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 : static unsigned long zone_reclaimable_pages(struct zone *zone)
152 : : {
153 : : int nr;
154 : :
155 : 611951 : nr = zone_page_state(zone, NR_ACTIVE_FILE) +
156 : : zone_page_state(zone, NR_INACTIVE_FILE);
157 : :
158 [ + - ]: 611951 : if (get_nr_swap_pages() > 0)
159 : 0 : 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 : 398149 : 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 : 644896 : 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->scan_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 : 28 : shrinker->flags &= ~SHRINKER_NUMA_AWARE;
225 : :
226 : : if (shrinker->flags & SHRINKER_NUMA_AWARE)
227 : : size *= nr_node_ids;
228 : :
229 : 28 : shrinker->nr_deferred = kzalloc(size, GFP_KERNEL);
230 [ + - ]: 28 : if (!shrinker->nr_deferred)
231 : : return -ENOMEM;
232 : :
233 : 28 : down_write(&shrinker_rwsem);
234 : 28 : list_add_tail(&shrinker->list, &shrinker_list);
235 : 28 : up_write(&shrinker_rwsem);
236 : 28 : 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 : 28 : down_write(&shrinker_rwsem);
255 : : list_del(&shrinker->list);
256 : 28 : up_write(&shrinker_rwsem);
257 : 28 : kfree(shrinker->nr_deferred);
258 : 28 : }
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 : 3258365 : int nid = shrinkctl->nid;
274 : 3258365 : long batch_size = shrinker->batch ? shrinker->batch
275 [ + + ]: 3258365 : : SHRINK_BATCH;
276 : :
277 : 3258365 : max_pass = shrinker->count_objects(shrinker, shrinkctl);
278 [ + + ]: 3257069 : 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 : 3560713 : nr = atomic_long_xchg(&shrinker->nr_deferred[nid], 0);
287 : :
288 : : total_scan = nr;
289 : 3409842 : delta = (4 * nr_pages_scanned) / shrinker->seeks;
290 : 3409842 : delta *= max_pass;
291 [ - + ][ # # ]: 3409842 : do_div(delta, lru_pages + 1);
[ - + ][ - + ]
[ - + ][ - + ]
[ - + ][ - + ]
[ - + ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ]
292 : 151477 : total_scan += delta;
293 [ - + ]: 151477 : 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 [ + + ]: 151297 : if (delta < max_pass / 4)
313 : 15715 : 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 [ + + ]: 151297 : if (total_scan > max_pass * 2)
321 : : total_scan = max_pass * 2;
322 : :
323 : 151297 : trace_mm_shrink_slab_start(shrinker, shrinkctl, nr,
324 : : nr_pages_scanned, lru_pages,
325 : : max_pass, delta, total_scan);
326 : :
327 : : /*
328 : : * Normally, we should not scan less than batch_size objects in one
329 : : * pass to avoid too frequent shrinker calls, but if the slab has less
330 : : * than batch_size objects in total and we are really tight on memory,
331 : : * we will try to reclaim all available objects, otherwise we can end
332 : : * up failing allocations although there are plenty of reclaimable
333 : : * objects spread over several slabs with usage less than the
334 : : * batch_size.
335 : : *
336 : : * We detect the "tight on memory" situations by looking at the total
337 : : * number of objects we want to scan (total_scan). If it is greater
338 : : * than the total number of objects on slab (max_pass), we must be
339 : : * scanning at high prio and therefore should try to reclaim as much as
340 : : * possible.
341 : : */
342 [ + + ]: 278661 : while (total_scan >= batch_size ||
343 : 278661 : total_scan >= max_pass) {
344 : : unsigned long ret;
345 : 126952 : unsigned long nr_to_scan = min(batch_size, total_scan);
346 : :
347 : 126952 : shrinkctl->nr_to_scan = nr_to_scan;
348 : 126952 : ret = shrinker->scan_objects(shrinker, shrinkctl);
349 [ + - ]: 127239 : if (ret == SHRINK_STOP)
350 : : break;
351 : 127239 : freed += ret;
352 : :
353 : : count_vm_events(SLABS_SCANNED, nr_to_scan);
354 : 127240 : total_scan -= nr_to_scan;
355 : :
356 : 127240 : cond_resched();
357 : : }
358 : :
359 : : /*
360 : : * move the unused scan count back into the shrinker in a
361 : : * manner that handles concurrent updates. If we exhausted the
362 : : * scan, there is no need to do an update.
