 680079 : total_scan += delta;
293 [ - + ]: 680079 : if (total_scan < 0) {
294 : 0 : printk(KERN_ERR
295 : : "shrink_slab: %pF negative objects to delete nr=%ld\n",
296 : : shrinker->scan_objects, total_scan);
297 : : total_scan = max_pass;
298 : : }
299 : :
300 : : /*
301 : : * We need to avoid excessive windup on filesystem shrinkers
302 : : * due to large numbers of GFP_NOFS allocations causing the
303 : : * shrinkers to return -1 all the time. This results in a large
304 : : * nr being built up so when a shrink that can do some work
305 : : * comes along it empties the entire cache due to nr >>>
306 : : * max_pass. This is bad for sustaining a working set in
307 : : * memory.
308 : : *
309 : : * Hence only allow the shrinker to scan the entire cache when
310 : : * a large delta change is calculated directly.
311 : : */
312 [ + + ]: 679779 : if (delta < max_pass / 4)
313 : 12157 : total_scan = min(total_scan, max_pass / 2);
314 : :
315 : : /*
316 : : * Avoid risking looping forever due to too large nr value:
317 : : * never try to free more than twice the estimate number of
318 : : * freeable entries.
319 : : */
320 [ + + ]: 679779 : if (total_scan > max_pass * 2)
321 : : total_scan = max_pass * 2;
322 : :
323 : 679779 : trace_mm_shrink_slab_start(shrinker, shrinkctl, nr,
324 : : nr_pages_scanned, lru_pages,
325 : : max_pass, delta, total_scan);
326 : :
327 [ + + ]: 679845 : while (total_scan >= batch_size) {
328 : : unsigned long ret;
329 : :
330 : 217 : shrinkctl->nr_to_scan = batch_size;
331 : 217 : ret = shrinker->scan_objects(shrinker, shrinkctl);
332 [ + - ]: 217 : if (ret == SHRINK_STOP)
333 : : break;
334 : 217 : freed += ret;
335 : :
336 : : count_vm_events(SLABS_SCANNED, batch_size);
337 : 217 : total_scan -= batch_size;
338 : :
339 : 217 : cond_resched();
340 : : }
341 : :
342 : : /*
343 : : * move the unused scan count back into the shrinker in a
344 : : * manner that handles concurrent updates. If we exhausted the
345 : : * scan, there is no need to do an update.
346 : : */
347 [ + + ]: 679628 : if (total_scan > 0)
348 : 679536 : new_nr = atomic_long_add_return(total_scan,
349 : 679536 : &shrinker->nr_deferred[nid]);
350 : : else
351 : 92 : new_nr = atomic_long_read(&shrinker->nr_deferred[nid]);
352 : :
353 : 679208 : trace_mm_shrink_slab_end(shrinker, freed, nr, new_nr);
354 : 679208 : return freed;
355 : : }
356 : :
357 : : /*
358 : : * Call the shrink functions to age shrinkable caches
359 : : *
360 : : * Here we assume it costs one seek to replace a lru page and that it also
361 : : * takes a seek to recreate a cache object. With this in mind we age equal
362 : : * percentages of the lru and ageable caches. This should balance the seeks
363 : : * generated by these structures.
364 : : *
365 : : * If the vm encountered mapped pages on the LRU it increase the pressure on
366 : : * slab to avoid swapping.
367 : : *
368 : : * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits.
369 : : *
370 : : * `lru_pages' represents the number of on-LRU pages in all the zones which
371 : : * are eligible for the caller's allocation attempt. It is used for balancing
372 : : * slab reclaim versus page reclaim.
373 : : *
374 : : * Returns the number of slab objects which we shrunk.
375 : : */
376 : 0 : unsigned long shrink_slab(struct shrink_control *shrinkctl,
377 : : unsigned long nr_pages_scanned,
378 : : unsigned long lru_pages)
379 : : {
380 : : struct shrinker *shrinker;
381 : : unsigned long freed = 0;
382 : :
383 [ + + ]: 170903 : if (nr_pages_scanned == 0)
384 : : nr_pages_scanned = SWAP_CLUSTER_MAX;
385 : :
386 [ + + ]: 170903 : if (!down_read_trylock(&shrinker_rwsem)) {
387 : : /*
388 : : * If we would return 0, our callers would understand that we
389 : : * have nothing else to shrink and give up trying. By returning
390 : : * 1 we keep it going and assume we'll be able to shrink next
391 : : * time.
392 : : */
393 : : freed = 1;
394 : : goto out;
395 : : }
396 : :
397 [ + + ]: 4772154 : list_for_each_entry(shrinker, &shrinker_list, list) {
398 [ + ][ + + ]: 9199179 : for_each_node_mask(shrinkctl->nid, shrinkctl->nodes_to_scan) {
399 [ + ]: 4601992 : if (!node_online(shrinkctl->nid))
400 : 0 : continue;
401 : :
402 [ + ][ + + ]: 4602097 : if (!(shrinker->flags & SHRINKER_NUMA_AWARE) &&
403 : : (shrinkctl->nid != 0))
404 : : break;
405 : :
406 : 4597025 : freed += shrink_slab_node(shrinkctl, shrinker,
407 : : nr_pages_scanned, lru_pages);
408 : :
409 : : }
410 : : }
411 : 170962 : up_read(&shrinker_rwsem);
412 : : out:
413 : 172026 : cond_resched();
414 : 170933 : return freed;
415 : : }
416 : :
417 : 288 : static inline int is_page_cache_freeable(struct page *page)
418 : : {
419 : : /*
420 : : * A freeable page cache page is referenced only by the caller
421 : : * that isolated the page, the page cache radix tree and
422 : : * optional buffer heads at page->private.
423 : : */
424 : 288 : return page_count(page) - page_has_private(page) == 2;
425 : : }
426 : :
427 : 288 : static int may_write_to_queue(struct backing_dev_info *bdi,
428 : : struct scan_control *sc)
429 : : {
430 [ - + ]: 288 : if (current->flags & PF_SWAPWRITE)
431 : : return 1;
432 [ # # ]: 0 : if (!bdi_write_congested(bdi))
433 : : return 1;
434 [ # # ]: 0 : if (bdi == current->backing_dev_info)
435 : : return 1;
436 : : return 0;
437 : : }
438 : :
439 : : /*
440 : : * We detected a synchronous write error writing a page out. Probably
441 : : * -ENOSPC. We need to propagate that into the address_space for a subsequent
442 : : * fsync(), msync() or close().
443 : : *
444 : : * The tricky part is that after writepage we cannot touch the mapping: nothing
445 : : * prevents it from being freed up. But we have a ref on the page and once
446 : : * that page is locked, the mapping is pinned.
447 : : *
448 : : * We're allowed to run sleeping lock_page() here because we know the caller has
449 : : * __GFP_FS.
450 : : */
451 : 0 : static void handle_write_error(struct address_space *mapping,
452 : : struct page *page, int error)
453 : : {
454 : : lock_page(page);
455 [ # # ]: 0 : if (page_mapping(page) == mapping)
456 : : mapping_set_error(mapping, error);
457 : 0 : unlock_page(page);
458 : 0 : }
459 : :
460 : : /* possible outcome of pageout() */
461 : : typedef enum {
462 : : /* failed to write page out, page is locked */
463 : : PAGE_KEEP,
464 : : /* move page to the active list, page is locked */
465 : : PAGE_ACTIVATE,
466 : : /* page has been sent to the disk successfully, page is unlocked */
467 : : PAGE_SUCCESS,
468 : : /* page is clean and locked */
469 : : PAGE_CLEAN,
470 : : } pageout_t;
471 : :
472 : : /*
473 : : * pageout is called by shrink_page_list() for each dirty page.
474 : : * Calls ->writepage().
475 : : */
476 : 0 : static pageout_t pageout(struct page *page, struct address_space *mapping,
477 : : struct scan_control *sc)
478 : : {
479 : : /*
480 : : * If the page is dirty, only perform writeback if that write
481 : : * will be non-blocking. To prevent this allocation from being
482 : : * stalled by pagecache activity. But note that there may be
483 : : * stalls if we need to run get_block(). We could test
484 : : * PagePrivate for that.
485 : : *
486 : : * If this process is currently in __generic_file_aio_write() against
487 : : * this page's queue, we can perform writeback even if that
488 : : * will block.
489 : : *
490 : : * If the page is swapcache, write it back even if that would
491 : : * block, for some throttling. This happens by accident, because
492 : : * swap_backing_dev_info is bust: it doesn't reflect the
493 : : * congestion state of the swapdevs. Easy to fix, if needed.
494 : : */
495 [ + - ]: 288 : if (!is_page_cache_freeable(page))
496 : : return PAGE_KEEP;
497 [ - + ]: 288 : if (!mapping) {
498 : : /*
499 : : * Some data journaling orphaned pages can have
500 : : * page->mapping == NULL while being dirty with clean buffers.
501 : : */
502 [ # # ]: 0 : if (page_has_private(page)) {
503 [ # # ]: 0 : if (try_to_free_buffers(page)) {
504 : : ClearPageDirty(page);
505 : 0 : printk("%s: orphaned page\n", __func__);
506 : : return PAGE_CLEAN;
507 : : }
508 : : }
509 : : return PAGE_KEEP;
510 : : }
511 [ + - ]: 288 : if (mapping->a_ops->writepage == NULL)
512 : : return PAGE_ACTIVATE;
513 [ + - ]: 288 : if (!may_write_to_queue(mapping->backing_dev_info, sc))
514 : : return PAGE_KEEP;
515 : :
516 [ + - ]: 288 : if (clear_page_dirty_for_io(page)) {
517 : : int res;
518 : 288 : struct writeback_control wbc = {
519 : : .sync_mode = WB_SYNC_NONE,
520 : : .nr_to_write = SWAP_CLUSTER_MAX,
521 : : .range_start = 0,
522 : : .range_end = LLONG_MAX,
523 : : .for_reclaim = 1,
524 : : };
525 : :
526 : : SetPageReclaim(page);
527 : 288 : res = mapping->a_ops->writepage(page, &wbc);
528 [ - + ]: 288 : if (res < 0)
529 : 0 : handle_write_error(mapping, page, res);
530 [ - + ]: 288 : if (res == AOP_WRITEPAGE_ACTIVATE) {
531 : : ClearPageReclaim(page);
532 : : return PAGE_ACTIVATE;
533 : : }
534 : :
535 [ + + ]: 288 : if (!PageWriteback(page)) {
536 : : /* synchronous write or broken a_ops? */
537 : : ClearPageReclaim(page);
538 : : }
539 [ - + ]: 576 : trace_mm_vmscan_writepage(page, trace_reclaim_flags(page));
540 : 288 : inc_zone_page_state(page, NR_VMSCAN_WRITE);
541 : : return PAGE_SUCCESS;
542 : : }
543 : :
544 : : return PAGE_CLEAN;
545 : : }
546 : :
547 : : /*
548 : : * Same as remove_mapping, but if the page is removed from the mapping, it
549 : : * gets returned with a refcount of 0.
550 : : */
551 : 0 : static int __remove_mapping(struct address_space *mapping, struct page *page)
552 : : {
553 [ - + ]: 276841 : BUG_ON(!PageLocked(page));
554 [ - + ]: 276841 : BUG_ON(mapping != page_mapping(page));
555 : :
556 : : spin_lock_irq(&mapping->tree_lock);
557 : : /*
558 : : * The non racy check for a busy page.
559 : : *
560 : : * Must be careful with the order of the tests. When someone has
561 : : * a ref to the page, it may be possible that they dirty it then
562 : : * drop the reference. So if PageDirty is tested before page_count
563 : : * here, then the following race may occur:
564 : : *
565 : : * get_user_pages(&page);
566 : : * [user mapping goes away]
567 : : * write_to(page);
568 : : * !PageDirty(page) [good]
569 : : * SetPageDirty(page);
570 : : * put_page(page);
571 : : * !page_count(page) [good, discard it]
572 : : *
573 : : * [oops, our write_to data is lost]
574 : : *
575 : : * Reversing the order of the tests ensures such a situation cannot
576 : : * escape unnoticed. The smp_rmb is needed to ensure the page->flags
577 : : * load is not satisfied before that of page->_count.
578 : : *
579 : : * Note that if SetPageDirty is always performed via set_page_dirty,
580 : : * and thus under tree_lock, then this ordering is not required.
581 : : */
582 [ + + ]: 276841 : if (!page_freeze_refs(page, 2))
583 : : goto cannot_free;
584 : : /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */
585 [ - + ]: 276674 : if (unlikely(PageDirty(page))) {
586 : : page_unfreeze_refs(page, 2);
587 : : goto cannot_free;
588 : : }
589 : :
590 [ - + ]: 276674 : if (PageSwapCache(page)) {
591 : 0 : swp_entry_t swap = { .val = page_private(page) };
592 : 0 : __delete_from_swap_cache(page);
593 : : spin_unlock_irq(&mapping->tree_lock);
594 : 0 : swapcache_free(swap, page);
595 : : } else {
596 : : void (*freepage)(struct page *);
597 : :
598 : 276674 : freepage = mapping->a_ops->freepage;
599 : :
600 : 276674 : __delete_from_page_cache(page);
601 : : spin_unlock_irq(&mapping->tree_lock);
602 : : mem_cgroup_uncharge_cache_page(page);
603 : :
604 [ - + ]: 276674 : if (freepage != NULL)
605 : 0 : freepage(page);
606 : : }
607 : :
608 : : return 1;
609 : :
610 : : cannot_free:
611 : : spin_unlock_irq(&mapping->tree_lock);
612 : 167 : return 0;
613 : : }
614 : :
615 : : /*
616 : : * Attempt to detach a locked page from its ->mapping. If it is dirty or if
617 : : * someone else has a ref on the page, abort and return 0. If it was
618 : : * successfully detached, return 1. Assumes the caller has a single ref on
619 : : * this page.
620 : : */
621 : 0 : int remove_mapping(struct address_space *mapping, struct page *page)
622 : : {
623 [ + + ]: 156436 : if (__remove_mapping(mapping, page)) {
624 : : /*
625 : : * Unfreezing the refcount with 1 rather than 2 effectively
626 : : * drops the pagecache ref for us without requiring another
627 : : * atomic operation.
628 : : */
629 : : page_unfreeze_refs(page, 1);
630 : 156401 : return 1;
631 : : }
632 : : return 0;
633 : : }
634 : :
635 : : /**
636 : : * putback_lru_page - put previously isolated page onto appropriate LRU list
637 : : * @page: page to be put back to appropriate lru list
638 : : *
639 : : * Add previously isolated @page to appropriate LRU list.
640 : : * Page may still be unevictable for other reasons.
641 : : *
642 : : * lru_lock must not be held, interrupts must be enabled.
643 : : */
644 : 0 : void putback_lru_page(struct page *page)
645 : : {
646 : : bool is_unevictable;
647 : : int was_unevictable = PageUnevictable(page);
648 : :
649 : : VM_BUG_ON(PageLRU(page));
650 : :
651 : : redo:
652 : : ClearPageUnevictable(page);
653 : :
654 [ + + ]: 5684 : if (page_evictable(page)) {
655 : : /*
656 : : * For evictable pages, we can use the cache.
657 : : * In event of a race, worst case is we end up with an
658 : : * unevictable page on [in]active list.
659 : : * We know how to handle that.
660 : : */
661 : : is_unevictable = false;
662 : 1902 : lru_cache_add(page);
663 : : } else {
664 : : /*
665 : : * Put unevictable pages directly on zone's unevictable
666 : : * list.
667 : : */
668 : : is_unevictable = true;
669 : 3782 : add_page_to_unevictable_list(page);
670 : : /*
671 : : * When racing with an mlock or AS_UNEVICTABLE clearing
672 : : * (page is unlocked) make sure that if the other thread
673 : : * does not observe our setting of PG_lru and fails
674 : : * isolation/check_move_unevictable_pages,
675 : : * we see PG_mlocked/AS_UNEVICTABLE cleared below and move
676 : : * the page back to the evictable list.
677 : : *
678 : : * The other side is TestClearPageMlocked() or shmem_lock().
679 : : */
680 : 3782 : smp_mb();
681 : : }
682 : :
683 : : /*
684 : : * page's status can change while we move it among lru. If an evictable
685 : : * page is on unevictable list, it never be freed. To avoid that,
686 : : * check after we added it to the list, again.
687 : : */
688 [ + + ][ - + ]: 5684 : if (is_unevictable && page_evictable(page)) {
689 [ # # ]: 0 : if (!isolate_lru_page(page)) {
690 : 0 : put_page(page);
691 : 0 : goto redo;
692 : : }
693 : : /* This means someone else dropped this page from LRU
694 : : * So, it will be freed or putback to LRU again. There is
695 : : * nothing to do here.
696 : : */
697 : : }
698 : :
699 [ + + ]: 5684 : if (was_unevictable && !is_unevictable)
700 : : count_vm_event(UNEVICTABLE_PGRESCUED);
701 [ + - ]: 3782 : else if (!was_unevictable && is_unevictable)
702 : : count_vm_event(UNEVICTABLE_PGCULLED);
703 : :
704 : 5684 : put_page(page); /* drop ref from isolate */
705 : 5684 : }
706 : :
707 : : enum page_references {
708 : : PAGEREF_RECLAIM,
709 : : PAGEREF_RECLAIM_CLEAN,
710 : : PAGEREF_KEEP,
711 : : PAGEREF_ACTIVATE,
712 : : };
713 : :
714 : 151090 : static enum page_references page_check_references(struct page *page,
715 : : struct scan_control *sc)
716 : : {
717 : : int referenced_ptes, referenced_page;
718 : : unsigned long vm_flags;
719 : :
720 : 151090 : referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup,
721 : : &vm_flags);
722 : : referenced_page = TestClearPageReferenced(page);
723 : :
724 : : /*
725 : : * Mlock lost the isolation race with us. Let try_to_unmap()
726 : : * move the page to the unevictable list.
727 : : */
728 [ + + ]: 151090 : if (vm_flags & VM_LOCKED)
729 : : return PAGEREF_RECLAIM;
730 : :
731 [ + + ]: 151089 : if (referenced_ptes) {
732 [ + - ]: 637 : if (PageSwapBacked(page))
733 : : return PAGEREF_ACTIVATE;
734 : : /*
735 : : * All mapped pages start out with page table
736 : : * references from the instantiating fault, so we need
737 : : * to look twice if a mapped file page is used more
738 : : * than once.
739 : : *
740 : : * Mark it and spare it for another trip around the
741 : : * inactive list. Another page table reference will
742 : : * lead to its activation.
743 : : *
744 : : * Note: the mark is set for activated pages as well
745 : : * so that recently deactivated but used pages are
746 : : * quickly recovered.
747 : : */
748 : : SetPageReferenced(page);
749 : :
750 [ + + ]: 637 : if (referenced_page || referenced_ptes > 1)
751 : : return PAGEREF_ACTIVATE;
752 : :
753 : : /*
754 : : * Activate file-backed executable pages after first usage.