363 : : */
364 [ + + ]: 151709 : if (total_scan > 0)
365 : 97018 : new_nr = atomic_long_add_return(total_scan,
366 : 97018 : &shrinker->nr_deferred[nid]);
367 : : else
368 : 54691 : new_nr = atomic_long_read(&shrinker->nr_deferred[nid]);
369 : :
370 : 151707 : trace_mm_shrink_slab_end(shrinker, freed, nr, new_nr);
371 : 151707 : return freed;
372 : : }
373 : :
374 : : /*
375 : : * Call the shrink functions to age shrinkable caches
376 : : *
377 : : * Here we assume it costs one seek to replace a lru page and that it also
378 : : * takes a seek to recreate a cache object. With this in mind we age equal
379 : : * percentages of the lru and ageable caches. This should balance the seeks
380 : : * generated by these structures.
381 : : *
382 : : * If the vm encountered mapped pages on the LRU it increase the pressure on
383 : : * slab to avoid swapping.
384 : : *
385 : : * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits.
386 : : *
387 : : * `lru_pages' represents the number of on-LRU pages in all the zones which
388 : : * are eligible for the caller's allocation attempt. It is used for balancing
389 : : * slab reclaim versus page reclaim.
390 : : *
391 : : * Returns the number of slab objects which we shrunk.
392 : : */
393 : 0 : unsigned long shrink_slab(struct shrink_control *shrinkctl,
394 : : unsigned long nr_pages_scanned,
395 : : unsigned long lru_pages)
396 : : {
397 : : struct shrinker *shrinker;
398 : : unsigned long freed = 0;
399 : :
400 [ + + ]: 125783 : if (nr_pages_scanned == 0)
401 : : nr_pages_scanned = SWAP_CLUSTER_MAX;
402 : :
403 [ + + ]: 125783 : if (!down_read_trylock(&shrinker_rwsem)) {
404 : : /*
405 : : * If we would return 0, our callers would understand that we
406 : : * have nothing else to shrink and give up trying. By returning
407 : : * 1 we keep it going and assume we'll be able to shrink next
408 : : * time.
409 : : */
410 : : freed = 1;
411 : : goto out;
412 : : }
413 : :
414 [ + + ]: 3381836 : list_for_each_entry(shrinker, &shrinker_list, list) {
415 [ + + ]: 3256045 : if (!(shrinker->flags & SHRINKER_NUMA_AWARE)) {
416 : 3255934 : shrinkctl->nid = 0;
417 : 3255934 : freed += shrink_slab_node(shrinkctl, shrinker,
418 : : nr_pages_scanned, lru_pages);
419 : 3258440 : continue;
420 : : }
421 : :
422 [ - + ][ # # ]: 111 : for_each_node_mask(shrinkctl->nid, shrinkctl->nodes_to_scan) {
423 [ # # ]: 0 : if (node_online(shrinkctl->nid))
424 : 0 : freed += shrink_slab_node(shrinkctl, shrinker,
425 : : nr_pages_scanned, lru_pages);
426 : :
427 : : }
428 : : }
429 : 125791 : up_read(&shrinker_rwsem);
430 : : out:
431 : 128297 : cond_resched();
432 : 125790 : return freed;
433 : : }
434 : :
435 : 4 : static inline int is_page_cache_freeable(struct page *page)
436 : : {
437 : : /*
438 : : * A freeable page cache page is referenced only by the caller
439 : : * that isolated the page, the page cache radix tree and
440 : : * optional buffer heads at page->private.