755 : : */
756 [ - + ]: 421 : if (vm_flags & VM_EXEC)
757 : : return PAGEREF_ACTIVATE;
758 : :
759 : : return PAGEREF_KEEP;
760 : : }
761 : :
762 : : /* Reclaim if clean, defer dirty pages to writeback */
763 [ + + ][ - + ]: 150452 : if (referenced_page && !PageSwapBacked(page))
764 : : return PAGEREF_RECLAIM_CLEAN;
765 : :
766 : : return PAGEREF_RECLAIM;
767 : : }
768 : :
769 : : /* Check if a page is dirty or under writeback */
770 : 0 : static void page_check_dirty_writeback(struct page *page,
771 : : bool *dirty, bool *writeback)
772 : : {
773 : : struct address_space *mapping;
774 : :
775 : : /*
776 : : * Anonymous pages are not handled by flushers and must be written
777 : : * from reclaim context. Do not stall reclaim based on them
778 : : */
779 [ - + ]: 151281 : if (!page_is_file_cache(page)) {
780 : 0 : *dirty = false;
781 : 0 : *writeback = false;
782 : 0 : return;
783 : : }
784 : :
785 : : /* By default assume that the page flags are accurate */
786 : 151281 : *dirty = PageDirty(page);
787 : 151281 : *writeback = PageWriteback(page);
788 : :
789 : : /* Verify dirty/writeback state if the filesystem supports it */
790 [ + + ]: 151281 : if (!page_has_private(page))
791 : : return;
792 : :
793 : 119006 : mapping = page_mapping(page);
794 [ + + ][ + + ]: 119006 : if (mapping && mapping->a_ops->is_dirty_writeback)
795 : 64954 : mapping->a_ops->is_dirty_writeback(page, dirty, writeback);
796 : : }
797 : :
798 : : /*
799 : : * shrink_page_list() returns the number of reclaimed pages
800 : : */
801 : 0 : static unsigned long shrink_page_list(struct list_head *page_list,
802 : : struct zone *zone,
803 : 151089 : struct scan_control *sc,
804 : : enum ttu_flags ttu_flags,
805 : : unsigned long *ret_nr_dirty,
806 : : unsigned long *ret_nr_unqueued_dirty,
807 : : unsigned long *ret_nr_congested,
808 : : unsigned long *ret_nr_writeback,
809 : : unsigned long *ret_nr_immediate,
810 : : bool force_reclaim)
811 : : {
812 : 7503 : LIST_HEAD(ret_pages);
813 : 7503 : LIST_HEAD(free_pages);
814 : : int pgactivate = 0;
815 : : unsigned long nr_unqueued_dirty = 0;
816 : : unsigned long nr_dirty = 0;
817 : : unsigned long nr_congested = 0;
818 : : unsigned long nr_reclaimed = 0;
819 : : unsigned long nr_writeback = 0;
820 : : unsigned long nr_immediate = 0;
821 : :
822 : 7503 : cond_resched();
823 : :
824 : : mem_cgroup_uncharge_start();
825 [ + + ]: 161179 : while (!list_empty(page_list)) {
826 : : struct address_space *mapping;
827 : 300026 : struct page *page;
828 : : int may_enter_fs;
829 : : enum page_references references = PAGEREF_RECLAIM_CLEAN;
830 : : bool dirty, writeback;
831 : :
832 : 153676 : cond_resched();
833 : :
834 : 153677 : page = lru_to_page(page_list);
835 : : list_del(&page->lru);
836 : :
837 [ + + ]: 153677 : if (!trylock_page(page))
838 : : goto keep;
839 : :
840 : : VM_BUG_ON(PageActive(page));
841 : : VM_BUG_ON(page_zone(page) != zone);
842 : :
843 : 151282 : sc->nr_scanned++;
844 : :
845 [ + - ]: 151282 : if (unlikely(!page_evictable(page)))
846 : : goto cull_mlocked;
847 : :
848 [ - + ][ # # ]: 151282 : if (!sc->may_unmap && page_mapped(page))
849 : : goto keep_locked;
850 : :
851 : : /* Double the slab pressure for mapped and swapcache pages */
852 [ + + ][ - + ]: 151281 : if (page_mapped(page) || PageSwapCache(page))
853 : 2183 : sc->nr_scanned++;
854 : :
855 [ - + ][ # # ]: 151281 : may_enter_fs = (sc->gfp_mask & __GFP_FS) ||
856 [ # # ]: 0 : (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO));
857 : :
858 : : /*
859 : : * The number of dirty pages determines if a zone is marked
860 : : * reclaim_congested which affects wait_iff_congested. kswapd
861 : : * will stall and start writing pages if the tail of the LRU
862 : : * is all dirty unqueued pages.
863 : : */
864 : 151281 : page_check_dirty_writeback(page, &dirty, &writeback);
865 [ + + ][ + + ]: 151282 : if (dirty || writeback)
866 : 1071 : nr_dirty++;
867 : :
868 [ + + ][ + - ]: 151282 : if (dirty && !writeback)
869 : 879 : nr_unqueued_dirty++;
870 : :
871 : : /*
872 : : * Treat this page as congested if the underlying BDI is or if
873 : : * pages are cycling through the LRU so quickly that the
874 : : * pages marked for immediate reclaim are making it to the
875 : : * end of the LRU a second time.
876 : : */
877 : 151282 : mapping = page_mapping(page);
878 [ + + ][ + - ]: 302562 : if ((mapping && bdi_write_congested(mapping->backing_dev_info)) ||
[ + + ]
879 [ + + ]: 192 : (writeback && PageReclaim(page)))
880 : 189 : nr_congested++;
881 : :
882 : : /*
883 : : * If a page at the tail of the LRU is under writeback, there
884 : : * are three cases to consider.
885 : : *
886 : : * 1) If reclaim is encountering an excessive number of pages
887 : : * under writeback and this page is both under writeback and
888 : : * PageReclaim then it indicates that pages are being queued
889 : : * for IO but are being recycled through the LRU before the
890 : : * IO can complete. Waiting on the page itself risks an
891 : : * indefinite stall if it is impossible to writeback the
892 : : * page due to IO error or disconnected storage so instead
893 : : * note that the LRU is being scanned too quickly and the
894 : : * caller can stall after page list has been processed.
895 : : *
896 : : * 2) Global reclaim encounters a page, memcg encounters a
897 : : * page that is not marked for immediate reclaim or
898 : : * the caller does not have __GFP_IO. In this case mark
899 : : * the page for immediate reclaim and continue scanning.
900 : : *
901 : : * __GFP_IO is checked because a loop driver thread might
902 : : * enter reclaim, and deadlock if it waits on a page for
903 : : * which it is needed to do the write (loop masks off
904 : : * __GFP_IO|__GFP_FS for this reason); but more thought
905 : : * would probably show more reasons.
906 : : *
907 : : * Don't require __GFP_FS, since we're not going into the
908 : : * FS, just waiting on its writeback completion. Worryingly,
909 : : * ext4 gfs2 and xfs allocate pages with
910 : : * grab_cache_page_write_begin(,,AOP_FLAG_NOFS), so testing
911 : : * may_enter_fs here is liable to OOM on them.
912 : : *
913 : : * 3) memcg encounters a page that is not already marked
914 : : * PageReclaim. memcg does not have any dirty pages
915 : : * throttling so we could easily OOM just because too many
916 : : * pages are in writeback and there is nothing else to
917 : : * reclaim. Wait for the writeback to complete.
918 : : */
919 [ + + ]: 151282 : if (PageWriteback(page)) {
920 : : /* Case 1 above */
921 [ + + ][ + + ]: 192 : if (current_is_kswapd() &&
922 [ - + ]: 128 : PageReclaim(page) &&
923 : : zone_is_reclaim_writeback(zone)) {
924 : 0 : nr_immediate++;
925 : 0 : goto keep_locked;
926 : :
927 : : /* Case 2 above */
928 : : } else if (global_reclaim(sc) ||
929 : : !PageReclaim(page) || !(sc->gfp_mask & __GFP_IO)) {
930 : : /*
931 : : * This is slightly racy - end_page_writeback()
932 : : * might have just cleared PageReclaim, then
933 : : * setting PageReclaim here end up interpreted
934 : : * as PageReadahead - but that does not matter
935 : : * enough to care. What we do want is for this
936 : : * page to have PageReclaim set next time memcg
937 : : * reclaim reaches the tests above, so it will
938 : : * then wait_on_page_writeback() to avoid OOM;
939 : : * and it's also appropriate in global reclaim.
940 : : */
941 : : SetPageReclaim(page);
942 : 192 : nr_writeback++;
943 : :
944 : 192 : goto keep_locked;
945 : :
946 : : /* Case 3 above */
947 : : } else {
948 : : wait_on_page_writeback(page);
949 : : }
950 : : }
951 : :
952 [ + + ]: 151090 : if (!force_reclaim)
953 : 151089 : references = page_check_references(page, sc);
954 : :
955 [ + + - ]: 151089 : switch (references) {
956 : : case PAGEREF_ACTIVATE:
957 : : goto activate_locked;
958 : : case PAGEREF_KEEP:
959 : : goto keep_locked;
960 : : case PAGEREF_RECLAIM:
961 : : case PAGEREF_RECLAIM_CLEAN:
962 : : ; /* try to reclaim the page below */
963 : : }
964 : :
965 : : /*
966 : : * Anonymous process memory has backing store?
967 : : * Try to allocate it some swap space here.
968 : : */
969 [ - + ][ # # ]: 150452 : if (PageAnon(page) && !PageSwapCache(page)) {
970 [ # # ]: 0 : if (!(sc->gfp_mask & __GFP_IO))
971 : : goto keep_locked;
972 [ # # ]: 0 : if (!add_to_swap(page, page_list))
973 : : goto activate_locked;
974 : : may_enter_fs = 1;
975 : :
976 : : /* Adding to swap updated mapping */
977 : 0 : mapping = page_mapping(page);
978 : : }
979 : :
980 : : /*
981 : : * The page is mapped into the page tables of one or more
982 : : * processes. Try to unmap it here.
983 : : */
984 [ + + ][ + - ]: 157956 : if (page_mapped(page) && mapping) {
985 [ + - - - ]: 1546 : switch (try_to_unmap(page, ttu_flags)) {
986 : : case SWAP_FAIL:
987 : : goto activate_locked;
988 : : case SWAP_AGAIN:
989 : : goto keep_locked;
990 : : case SWAP_MLOCK:
991 : : goto cull_mlocked;
992 : : case SWAP_SUCCESS:
993 : : ; /* try to free the page below */
994 : : }
995 : : }
996 : :
997 [ + + ]: 150453 : if (PageDirty(page)) {
998 : : /*
999 : : * Only kswapd can writeback filesystem pages to
1000 : : * avoid risk of stack overflow but only writeback
1001 : : * if many dirty pages have been encountered.
1002 : : */
1003 [ + + ][ + + ]: 879 : if (page_is_file_cache(page) &&
1004 [ + + ]: 511 : (!current_is_kswapd() ||
1005 : : !zone_is_reclaim_dirty(zone))) {
1006 : : /*
1007 : : * Immediately reclaim when written back.
1008 : : * Similar in principal to deactivate_page()
1009 : : * except we already have the page isolated
1010 : : * and know it's dirty
1011 : : */
1012 : 581 : inc_zone_page_state(page, NR_VMSCAN_IMMEDIATE);
1013 : : SetPageReclaim(page);
1014 : :
1015 : : goto keep_locked;
1016 : : }
1017 : :
1018 [ + + ]: 298 : if (references == PAGEREF_RECLAIM_CLEAN)
1019 : : goto keep_locked;
1020 [ + - ]: 288 : if (!may_enter_fs)
1021 : : goto keep_locked;
1022 [ + - ]: 288 : if (!sc->may_writepage)
1023 : : goto keep_locked;
1024 : :
1025 : : /* Page is dirty, try to write it out here */
1026 [ + - - - ]: 288 : switch (pageout(page, mapping, sc)) {
1027 : : case PAGE_KEEP:
1028 : : goto keep_locked;
1029 : : case PAGE_ACTIVATE:
1030 : : goto activate_locked;
1031 : : case PAGE_SUCCESS:
1032 [ + + ]: 288 : if (PageWriteback(page))
1033 : : goto keep;
1034 [ - + ]: 273 : if (PageDirty(page))
1035 : : goto keep;
1036 : :
1037 : : /*
1038 : : * A synchronous write - probably a ramdisk. Go
1039 : : * ahead and try to reclaim the page.
1040 : : */
1041 [ # # ]: 0 : if (!trylock_page(page))
1042 : : goto keep;
1043 [ # # ][ # # ]: 0 : if (PageDirty(page) || PageWriteback(page))
1044 : : goto keep_locked;
1045 : 0 : mapping = page_mapping(page);
1046 : : case PAGE_CLEAN:
1047 : : ; /* try to free the page below */
1048 : : }
1049 : : }
1050 : :
1051 : : /*
1052 : : * If the page has buffers, try to free the buffer mappings
1053 : : * associated with this page. If we succeed we try to free
1054 : : * the page as well.
1055 : : *
1056 : : * We do this even if the page is PageDirty().
1057 : : * try_to_release_page() does not perform I/O, but it is
1058 : : * possible for a page to have PageDirty set, but it is actually
1059 : : * clean (all its buffers are clean). This happens if the
1060 : : * buffers were written out directly, with submit_bh(). ext3
1061 : : * will do this, as well as the blockdev mapping.
1062 : : * try_to_release_page() will discover that cleanness and will
1063 : : * drop the buffers and mark the page clean - it can be freed.
1064 : : *
1065 : : * Rarely, pages can have buffers and no ->mapping. These are
1066 : : * the pages which were not successfully invalidated in
1067 : : * truncate_complete_page(). We try to drop those buffers here
1068 : : * and if that worked, and the page is no longer mapped into
1069 : : * process address space (page_count == 1) it can be freed.
1070 : : * Otherwise, leave the page on the LRU so it is swappable.
1071 : : */
1072 [ + + ]: 149574 : if (page_has_private(page)) {
1073 [ + + ]: 117936 : if (!try_to_release_page(page, sc->gfp_mask))
1074 : : goto activate_locked;
1075 [ + + ][ + - ]: 88769 : if (!mapping && page_count(page) == 1) {
1076 : 1 : unlock_page(page);
1077 [ - + ]: 1 : if (put_page_testzero(page))
1078 : : goto free_it;
1079 : : else {
1080 : : /*
1081 : : * rare race with speculative reference.
1082 : : * the speculative reference will free
1083 : : * this page shortly, so we may
1084 : : * increment nr_reclaimed here (and
1085 : : * leave it off the LRU).
1086 : : */
1087 : 0 : nr_reclaimed++;
1088 : 153677 : continue;
1089 : : }
1090 : : }
1091 : : }
1092 : :
1093 [ + - ][ + + ]: 120405 : if (!mapping || !__remove_mapping(mapping, page))
1094 : : goto keep_locked;
1095 : :
1096 : : /*
1097 : : * At this point, we have no other references and there is
1098 : : * no way to pick any more up (removed from LRU, removed
1099 : : * from pagecache). Can use non-atomic bitops now (and
1100 : : * we obviously don't have to worry about waking up a process
1101 : : * waiting on the page lock, because there are no references.
1102 : : */
1103 : : __clear_page_locked(page);
1104 : : free_it:
1105 : 120274 : nr_reclaimed++;
1106 : :
1107 : : /*
1108 : : * Is there need to periodically free_page_list? It would
1109 : : * appear not as the counts should be low
1110 : : */
1111 : 120274 : list_add(&page->lru, &free_pages);
1112 : 120274 : continue;
1113 : :
1114 : : cull_mlocked:
1115 [ # # ]: 0 : if (PageSwapCache(page))
1116 : 0 : try_to_free_swap(page);
1117 : 0 : unlock_page(page);
1118 : 0 : putback_lru_page(page);
1119 : 0 : continue;
1120 : :
1121 : : activate_locked:
1122 : : /* Not a candidate for swapping, so reclaim swap space. */
1123 [ - + ][ # # ]: 29805 : if (PageSwapCache(page) && vm_swap_full())
1124 : 0 : try_to_free_swap(page);
1125 : : VM_BUG_ON(PageActive(page));
1126 : : SetPageActive(page);
1127 : 29805 : pgactivate++;
1128 : : keep_locked:
1129 : 30721 : unlock_page(page);
1130 : : keep:
1131 : 33403 : list_add(&page->lru, &ret_pages);
1132 : : VM_BUG_ON(PageLRU(page) || PageUnevictable(page));
1133 : : }
1134 : :
1135 : 7503 : free_hot_cold_page_list(&free_pages, 1);
1136 : :
1137 : : list_splice(&ret_pages, page_list);
1138 : : count_vm_events(PGACTIVATE, pgactivate);
1139 : : mem_cgroup_uncharge_end();
1140 : 7503 : *ret_nr_dirty += nr_dirty;
1141 : 7503 : *ret_nr_congested += nr_congested;
1142 : 7503 : *ret_nr_unqueued_dirty += nr_unqueued_dirty;
1143 : 7503 : *ret_nr_writeback += nr_writeback;
1144 : 7503 : *ret_nr_immediate += nr_immediate;
1145 : 7503 : return nr_reclaimed;
1146 : : }
1147 : :
1148 : 0 : unsigned long reclaim_clean_pages_from_list(struct zone *zone,
1149 : : struct list_head *page_list)
1150 : : {
1151 : 0 : struct scan_control sc = {
1152 : : .gfp_mask = GFP_KERNEL,
1153 : : .priority = DEF_PRIORITY,
1154 : : .may_unmap = 1,
1155 : : };
1156 : : unsigned long ret, dummy1, dummy2, dummy3, dummy4, dummy5;
1157 : : struct page *page, *next;
1158 : 0 : LIST_HEAD(clean_pages);
1159 : :
1160 [ # # ]: 0 : list_for_each_entry_safe(page, next, page_list, lru) {
1161 [ # # ][ # # ]: 0 : if (page_is_file_cache(page) && !PageDirty(page) &&
1162 : : !isolated_balloon_page(page)) {
1163 : : ClearPageActive(page);
1164 : : list_move(&page->lru, &clean_pages);
1165 : : }
1166 : : }
1167 : :
1168 : 0 : ret = shrink_page_list(&clean_pages, zone, &sc,
1169 : : TTU_UNMAP|TTU_IGNORE_ACCESS,
1170 : : &dummy1, &dummy2, &dummy3, &dummy4, &dummy5, true);
1171 : : list_splice(&clean_pages, page_list);
1172 : 0 : __mod_zone_page_state(zone, NR_ISOLATED_FILE, -ret);
1173 : 0 : return ret;
1174 : : }
1175 : :
1176 : : /*
1177 : : * Attempt to remove the specified page from its LRU. Only take this page
1178 : : * if it is of the appropriate PageActive status. Pages which are being
1179 : : * freed elsewhere are also ignored.
1180 : : *
1181 : : * page: page to consider
1182 : : * mode: one of the LRU isolation modes defined above
1183 : : *
1184 : : * returns 0 on success, -ve errno on failure.
1185 : : */
1186 : 0 : int __isolate_lru_page(struct page *page, isolate_mode_t mode)
1187 : : {
1188 : : int ret = -EINVAL;
1189 : :
1190 : : /* Only take pages on the LRU. */
1191 [ + - ]: 227812 : if (!PageLRU(page))
1192 : : return ret;
1193 : :
1194 : : /* Compaction should not handle unevictable pages but CMA can do so */
1195 [ - + ][ # # ]: 227812 : if (PageUnevictable(page) && !(mode & ISOLATE_UNEVICTABLE))
1196 : : return ret;
1197 : :
1198 : : ret = -EBUSY;
1199 : :
1200 : : /*
1201 : : * To minimise LRU disruption, the caller can indicate that it only
1202 : : * wants to isolate pages it will be able to operate on without
1203 : : * blocking - clean pages for the most part.