441 : : */
442 : 4 : return page_count(page) - page_has_private(page) == 2;
443 : : }
444 : :
445 : 4 : static int may_write_to_queue(struct backing_dev_info *bdi,
446 : : struct scan_control *sc)
447 : : {
448 [ - + ]: 4 : if (current->flags & PF_SWAPWRITE)
449 : : return 1;
450 [ # # ]: 0 : if (!bdi_write_congested(bdi))
451 : : return 1;
452 [ # # ]: 0 : if (bdi == current->backing_dev_info)
453 : : return 1;
454 : : return 0;
455 : : }
456 : :
457 : : /*
458 : : * We detected a synchronous write error writing a page out. Probably
459 : : * -ENOSPC. We need to propagate that into the address_space for a subsequent
460 : : * fsync(), msync() or close().
461 : : *
462 : : * The tricky part is that after writepage we cannot touch the mapping: nothing
463 : : * prevents it from being freed up. But we have a ref on the page and once
464 : : * that page is locked, the mapping is pinned.
465 : : *
466 : : * We're allowed to run sleeping lock_page() here because we know the caller has
467 : : * __GFP_FS.
468 : : */
469 : 0 : static void handle_write_error(struct address_space *mapping,
470 : : struct page *page, int error)
471 : : {
472 : : lock_page(page);
473 [ # # ]: 0 : if (page_mapping(page) == mapping)
474 : : mapping_set_error(mapping, error);
475 : 0 : unlock_page(page);
476 : 0 : }
477 : :
478 : : /* possible outcome of pageout() */
479 : : typedef enum {
480 : : /* failed to write page out, page is locked */
481 : : PAGE_KEEP,
482 : : /* move page to the active list, page is locked */
483 : : PAGE_ACTIVATE,
484 : : /* page has been sent to the disk successfully, page is unlocked */
485 : : PAGE_SUCCESS,
486 : : /* page is clean and locked */
487 : : PAGE_CLEAN,
488 : : } pageout_t;
489 : :
490 : : /*
491 : : * pageout is called by shrink_page_list() for each dirty page.
492 : : * Calls ->writepage().
493 : : */
494 : 0 : static pageout_t pageout(struct page *page, struct address_space *mapping,
495 : : struct scan_control *sc)
496 : : {
497 : : /*
498 : : * If the page is dirty, only perform writeback if that write
499 : : * will be non-blocking. To prevent this allocation from being
500 : : * stalled by pagecache activity. But note that there may be
501 : : * stalls if we need to run get_block(). We could test
502 : : * PagePrivate for that.
503 : : *
504 : : * If this process is currently in __generic_file_aio_write() against
505 : : * this page's queue, we can perform writeback even if that
506 : : * will block.
507 : : *
508 : : * If the page is swapcache, write it back even if that would
509 : : * block, for some throttling. This happens by accident, because
510 : : * swap_backing_dev_info is bust: it doesn't reflect the
511 : : * congestion state of the swapdevs. Easy to fix, if needed.
512 : : */
513 [ + - ]: 4 : if (!is_page_cache_freeable(page))
514 : : return PAGE_KEEP;
515 [ - + ]: 4 : if (!mapping) {
516 : : /*
517 : : * Some data journaling orphaned pages can have
518 : : * page->mapping == NULL while being dirty with clean buffers.
519 : : */
520 [ # # ]: 0 : if (page_has_private(page)) {
521 [ # # ]: 0 : if (try_to_free_buffers(page)) {
522 : : ClearPageDirty(page);
523 : 0 : printk("%s: orphaned page\n", __func__);
524 : : return PAGE_CLEAN;
525 : : }
526 : : }
527 : : return PAGE_KEEP;
528 : : }
529 [ + - ]: 4 : if (mapping->a_ops->writepage == NULL)
530 : : return PAGE_ACTIVATE;
531 [ + - ]: 4 : if (!may_write_to_queue(mapping->backing_dev_info, sc))
532 : : return PAGE_KEEP;
533 : :
534 [ + - ]: 4 : if (clear_page_dirty_for_io(page)) {
535 : : int res;
536 : 4 : struct writeback_control wbc = {
537 : : .sync_mode = WB_SYNC_NONE,
538 : : .nr_to_write = SWAP_CLUSTER_MAX,
539 : : .range_start = 0,
540 : : .range_end = LLONG_MAX,
541 : : .for_reclaim = 1,
542 : : };
543 : :
544 : : SetPageReclaim(page);
545 : 4 : res = mapping->a_ops->writepage(page, &wbc);
546 [ - + ]: 4 : if (res < 0)
547 : 0 : handle_write_error(mapping, page, res);
548 [ - + ]: 4 : if (res == AOP_WRITEPAGE_ACTIVATE) {
549 : : ClearPageReclaim(page);
550 : : return PAGE_ACTIVATE;
551 : : }
552 : :
553 [ + - ]: 4 : if (!PageWriteback(page)) {
554 : : /* synchronous write or broken a_ops? */
555 : : ClearPageReclaim(page);
556 : : }
557 [ - + ]: 8 : trace_mm_vmscan_writepage(page, trace_reclaim_flags(page));
558 : 4 : inc_zone_page_state(page, NR_VMSCAN_WRITE);
559 : : return PAGE_SUCCESS;
560 : : }
561 : :
562 : : return PAGE_CLEAN;
563 : : }
564 : :
565 : : /*
566 : : * Same as remove_mapping, but if the page is removed from the mapping, it
567 : : * gets returned with a refcount of 0.
568 : : */
569 : 0 : static int __remove_mapping(struct address_space *mapping, struct page *page)
570 : : {
571 [ - + ]: 155827 : BUG_ON(!PageLocked(page));
572 [ - + ]: 155827 : BUG_ON(mapping != page_mapping(page));
573 : :
574 : : spin_lock_irq(&mapping->tree_lock);
575 : : /*
576 : : * The non racy check for a busy page.
577 : : *
578 : : * Must be careful with the order of the tests. When someone has
579 : : * a ref to the page, it may be possible that they dirty it then
580 : : * drop the reference. So if PageDirty is tested before page_count
581 : : * here, then the following race may occur:
582 : : *
583 : : * get_user_pages(&page);
584 : : * [user mapping goes away]
585 : : * write_to(page);
586 : : * !PageDirty(page) [good]
587 : : * SetPageDirty(page);
588 : : * put_page(page);
589 : : * !page_count(page) [good, discard it]
590 : : *
591 : : * [oops, our write_to data is lost]
592 : : *
593 : : * Reversing the order of the tests ensures such a situation cannot
594 : : * escape unnoticed. The smp_rmb is needed to ensure the page->flags
595 : : * load is not satisfied before that of page->_count.
596 : : *
597 : : * Note that if SetPageDirty is always performed via set_page_dirty,
598 : : * and thus under tree_lock, then this ordering is not required.
599 : : */
600 [ + + ]: 155827 : if (!page_freeze_refs(page, 2))
601 : : goto cannot_free;
602 : : /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */
603 [ - + ]: 155726 : if (unlikely(PageDirty(page))) {
604 : : page_unfreeze_refs(page, 2);
605 : : goto cannot_free;
606 : : }
607 : :
608 [ - + ]: 155726 : if (PageSwapCache(page)) {
609 : 0 : swp_entry_t swap = { .val = page_private(page) };
610 : 0 : __delete_from_swap_cache(page);
611 : : spin_unlock_irq(&mapping->tree_lock);
612 : 0 : swapcache_free(swap, page);
613 : : } else {
614 : : void (*freepage)(struct page *);
615 : :
616 : 155726 : freepage = mapping->a_ops->freepage;
617 : :
618 : 155726 : __delete_from_page_cache(page);
619 : : spin_unlock_irq(&mapping->tree_lock);
620 : : mem_cgroup_uncharge_cache_page(page);
621 : :
622 [ - + ]: 155726 : if (freepage != NULL)
623 : 0 : freepage(page);
624 : : }
625 : :
626 : : return 1;
627 : :
628 : : cannot_free:
629 : : spin_unlock_irq(&mapping->tree_lock);
630 : 101 : return 0;
631 : : }
632 : :
633 : : /*
634 : : * Attempt to detach a locked page from its ->mapping. If it is dirty or if
635 : : * someone else has a ref on the page, abort and return 0. If it was
636 : : * successfully detached, return 1. Assumes the caller has a single ref on
637 : : * this page.