1204 : : *
1205 : : * ISOLATE_CLEAN means that only clean pages should be isolated. This
1206 : : * is used by reclaim when it is cannot write to backing storage
1207 : : *
1208 : : * ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages
1209 : : * that it is possible to migrate without blocking
1210 : : */
1211 [ - + ]: 227812 : if (mode & (ISOLATE_CLEAN|ISOLATE_ASYNC_MIGRATE)) {
1212 : : /* All the caller can do on PageWriteback is block */
1213 [ # # ]: 0 : if (PageWriteback(page))
1214 : : return ret;
1215 : :
1216 [ # # ]: 0 : if (PageDirty(page)) {
1217 : : struct address_space *mapping;
1218 : :
1219 : : /* ISOLATE_CLEAN means only clean pages */
1220 [ # # ]: 0 : if (mode & ISOLATE_CLEAN)
1221 : : return ret;
1222 : :
1223 : : /*
1224 : : * Only pages without mappings or that have a
1225 : : * ->migratepage callback are possible to migrate
1226 : : * without blocking
1227 : : */
1228 : 0 : mapping = page_mapping(page);
1229 [ # # ][ # # ]: 0 : if (mapping && !mapping->a_ops->migratepage)
1230 : : return ret;
1231 : : }
1232 : : }
1233 : :
1234 [ - + ][ # # ]: 227812 : if ((mode & ISOLATE_UNMAPPED) && page_mapped(page))
1235 : : return ret;
1236 : :
1237 [ + - ]: 227811 : if (likely(get_page_unless_zero(page))) {
1238 : : /*
1239 : : * Be careful not to clear PageLRU until after we're
1240 : : * sure the page is not being freed elsewhere -- the
1241 : : * page release code relies on it.
1242 : : */
1243 : : ClearPageLRU(page);
1244 : : ret = 0;
1245 : : }
1246 : :
1247 : 227812 : return ret;
1248 : : }
1249 : :
1250 : : /*
1251 : : * zone->lru_lock is heavily contended. Some of the functions that
1252 : : * shrink the lists perform better by taking out a batch of pages
1253 : : * and working on them outside the LRU lock.
1254 : : *
1255 : : * For pagecache intensive workloads, this function is the hottest
1256 : : * spot in the kernel (apart from copy_*_user functions).
1257 : : *
1258 : : * Appropriate locks must be held before calling this function.
1259 : : *
1260 : : * @nr_to_scan: The number of pages to look through on the list.
1261 : : * @lruvec: The LRU vector to pull pages from.
1262 : : * @dst: The temp list to put pages on to.
1263 : : * @nr_scanned: The number of pages that were scanned.
1264 : : * @sc: The scan_control struct for this reclaim session
1265 : : * @mode: One of the LRU isolation modes
1266 : : * @lru: LRU list id for isolating
1267 : : *
1268 : : * returns how many pages were moved onto *@dst.
1269 : : */
1270 : 0 : static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
1271 : : struct lruvec *lruvec, struct list_head *dst,
1272 : : unsigned long *nr_scanned, struct scan_control *sc,
1273 : : isolate_mode_t mode, enum lru_list lru)
1274 : : {
1275 : 69794 : struct list_head *src = &lruvec->lists[lru];
1276 : : unsigned long nr_taken = 0;
1277 : : unsigned long scan;
1278 : :
1279 [ + + ][ + + ]: 297605 : for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) {
1280 : : struct page *page;
1281 : : int nr_pages;
1282 : :
1283 : 227812 : page = lru_to_page(src);
1284 [ + + ]: 227812 : prefetchw_prev_lru_page(page, src, flags);
1285 : :
1286 : : VM_BUG_ON(!PageLRU(page));
1287 : :
1288 [ + - - ]: 227812 : switch (__isolate_lru_page(page, mode)) {
1289 : : case 0:
1290 : : nr_pages = hpage_nr_pages(page);
1291 : : mem_cgroup_update_lru_size(lruvec, lru, -nr_pages);
1292 : 227811 : list_move(&page->lru, dst);
1293 : 227811 : nr_taken += nr_pages;
1294 : : break;
1295 : :
1296 : : case -EBUSY:
1297 : : /* else it is being freed elsewhere */
1298 : 0 : list_move(&page->lru, src);
1299 : 0 : continue;
1300 : :
1301 : : default:
1302 : 0 : BUG();
1303 : : }
1304 : : }
1305 : :
1306 : 69793 : *nr_scanned = scan;
1307 : 69793 : trace_mm_vmscan_lru_isolate(sc->order, nr_to_scan, scan,
1308 : : nr_taken, mode, is_file_lru(lru));
1309 : 0 : return nr_taken;
1310 : : }
1311 : :
1312 : : /**
1313 : : * isolate_lru_page - tries to isolate a page from its LRU list
1314 : : * @page: page to isolate from its LRU list
1315 : : *
1316 : : * Isolates a @page from an LRU list, clears PageLRU and adjusts the
1317 : : * vmstat statistic corresponding to whatever LRU list the page was on.
1318 : : *
1319 : : * Returns 0 if the page was removed from an LRU list.
1320 : : * Returns -EBUSY if the page was not on an LRU list.
1321 : : *
1322 : : * The returned page will have PageLRU() cleared. If it was found on
1323 : : * the active list, it will have PageActive set. If it was found on
1324 : : * the unevictable list, it will have the PageUnevictable bit set. That flag
1325 : : * may need to be cleared by the caller before letting the page go.
1326 : : *
1327 : : * The vmstat statistic corresponding to the list on which the page was
1328 : : * found will be decremented.
1329 : : *
1330 : : * Restrictions:
1331 : : * (1) Must be called with an elevated refcount on the page. This is a
1332 : : * fundamentnal difference from isolate_lru_pages (which is called
1333 : : * without a stable reference).
1334 : : * (2) the lru_lock must not be held.
1335 : : * (3) interrupts must be enabled.
1336 : : */
1337 : 0 : int isolate_lru_page(struct page *page)
1338 : : {
1339 : : int ret = -EBUSY;
1340 : :
1341 : : VM_BUG_ON(!page_count(page));
1342 : :
1343 [ + - ]: 3795 : if (PageLRU(page)) {
1344 : 3795 : struct zone *zone = page_zone(page);
1345 : : struct lruvec *lruvec;
1346 : :
1347 : : spin_lock_irq(&zone->lru_lock);
1348 : : lruvec = mem_cgroup_page_lruvec(page, zone);
1349 [ + - ]: 3795 : if (PageLRU(page)) {
1350 : : int lru = page_lru(page);
1351 : : get_page(page);
1352 : : ClearPageLRU(page);
1353 : : del_page_from_lru_list(page, lruvec, lru);
1354 : : ret = 0;
1355 : : }
1356 : : spin_unlock_irq(&zone->lru_lock);
1357 : : }
1358 : 3795 : return ret;
1359 : : }
1360 : :
1361 : : /*
1362 : : * A direct reclaimer may isolate SWAP_CLUSTER_MAX pages from the LRU list and
1363 : : * then get resheduled. When there are massive number of tasks doing page
1364 : : * allocation, such sleeping direct reclaimers may keep piling up on each CPU,
1365 : : * the LRU list will go small and be scanned faster than necessary, leading to
1366 : : * unnecessary swapping, thrashing and OOM.
1367 : : */
1368 : 58835 : static int too_many_isolated(struct zone *zone, int file,
1369 : : struct scan_control *sc)
1370 : : {
1371 : : unsigned long inactive, isolated;
1372 : :
1373 [ + ]: 58835 : if (current_is_kswapd())
1374 : : return 0;
1375 : :
1376 : : if (!global_reclaim(sc))
1377 : : return 0;
1378 : :
1379 [ + - ]: 106835 : if (file) {
1380 : : inactive = zone_page_state(zone, NR_INACTIVE_FILE);
1381 : : isolated = zone_page_state(zone, NR_ISOLATED_FILE);
1382 : : } else {
1383 : : inactive = zone_page_state(zone, NR_INACTIVE_ANON);
1384 : : isolated = zone_page_state(zone, NR_ISOLATED_ANON);
1385 : : }
1386 : :
1387 : : /*
1388 : : * GFP_NOIO/GFP_NOFS callers are allowed to isolate more pages, so they
1389 : : * won't get blocked by normal direct-reclaimers, forming a circular
1390 : : * deadlock.
1391 : : */
1392 [ + + ]: 48000 : if ((sc->gfp_mask & GFP_IOFS) == GFP_IOFS)
1393 : 47997 : inactive >>= 3;
1394 : :
1395 : 48000 : return isolated > inactive;
1396 : : }
1397 : :
1398 : : static noinline_for_stack void
1399 : 0 : putback_inactive_pages(struct lruvec *lruvec, struct list_head *page_list)
1400 : : {
1401 : : struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
1402 : : struct zone *zone = lruvec_zone(lruvec);
1403 : 7503 : LIST_HEAD(pages_to_free);
1404 : :
1405 : : /*
1406 : : * Put back any unfreeable pages.
1407 : : */
1408 [ + + ]: 40906 : while (!list_empty(page_list)) {
1409 : 33403 : struct page *page = lru_to_page(page_list);
1410 : : int lru;
1411 : :
1412 : : VM_BUG_ON(PageLRU(page));
1413 : : list_del(&page->lru);
1414 [ - + ]: 33403 : if (unlikely(!page_evictable(page))) {
1415 : : spin_unlock_irq(&zone->lru_lock);
1416 : 0 : putback_lru_page(page);
1417 : : spin_lock_irq(&zone->lru_lock);
1418 : 0 : continue;
1419 : : }
1420 : :
1421 : : lruvec = mem_cgroup_page_lruvec(page, zone);
1422 : :
1423 : : SetPageLRU(page);
1424 : : lru = page_lru(page);
1425 : : add_page_to_lru_list(page, lruvec, lru);
1426 : :
1427 [ + + ]: 33403 : if (is_active_lru(lru)) {
1428 : : int file = is_file_lru(lru);
1429 : : int numpages = hpage_nr_pages(page);
1430 : 29805 : reclaim_stat->recent_rotated[file] += numpages;
1431 : : }
1432 [ - + ]: 33403 : if (put_page_testzero(page)) {
1433 : : __ClearPageLRU(page);
1434 : : __ClearPageActive(page);
1435 : : del_page_from_lru_list(page, lruvec, lru);
1436 : :
1437 [ # # ]: 0 : if (unlikely(PageCompound(page))) {
1438 : : spin_unlock_irq(&zone->lru_lock);
1439 : 0 : (*get_compound_page_dtor(page))(page);
1440 : : spin_lock_irq(&zone->lru_lock);
1441 : : } else
1442 : : list_add(&page->lru, &pages_to_free);
1443 : : }
1444 : : }
1445 : :
1446 : : /*
1447 : : * To save our caller's stack, now use input list for pages to free.
1448 : : */
1449 : : list_splice(&pages_to_free, page_list);
1450 : 7503 : }
1451 : :
1452 : : /*
1453 : : * shrink_inactive_list() is a helper for shrink_zone(). It returns the number
1454 : : * of reclaimed pages
1455 : : */
1456 : : static noinline_for_stack unsigned long
1457 : 0 : shrink_inactive_list(unsigned long nr_to_scan, struct lruvec *lruvec,
1458 : : struct scan_control *sc, enum lru_list lru)
1459 : : {
1460 : 58834 : LIST_HEAD(page_list);
1461 : : unsigned long nr_scanned;
1462 : : unsigned long nr_reclaimed = 0;
1463 : : unsigned long nr_taken;
1464 : 58834 : unsigned long nr_dirty = 0;
1465 : 58834 : unsigned long nr_congested = 0;
1466 : 58834 : unsigned long nr_unqueued_dirty = 0;
1467 : 58834 : unsigned long nr_writeback = 0;
1468 : 58834 : unsigned long nr_immediate = 0;
1469 : : isolate_mode_t isolate_mode = 0;
1470 : : int file = is_file_lru(lru);
1471 : 58834 : struct zone *zone = lruvec_zone(lruvec);
1472 : : struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
1473 : :
1474 [ - + ]: 58834 : while (unlikely(too_many_isolated(zone, file, sc))) {
1475 : 0 : congestion_wait(BLK_RW_ASYNC, HZ/10);
1476 : :
1477 : : /* We are about to die and free our memory. Return now. */
1478 [ # # ]: 0 : if (fatal_signal_pending(current))
1479 : : return SWAP_CLUSTER_MAX;
1480 : : }
1481 : :
1482 : 58839 : lru_add_drain();
1483 : :
1484 [ - + ]: 58843 : if (!sc->may_unmap)
1485 : : isolate_mode |= ISOLATE_UNMAPPED;
1486 [ - + ]: 58843 : if (!sc->may_writepage)
1487 : 0 : isolate_mode |= ISOLATE_CLEAN;
1488 : :
1489 : : spin_lock_irq(&zone->lru_lock);
1490 : :
1491 : 58846 : nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &page_list,
1492 : : &nr_scanned, sc, isolate_mode, lru);
1493 : :
1494 : 58843 : __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken);
1495 : 58845 : __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken);
1496 : :
1497 : : if (global_reclaim(sc)) {
1498 : 58845 : zone->pages_scanned += nr_scanned;
1499 [ + + ]: 58845 : if (current_is_kswapd())
1500 : 10842 : __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scanned);
1501 : : else
1502 : 48003 : __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scanned);
1503 : : }
1504 : : spin_unlock_irq(&zone->lru_lock);
1505 : :
1506 [ + + ]: 58846 : if (nr_taken == 0)
1507 : : return 0;
1508 : :
1509 : 7503 : nr_reclaimed = shrink_page_list(&page_list, zone, sc, TTU_UNMAP,
1510 : : &nr_dirty, &nr_unqueued_dirty, &nr_congested,
1511 : : &nr_writeback, &nr_immediate,
1512 : : false);
1513 : :
1514 : : spin_lock_irq(&zone->lru_lock);
1515 : :
1516 : 7503 : reclaim_stat->recent_scanned[file] += nr_taken;
1517 : :
1518 : : if (global_reclaim(sc)) {
1519 [ + + ]: 7503 : if (current_is_kswapd())
1520 : 7000 : __count_zone_vm_events(PGSTEAL_KSWAPD, zone,
1521 : : nr_reclaimed);
1522 : : else
1523 : 503 : __count_zone_vm_events(PGSTEAL_DIRECT, zone,
1524 : : nr_reclaimed);
1525 : : }
1526 : :
1527 : 7503 : putback_inactive_pages(lruvec, &page_list);
1528 : :
1529 : 7503 : __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken);
1530 : :
1531 : : spin_unlock_irq(&zone->lru_lock);
1532 : :
1533 : 7503 : free_hot_cold_page_list(&page_list, 1);
1534 : :
1535 : : /*
1536 : : * If reclaim is isolating dirty pages under writeback, it implies
1537 : : * that the long-lived page allocation rate is exceeding the page
1538 : : * laundering rate. Either the global limits are not being effective
1539 : : * at throttling processes due to the page distribution throughout
1540 : : * zones or there is heavy usage of a slow backing device. The
1541 : : * only option is to throttle from reclaim context which is not ideal
1542 : : * as there is no guarantee the dirtying process is throttled in the
1543 : : * same way balance_dirty_pages() manages.
1544 : : *
1545 : : * Once a zone is flagged ZONE_WRITEBACK, kswapd will count the number
1546 : : * of pages under pages flagged for immediate reclaim and stall if any
1547 : : * are encountered in the nr_immediate check below.
1548 : : */
1549 [ + + ][ + + ]: 7503 : if (nr_writeback && nr_writeback == nr_taken)
1550 : : zone_set_flag(zone, ZONE_WRITEBACK);
1551 : :
1552 : : /*
1553 : : * memcg will stall in page writeback so only consider forcibly
1554 : : * stalling for global reclaim
1555 : : */
1556 : : if (global_reclaim(sc)) {
1557 : : /*
1558 : : * Tag a zone as congested if all the dirty pages scanned were
1559 : : * backed by a congested BDI and wait_iff_congested will stall.
1560 : : */
1561 [ + + ][ + + ]: 7503 : if (nr_dirty && nr_dirty == nr_congested)
1562 : : zone_set_flag(zone, ZONE_CONGESTED);
1563 : :
1564 : : /*
1565 : : * If dirty pages are scanned that are not queued for IO, it
1566 : : * implies that flushers are not keeping up. In this case, flag
1567 : : * the zone ZONE_TAIL_LRU_DIRTY and kswapd will start writing
1568 : : * pages from reclaim context. It will forcibly stall in the
1569 : : * next check.
1570 : : */
1571 [ + + ]: 7503 : if (nr_unqueued_dirty == nr_taken)
1572 : : zone_set_flag(zone, ZONE_TAIL_LRU_DIRTY);
1573 : :
1574 : : /*
1575 : : * In addition, if kswapd scans pages marked marked for
1576 : : * immediate reclaim and under writeback (nr_immediate), it
1577 : : * implies that pages are cycling through the LRU faster than
1578 : : * they are written so also forcibly stall.
1579 : : */
1580 [ + + ][ + ]: 7503 : if (nr_unqueued_dirty == nr_taken || nr_immediate)
1581 : 0 : congestion_wait(BLK_RW_ASYNC, HZ/10);
1582 : : }
1583 : :
1584 : : /*
1585 : : * Stall direct reclaim for IO completions if underlying BDIs or zone
1586 : : * is congested. Allow kswapd to continue until it starts encountering
1587 : : * unqueued dirty pages or cycling through the LRU too quickly.
1588 : : */
1589 [ + - ][ + + ]: 66337 : if (!sc->hibernation_mode && !current_is_kswapd())
1590 : 503 : wait_iff_congested(zone, BLK_RW_ASYNC, HZ/10);
1591 : :
1592 [ - + ]: 7503 : trace_mm_vmscan_lru_shrink_inactive(zone->zone_pgdat->node_id,
1593 : 7503 : zone_idx(zone),
1594 : : nr_scanned, nr_reclaimed,
1595 : : sc->priority,
1596 : : trace_shrink_flags(file));
1597 : 7503 : return nr_reclaimed;
1598 : : }
1599 : :
1600 : : /*
1601 : : * This moves pages from the active list to the inactive list.
1602 : : *
1603 : : * We move them the other way if the page is referenced by one or more
1604 : : * processes, from rmap.
1605 : : *
1606 : : * If the pages are mostly unmapped, the processing is fast and it is
1607 : : * appropriate to hold zone->lru_lock across the whole operation. But if
1608 : : * the pages are mapped, the processing is slow (page_referenced()) so we
1609 : : * should drop zone->lru_lock around each page. It's impossible to balance
1610 : : * this, so instead we remove the pages from the LRU while processing them.
1611 : : * It is safe to rely on PG_active against the non-LRU pages in here because
1612 : : * nobody will play with that bit on a non-LRU page.
1613 : : *
1614 : : * The downside is that we have to touch page->_count against each page.
1615 : : * But we had to alter page->flags anyway.
1616 : : */
1617 : :
1618 : 0 : static void move_active_pages_to_lru(struct lruvec *lruvec,
1619 : : struct list_head *list,
1620 : : struct list_head *pages_to_free,
1621 : : enum lru_list lru)
1622 : : {
1623 : 21916 : struct zone *zone = lruvec_zone(lruvec);
1624 : : unsigned long pgmoved = 0;
1625 : 0 : struct page *page;
1626 : : int nr_pages;
1627 : :
1628 [ + + ]: 96051 : while (!list_empty(list)) {
1629 : 74135 : page = lru_to_page(list);
1630 : : lruvec = mem_cgroup_page_lruvec(page, zone);
1631 : :
1632 : : VM_BUG_ON(PageLRU(page));
1633 : : SetPageLRU(page);
1634 : :
1635 : : nr_pages = hpage_nr_pages(page);
1636 : : mem_cgroup_update_lru_size(lruvec, lru, nr_pages);
1637 : 74135 : list_move(&page->lru, &lruvec->lists[lru]);
1638 : 74135 : pgmoved += nr_pages;
1639 : :
1640 [ - + ]: 74135 : if (put_page_testzero(page)) {
1641 : : __ClearPageLRU(page);
1642 : : __ClearPageActive(page);
1643 : : del_page_from_lru_list(page, lruvec, lru);
1644 : :
1645 [ # # ]: 0 : if (unlikely(PageCompound(page))) {
1646 : : spin_unlock_irq(&zone->lru_lock);
1647 : 0 : (*get_compound_page_dtor(page))(page);
1648 : : spin_lock_irq(&zone->lru_lock);
1649 : : } else
1650 : : list_add(&page->lru, pages_to_free);
1651 : : }
1652 : : }
1653 : 21916 : __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
1654 [ + + ]: 21915 : if (!is_active_lru(lru))
1655 : : __count_vm_events(PGDEACTIVATE, pgmoved);
1656 : 21915 : }
1657 : :
1658 : 0 : static void shrink_active_list(unsigned long nr_to_scan,
1659 : : struct lruvec *lruvec,
1660 : : struct scan_control *sc,
1661 : : enum lru_list lru)
1662 : : {
1663 : : unsigned long nr_taken;
1664 : : unsigned long nr_scanned;
1665 : : unsigned long vm_flags;
1666 : 10958 : LIST_HEAD(l_hold); /* The pages which were snipped off */
1667 : 10958 : LIST_HEAD(l_active);
1668 : 10958 : LIST_HEAD(l_inactive);
1669 : 0 : struct page *page;
1670 : : struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
1671 : : unsigned long nr_rotated = 0;
1672 : : isolate_mode_t isolate_mode = 0;
1673 : : int file = is_file_lru(lru);
1674 : 10958 : struct zone *zone = lruvec_zone(lruvec);
1675 : :
1676 : 10958 : lru_add_drain();
1677 : :
1678 [ - + ]: 10958 : if (!sc->may_unmap)
1679 : : isolate_mode |= ISOLATE_UNMAPPED;
1680 [ - + ]: 10958 : if (!sc->may_writepage)
1681 : 0 : isolate_mode |= ISOLATE_CLEAN;
1682 : :
1683 : : spin_lock_irq(&zone->lru_lock);
1684 : :
1685 : 10958 : nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &l_hold,
1686 : : &nr_scanned, sc, isolate_mode, lru);
1687 : : if (global_reclaim(sc))
1688 : 10958 : zone->pages_scanned += nr_scanned;
1689 : :
1690 : 10958 : reclaim_stat->recent_scanned[file] += nr_taken;
1691 : :
1692 : 10958 : __count_zone_vm_events(PGREFILL, zone, nr_scanned);
1693 : 10958 : __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken);
1694 : 10958 : __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken);
1695 : : spin_unlock_irq(&zone->lru_lock);
1696 : :
1697 [ + + ]: 85090 : while (!list_empty(&l_hold)) {
1698 : 74134 : cond_resched();
1699 : 74135 : page = lru_to_page(&l_hold);
1700 : : list_del(&page->lru);
1701 : :
1702 [ - + ]: 74135 : if (unlikely(!page_evictable(page))) {
1703 : 0 : putback_lru_page(page);
1704 : 0 : continue;
1705 : : }
1706 : :
1707 [ - + ]: 74135 : if (unlikely(buffer_heads_over_limit)) {
1708 [ # # # # ]: 0 : if (page_has_private(page) && trylock_page(page)) {
1709 [ # # ]: 0 : if (page_has_private(page))
1710 : 0 : try_to_release_page(page, 0);
1711 : 0 : unlock_page(page);
1712 : : }
1713 : : }
1714 : :
1715 [ + + ]: 74135 : if (page_referenced(page, 0, sc->target_mem_cgroup,
1716 : : &vm_flags)) {
1717 : 1051 : nr_rotated += hpage_nr_pages(page);
1718 : : /*
1719 : : * Identify referenced, file-backed active pages and
1720 : : * give them one more trip around the active list. So
1721 : : * that executable code get better chances to stay in
1722 : : * memory under moderate memory pressure. Anon pages
1723 : : * are not likely to be evicted by use-once streaming
1724 : : * IO, plus JVM can create lots of anon VM_EXEC pages,
1725 : : * so we ignore them here.
1726 : : */
1727 [ + + ][ + - ]: 1051 : if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) {
1728 : 1026 : list_add(&page->lru, &l_active);
1729 : 1026 : continue;
1730 : : }
1731 : : }
1732 : :
1733 : : ClearPageActive(page); /* we are de-activating */
1734 : 73109 : list_add(&page->lru, &l_inactive);
1735 : : }
1736 : :
1737 : : /*
1738 : : * Move pages back to the lru list.
1739 : : */
1740 : : spin_lock_irq(&zone->lru_lock);
1741 : : /*
1742 : : * Count referenced pages from currently used mappings as rotated,
1743 : : * even though only some of them are actually re-activated. This
1744 : : * helps balance scan pressure between file and anonymous pages in
1745 : : * get_scan_ratio.
1746 : : */
1747 : 10957 : reclaim_stat->recent_rotated[file] += nr_rotated;
1748 : :
1749 : 10957 : move_active_pages_to_lru(lruvec, &l_active, &l_hold, lru);
1750 : 10958 : move_active_pages_to_lru(lruvec, &l_inactive, &l_hold, lru - LRU_ACTIVE);
1751 : 10958 : __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken);
1752 : : spin_unlock_irq(&zone->lru_lock);
1753 : :
1754 : 10958 : free_hot_cold_page_list(&l_hold, 1);
1755 : 10958 : }
1756 : :
1757 : : #ifdef CONFIG_SWAP
1758 : : static int inactive_anon_is_low_global(struct zone *zone)
1759 : : {
1760 : : unsigned long active, inactive;
1761 : :
1762 : : active = zone_page_state(zone, NR_ACTIVE_ANON);
1763 : : inactive = zone_page_state(zone, NR_INACTIVE_ANON);
1764 : :
1765 [ # # ][ # # ]: 0 : if (inactive * zone->inactive_ratio < active)
[ # # ]
1766 : : return 1;
1767 : :
1768 : : return 0;
1769 : : }
1770 : :
1771 : : /**
1772 : : * inactive_anon_is_low - check if anonymous pages need to be deactivated
1773 : : * @lruvec: LRU vector to check
1774 : : *
1775 : : * Returns true if the zone does not have enough inactive anon pages,
1776 : : * meaning some active anon pages need to be deactivated.
1777 : : */
1778 : : static int inactive_anon_is_low(struct lruvec *lruvec)
1779 : : {
1780 : : /*
1781 : : * If we don't have swap space, anonymous page deactivation
1782 : : * is pointless.
1783 : : */
1784 [ # # - + ]: 203477 : if (!total_swap_pages)
[ # # ]
1785 : : return 0;
1786 : :
1787 : : if (!mem_cgroup_disabled())
1788 : : return mem_cgroup_inactive_anon_is_low(lruvec);
1789 : :
1790 : : return inactive_anon_is_low_global(lruvec_zone(lruvec));
1791 : : }
1792 : : #else
1793 : : static inline int inactive_anon_is_low(struct lruvec *lruvec)
1794 : : {
1795 : : return 0;
1796 : : }
1797 : : #endif
1798 : :
1799 : : /**
1800 : : * inactive_file_is_low - check if file pages need to be deactivated
1801 : : * @lruvec: LRU vector to check
1802 : : *
1803 : : * When the system is doing streaming IO, memory pressure here
1804 : : * ensures that active file pages get deactivated, until more
1805 : : * than half of the file pages are on the inactive list.
1806 : : *
1807 : : * Once we get to that situation, protect the system's working
1808 : : * set from being evicted by disabling active file page aging.
1809 : : *
1810 : : * This uses a different ratio than the anonymous pages, because
1811 : : * the page cache uses a use-once replacement algorithm.
1812 : : */
1813 : : static int inactive_file_is_low(struct lruvec *lruvec)
1814 : : {
1815 : : unsigned long inactive;
1816 : : unsigned long active;
1817 : :
1818 : : inactive = get_lru_size(lruvec, LRU_INACTIVE_FILE);
1819 : : active = get_lru_size(lruvec, LRU_ACTIVE_FILE);
1820 : :
1821 : 60570 : return active > inactive;
1822 : : }
1823 : :
1824 : 0 : static int inactive_list_is_low(struct lruvec *lruvec, enum lru_list lru)
1825 : : {
1826 [ + - ]: 60570 : if (is_file_lru(lru))
1827 : 60570 : return inactive_file_is_low(lruvec);
1828 : : else
1829 : 0 : return inactive_anon_is_low(lruvec);
1830 : : }
1831 : :
1832 : 0 : static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
1833 : : struct lruvec *lruvec, struct scan_control *sc)
1834 : : {
1835 [ + + ]: 119398 : if (is_active_lru(lru)) {
1836 [ + + ]: 60571 : if (inactive_list_is_low(lruvec, lru))
1837 : 10957 : shrink_active_list(nr_to_scan, lruvec, sc, lru);
1838 : : return 0;
1839 : : }
1840 : :
1841 : 58827 : return shrink_inactive_list(nr_to_scan, lruvec, sc, lru);
1842 : : }
1843 : :
1844 : : static int vmscan_swappiness(struct scan_control *sc)
1845 : : {
1846 : : if (global_reclaim(sc))
1847 : 0 : return vm_swappiness;
1848 : : return mem_cgroup_swappiness(sc->target_mem_cgroup);
1849 : : }
1850 : :
1851 : : enum scan_balance {
1852 : : SCAN_EQUAL,
1853 : : SCAN_FRACT,
1854 : : SCAN_ANON,
1855 : : SCAN_FILE,
1856 : : };
1857 : :
1858 : : /*
1859 : : * Determine how aggressively the anon and file LRU lists should be
1860 : : * scanned. The relative value of each set of LRU lists is determined
1861 : : * by looking at the fraction of the pages scanned we did rotate back
1862 : : * onto the active list instead of evict.
1863 : : *
1864 : : * nr[0] = anon inactive pages to scan; nr[1] = anon active pages to scan
1865 : : * nr[2] = file inactive pages to scan; nr[3] = file active pages to scan
1866 : : */
1867 : 0 : static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
1868 : : unsigned long *nr)
1869 : : {
1870 : : struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
1871 : : u64 fraction[2];
1872 : : u64 denominator = 0; /* gcc */
1873 : 202653 : struct zone *zone = lruvec_zone(lruvec);
1874 : : unsigned long anon_prio, file_prio;
1875 : : enum scan_balance scan_balance;
1876 : : unsigned long anon, file, free;
1877 : : bool force_scan = false;
1878 : : unsigned long ap, fp;
1879 : : enum lru_list lru;
1880 : :
1881 : : /*
1882 : : * If the zone or memcg is small, nr[l] can be 0. This
1883 : : * results in no scanning on this priority and a potential
1884 : : * priority drop. Global direct reclaim can go to the next
1885 : : * zone and tends to have no problems. Global kswapd is for
1886 : : * zone balancing and it needs to scan a minimum amount. When
1887 : : * reclaiming for a memcg, a priority drop can cause high
1888 : : * latencies, so it's better to scan a minimum amount there as
1889 : : * well.
1890 : : */
1891 [ + + + + ]: 241029 : if (current_is_kswapd() && !zone_reclaimable(zone))
1892 : : force_scan = true;
1893 : : if (!global_reclaim(sc))
1894 : : force_scan = true;
1895 : :
1896 : : /* If we have no swap space, do not bother scanning anon pages. */
1897 [ + ][ - + ]: 202653 : if (!sc->may_swap || (get_nr_swap_pages() <= 0)) {
1898 : : scan_balance = SCAN_FILE;
1899 : : goto out;
1900 : : }
1901 : :
1902 : : /*
1903 : : * Global reclaim will swap to prevent OOM even with no
1904 : : * swappiness, but memcg users want to use this knob to
1905 : : * disable swapping for individual groups completely when
1906 : : * using the memory controller's swap limit feature would be
1907 : : * too expensive.
1908 : : */
1909 : : if (!global_reclaim(sc) && !vmscan_swappiness(sc)) {
1910 : : scan_balance = SCAN_FILE;
1911 : : goto out;
1912 : : }
1913 : :
1914 : : /*
1915 : : * Do not apply any pressure balancing cleverness when the
1916 : : * system is close to OOM, scan both anon and file equally
1917 : : * (unless the swappiness setting disagrees with swapping).
1918 : : */
1919 [ # # ][ # # ]: 0 : if (!sc->priority && vmscan_swappiness(sc)) {
1920 : : scan_balance = SCAN_EQUAL;
1921 : : goto out;
1922 : : }
1923 : :
1924 : 0 : anon = get_lru_size(lruvec, LRU_ACTIVE_ANON) +
1925 : : get_lru_size(lruvec, LRU_INACTIVE_ANON);
1926 : 0 : file = get_lru_size(lruvec, LRU_ACTIVE_FILE) +
1927 : : get_lru_size(lruvec, LRU_INACTIVE_FILE);
1928 : :
1929 : : /*
1930 : : * If it's foreseeable that reclaiming the file cache won't be
1931 : : * enough to get the zone back into a desirable shape, we have
1932 : : * to swap. Better start now and leave the - probably heavily
1933 : : * thrashing - remaining file pages alone.
1934 : : */
1935 : : if (global_reclaim(sc)) {
1936 : : free = zone_page_state(zone, NR_FREE_PAGES);
1937 [ # # ]: 0 : if (unlikely(file + free <= high_wmark_pages(zone))) {
1938 : : scan_balance = SCAN_ANON;
1939 : : goto out;
1940 : : }
1941 : : }
1942 : :
1943 : : /*
1944 : : * There is enough inactive page cache, do not reclaim
1945 : : * anything from the anonymous working set right now.
1946 : : */
1947 [ # # ]: 0 : if (!inactive_file_is_low(lruvec)) {
1948 : : scan_balance = SCAN_FILE;
1949 : : goto out;
1950 : : }
1951 : :
1952 : : scan_balance = SCAN_FRACT;
1953 : :
1954 : : /*
1955 : : * With swappiness at 100, anonymous and file have the same priority.
1956 : : * This scanning priority is essentially the inverse of IO cost.
1957 : : */
1958 : 0 : anon_prio = vmscan_swappiness(sc);
1959 : 0 : file_prio = 200 - anon_prio;
1960 : :
1961 : : /*
1962 : : * OK, so we have swap space and a fair amount of page cache
1963 : : * pages. We use the recently rotated / recently scanned
1964 : : * ratios to determine how valuable each cache is.
1965 : : *
1966 : : * Because workloads change over time (and to avoid overflow)
1967 : : * we keep these statistics as a floating average, which ends
1968 : : * up weighing recent references more than old ones.
1969 : : *
1970 : : * anon in [0], file in [1]
1971 : : */
1972 : : spin_lock_irq(&zone->lru_lock);
1973 [ # # ]: 0 : if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) {
1974 : 0 : reclaim_stat->recent_scanned[0] /= 2;
1975 : 0 : reclaim_stat->recent_rotated[0] /= 2;
1976 : : }
1977 : :
1978 [ # # ]: 0 : if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) {
1979 : 0 : reclaim_stat->recent_scanned[1] /= 2;
1980 : 0 : reclaim_stat->recent_rotated[1] /= 2;
1981 : : }
1982 : :
1983 : : /*
1984 : : * The amount of pressure on anon vs file pages is inversely
1985 : : * proportional to the fraction of recently scanned pages on
1986 : : * each list that were recently referenced and in active use.
1987 : : */
1988 : 0 : ap = anon_prio * (reclaim_stat->recent_scanned[0] + 1);
1989 : 0 : ap /= reclaim_stat->recent_rotated[0] + 1;
1990 : :
1991 : 0 : fp = file_prio * (reclaim_stat->recent_scanned[1] + 1);
1992 : 0 : fp /= reclaim_stat->recent_rotated[1] + 1;
1993 : : spin_unlock_irq(&zone->lru_lock);
1994 : :
1995 : 0 : fraction[0] = ap;
1996 : 0 : fraction[1] = fp;
1997 : 202653 : denominator = ap + fp + 1;
1998 : : out:
1999 [ + + ]: 1015014 : for_each_evictable_lru(lru) {
2000 : : int file = is_file_lru(lru);
2001 : : unsigned long size;
2002 : : unsigned long scan;
2003 : :
2004 : : size = get_lru_size(lruvec, lru);
2005 : 811779 : scan = size >> sc->priority;
2006 : :
2007 [ + + ]: 811779 : if (!scan && force_scan)
2008 : 5715 : scan = min(size, SWAP_CLUSTER_MAX);
2009 : :
2010 [ - + + ]: 811779 : switch (scan_balance) {
2011 : : case SCAN_EQUAL:
2012 : : /* Scan lists relative to size */
2013 : : break;
2014 : : case SCAN_FRACT:
2015 : : /*
2016 : : * Scan types proportional to swappiness and
2017 : : * their relative recent reclaim efficiency.
2018 : : */
2019 : 0 : scan = div64_u64(scan * fraction[file], denominator);
2020 : : break;
2021 : : case SCAN_FILE:
2022 : : case SCAN_ANON:
2023 : : /* Scan one type exclusively */
2024 [ + + ]: 811834 : if ((scan_balance == SCAN_FILE) != file)
2025 : : scan = 0;
2026 : : break;
2027 : : default:
2028 : : /* Look ma, no brain */
2029 : 0 : BUG();
2030 : : }
2031 : 812361 : nr[lru] = scan;
2032 : : }
2033 : 203235 : }
2034 : :
2035 : : /*
2036 : : * This is a basic per-zone page freer. Used by both kswapd and direct reclaim.
2037 : : */
2038 : 0 : static void shrink_lruvec(struct lruvec *lruvec, struct scan_control *sc)
2039 : : {
2040 : : unsigned long nr[NR_LRU_LISTS];
2041 : : unsigned long targets[NR_LRU_LISTS];
2042 : : unsigned long nr_to_scan;
2043 : : enum lru_list lru;
2044 : : unsigned long nr_reclaimed = 0;
2045 : 203240 : unsigned long nr_to_reclaim = sc->nr_to_reclaim;
2046 : : struct blk_plug plug;
2047 : : bool scan_adjusted = false;
2048 : :
2049 : 203240 : get_scan_count(lruvec, sc, nr);
2050 : :
2051 : : /* Record the original scan target for proportional adjustments later */
2052 : 203407 : memcpy(targets, nr, sizeof(nr));
2053 : :
2054 : 203407 : blk_start_plug(&plug);
2055 [ + + ][ + + ]: 268317 : while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] ||
[ + + ]
2056 : 207792 : nr[LRU_INACTIVE_FILE]) {
2057 : : unsigned long nr_anon, nr_file, percentage;
2058 : : unsigned long nr_scanned;
2059 : :
2060 [ + + ]: 324137 : for_each_evictable_lru(lru) {
2061 [ + + ]: 259295 : if (nr[lru]) {
2062 : 0 : nr_to_scan = min(nr[lru], SWAP_CLUSTER_MAX);
2063 : 0 : nr[lru] -= nr_to_scan;
2064 : :
2065 : 119399 : nr_reclaimed += shrink_list(lru, nr_to_scan,
2066 : : lruvec, sc);
2067 : : }
2068 : : }
2069 : :
2070 [ + - ]: 64842 : if (nr_reclaimed < nr_to_reclaim || scan_adjusted)
2071 : 64842 : continue;
2072 : :
2073 : : /*
2074 : : * For global direct reclaim, reclaim only the number of pages
2075 : : * requested. Less care is taken to scan proportionally as it
2076 : : * is more important to minimise direct reclaim stall latency
2077 : : * than it is to properly age the LRU lists.