638 : : */
639 : 0 : int remove_mapping(struct address_space *mapping, struct page *page)
640 : : {
641 [ + + ]: 72844 : if (__remove_mapping(mapping, page)) {
642 : : /*
643 : : * Unfreezing the refcount with 1 rather than 2 effectively
644 : : * drops the pagecache ref for us without requiring another
645 : : * atomic operation.
646 : : */
647 : : page_unfreeze_refs(page, 1);
648 : 72843 : return 1;
649 : : }
650 : : return 0;
651 : : }
652 : :
653 : : /**
654 : : * putback_lru_page - put previously isolated page onto appropriate LRU list
655 : : * @page: page to be put back to appropriate lru list
656 : : *
657 : : * Add previously isolated @page to appropriate LRU list.
658 : : * Page may still be unevictable for other reasons.
659 : : *
660 : : * lru_lock must not be held, interrupts must be enabled.
661 : : */
662 : 0 : void putback_lru_page(struct page *page)
663 : : {
664 : : bool is_unevictable;
665 : : int was_unevictable = PageUnevictable(page);
666 : :
667 : : VM_BUG_ON_PAGE(PageLRU(page), page);
668 : :
669 : : redo:
670 : : ClearPageUnevictable(page);
671 : :
672 [ + + ]: 5666 : if (page_evictable(page)) {
673 : : /*
674 : : * For evictable pages, we can use the cache.
675 : : * In event of a race, worst case is we end up with an
676 : : * unevictable page on [in]active list.
677 : : * We know how to handle that.
678 : : */
679 : : is_unevictable = false;
680 : 1884 : lru_cache_add(page);
681 : : } else {
682 : : /*
683 : : * Put unevictable pages directly on zone's unevictable
684 : : * list.
685 : : */
686 : : is_unevictable = true;
687 : 3782 : add_page_to_unevictable_list(page);
688 : : /*
689 : : * When racing with an mlock or AS_UNEVICTABLE clearing
690 : : * (page is unlocked) make sure that if the other thread
691 : : * does not observe our setting of PG_lru and fails
692 : : * isolation/check_move_unevictable_pages,
693 : : * we see PG_mlocked/AS_UNEVICTABLE cleared below and move
694 : : * the page back to the evictable list.
695 : : *
696 : : * The other side is TestClearPageMlocked() or shmem_lock().
697 : : */
698 : 3782 : smp_mb();
699 : : }
700 : :
701 : : /*
702 : : * page's status can change while we move it among lru. If an evictable
703 : : * page is on unevictable list, it never be freed. To avoid that,
704 : : * check after we added it to the list, again.
705 : : */
706 [ + + ][ - + ]: 5666 : if (is_unevictable && page_evictable(page)) {
707 [ # # ]: 0 : if (!isolate_lru_page(page)) {
708 : 0 : put_page(page);
709 : 0 : goto redo;
710 : : }
711 : : /* This means someone else dropped this page from LRU
712 : : * So, it will be freed or putback to LRU again. There is
713 : : * nothing to do here.
714 : : */
715 : : }
716 : :
717 [ + + ]: 5666 : if (was_unevictable && !is_unevictable)
718 : : count_vm_event(UNEVICTABLE_PGRESCUED);
719 [ + - ]: 3782 : else if (!was_unevictable && is_unevictable)
720 : : count_vm_event(UNEVICTABLE_PGCULLED);
721 : :
722 : 5666 : put_page(page); /* drop ref from isolate */
723 : 5666 : }
724 : :
725 : : enum page_references {
726 : : PAGEREF_RECLAIM,
727 : : PAGEREF_RECLAIM_CLEAN,
728 : : PAGEREF_KEEP,
729 : : PAGEREF_ACTIVATE,
730 : : };
731 : :
732 : 100668 : static enum page_references page_check_references(struct page *page,
733 : : struct scan_control *sc)
734 : : {
735 : : int referenced_ptes, referenced_page;
736 : : unsigned long vm_flags;
737 : :
738 : 100668 : referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup,
739 : : &vm_flags);
740 : : referenced_page = TestClearPageReferenced(page);
741 : :
742 : : /*
743 : : * Mlock lost the isolation race with us. Let try_to_unmap()
744 : : * move the page to the unevictable list.