2078 : : */
2079 [ # # ]: 0 : if (global_reclaim(sc) && !current_is_kswapd())
2080 : : break;
2081 : :
2082 : : /*
2083 : : * For kswapd and memcg, reclaim at least the number of pages
2084 : : * requested. Ensure that the anon and file LRUs shrink
2085 : : * proportionally what was requested by get_scan_count(). We
2086 : : * stop reclaiming one LRU and reduce the amount scanning
2087 : : * proportional to the original scan target.
2088 : : */
2089 : 0 : nr_file = nr[LRU_INACTIVE_FILE] + nr[LRU_ACTIVE_FILE];
2090 : 0 : nr_anon = nr[LRU_INACTIVE_ANON] + nr[LRU_ACTIVE_ANON];
2091 : :
2092 [ # # ]: 0 : if (nr_file > nr_anon) {
2093 : 0 : unsigned long scan_target = targets[LRU_INACTIVE_ANON] +
2094 : 0 : targets[LRU_ACTIVE_ANON] + 1;
2095 : : lru = LRU_BASE;
2096 : 0 : percentage = nr_anon * 100 / scan_target;
2097 : : } else {
2098 : 0 : unsigned long scan_target = targets[LRU_INACTIVE_FILE] +
2099 : 0 : targets[LRU_ACTIVE_FILE] + 1;
2100 : : lru = LRU_FILE;
2101 : 0 : percentage = nr_file * 100 / scan_target;
2102 : : }
2103 : :
2104 : : /* Stop scanning the smaller of the LRU */
2105 : 0 : nr[lru] = 0;
2106 : 0 : nr[lru + LRU_ACTIVE] = 0;
2107 : :
2108 : : /*
2109 : : * Recalculate the other LRU scan count based on its original
2110 : : * scan target and the percentage scanning already complete
2111 : : */
2112 [ # # ]: 0 : lru = (lru == LRU_FILE) ? LRU_BASE : LRU_FILE;
2113 : 0 : nr_scanned = targets[lru] - nr[lru];
2114 : 0 : nr[lru] = targets[lru] * (100 - percentage) / 100;
2115 : 0 : nr[lru] -= min(nr[lru], nr_scanned);
2116 : :
2117 : 0 : lru += LRU_ACTIVE;
2118 : 0 : nr_scanned = targets[lru] - nr[lru];
2119 : 0 : nr[lru] = targets[lru] * (100 - percentage) / 100;
2120 : 64842 : nr[lru] -= min(nr[lru], nr_scanned);
2121 : :
2122 : : scan_adjusted = true;
2123 : : }
2124 : 203488 : blk_finish_plug(&plug);
2125 : 203477 : sc->nr_reclaimed += nr_reclaimed;
2126 : :
2127 : : /*
2128 : : * Even if we did not try to evict anon pages at all, we want to
2129 : : * rebalance the anon lru active/inactive ratio.
2130 : : */
2131 [ - + ]: 203477 : if (inactive_anon_is_low(lruvec))
2132 : 0 : shrink_active_list(SWAP_CLUSTER_MAX, lruvec,
2133 : : sc, LRU_ACTIVE_ANON);
2134 : :
2135 : 203477 : throttle_vm_writeout(sc->gfp_mask);
2136 : 202181 : }
2137 : :
2138 : : /* Use reclaim/compaction for costly allocs or under memory pressure */
2139 : 202376 : static bool in_reclaim_compaction(struct scan_control *sc)
2140 : : {
2141 [ + + ][ + - ]: 202376 : if (IS_ENABLED(CONFIG_COMPACTION) && sc->order &&
2142 [ + ]: 202376 : (sc->order > PAGE_ALLOC_COSTLY_ORDER ||
2143 : 2 : sc->priority < DEF_PRIORITY - 2))
2144 : : return true;
2145 : :
2146 : : return false;
2147 : : }
2148 : :
2149 : : /*
2150 : : * Reclaim/compaction is used for high-order allocation requests. It reclaims
2151 : : * order-0 pages before compacting the zone. should_continue_reclaim() returns
2152 : : * true if more pages should be reclaimed such that when the page allocator
2153 : : * calls try_to_compact_zone() that it will have enough free pages to succeed.
2154 : : * It will give up earlier than that if there is difficulty reclaiming pages.
2155 : : */
2156 : : static inline bool should_continue_reclaim(struct zone *zone,
2157 : : unsigned long nr_reclaimed,
2158 : : unsigned long nr_scanned,
2159 : 202229 : struct scan_control *sc)
2160 : : {
2161 : : unsigned long pages_for_compaction;
2162 : : unsigned long inactive_lru_pages;
2163 : :
2164 : : /* If not in reclaim/compaction mode, stop */
2165 [ - + ]: 202229 : if (!in_reclaim_compaction(sc))
2166 : : return false;
2167 : :
2168 : : /* Consider stopping depending on scan and reclaim activity */
2169 [ # # ]: 0 : if (sc->gfp_mask & __GFP_REPEAT) {
2170 : : /*
2171 : : * For __GFP_REPEAT allocations, stop reclaiming if the
2172 : : * full LRU list has been scanned and we are still failing
2173 : : * to reclaim pages. This full LRU scan is potentially
2174 : : * expensive but a __GFP_REPEAT caller really wants to succeed
2175 : : */
2176 [ # # ]: 0 : if (!nr_reclaimed && !nr_scanned)
2177 : : return false;
2178 : : } else {
2179 : : /*
2180 : : * For non-__GFP_REPEAT allocations which can presumably
2181 : : * fail without consequence, stop if we failed to reclaim
2182 : : * any pages from the last SWAP_CLUSTER_MAX number of
2183 : : * pages that were scanned. This will return to the
2184 : : * caller faster at the risk reclaim/compaction and
2185 : : * the resulting allocation attempt fails
2186 : : */
2187 [ # # ]: 0 : if (!nr_reclaimed)
2188 : : return false;
2189 : : }
2190 : :
2191 : : /*
2192 : : * If we have not reclaimed enough pages for compaction and the
2193 : : * inactive lists are large enough, continue reclaiming
2194 : : */
2195 : 107 : pages_for_compaction = (2UL << sc->order);
2196 : : inactive_lru_pages = zone_page_state(zone, NR_INACTIVE_FILE);
2197 [ - + ]: 107 : if (get_nr_swap_pages() > 0)
2198 : 0 : inactive_lru_pages += zone_page_state(zone, NR_INACTIVE_ANON);
2199 [ - ][ # # ]: 107 : if (sc->nr_reclaimed < pages_for_compaction &&
2200 : : inactive_lru_pages > pages_for_compaction)
2201 : : return true;
2202 : :
2203 : : /* If compaction would go ahead or the allocation would succeed, stop */
2204 [ # # ]: 0 : switch (compaction_suitable(zone, sc->order)) {
2205 : : case COMPACT_PARTIAL:
2206 : : case COMPACT_CONTINUE:
2207 : : return false;
2208 : : default:
2209 : : return true;
2210 : : }
2211 : : }
2212 : :
2213 : 203303 : static void shrink_zone(struct zone *zone, struct scan_control *sc)
2214 : : {
2215 : : unsigned long nr_reclaimed, nr_scanned;
2216 : :
2217 : : do {
2218 : : struct mem_cgroup *root = sc->target_mem_cgroup;
2219 : : struct mem_cgroup_reclaim_cookie reclaim = {
2220 : : .zone = zone,
2221 : : .priority = sc->priority,
2222 : : };
2223 : : struct mem_cgroup *memcg;
2224 : :
2225 : 203303 : nr_reclaimed = sc->nr_reclaimed;
2226 : 203303 : nr_scanned = sc->nr_scanned;
2227 : :
2228 : : memcg = mem_cgroup_iter(root, NULL, &reclaim);
2229 : : do {
2230 : : struct lruvec *lruvec;
2231 : :
2232 : 203303 : lruvec = mem_cgroup_zone_lruvec(zone, memcg);
2233 : :
2234 : 203303 : shrink_lruvec(lruvec, sc);
2235 : :
2236 : : /*
2237 : : * Direct reclaim and kswapd have to scan all memory
2238 : : * cgroups to fulfill the overall scan target for the
2239 : : * zone.
2240 : : *
2241 : : * Limit reclaim, on the other hand, only cares about
2242 : : * nr_to_reclaim pages to be reclaimed and it will
2243 : : * retry with decreasing priority if one round over the
2244 : : * whole hierarchy is not sufficient.
2245 : : */
2246 : : if (!global_reclaim(sc) &&
2247 : : sc->nr_reclaimed >= sc->nr_to_reclaim) {
2248 : : mem_cgroup_iter_break(root, memcg);
2249 : : break;
2250 : : }
2251 : : memcg = mem_cgroup_iter(root, memcg, &reclaim);
2252 : : } while (memcg);
2253 : :
2254 : : vmpressure(sc->gfp_mask, sc->target_mem_cgroup,
2255 : : sc->nr_scanned - nr_scanned,
2256 : : sc->nr_reclaimed - nr_reclaimed);
2257 : :
2258 : 404458 : } while (should_continue_reclaim(zone, sc->nr_reclaimed - nr_reclaimed,
2259 [ - + ]: 203566 : sc->nr_scanned - nr_scanned, sc));
2260 : 203566 : }
2261 : :
2262 : : /* Returns true if compaction should go ahead for a high-order request */
2263 : : static inline bool compaction_ready(struct zone *zone, struct scan_control *sc)
2264 : : {
2265 : : unsigned long balance_gap, watermark;
2266 : : bool watermark_ok;
2267 : :
2268 : : /* Do not consider compaction for orders reclaim is meant to satisfy */
2269 [ - + ]: 165322 : if (sc->order <= PAGE_ALLOC_COSTLY_ORDER)
2270 : : return false;
2271 : :
2272 : : /*
2273 : : * Compaction takes time to run and there are potentially other
2274 : : * callers using the pages just freed. Continue reclaiming until
2275 : : * there is a buffer of free pages available to give compaction
2276 : : * a reasonable chance of completing and allocating the page
2277 : : */
2278 : 0 : balance_gap = min(low_wmark_pages(zone),
2279 : : (zone->managed_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) /
2280 : : KSWAPD_ZONE_BALANCE_GAP_RATIO);
2281 : 0 : watermark = high_wmark_pages(zone) + balance_gap + (2UL << sc->order);
2282 : 0 : watermark_ok = zone_watermark_ok_safe(zone, 0, watermark, 0, 0);
2283 : :
2284 : : /*
2285 : : * If compaction is deferred, reclaim up to a point where
2286 : : * compaction will have a chance of success when re-enabled
2287 : : */
2288 [ # # ]: 0 : if (compaction_deferred(zone, sc->order))
2289 : : return watermark_ok;
2290 : :
2291 : : /* If compaction is not ready to start, keep reclaiming */
2292 [ # # ]: 0 : if (!compaction_suitable(zone, sc->order))
2293 : : return false;
2294 : :
2295 : : return watermark_ok;
2296 : : }
2297 : :
2298 : : /*
2299 : : * This is the direct reclaim path, for page-allocating processes. We only
2300 : : * try to reclaim pages from zones which will satisfy the caller's allocation
2301 : : * request.
2302 : : *
2303 : : * We reclaim from a zone even if that zone is over high_wmark_pages(zone).
2304 : : * Because:
2305 : : * a) The caller may be trying to free *extra* pages to satisfy a higher-order
2306 : : * allocation or
2307 : : * b) The target zone may be at high_wmark_pages(zone) but the lower zones
2308 : : * must go *over* high_wmark_pages(zone) to satisfy the `incremental min'
2309 : : * zone defense algorithm.
2310 : : *
2311 : : * If a zone is deemed to be full of pinned pages then just give it a light
2312 : : * scan then give up on it.
2313 : : *
2314 : : * This function returns true if a zone is being reclaimed for a costly
2315 : : * high-order allocation and compaction is ready to begin. This indicates to
2316 : : * the caller that it should consider retrying the allocation instead of
2317 : : * further reclaim.
2318 : : */
2319 : 0 : static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
2320 : : {
2321 : : struct zoneref *z;
2322 : : struct zone *zone;
2323 : : unsigned long nr_soft_reclaimed;
2324 : : unsigned long nr_soft_scanned;
2325 : : bool aborted_reclaim = false;
2326 : :
2327 : : /*
2328 : : * If the number of buffer_heads in the machine exceeds the maximum
2329 : : * allowed level, force direct reclaim to scan the highmem zone as
2330 : : * highmem pages could be pinning lowmem pages storing buffer_heads
2331 : : */
2332 [ - + ]: 132528 : if (buffer_heads_over_limit)
2333 : 0 : sc->gfp_mask |= __GFP_HIGHMEM;
2334 : :
2335 [ + + ]: 264241 : for_each_zone_zonelist_nodemask(zone, z, zonelist,
2336 : : gfp_zone(sc->gfp_mask), sc->nodemask) {
2337 [ + ]: 264574 : if (!populated_zone(zone))
2338 : 0 : continue;
2339 : : /*
2340 : : * Take care memory controller reclaiming has small influence
2341 : : * to global LRU.
2342 : : */
2343 : : if (global_reclaim(sc)) {
2344 : : if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
2345 : : continue;
2346 [ + + + + ]: 641935 : if (sc->priority != DEF_PRIORITY &&
2347 : : !zone_reclaimable(zone))
2348 : 99657 : continue; /* Let kswapd poll it */
2349 : : if (IS_ENABLED(CONFIG_COMPACTION)) {
2350 : : /*
2351 : : * If we already have plenty of memory free for
2352 : : * compaction in this zone, don't free any more.
2353 : : * Even though compaction is invoked for any
2354 : : * non-zero order, only frequent costly order
2355 : : * reclamation is disruptive enough to become a
2356 : : * noticeable problem, like transparent huge
2357 : : * page allocations.
2358 : : */
2359 [ - + ]: 165322 : if (compaction_ready(zone, sc)) {
2360 : : aborted_reclaim = true;
2361 : 0 : continue;
2362 : : }
2363 : : }
2364 : : /*
2365 : : * This steals pages from memory cgroups over softlimit
2366 : : * and returns the number of reclaimed pages and
2367 : : * scanned pages. This works for global memory pressure
2368 : : * and balancing, not for a memcg's limit.
2369 : : */
2370 : : nr_soft_scanned = 0;
2371 : : nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone,
2372 : : sc->order, sc->gfp_mask,
2373 : : &nr_soft_scanned);
2374 : : sc->nr_reclaimed += nr_soft_reclaimed;
2375 : : sc->nr_scanned += nr_soft_scanned;
2376 : : /* need some check for avoid more shrink_zone() */
2377 : : }
2378 : :
2379 : 165322 : shrink_zone(zone, sc);
2380 : : }
2381 : :
2382 : 132565 : return aborted_reclaim;
2383 : : }
2384 : :
2385 : : /* All zones in zonelist are unreclaimable? */
2386 : 10148 : static bool all_unreclaimable(struct zonelist *zonelist,
2387 : : struct scan_control *sc)
2388 : : {
2389 : : struct zoneref *z;
2390 : : struct zone *zone;
2391 : :
2392 [ + + ]: 17530 : for_each_zone_zonelist_nodemask(zone, z, zonelist,
2393 : : gfp_zone(sc->gfp_mask), sc->nodemask) {
2394 [ - + ]: 17133 : if (!populated_zone(zone))
2395 : 0 : continue;
2396 : : if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
2397 : : continue;
2398 [ + + ]: 17133 : if (zone_reclaimable(zone))
2399 : : return false;
2400 : : }
2401 : :
2402 : : return true;
2403 : : }
2404 : :
2405 : : /*
2406 : : * This is the main entry point to direct page reclaim.
2407 : : *
2408 : : * If a full scan of the inactive list fails to free enough memory then we
2409 : : * are "out of memory" and something needs to be killed.
2410 : : *
2411 : : * If the caller is !__GFP_FS then the probability of a failure is reasonably
2412 : : * high - the zone may be full of dirty or under-writeback pages, which this
2413 : : * caller can't do much about. We kick the writeback threads and take explicit
2414 : : * naps in the hope that some of these pages can be written. But if the
2415 : : * allocating task holds filesystem locks which prevent writeout this might not
2416 : : * work, and the allocation attempt will fail.
2417 : : *
2418 : : * returns: 0, if no pages reclaimed
2419 : : * else, the number of pages reclaimed
2420 : : */
2421 : 0 : static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
2422 : : struct scan_control *sc,
2423 : : struct shrink_control *shrink)
2424 : : {
2425 : : unsigned long total_scanned = 0;
2426 : 10198 : struct reclaim_state *reclaim_state = current->reclaim_state;
2427 : : struct zoneref *z;
2428 : : struct zone *zone;
2429 : : unsigned long writeback_threshold;
2430 : : bool aborted_reclaim;
2431 : :
2432 : : delayacct_freepages_start();
2433 : :
2434 : : if (global_reclaim(sc))
2435 : : count_vm_event(ALLOCSTALL);
2436 : :
2437 : : do {
2438 : : vmpressure_prio(sc->gfp_mask, sc->target_mem_cgroup,
2439 : : sc->priority);
2440 : 132525 : sc->nr_scanned = 0;
2441 : 132525 : aborted_reclaim = shrink_zones(zonelist, sc);
2442 : :
2443 : : /*
2444 : : * Don't shrink slabs when reclaiming memory from over limit
2445 : : * cgroups but do shrink slab at least once when aborting
2446 : : * reclaim for compaction to avoid unevenly scanning file/anon
2447 : : * LRU pages over slab pages.
2448 : : */
2449 : : if (global_reclaim(sc)) {
2450 : : unsigned long lru_pages = 0;
2451 : :
2452 : : nodes_clear(shrink->nodes_to_scan);
2453 [ + + ]: 397699 : for_each_zone_zonelist(zone, z, zonelist,
2454 : : gfp_zone(sc->gfp_mask)) {
2455 : : if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
2456 : : continue;
2457 : :
2458 : 265146 : lru_pages += zone_reclaimable_pages(zone);
2459 : : node_set(zone_to_nid(zone),
2460 : : shrink->nodes_to_scan);
2461 : : }
2462 : :
2463 : 132574 : shrink_slab(shrink, sc->nr_scanned, lru_pages);
2464 [ + ]: 132574 : if (reclaim_state) {
2465 : 132581 : sc->nr_reclaimed += reclaim_state->reclaimed_slab;
2466 : 132581 : reclaim_state->reclaimed_slab = 0;
2467 : : }
2468 : : }
2469 : 132574 : total_scanned += sc->nr_scanned;
2470 [ + ]: 132574 : if (sc->nr_reclaimed >= sc->nr_to_reclaim)
2471 : : goto out;
2472 : :
2473 : : /*
2474 : : * If we're getting trouble reclaiming, start doing
2475 : : * writepage even in laptop mode.
2476 : : */
2477 [ + + ]: 132584 : if (sc->priority < DEF_PRIORITY - 2)
2478 : 101987 : sc->may_writepage = 1;
2479 : :
2480 : : /*
2481 : : * Try to write back as many pages as we just scanned. This
2482 : : * tends to cause slow streaming writers to write data to the
2483 : : * disk smoothly, at the dirtying rate, which is nice. But
2484 : : * that's undesirable in laptop mode, where we *want* lumpy
2485 : : * writeout. So in laptop mode, write out the whole world.
2486 : : */
2487 : 132584 : writeback_threshold = sc->nr_to_reclaim + sc->nr_to_reclaim / 2;
2488 [ + + ]: 132584 : if (total_scanned > writeback_threshold) {
2489 [ + - ]: 13 : wakeup_flusher_threads(laptop_mode ? 0 : total_scanned,
2490 : : WB_REASON_TRY_TO_FREE_PAGES);
2491 : 13 : sc->may_writepage = 1;
2492 : : }
2493 [ + + ][ + ]: 132584 : } while (--sc->priority >= 0 && !aborted_reclaim);
2494 : :
2495 : : out:
2496 : : delayacct_freepages_end();
2497 : :
2498 [ + ]: 10247 : if (sc->nr_reclaimed)
2499 : : return sc->nr_reclaimed;
2500 : :
2501 : : /*
2502 : : * As hibernation is going on, kswapd is freezed so that it can't mark
2503 : : * the zone into all_unreclaimable. Thus bypassing all_unreclaimable
2504 : : * check.
2505 : : */
2506 [ + - ]: 10148 : if (oom_killer_disabled)
2507 : : return 0;
2508 : :
2509 : : /* Aborted reclaim to try compaction? don't OOM, then */
2510 [ + - ]: 10148 : if (aborted_reclaim)
2511 : : return 1;
2512 : :
2513 : : /* top priority shrink_zones still had more to do? don't OOM, then */
2514 [ + + ]: 10148 : if (global_reclaim(sc) && !all_unreclaimable(zonelist, sc))
2515 : : return 1;
2516 : :
2517 : : return 0;
2518 : : }
2519 : :
2520 : 0 : static bool pfmemalloc_watermark_ok(pg_data_t *pgdat)
2521 : : {
2522 : : struct zone *zone;
2523 : : unsigned long pfmemalloc_reserve = 0;
2524 : : unsigned long free_pages = 0;
2525 : : int i;
2526 : : bool wmark_ok;
2527 : :
2528 [ + + ]: 20397 : for (i = 0; i <= ZONE_NORMAL; i++) {
2529 : 10199 : zone = &pgdat->node_zones[i];
2530 : 10199 : pfmemalloc_reserve += min_wmark_pages(zone);
2531 : 10199 : free_pages += zone_page_state(zone, NR_FREE_PAGES);
2532 : : }
2533 : :
2534 : 10198 : wmark_ok = free_pages > pfmemalloc_reserve / 2;
2535 : :
2536 : : /* kswapd must be awake if processes are being throttled */
2537 [ - + ][ # # ]: 10198 : if (!wmark_ok && waitqueue_active(&pgdat->kswapd_wait)) {
2538 : 0 : pgdat->classzone_idx = min(pgdat->classzone_idx,
2539 : : (enum zone_type)ZONE_NORMAL);
2540 : 0 : wake_up_interruptible(&pgdat->kswapd_wait);
2541 : : }
2542 : :
2543 : 0 : return wmark_ok;
2544 : : }
2545 : :
2546 : : /*
2547 : : * Throttle direct reclaimers if backing storage is backed by the network
2548 : : * and the PFMEMALLOC reserve for the preferred node is getting dangerously
2549 : : * depleted. kswapd will continue to make progress and wake the processes
2550 : : * when the low watermark is reached.
2551 : : *
2552 : : * Returns true if a fatal signal was delivered during throttling. If this
2553 : : * happens, the page allocator should not consider triggering the OOM killer.
2554 : : */
2555 : 0 : static bool throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist,
2556 : : nodemask_t *nodemask)
2557 : : {
2558 : : struct zone *zone;
2559 : : int high_zoneidx = gfp_zone(gfp_mask);
2560 : : pg_data_t *pgdat;
2561 : :
2562 : : /*
2563 : : * Kernel threads should not be throttled as they may be indirectly
2564 : : * responsible for cleaning pages necessary for reclaim to make forward
2565 : : * progress. kjournald for example may enter direct reclaim while
2566 : : * committing a transaction where throttling it could forcing other
2567 : : * processes to block on log_wait_commit().
2568 : : */
2569 [ + + ]: 10199 : if (current->flags & PF_KTHREAD)
2570 : : goto out;
2571 : :
2572 : : /*
2573 : : * If a fatal signal is pending, this process should not throttle.
2574 : : * It should return quickly so it can exit and free its memory
2575 : : */
2576 [ + + ]: 10198 : if (fatal_signal_pending(current))
2577 : : goto out;
2578 : :
2579 : : /* Check if the pfmemalloc reserves are ok */
2580 : : first_zones_zonelist(zonelist, high_zoneidx, NULL, &zone);
2581 : 10199 : pgdat = zone->zone_pgdat;
2582 [ - + ]: 10199 : if (pfmemalloc_watermark_ok(pgdat))
2583 : : goto out;
2584 : :
2585 : : /* Account for the throttling */
2586 : : count_vm_event(PGSCAN_DIRECT_THROTTLE);
2587 : :
2588 : : /*
2589 : : * If the caller cannot enter the filesystem, it's possible that it
2590 : : * is due to the caller holding an FS lock or performing a journal
2591 : : * transaction in the case of a filesystem like ext[3|4]. In this case,
2592 : : * it is not safe to block on pfmemalloc_wait as kswapd could be
2593 : : * blocked waiting on the same lock. Instead, throttle for up to a
2594 : : * second before continuing.
2595 : : */
2596 [ # # ]: 0 : if (!(gfp_mask & __GFP_FS)) {
2597 [ # # ][ # # ]: 0 : wait_event_interruptible_timeout(pgdat->pfmemalloc_wait,
[ # # ][ # # ]
2598 : : pfmemalloc_watermark_ok(pgdat), HZ);
2599 : :
2600 : : goto check_pending;
2601 : : }
2602 : :
2603 : : /* Throttle until kswapd wakes the process */
2604 [ # # ][ # # ]: 0 : wait_event_killable(zone->zone_pgdat->pfmemalloc_wait,
[ # # ]
2605 : : pfmemalloc_watermark_ok(pgdat));
2606 : :
2607 : : check_pending:
2608 [ # # ]: 0 : if (fatal_signal_pending(current))
2609 : : return true;
2610 : :
2611 : : out:
2612 : : return false;
2613 : : }
2614 : :
2615 : 0 : unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
2616 : : gfp_t gfp_mask, nodemask_t *nodemask)
2617 : : {
2618 : : unsigned long nr_reclaimed;
2619 : 30591 : struct scan_control sc = {
2620 : : .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)),
2621 : 10197 : .may_writepage = !laptop_mode,
2622 : : .nr_to_reclaim = SWAP_CLUSTER_MAX,
2623 : : .may_unmap = 1,
2624 : : .may_swap = 1,
2625 : : .order = order,
2626 : : .priority = DEF_PRIORITY,
2627 : : .target_mem_cgroup = NULL,
2628 : : .nodemask = nodemask,
2629 : : };
2630 : 10197 : struct shrink_control shrink = {
2631 : : .gfp_mask = sc.gfp_mask,
2632 : : };
2633 : :
2634 : : /*
2635 : : * Do not enter reclaim if fatal signal was delivered while throttled.
2636 : : * 1 is returned so that the page allocator does not OOM kill at this
2637 : : * point.
2638 : : */
2639 [ + + ]: 10197 : if (throttle_direct_reclaim(gfp_mask, zonelist, nodemask))
2640 : : return 1;
2641 : :
2642 : 10197 : trace_mm_vmscan_direct_reclaim_begin(order,
2643 : : sc.may_writepage,
2644 : : gfp_mask);
2645 : :
2646 : 10197 : nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink);
2647 : :
2648 : : trace_mm_vmscan_direct_reclaim_end(nr_reclaimed);
2649 : :
2650 : 10198 : return nr_reclaimed;
2651 : : }
2652 : :
2653 : : #ifdef CONFIG_MEMCG
2654 : :
2655 : : unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *memcg,
2656 : : gfp_t gfp_mask, bool noswap,
2657 : : struct zone *zone,
2658 : : unsigned long *nr_scanned)
2659 : : {
2660 : : struct scan_control sc = {
2661 : : .nr_scanned = 0,
2662 : : .nr_to_reclaim = SWAP_CLUSTER_MAX,
2663 : : .may_writepage = !laptop_mode,
2664 : : .may_unmap = 1,
2665 : : .may_swap = !noswap,
2666 : : .order = 0,
2667 : : .priority = 0,
2668 : : .target_mem_cgroup = memcg,
2669 : : };
2670 : : struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg);
2671 : :
2672 : : sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
2673 : : (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK);
2674 : :
2675 : : trace_mm_vmscan_memcg_softlimit_reclaim_begin(sc.order,
2676 : : sc.may_writepage,
2677 : : sc.gfp_mask);
2678 : :
2679 : : /*
2680 : : * NOTE: Although we can get the priority field, using it
2681 : : * here is not a good idea, since it limits the pages we can scan.
2682 : : * if we don't reclaim here, the shrink_zone from balance_pgdat
2683 : : * will pick up pages from other mem cgroup's as well. We hack
2684 : : * the priority and make it zero.
2685 : : */
2686 : : shrink_lruvec(lruvec, &sc);
2687 : :
2688 : : trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed);
2689 : :
2690 : : *nr_scanned = sc.nr_scanned;
2691 : : return sc.nr_reclaimed;
2692 : : }
2693 : :
2694 : : unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg,
2695 : : gfp_t gfp_mask,
2696 : : bool noswap)
2697 : : {
2698 : : struct zonelist *zonelist;
2699 : : unsigned long nr_reclaimed;
2700 : : int nid;
2701 : : struct scan_control sc = {
2702 : : .may_writepage = !laptop_mode,
2703 : : .may_unmap = 1,
2704 : : .may_swap = !noswap,
2705 : : .nr_to_reclaim = SWAP_CLUSTER_MAX,
2706 : : .order = 0,
2707 : : .priority = DEF_PRIORITY,
2708 : : .target_mem_cgroup = memcg,
2709 : : .nodemask = NULL, /* we don't care the placement */
2710 : : .gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
2711 : : (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK),
2712 : : };
2713 : : struct shrink_control shrink = {
2714 : : .gfp_mask = sc.gfp_mask,
2715 : : };
2716 : :
2717 : : /*
2718 : : * Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't
2719 : : * take care of from where we get pages. So the node where we start the
2720 : : * scan does not need to be the current node.
2721 : : */
2722 : : nid = mem_cgroup_select_victim_node(memcg);
2723 : :
2724 : : zonelist = NODE_DATA(nid)->node_zonelists;
2725 : :
2726 : : trace_mm_vmscan_memcg_reclaim_begin(0,
2727 : : sc.may_writepage,
2728 : : sc.gfp_mask);
2729 : :
2730 : : nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink);
2731 : :
2732 : : trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed);
2733 : :
2734 : : return nr_reclaimed;
2735 : : }
2736 : : #endif
2737 : :
2738 : 0 : static void age_active_anon(struct zone *zone, struct scan_control *sc)
2739 : : {
2740 : : struct mem_cgroup *memcg;
2741 : :
2742 [ - + ]: 26045 : if (!total_swap_pages)
2743 : 0 : return;
2744 : :
2745 : : memcg = mem_cgroup_iter(NULL, NULL, NULL);
2746 : : do {
2747 : 0 : struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg);
2748 : :
2749 [ - ]: 0 : if (inactive_anon_is_low(lruvec))
2750 : 0 : shrink_active_list(SWAP_CLUSTER_MAX, lruvec,
2751 : : sc, LRU_ACTIVE_ANON);
2752 : :
2753 : : memcg = mem_cgroup_iter(NULL, memcg, NULL);
2754 : : } while (memcg);
2755 : : }
2756 : :
2757 : 0 : static bool zone_balanced(struct zone *zone, int order,
2758 : : unsigned long balance_gap, int classzone_idx)
2759 : : {
2760 [ + + ]: 174493 : if (!zone_watermark_ok_safe(zone, order, high_wmark_pages(zone) +
2761 : : balance_gap, classzone_idx, 0))
2762 : : return false;
2763 : :
2764 [ + + + - ]: 47547 : if (IS_ENABLED(CONFIG_COMPACTION) && order &&
2765 : 921 : !compaction_suitable(zone, order))
2766 : : return false;
2767 : :
2768 : : return true;
2769 : : }
2770 : :
2771 : : /*
2772 : : * pgdat_balanced() is used when checking if a node is balanced.
2773 : : *
2774 : : * For order-0, all zones must be balanced!
2775 : : *
2776 : : * For high-order allocations only zones that meet watermarks and are in a
2777 : : * zone allowed by the callers classzone_idx are added to balanced_pages. The
2778 : : * total of balanced pages must be at least 25% of the zones allowed by
2779 : : * classzone_idx for the node to be considered balanced. Forcing all zones to
2780 : : * be balanced for high orders can cause excessive reclaim when there are
2781 : : * imbalanced zones.
2782 : : * The choice of 25% is due to
2783 : : * o a 16M DMA zone that is balanced will not balance a zone on any
2784 : : * reasonable sized machine
2785 : : * o On all other machines, the top zone must be at least a reasonable
2786 : : * percentage of the middle zones. For example, on 32-bit x86, highmem
2787 : : * would need to be at least 256M for it to be balance a whole node.
2788 : : * Similarly, on x86-64 the Normal zone would need to be at least 1G
2789 : : * to balance a node on its own. These seemed like reasonable ratios.
2790 : : */
2791 : 0 : static bool pgdat_balanced(pg_data_t *pgdat, int order, int classzone_idx)
2792 : : {
2793 : : unsigned long managed_pages = 0;
2794 : : unsigned long balanced_pages = 0;
2795 : : int i;
2796 : :
2797 : : /* Check the watermark levels */
2798 [ + + ]: 58143 : for (i = 0; i <= classzone_idx; i++) {
2799 : 57171 : struct zone *zone = pgdat->node_zones + i;
2800 : :
2801 [ - + ]: 57171 : if (!populated_zone(zone))
2802 : 0 : continue;
2803 : :
2804 : 57171 : managed_pages += zone->managed_pages;
2805 : :
2806 : : /*
2807 : : * A special case here:
2808 : : *
2809 : : * balance_pgdat() skips over all_unreclaimable after
2810 : : * DEF_PRIORITY. Effectively, it considers them balanced so
2811 : : * they must be considered balanced here as well!
2812 : : */
2813 [ + + ]: 57171 : if (!zone_reclaimable(zone)) {
2814 : 15590 : balanced_pages += zone->managed_pages;
2815 : 15590 : continue;
2816 : : }
2817 : :
2818 [ + + ]: 41581 : if (zone_balanced(zone, order, 0, i))
2819 : 13966 : balanced_pages += zone->managed_pages;
2820 [ + + ]: 27615 : else if (!order)
2821 : : return false;
2822 : : }
2823 : :
2824 [ + + ]: 972 : if (order)
2825 : 2 : return balanced_pages >= (managed_pages >> 2);
2826 : : else
2827 : : return true;
2828 : : }
2829 : :
2830 : : /*
2831 : : * Prepare kswapd for sleeping. This verifies that there are no processes
2832 : : * waiting in throttle_direct_reclaim() and that watermarks have been met.
2833 : : *
2834 : : * Returns true if kswapd is ready to sleep
2835 : : */
2836 : 0 : static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, long remaining,
2837 : : int classzone_idx)
2838 : : {
2839 : : /* If a direct reclaimer woke kswapd within HZ/10, it's premature */
2840 [ + + ]: 5060 : if (remaining)
2841 : : return false;
2842 : :
2843 : : /*
2844 : : * There is a potential race between when kswapd checks its watermarks
2845 : : * and a process gets throttled. There is also a potential race if
2846 : : * processes get throttled, kswapd wakes, a large process exits therby
2847 : : * balancing the zones that causes kswapd to miss a wakeup. If kswapd
2848 : : * is going to sleep, no process should be sleeping on pfmemalloc_wait
2849 : : * so wake them now if necessary. If necessary, processes will wake
2850 : : * kswapd and get throttled again
2851 : : */
2852 [ - + ]: 4592 : if (waitqueue_active(&pgdat->pfmemalloc_wait)) {
2853 : 0 : wake_up(&pgdat->pfmemalloc_wait);
2854 : 0 : return false;
2855 : : }
2856 : :
2857 : 4592 : return pgdat_balanced(pgdat, order, classzone_idx);
2858 : : }
2859 : :
2860 : : /*
2861 : : * kswapd shrinks the zone by the number of pages required to reach
2862 : : * the high watermark.
2863 : : *
2864 : : * Returns true if kswapd scanned at least the requested number of pages to
2865 : : * reclaim or if the lack of progress was due to pages under writeback.
2866 : : * This is used to determine if the scanning priority needs to be raised.
2867 : : */
2868 : 0 : static bool kswapd_shrink_zone(struct zone *zone,
2869 : : int classzone_idx,
2870 : : struct scan_control *sc,
2871 : : unsigned long lru_pages,
2872 : : unsigned long *nr_attempted)
2873 : : {
2874 : 40328 : int testorder = sc->order;
2875 : : unsigned long balance_gap;
2876 : 40328 : struct reclaim_state *reclaim_state = current->reclaim_state;
2877 : 80656 : struct shrink_control shrink = {
2878 : 40328 : .gfp_mask = sc->gfp_mask,
2879 : : };
2880 : : bool lowmem_pressure;
2881 : :
2882 : : /* Reclaim above the high watermark. */
2883 : 40328 : sc->nr_to_reclaim = max(SWAP_CLUSTER_MAX, high_wmark_pages(zone));
2884 : :
2885 : : /*
2886 : : * Kswapd reclaims only single pages with compaction enabled. Trying
2887 : : * too hard to reclaim until contiguous free pages have become
2888 : : * available can hurt performance by evicting too much useful data
2889 : : * from memory. Do not reclaim more than needed for compaction.
2890 : : */
2891 [ + + + + ]: 40330 : if (IS_ENABLED(CONFIG_COMPACTION) && sc->order &&
2892 : 2 : compaction_suitable(zone, sc->order) !=
2893 : : COMPACT_SKIPPED)
2894 : : testorder = 0;
2895 : :
2896 : : /*
2897 : : * We put equal pressure on every zone, unless one zone has way too
2898 : : * many pages free already. The "too many pages" is defined as the
2899 : : * high wmark plus a "gap" where the gap is either the low
2900 : : * watermark or 1% of the zone, whichever is smaller.
2901 : : */
2902 : 40328 : balance_gap = min(low_wmark_pages(zone),
2903 : : (zone->managed_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) /
2904 : : KSWAPD_ZONE_BALANCE_GAP_RATIO);
2905 : :
2906 : : /*
2907 : : * If there is no low memory pressure or the zone is balanced then no
2908 : : * reclaim is necessary
2909 : : */
2910 [ - + ][ # # ]: 40328 : lowmem_pressure = (buffer_heads_over_limit && is_highmem(zone));
2911 [ + - ][ + + ]: 40328 : if (!lowmem_pressure && zone_balanced(zone, testorder,
2912 : : balance_gap, classzone_idx))
2913 : : return true;
2914 : :
2915 : 38376 : shrink_zone(zone, sc);
2916 : : nodes_clear(shrink.nodes_to_scan);
2917 : : node_set(zone_to_nid(zone), shrink.nodes_to_scan);
2918 : :
2919 : 38376 : reclaim_state->reclaimed_slab = 0;
2920 : 38376 : shrink_slab(&shrink, sc->nr_scanned, lru_pages);
2921 : 38376 : sc->nr_reclaimed += reclaim_state->reclaimed_slab;
2922 : :
2923 : : /* Account for the number of pages attempted to reclaim */
2924 : 38376 : *nr_attempted += sc->nr_to_reclaim;
2925 : :
2926 : : zone_clear_flag(zone, ZONE_WRITEBACK);
2927 : :
2928 : : /*
2929 : : * If a zone reaches its high watermark, consider it to be no longer
2930 : : * congested. It's possible there are dirty pages backed by congested
2931 : : * BDIs but as pressure is relieved, speculatively avoid congestion
2932 : : * waits.