745 : : */
746 [ + ]: 100662 : if (vm_flags & VM_LOCKED)
747 : : return PAGEREF_RECLAIM;
748 : :
749 [ + + ]: 100665 : if (referenced_ptes) {
750 [ + - ]: 754 : if (PageSwapBacked(page))
751 : : return PAGEREF_ACTIVATE;
752 : : /*
753 : : * All mapped pages start out with page table
754 : : * references from the instantiating fault, so we need
755 : : * to look twice if a mapped file page is used more
756 : : * than once.
757 : : *
758 : : * Mark it and spare it for another trip around the
759 : : * inactive list. Another page table reference will
760 : : * lead to its activation.
761 : : *
762 : : * Note: the mark is set for activated pages as well
763 : : * so that recently deactivated but used pages are
764 : : * quickly recovered.
765 : : */
766 : : SetPageReferenced(page);
767 : :
768 [ + + ]: 754 : if (referenced_page || referenced_ptes > 1)
769 : : return PAGEREF_ACTIVATE;
770 : :
771 : : /*
772 : : * Activate file-backed executable pages after first usage.
773 : : */
774 [ - + ]: 419 : if (vm_flags & VM_EXEC)
775 : : return PAGEREF_ACTIVATE;
776 : :
777 : : return PAGEREF_KEEP;
778 : : }
779 : :
780 : : /* Reclaim if clean, defer dirty pages to writeback */
781 [ + + ][ - + ]: 99911 : if (referenced_page && !PageSwapBacked(page))
782 : : return PAGEREF_RECLAIM_CLEAN;
783 : :
784 : : return PAGEREF_RECLAIM;
785 : : }
786 : :
787 : : /* Check if a page is dirty or under writeback */
788 : 0 : static void page_check_dirty_writeback(struct page *page,
789 : : bool *dirty, bool *writeback)
790 : : {
791 : : struct address_space *mapping;
792 : :
793 : : /*
794 : : * Anonymous pages are not handled by flushers and must be written
795 : : * from reclaim context. Do not stall reclaim based on them
796 : : */
797 [ - + ]: 101827 : if (!page_is_file_cache(page)) {
798 : 0 : *dirty = false;
799 : 0 : *writeback = false;
800 : 0 : return;
801 : : }
802 : :
803 : : /* By default assume that the page flags are accurate */
804 : 101827 : *dirty = PageDirty(page);
805 : 101827 : *writeback = PageWriteback(page);
806 : :
807 : : /* Verify dirty/writeback state if the filesystem supports it */
808 [ + + ]: 101827 : if (!page_has_private(page))
809 : : return;
810 : :
811 : 85652 : mapping = page_mapping(page);
812 [ + + ][ + + ]: 85653 : if (mapping && mapping->a_ops->is_dirty_writeback)
813 : 51736 : mapping->a_ops->is_dirty_writeback(page, dirty, writeback);
814 : : }
815 : :
816 : : /*
817 : : * shrink_page_list() returns the number of reclaimed pages
818 : : */
819 : 0 : static unsigned long shrink_page_list(struct list_head *page_list,
820 : : struct zone *zone,
821 : 100667 : struct scan_control *sc,
822 : : enum ttu_flags ttu_flags,
823 : : unsigned long *ret_nr_dirty,
824 : : unsigned long *ret_nr_unqueued_dirty,
825 : : unsigned long *ret_nr_congested,
826 : : unsigned long *ret_nr_writeback,
827 : : unsigned long *ret_nr_immediate,
828 : : bool force_reclaim)
829 : : {
830 : 6922 : LIST_HEAD(ret_pages);
831 : 6922 : LIST_HEAD(free_pages);
832 : : int pgactivate = 0;
833 : : unsigned long nr_unqueued_dirty = 0;
834 : : unsigned long nr_dirty = 0;
835 : : unsigned long nr_congested = 0;
836 : : unsigned long nr_reclaimed = 0;
837 : : unsigned long nr_writeback = 0;
838 : : unsigned long nr_immediate = 0;
839 : :
840 : 6922 : cond_resched();
841 : :
842 : : mem_cgroup_uncharge_start();
843 [ + + ]: 112592 : while (!list_empty(page_list)) {