2933 : : */
2934 [ + + + + ]: 74837 : if (zone_reclaimable(zone) &&
2935 : 36461 : zone_balanced(zone, testorder, 0, classzone_idx)) {
2936 : : zone_clear_flag(zone, ZONE_CONGESTED);
2937 : : zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY);
2938 : : }
2939 : :
2940 : 38376 : return sc->nr_scanned >= sc->nr_to_reclaim;
2941 : : }
2942 : :
2943 : : /*
2944 : : * For kswapd, balance_pgdat() will work across all this node's zones until
2945 : : * they are all at high_wmark_pages(zone).
2946 : : *
2947 : : * Returns the final order kswapd was reclaiming at
2948 : : *
2949 : : * There is special handling here for zones which are full of pinned pages.
2950 : : * This can happen if the pages are all mlocked, or if they are all used by
2951 : : * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb.
2952 : : * What we do is to detect the case where all pages in the zone have been
2953 : : * scanned twice and there has been zero successful reclaim. Mark the zone as
2954 : : * dead and from now on, only perform a short scan. Basically we're polling
2955 : : * the zone for when the problem goes away.
2956 : : *
2957 : : * kswapd scans the zones in the highmem->normal->dma direction. It skips
2958 : : * zones which have free_pages > high_wmark_pages(zone), but once a zone is
2959 : : * found to have free_pages <= high_wmark_pages(zone), we scan that zone and the
2960 : : * lower zones regardless of the number of free pages in the lower zones. This
2961 : : * interoperates with the page allocator fallback scheme to ensure that aging
2962 : : * of pages is balanced across the zones.
2963 : : */
2964 : 0 : static unsigned long balance_pgdat(pg_data_t *pgdat, int order,
2965 : : int *classzone_idx)
2966 : : {
2967 : : int i;
2968 : : int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */
2969 : : unsigned long nr_soft_reclaimed;
2970 : : unsigned long nr_soft_scanned;
2971 : 5060 : struct scan_control sc = {
2972 : : .gfp_mask = GFP_KERNEL,
2973 : : .priority = DEF_PRIORITY,
2974 : : .may_unmap = 1,
2975 : : .may_swap = 1,
2976 : 2530 : .may_writepage = !laptop_mode,
2977 : : .order = order,
2978 : : .target_mem_cgroup = NULL,
2979 : : };
2980 : : count_vm_event(PAGEOUTRUN);
2981 : :
2982 : : do {
2983 : : unsigned long lru_pages = 0;
2984 : 26045 : unsigned long nr_attempted = 0;
2985 : : bool raise_priority = true;
2986 : 26045 : bool pgdat_needs_compaction = (order > 0);
2987 : :
2988 : 26045 : sc.nr_reclaimed = 0;
2989 : :
2990 : : /*
2991 : : * Scan in the highmem->dma direction for the highest
2992 : : * zone which needs scanning
2993 : : */
2994 [ + - ]: 26045 : for (i = pgdat->nr_zones - 1; i >= 0; i--) {
2995 : 26045 : struct zone *zone = pgdat->node_zones + i;
2996 : :
2997 [ - + ]: 26045 : if (!populated_zone(zone))
2998 : 0 : continue;
2999 : :
3000 [ + + - + ]: 49087 : if (sc.priority != DEF_PRIORITY &&
3001 : : !zone_reclaimable(zone))
3002 : 0 : continue;
3003 : :
3004 : : /*
3005 : : * Do some background aging of the anon list, to give
3006 : : * pages a chance to be referenced before reclaiming.
3007 : : */
3008 : 26045 : age_active_anon(zone, &sc);
3009 : :
3010 : : /*
3011 : : * If the number of buffer_heads in the machine
3012 : : * exceeds the maximum allowed level and this node
3013 : : * has a highmem zone, force kswapd to reclaim from
3014 : : * it to relieve lowmem pressure.
3015 : : */
3016 [ - + ][ # # ]: 26045 : if (buffer_heads_over_limit && is_highmem_idx(i)) {
3017 : : end_zone = i;
3018 : : break;
3019 : : }
3020 : :
3021 [ - + ]: 26045 : if (!zone_balanced(zone, order, 0, 0)) {
3022 : : end_zone = i;
3023 : : break;
3024 : : } else {
3025 : : /*
3026 : : * If balanced, clear the dirty and congested
3027 : : * flags
3028 : : */
3029 : : zone_clear_flag(zone, ZONE_CONGESTED);
3030 : : zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY);
3031 : : }
3032 : : }
3033 : :
3034 [ + - ]: 26045 : if (i < 0)
3035 : : goto out;
3036 : :
3037 [ + + ]: 78135 : for (i = 0; i <= end_zone; i++) {
3038 : 52090 : struct zone *zone = pgdat->node_zones + i;
3039 : :
3040 [ - + ]: 52090 : if (!populated_zone(zone))
3041 : 0 : continue;
3042 : :
3043 : 52090 : lru_pages += zone_reclaimable_pages(zone);
3044 : :
3045 : : /*
3046 : : * If any zone is currently balanced then kswapd will
3047 : : * not call compaction as it is expected that the
3048 : : * necessary pages are already available.
3049 : : */
3050 [ + + + - ]: 52091 : if (pgdat_needs_compaction &&
3051 : 1 : zone_watermark_ok(zone, order,
3052 : : low_wmark_pages(zone),
3053 : : *classzone_idx, 0))
3054 : : pgdat_needs_compaction = false;
3055 : : }
3056 : :
3057 : : /*
3058 : : * If we're getting trouble reclaiming, start doing writepage
3059 : : * even in laptop mode.
3060 : : */
3061 [ + + ]: 26045 : if (sc.priority < DEF_PRIORITY - 2)
3062 : 26045 : sc.may_writepage = 1;
3063 : :
3064 : : /*
3065 : : * Now scan the zone in the dma->highmem direction, stopping
3066 : : * at the last zone which needs scanning.
3067 : : *
3068 : : * We do this because the page allocator works in the opposite
3069 : : * direction. This prevents the page allocator from allocating
3070 : : * pages behind kswapd's direction of progress, which would
3071 : : * cause too much scanning of the lower zones.
3072 : : */
3073 [ + + ]: 78135 : for (i = 0; i <= end_zone; i++) {
3074 : 52090 : struct zone *zone = pgdat->node_zones + i;
3075 : :
3076 [ - + ]: 52090 : if (!populated_zone(zone))
3077 : 0 : continue;
3078 : :
3079 [ + + + + ]: 98174 : if (sc.priority != DEF_PRIORITY &&
3080 : : !zone_reclaimable(zone))
3081 : 11762 : continue;
3082 : :
3083 : 40328 : sc.nr_scanned = 0;
3084 : :
3085 : : nr_soft_scanned = 0;
3086 : : /*
3087 : : * Call soft limit reclaim before calling shrink_zone.
3088 : : */
3089 : : nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone,
3090 : : order, sc.gfp_mask,
3091 : : &nr_soft_scanned);
3092 : : sc.nr_reclaimed += nr_soft_reclaimed;
3093 : :
3094 : : /*
3095 : : * There should be no need to raise the scanning
3096 : : * priority if enough pages are already being scanned
3097 : : * that that high watermark would be met at 100%
3098 : : * efficiency.
3099 : : */
3100 [ + + ]: 40328 : if (kswapd_shrink_zone(zone, end_zone, &sc,
3101 : : lru_pages, &nr_attempted))
3102 : : raise_priority = false;
3103 : : }
3104 : :
3105 : : /*
3106 : : * If the low watermark is met there is no need for processes
3107 : : * to be throttled on pfmemalloc_wait as they should not be
3108 : : * able to safely make forward progress. Wake them
3109 : : */
3110 [ - + # # ]: 26045 : if (waitqueue_active(&pgdat->pfmemalloc_wait) &&
3111 : 0 : pfmemalloc_watermark_ok(pgdat))
3112 : 0 : wake_up(&pgdat->pfmemalloc_wait);
3113 : :
3114 : : /*
3115 : : * Fragmentation may mean that the system cannot be rebalanced
3116 : : * for high-order allocations in all zones. If twice the
3117 : : * allocation size has been reclaimed and the zones are still
3118 : : * not balanced then recheck the watermarks at order-0 to
3119 : : * prevent kswapd reclaiming excessively. Assume that a
3120 : : * process requested a high-order can direct reclaim/compact.
3121 : : */
3122 [ + + ][ - + ]: 26045 : if (order && sc.nr_reclaimed >= 2UL << order)
3123 : 0 : order = sc.order = 0;
3124 : :
3125 : : /* Check if kswapd should be suspending */
3126 [ + - ][ + - ]: 26045 : if (try_to_freeze() || kthread_should_stop())
3127 : : break;
3128 : :
3129 : : /*
3130 : : * Compact if necessary and kswapd is reclaiming at least the
3131 : : * high watermark number of pages as requsted
3132 : : */
3133 [ - + ][ # # ]: 26045 : if (pgdat_needs_compaction && sc.nr_reclaimed > nr_attempted)
3134 : 0 : compact_pgdat(pgdat, order);
3135 : :
3136 : : /*
3137 : : * Raise priority if scanning rate is too low or there was no
3138 : : * progress in reclaiming pages
3139 : : */
3140 [ + + ][ + + ]: 26045 : if (raise_priority || !sc.nr_reclaimed)
3141 : 25319 : sc.priority--;
3142 [ + + ]: 23994 : } while (sc.priority >= 1 &&
3143 [ + + ]: 26045 : !pgdat_balanced(pgdat, order, *classzone_idx));
3144 : :
3145 : : out:
3146 : : /*
3147 : : * Return the order we were reclaiming at so prepare_kswapd_sleep()
3148 : : * makes a decision on the order we were last reclaiming at. However,
3149 : : * if another caller entered the allocator slow path while kswapd
3150 : : * was awake, order will remain at the higher level
3151 : : */
3152 : 2530 : *classzone_idx = end_zone;
3153 : 2530 : return order;
3154 : : }
3155 : :
3156 : 0 : static void kswapd_try_to_sleep(pg_data_t *pgdat, int order, int classzone_idx)
3157 : : {
3158 : : long remaining = 0;
3159 : 5060 : DEFINE_WAIT(wait);
3160 : :
3161 [ + - ][ + - ]: 2530 : if (freezing(current) || kthread_should_stop())
3162 : 0 : return;
3163 : :
3164 : 2530 : prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE);
3165 : :
3166 : : /* Try to sleep for a short interval */
3167 [ + + ]: 2530 : if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) {
3168 : 482 : remaining = schedule_timeout(HZ/10);
3169 : 482 : finish_wait(&pgdat->kswapd_wait, &wait);
3170 : 482 : prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE);
3171 : : }
3172 : :
3173 : : /*
3174 : : * After a short sleep, check if it was a premature sleep. If not, then
3175 : : * go fully to sleep until explicitly woken up.
3176 : : */
3177 [ + + ]: 2530 : if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) {
3178 : 11 : trace_mm_vmscan_kswapd_sleep(pgdat->node_id);
3179 : :
3180 : : /*
3181 : : * vmstat counters are not perfectly accurate and the estimated
3182 : : * value for counters such as NR_FREE_PAGES can deviate from the
3183 : : * true value by nr_online_cpus * threshold. To avoid the zone
3184 : : * watermarks being breached while under pressure, we reduce the
3185 : : * per-cpu vmstat threshold while kswapd is awake and restore
3186 : : * them before going back to sleep.
3187 : : */
3188 : 11 : set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold);
3189 : :
3190 : : /*
3191 : : * Compaction records what page blocks it recently failed to
3192 : : * isolate pages from and skips them in the future scanning.
3193 : : * When kswapd is going to sleep, it is reasonable to assume
3194 : : * that pages and compaction may succeed so reset the cache.
3195 : : */
3196 : 11 : reset_isolation_suitable(pgdat);
3197 : :
3198 [ + - ]: 11 : if (!kthread_should_stop())
3199 : 11 : schedule();
3200 : :
3201 : 11 : set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold);
3202 : : } else {
3203 [ + + ]: 2519 : if (remaining)
3204 : : count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY);
3205 : : else
3206 : : count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY);
3207 : : }
3208 : 2530 : finish_wait(&pgdat->kswapd_wait, &wait);
3209 : : }
3210 : :
3211 : : /*
3212 : : * The background pageout daemon, started as a kernel thread
3213 : : * from the init process.
3214 : : *
3215 : : * This basically trickles out pages so that we have _some_
3216 : : * free memory available even if there is no other activity
3217 : : * that frees anything up. This is needed for things like routing
3218 : : * etc, where we otherwise might have all activity going on in
3219 : : * asynchronous contexts that cannot page things out.
3220 : : *
3221 : : * If there are applications that are active memory-allocators
3222 : : * (most normal use), this basically shouldn't matter.
3223 : : */
3224 : 0 : static int kswapd(void *p)
3225 : : {
3226 : : unsigned long order, new_order;
3227 : : unsigned balanced_order;
3228 : : int classzone_idx, new_classzone_idx;
3229 : : int balanced_classzone_idx;
3230 : : pg_data_t *pgdat = (pg_data_t*)p;
3231 : 0 : struct task_struct *tsk = current;
3232 : :
3233 : 0 : struct reclaim_state reclaim_state = {
3234 : : .reclaimed_slab = 0,
3235 : : };
3236 : 0 : const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
3237 : :
3238 : : lockdep_set_current_reclaim_state(GFP_KERNEL);
3239 : :
3240 [ # # ]: 0 : if (!cpumask_empty(cpumask))
3241 : 0 : set_cpus_allowed_ptr(tsk, cpumask);
3242 : 0 : current->reclaim_state = &reclaim_state;
3243 : :
3244 : : /*
3245 : : * Tell the memory management that we're a "memory allocator",
3246 : : * and that if we need more memory we should get access to it
3247 : : * regardless (see "__alloc_pages()"). "kswapd" should
3248 : : * never get caught in the normal page freeing logic.
3249 : : *
3250 : : * (Kswapd normally doesn't need memory anyway, but sometimes
3251 : : * you need a small amount of memory in order to be able to
3252 : : * page out something else, and this flag essentially protects
3253 : : * us from recursively trying to free more memory as we're
3254 : : * trying to free the first piece of memory in the first place).
3255 : : */
3256 : 0 : tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
3257 : 0 : set_freezable();
3258 : :
3259 : : order = new_order = 0;
3260 : : balanced_order = 0;
3261 : 0 : classzone_idx = new_classzone_idx = pgdat->nr_zones - 1;
3262 : 2530 : balanced_classzone_idx = classzone_idx;
3263 : : for ( ; ; ) {
3264 : : bool ret;
3265 : :
3266 : : /*
3267 : : * If the last balance_pgdat was unsuccessful it's unlikely a
3268 : : * new request of a similar or harder type will succeed soon
3269 : : * so consider going to sleep on the basis we reclaimed at
3270 : : */
3271 [ + - ]: 2530 : if (balanced_classzone_idx >= new_classzone_idx &&
3272 : 2530 : balanced_order == new_order) {
3273 : 2530 : new_order = pgdat->kswapd_max_order;
3274 : 2530 : new_classzone_idx = pgdat->classzone_idx;
3275 : 2530 : pgdat->kswapd_max_order = 0;
3276 : 2530 : pgdat->classzone_idx = pgdat->nr_zones - 1;
3277 : : }
3278 : :
3279 [ + - ]: 2530 : if (order < new_order || classzone_idx > new_classzone_idx) {
3280 : : /*
3281 : : * Don't sleep if someone wants a larger 'order'
3282 : : * allocation or has tigher zone constraints
3283 : : */
3284 : : order = new_order;
3285 : : classzone_idx = new_classzone_idx;
3286 : : } else {
3287 : 2530 : kswapd_try_to_sleep(pgdat, balanced_order,
3288 : : balanced_classzone_idx);
3289 : 2530 : order = pgdat->kswapd_max_order;
3290 : 2530 : classzone_idx = pgdat->classzone_idx;
3291 : : new_order = order;
3292 : : new_classzone_idx = classzone_idx;
3293 : 2530 : pgdat->kswapd_max_order = 0;
3294 : 2530 : pgdat->classzone_idx = pgdat->nr_zones - 1;
3295 : : }
3296 : :
3297 : : ret = try_to_freeze();
3298 [ + - ]: 2530 : if (kthread_should_stop())
3299 : : break;
3300 : :
3301 : : /*
3302 : : * We can speed up thawing tasks if we don't call balance_pgdat
3303 : : * after returning from the refrigerator
3304 : : */
3305 [ - + ]: 2530 : if (!ret) {
3306 : 2530 : trace_mm_vmscan_kswapd_wake(pgdat->node_id, order);
3307 : 2530 : balanced_classzone_idx = classzone_idx;
3308 : 2530 : balanced_order = balance_pgdat(pgdat, order,
3309 : : &balanced_classzone_idx);
3310 : : }
3311 : : }
3312 : :
3313 : 0 : current->reclaim_state = NULL;
3314 : 0 : return 0;
3315 : : }
3316 : :
3317 : : /*
3318 : : * A zone is low on free memory, so wake its kswapd task to service it.
3319 : : */
3320 : 0 : void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx)
3321 : : {
3322 : : pg_data_t *pgdat;
3323 : :
3324 [ + + ]: 75713 : if (!populated_zone(zone))
3325 : : return;
3326 : :
3327 : : if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
3328 : : return;
3329 : 75688 : pgdat = zone->zone_pgdat;
3330 [ + + ]: 75688 : if (pgdat->kswapd_max_order < order) {
3331 : 3 : pgdat->kswapd_max_order = order;
3332 : 3 : pgdat->classzone_idx = min(pgdat->classzone_idx, classzone_idx);
3333 : : }
3334 [ + + ]: 75688 : if (!waitqueue_active(&pgdat->kswapd_wait))
3335 : : return;
3336 [ + + ]: 30068 : if (zone_balanced(zone, order, 0, 0))
3337 : : return;
3338 : :
3339 : 484 : trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order);
3340 : 484 : wake_up_interruptible(&pgdat->kswapd_wait);
3341 : : }
3342 : :
3343 : : /*
3344 : : * The reclaimable count would be mostly accurate.
3345 : : * The less reclaimable pages may be
3346 : : * - mlocked pages, which will be moved to unevictable list when encountered
3347 : : * - mapped pages, which may require several travels to be reclaimed
3348 : : * - dirty pages, which is not "instantly" reclaimable
3349 : : */
3350 : 0 : unsigned long global_reclaimable_pages(void)
3351 : : {
3352 : : int nr;
3353 : :
3354 : 234717 : nr = global_page_state(NR_ACTIVE_FILE) +
3355 : : global_page_state(NR_INACTIVE_FILE);
3356 : :
3357 [ + ]: 234717 : if (get_nr_swap_pages() > 0)
3358 : 16 : nr += global_page_state(NR_ACTIVE_ANON) +
3359 : : global_page_state(NR_INACTIVE_ANON);
3360 : :
3361 : 0 : return nr;
3362 : : }
3363 : :
3364 : : #ifdef CONFIG_HIBERNATION
3365 : : /*
3366 : : * Try to free `nr_to_reclaim' of memory, system-wide, and return the number of
3367 : : * freed pages.
3368 : : *
3369 : : * Rather than trying to age LRUs the aim is to preserve the overall
3370 : : * LRU order by reclaiming preferentially
3371 : : * inactive > active > active referenced > active mapped
3372 : : */
3373 : : unsigned long shrink_all_memory(unsigned long nr_to_reclaim)
3374 : : {
3375 : : struct reclaim_state reclaim_state;
3376 : : struct scan_control sc = {
3377 : : .gfp_mask = GFP_HIGHUSER_MOVABLE,
3378 : : .may_swap = 1,
3379 : : .may_unmap = 1,
3380 : : .may_writepage = 1,
3381 : : .nr_to_reclaim = nr_to_reclaim,
3382 : : .hibernation_mode = 1,
3383 : : .order = 0,
3384 : : .priority = DEF_PRIORITY,
3385 : : };
3386 : : struct shrink_control shrink = {
3387 : : .gfp_mask = sc.gfp_mask,
3388 : : };
3389 : : struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask);
3390 : : struct task_struct *p = current;
3391 : : unsigned long nr_reclaimed;
3392 : :
3393 : : p->flags |= PF_MEMALLOC;
3394 : : lockdep_set_current_reclaim_state(sc.gfp_mask);
3395 : : reclaim_state.reclaimed_slab = 0;
3396 : : p->reclaim_state = &reclaim_state;
3397 : :
3398 : : nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink);
3399 : :
3400 : : p->reclaim_state = NULL;
3401 : : lockdep_clear_current_reclaim_state();
3402 : : p->flags &= ~PF_MEMALLOC;
3403 : :
3404 : : return nr_reclaimed;
3405 : : }
3406 : : #endif /* CONFIG_HIBERNATION */
3407 : :
3408 : : /* It's optimal to keep kswapds on the same CPUs as their memory, but
3409 : : not required for correctness. So if the last cpu in a node goes
3410 : : away, we get changed to run anywhere: as the first one comes back,
3411 : : restore their cpu bindings. */
3412 : 0 : static int cpu_callback(struct notifier_block *nfb, unsigned long action,
3413 : : void *hcpu)
3414 : : {
3415 : : int nid;
3416 : :
3417 [ # # ]: 0 : if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) {
3418 [ # # ]: 0 : for_each_node_state(nid, N_MEMORY) {
3419 : : pg_data_t *pgdat = NODE_DATA(nid);
3420 : : const struct cpumask *mask;
3421 : :
3422 : 0 : mask = cpumask_of_node(pgdat->node_id);
3423 : :
3424 [ # # ]: 0 : if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids)
3425 : : /* One of our CPUs online: restore mask */
3426 : 0 : set_cpus_allowed_ptr(pgdat->kswapd, mask);
3427 : : }
3428 : : }
3429 : 0 : return NOTIFY_OK;
3430 : : }
3431 : :
3432 : : /*
3433 : : * This kswapd start function will be called by init and node-hot-add.
3434 : : * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added.
3435 : : */
3436 : 0 : int kswapd_run(int nid)
3437 : : {
3438 : : pg_data_t *pgdat = NODE_DATA(nid);
3439 : : int ret = 0;
3440 : :
3441 [ # # ]: 0 : if (pgdat->kswapd)
3442 : : return 0;
3443 : :
3444 [ # # ]: 0 : pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid);
3445 [ # # ]: 0 : if (IS_ERR(pgdat->kswapd)) {
3446 : : /* failure at boot is fatal */
3447 [ # # ]: 0 : BUG_ON(system_state == SYSTEM_BOOTING);
3448 : 0 : pr_err("Failed to start kswapd on node %d\n", nid);
3449 : 0 : ret = PTR_ERR(pgdat->kswapd);
3450 : 0 : pgdat->kswapd = NULL;
3451 : : }
3452 : 0 : return ret;
3453 : : }
3454 : :
3455 : : /*
3456 : : * Called by memory hotplug when all memory in a node is offlined. Caller must
3457 : : * hold lock_memory_hotplug().
3458 : : */
3459 : 0 : void kswapd_stop(int nid)
3460 : : {
3461 : 0 : struct task_struct *kswapd = NODE_DATA(nid)->kswapd;
3462 : :
3463 [ # # ]: 0 : if (kswapd) {
3464 : 0 : kthread_stop(kswapd);
3465 : 0 : NODE_DATA(nid)->kswapd = NULL;
3466 : : }
3467 : 0 : }
3468 : :
3469 : 0 : static int __init kswapd_init(void)
3470 : : {
3471 : : int nid;
3472 : :
3473 : 0 : swap_setup();
3474 [ # # ]: 0 : for_each_node_state(nid, N_MEMORY)
3475 : 0 : kswapd_run(nid);
3476 : 0 : hotcpu_notifier(cpu_callback, 0);
3477 : 0 : return 0;
3478 : : }
3479 : :
3480 : : module_init(kswapd_init)
3481 : :
3482 : : #ifdef CONFIG_NUMA
3483 : : /*
3484 : : * Zone reclaim mode
3485 : : *
3486 : : * If non-zero call zone_reclaim when the number of free pages falls below
3487 : : * the watermarks.
3488 : : */
3489 : : int zone_reclaim_mode __read_mostly;
3490 : :
3491 : : #define RECLAIM_OFF 0
3492 : : #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */
3493 : : #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */
3494 : : #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */
3495 : :
3496 : : /*
3497 : : * Priority for ZONE_RECLAIM. This determines the fraction of pages
3498 : : * of a node considered for each zone_reclaim. 4 scans 1/16th of
3499 : : * a zone.
3500 : : */
3501 : : #define ZONE_RECLAIM_PRIORITY 4
3502 : :
3503 : : /*
3504 : : * Percentage of pages in a zone that must be unmapped for zone_reclaim to
3505 : : * occur.
3506 : : */
3507 : : int sysctl_min_unmapped_ratio = 1;
3508 : :
3509 : : /*
3510 : : * If the number of slab pages in a zone grows beyond this percentage then
3511 : : * slab reclaim needs to occur.
3512 : : */
3513 : : int sysctl_min_slab_ratio = 5;
3514 : :
3515 : : static inline unsigned long zone_unmapped_file_pages(struct zone *zone)
3516 : : {
3517 : : unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED);
3518 : : unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) +
3519 : : zone_page_state(zone, NR_ACTIVE_FILE);
3520 : :
3521 : : /*
3522 : : * It's possible for there to be more file mapped pages than
3523 : : * accounted for by the pages on the file LRU lists because
3524 : : * tmpfs pages accounted for as ANON can also be FILE_MAPPED
3525 : : */
3526 : : return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0;
3527 : : }
3528 : :
3529 : : /* Work out how many page cache pages we can reclaim in this reclaim_mode */
3530 : : static long zone_pagecache_reclaimable(struct zone *zone)
3531 : : {
3532 : : long nr_pagecache_reclaimable;
3533 : : long delta = 0;
3534 : :
3535 : : /*
3536 : : * If RECLAIM_SWAP is set, then all file pages are considered
3537 : : * potentially reclaimable. Otherwise, we have to worry about
3538 : : * pages like swapcache and zone_unmapped_file_pages() provides
3539 : : * a better estimate
3540 : : */
3541 : : if (zone_reclaim_mode & RECLAIM_SWAP)
3542 : : nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES);
3543 : : else
3544 : : nr_pagecache_reclaimable = zone_unmapped_file_pages(zone);
3545 : :
3546 : : /* If we can't clean pages, remove dirty pages from consideration */
3547 : : if (!(zone_reclaim_mode & RECLAIM_WRITE))
3548 : : delta += zone_page_state(zone, NR_FILE_DIRTY);
3549 : :
3550 : : /* Watch for any possible underflows due to delta */
3551 : : if (unlikely(delta > nr_pagecache_reclaimable))
3552 : : delta = nr_pagecache_reclaimable;
3553 : :
3554 : : return nr_pagecache_reclaimable - delta;
3555 : : }
3556 : :
3557 : : /*
3558 : : * Try to free up some pages from this zone through reclaim.
3559 : : */
3560 : : static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
3561 : : {
3562 : : /* Minimum pages needed in order to stay on node */
3563 : : const unsigned long nr_pages = 1 << order;
3564 : : struct task_struct *p = current;
3565 : : struct reclaim_state reclaim_state;
3566 : : struct scan_control sc = {
3567 : : .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE),
3568 : : .may_unmap = !!(zone_reclaim_mode & RECLAIM_SWAP),
3569 : : .may_swap = 1,
3570 : : .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX),
3571 : : .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)),
3572 : : .order = order,
3573 : : .priority = ZONE_RECLAIM_PRIORITY,
3574 : : };
3575 : : struct shrink_control shrink = {
3576 : : .gfp_mask = sc.gfp_mask,
3577 : : };
3578 : : unsigned long nr_slab_pages0, nr_slab_pages1;
3579 : :
3580 : : cond_resched();
3581 : : /*
3582 : : * We need to be able to allocate from the reserves for RECLAIM_SWAP
3583 : : * and we also need to be able to write out pages for RECLAIM_WRITE
3584 : : * and RECLAIM_SWAP.
3585 : : */
3586 : : p->flags |= PF_MEMALLOC | PF_SWAPWRITE;
3587 : : lockdep_set_current_reclaim_state(gfp_mask);
3588 : : reclaim_state.reclaimed_slab = 0;
3589 : : p->reclaim_state = &reclaim_state;
3590 : :
3591 : : if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) {
3592 : : /*
3593 : : * Free memory by calling shrink zone with increasing
3594 : : * priorities until we have enough memory freed.
3595 : : */
3596 : : do {
3597 : : shrink_zone(zone, &sc);
3598 : : } while (sc.nr_reclaimed < nr_pages && --sc.priority >= 0);
3599 : : }
3600 : :
3601 : : nr_slab_pages0 = zone_page_state(zone, NR_SLAB_RECLAIMABLE);
3602 : : if (nr_slab_pages0 > zone->min_slab_pages) {
3603 : : /*
3604 : : * shrink_slab() does not currently allow us to determine how
3605 : : * many pages were freed in this zone. So we take the current
3606 : : * number of slab pages and shake the slab until it is reduced
3607 : : * by the same nr_pages that we used for reclaiming unmapped
3608 : : * pages.
3609 : : */
3610 : : nodes_clear(shrink.nodes_to_scan);
3611 : : node_set(zone_to_nid(zone), shrink.nodes_to_scan);
3612 : : for (;;) {
3613 : : unsigned long lru_pages = zone_reclaimable_pages(zone);
3614 : :
3615 : : /* No reclaimable slab or very low memory pressure */
3616 : : if (!shrink_slab(&shrink, sc.nr_scanned, lru_pages))
3617 : : break;
3618 : :
3619 : : /* Freed enough memory */
3620 : : nr_slab_pages1 = zone_page_state(zone,
3621 : : NR_SLAB_RECLAIMABLE);
3622 : : if (nr_slab_pages1 + nr_pages <= nr_slab_pages0)
3623 : : break;
3624 : : }
3625 : :
3626 : : /*
3627 : : * Update nr_reclaimed by the number of slab pages we
3628 : : * reclaimed from this zone.
3629 : : */
3630 : : nr_slab_pages1 = zone_page_state(zone, NR_SLAB_RECLAIMABLE);
3631 : : if (nr_slab_pages1 < nr_slab_pages0)
3632 : : sc.nr_reclaimed += nr_slab_pages0 - nr_slab_pages1;
3633 : : }
3634 : :
3635 : : p->reclaim_state = NULL;
3636 : : current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE);
3637 : : lockdep_clear_current_reclaim_state();
3638 : : return sc.nr_reclaimed >= nr_pages;
3639 : : }
3640 : :
3641 : : int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
3642 : : {
3643 : : int node_id;
3644 : : int ret;
3645 : :
3646 : : /*
3647 : : * Zone reclaim reclaims unmapped file backed pages and
3648 : : * slab pages if we are over the defined limits.
3649 : : *
3650 : : * A small portion of unmapped file backed pages is needed for
3651 : : * file I/O otherwise pages read by file I/O will be immediately
3652 : : * thrown out if the zone is overallocated. So we do not reclaim
3653 : : * if less than a specified percentage of the zone is used by
3654 : : * unmapped file backed pages.
3655 : : */
3656 : : if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages &&
3657 : : zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages)
3658 : : return ZONE_RECLAIM_FULL;
3659 : :
3660 : : if (!zone_reclaimable(zone))
3661 : : return ZONE_RECLAIM_FULL;
3662 : :
3663 : : /*
3664 : : * Do not scan if the allocation should not be delayed.
3665 : : */
3666 : : if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC))
3667 : : return ZONE_RECLAIM_NOSCAN;
3668 : :
3669 : : /*
3670 : : * Only run zone reclaim on the local zone or on zones that do not
3671 : : * have associated processors. This will favor the local processor
3672 : : * over remote processors and spread off node memory allocations
3673 : : * as wide as possible.
3674 : : */
3675 : : node_id = zone_to_nid(zone);
3676 : : if (node_state(node_id, N_CPU) && node_id != numa_node_id())
3677 : : return ZONE_RECLAIM_NOSCAN;
3678 : :
3679 : : if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED))
3680 : : return ZONE_RECLAIM_NOSCAN;
3681 : :
3682 : : ret = __zone_reclaim(zone, gfp_mask, order);
3683 : : zone_clear_flag(zone, ZONE_RECLAIM_LOCKED);
3684 : :
3685 : : if (!ret)
3686 : : count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED);
3687 : :
3688 : : return ret;
3689 : : }
3690 : : #endif
3691 : :
3692 : : /*
3693 : : * page_evictable - test whether a page is evictable
3694 : : * @page: the page to test
3695 : : *
3696 : : * Test whether page is evictable--i.e., should be placed on active/inactive
3697 : : * lists vs unevictable list.
3698 : : *
3699 : : * Reasons page might not be evictable:
3700 : : * (1) page's mapping marked unevictable
3701 : : * (2) page is part of an mlocked VMA
3702 : : *
3703 : : */
3704 : 0 : int page_evictable(struct page *page)
3705 : : {
3706 [ + + ][ + + ]: 552878 : return !mapping_unevictable(page_mapping(page)) && !PageMlocked(page);
3707 : : }
3708 : :
3709 : : #ifdef CONFIG_SHMEM
3710 : : /**
3711 : : * check_move_unevictable_pages - check pages for evictability and move to appropriate zone lru list
3712 : : * @pages: array of pages to check
3713 : : * @nr_pages: number of pages to check
3714 : : *
3715 : : * Checks pages for evictability and moves them to the appropriate lru list.
3716 : : *
3717 : : * This function is only used for SysV IPC SHM_UNLOCK.
3718 : : */
3719 : 0 : void check_move_unevictable_pages(struct page **pages, int nr_pages)
3720 : : {
3721 : : struct lruvec *lruvec;
3722 : : struct zone *zone = NULL;
3723 : : int pgscanned = 0;
3724 : : int pgrescued = 0;
3725 : : int i;
3726 : :
3727 [ + + ]: 2 : for (i = 0; i < nr_pages; i++) {
3728 : 1 : struct page *page = pages[i];
3729 : : struct zone *pagezone;
3730 : :
3731 : 1 : pgscanned++;
3732 : 1 : pagezone = page_zone(page);
3733 [ + - ]: 1 : if (pagezone != zone) {
3734 [ - + ]: 1 : if (zone)
3735 : : spin_unlock_irq(&zone->lru_lock);
3736 : : zone = pagezone;
3737 : : spin_lock_irq(&zone->lru_lock);
3738 : : }
3739 : : lruvec = mem_cgroup_page_lruvec(page, zone);
3740 : :
3741 [ + - ][ + - ]: 2 : if (!PageLRU(page) || !PageUnevictable(page))
3742 : 1 : continue;
3743 : :
3744 [ # # ]: 0 : if (page_evictable(page)) {
3745 : : enum lru_list lru = page_lru_base_type(page);
3746 : :
3747 : : VM_BUG_ON(PageActive(page));
3748 : : ClearPageUnevictable(page);
3749 : : del_page_from_lru_list(page, lruvec, LRU_UNEVICTABLE);
3750 : : add_page_to_lru_list(page, lruvec, lru);
3751 : 0 : pgrescued++;
3752 : : }
3753 : : }
3754 : :
3755 [ + - ]: 1 : if (zone) {
3756 : : __count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
3757 : : __count_vm_events(UNEVICTABLE_PGSCANNED, pgscanned);
3758 : : spin_unlock_irq(&zone->lru_lock);
3759 : : }
3760 : 1 : }
3761 : : #endif /* CONFIG_SHMEM */
3762 : :
3763 : 0 : static void warn_scan_unevictable_pages(void)
3764 : : {
3765 [ + + ]: 2 : printk_once(KERN_WARNING
3766 : : "%s: The scan_unevictable_pages sysctl/node-interface has been "
3767 : : "disabled for lack of a legitimate use case. If you have "
3768 : : "one, please send an email to linux-mm@kvack.org.\n",
3769 : : current->comm);
3770 : 0 : }
3771 : :
3772 : : /*
3773 : : * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of
3774 : : * all nodes' unevictable lists for evictable pages
3775 : : */
3776 : : unsigned long scan_unevictable_pages;
3777 : :
3778 : 0 : int scan_unevictable_handler(struct ctl_table *table, int write,
3779 : : void __user *buffer,
3780 : : size_t *length, loff_t *ppos)
3781 : : {
3782 : 2 : warn_scan_unevictable_pages();
3783 : 2 : proc_doulongvec_minmax(table, write, buffer, length, ppos);
3784 : 2 : scan_unevictable_pages = 0;
3785 : 2 : return 0;
3786 : : }
3787 : :
3788 : : #ifdef CONFIG_NUMA
3789 : : /*
3790 : : * per node 'scan_unevictable_pages' attribute. On demand re-scan of
3791 : : * a specified node's per zone unevictable lists for evictable pages.
3792 : : */
3793 : :
3794 : : static ssize_t read_scan_unevictable_node(struct device *dev,
3795 : : struct device_attribute *attr,
3796 : : char *buf)
3797 : : {
3798 : : warn_scan_unevictable_pages();
3799 : : return sprintf(buf, "0\n"); /* always zero; should fit... */
3800 : : }
3801 : :
3802 : : static ssize_t write_scan_unevictable_node(struct device *dev,
3803 : : struct device_attribute *attr,
3804 : : const char *buf, size_t count)
3805 : : {
3806 : : warn_scan_unevictable_pages();
3807 : : return 1;
3808 : : }
3809 : :
3810 : :
3811 : : static DEVICE_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR,
3812 : : read_scan_unevictable_node,
3813 : : write_scan_unevictable_node);
3814 : :
3815 : : int scan_unevictable_register_node(struct node *node)
3816 : : {
3817 : : return device_create_file(&node->dev, &dev_attr_scan_unevictable_pages);
3818 : : }
3819 : :
3820 : : void scan_unevictable_unregister_node(struct node *node)
3821 : : {
3822 : : device_remove_file(&node->dev, &dev_attr_scan_unevictable_pages);
3823 : : }
3824 : : #endif
|