844 : : struct address_space *mapping;
845 : 199802 : struct page *page;
846 : : int may_enter_fs;
847 : : enum page_references references = PAGEREF_RECLAIM_CLEAN;
848 : : bool dirty, writeback;
849 : :
850 : 105671 : cond_resched();
851 : :
852 : 105674 : page = lru_to_page(page_list);
853 : : list_del(&page->lru);
854 : :
855 [ + + ]: 105670 : if (!trylock_page(page))
856 : : goto keep;
857 : :
858 : : VM_BUG_ON_PAGE(PageActive(page), page);
859 : : VM_BUG_ON_PAGE(page_zone(page) != zone, page);
860 : :
861 : 101823 : sc->nr_scanned++;
862 : :
863 [ + - ]: 101823 : if (unlikely(!page_evictable(page)))
864 : : goto cull_mlocked;
865 : :
866 [ - + ][ # # ]: 101828 : if (!sc->may_unmap && page_mapped(page))
867 : : goto keep_locked;
868 : :
869 : : /* Double the slab pressure for mapped and swapcache pages */
870 [ + + ][ - + ]: 101828 : if (page_mapped(page) || PageSwapCache(page))
871 : 2273 : sc->nr_scanned++;
872 : :
873 [ - + ][ # # ]: 101828 : may_enter_fs = (sc->gfp_mask & __GFP_FS) ||
874 [ # # ]: 0 : (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO));
875 : :
876 : : /*
877 : : * The number of dirty pages determines if a zone is marked
878 : : * reclaim_congested which affects wait_iff_congested. kswapd
879 : : * will stall and start writing pages if the tail of the LRU
880 : : * is all dirty unqueued pages.
881 : : */
882 : 101828 : page_check_dirty_writeback(page, &dirty, &writeback);
883 [ + + ][ + + ]: 101827 : if (dirty || writeback)
884 : 1180 : nr_dirty++;
885 : :
886 [ + + ][ + - ]: 101827 : if (dirty && !writeback)
887 : 20 : nr_unqueued_dirty++;
888 : :
889 : : /*
890 : : * Treat this page as congested if the underlying BDI is or if
891 : : * pages are cycling through the LRU so quickly that the
892 : : * pages marked for immediate reclaim are making it to the
893 : : * end of the LRU a second time.
894 : : */
895 : 101827 : mapping = page_mapping(page);
896 [ + + ][ + - ]: 203655 : if ((mapping && bdi_write_congested(mapping->backing_dev_info)) ||
[ + + ]
897 [ + + ]: 1160 : (writeback && PageReclaim(page)))
898 : 1153 : nr_congested++;
899 : :
900 : : /*
901 : : * If a page at the tail of the LRU is under writeback, there
902 : : * are three cases to consider.
903 : : *
904 : : * 1) If reclaim is encountering an excessive number of pages
905 : : * under writeback and this page is both under writeback and
906 : : * PageReclaim then it indicates that pages are being queued
907 : : * for IO but are being recycled through the LRU before the
908 : : * IO can complete. Waiting on the page itself risks an
909 : : * indefinite stall if it is impossible to writeback the
910 : : * page due to IO error or disconnected storage so instead
911 : : * note that the LRU is being scanned too quickly and the
912 : : * caller can stall after page list has been processed.
913 : : *
914 : : * 2) Global reclaim encounters a page, memcg encounters a
915 : : * page that is not marked for immediate reclaim or
916 : : * the caller does not have __GFP_IO. In this case mark
917 : : * the page for immediate reclaim and continue scanning.
918 : : *
919 : : * __GFP_IO is checked because a loop driver thread might
920 : : * enter reclaim, and deadlock if it waits on a page for
921 : : * which it is needed to do the write (loop masks off
922 : : * __GFP_IO|__GFP_FS for this reason); but more thought
923 : : * would probably show more reasons.
924 : : *
925 : : * Don't require __GFP_FS, since we're not going into the
926 : : * FS, just waiting on its writeback completion. Worryingly,
927 : : * ext4 gfs2 and xfs allocate pages with
928 : : * grab_cache_page_write_begin(,,AOP_FLAG_NOFS), so testing
929 : : * may_enter_fs here is liable to OOM on them.
930 : : *
931 : : * 3) memcg encounters a page that is not already marked
932 : : * PageReclaim. memcg does not have any dirty pages
933 : : * throttling so we could easily OOM just because too many
934 : : * pages are in writeback and there is nothing else to
935 : : * reclaim. Wait for the writeback to complete.
936 : : */
937 [ + + ]: 101828 : if (PageWriteback(page)) {
938 : : /* Case 1 above */
939 [ + + ][ + + ]: 1160 : if (current_is_kswapd() &&
940 [ + + ]: 175 : PageReclaim(page) &&
941 : : zone_is_reclaim_writeback(zone)) {
942 : 8 : nr_immediate++;
943 : 8 : goto keep_locked;
944 : :
945 : : /* Case 2 above */
946 : : } else if (global_reclaim(sc) ||
947 : : !PageReclaim(page) || !(sc->gfp_mask & __GFP_IO)) {
948 : : /*
949 : : * This is slightly racy - end_page_writeback()
950 : : * might have just cleared PageReclaim, then
951 : : * setting PageReclaim here end up interpreted
952 : : * as PageReadahead - but that does not matter
953 : : * enough to care. What we do want is for this
954 : : * page to have PageReclaim set next time memcg
955 : : * reclaim reaches the tests above, so it will
956 : : * then wait_on_page_writeback() to avoid OOM;
957 : : * and it's also appropriate in global reclaim.
958 : : */
959 : : SetPageReclaim(page);
960 : 1152 : nr_writeback++;
961 : :
962 : 1152 : goto keep_locked;
963 : :
964 : : /* Case 3 above */
965 : : } else {
966 : : wait_on_page_writeback(page);
967 : : }
968 : : }
969 : :
970 [ + + ]: 100668 : if (!force_reclaim)
971 : 100667 : references = page_check_references(page, sc);
972 : :
973 [ + + - ]: 100664 : switch (references) {
974 : : case PAGEREF_ACTIVATE:
975 : : goto activate_locked;
976 : : case PAGEREF_KEEP:
977 : : goto keep_locked;
978 : : case PAGEREF_RECLAIM:
979 : : case PAGEREF_RECLAIM_CLEAN:
980 : : ; /* try to reclaim the page below */
981 : : }
982 : :
983 : : /*
984 : : * Anonymous process memory has backing store?
985 : : * Try to allocate it some swap space here.
986 : : */
987 [ - + ][ # # ]: 99910 : if (PageAnon(page) && !PageSwapCache(page)) {
988 [ # # ]: 0 : if (!(sc->gfp_mask & __GFP_IO))
989 : : goto keep_locked;
990 [ # # ]: 0 : if (!add_to_swap(page, page_list))
991 : : goto activate_locked;
992 : : may_enter_fs = 1;
993 : :
994 : : /* Adding to swap updated mapping */
995 : 0 : mapping = page_mapping(page);
996 : : }
997 : :
998 : : /*
999 : : * The page is mapped into the page tables of one or more
1000 : : * processes. Try to unmap it here.
1001 : : */
1002 [ + + ][ + - ]: 106834 : if (page_mapped(page) && mapping) {
1003 [ + - - - ]: 1519 : switch (try_to_unmap(page, ttu_flags)) {
1004 : : case SWAP_FAIL:
1005 : : goto activate_locked;
1006 : : case SWAP_AGAIN:
1007 : : goto keep_locked;
1008 : : case SWAP_MLOCK:
1009 : : goto cull_mlocked;
1010 : : case SWAP_SUCCESS:
1011 : : ; /* try to free the page below */
1012 : : }
1013 : : }
1014 : :
1015 [ + + ]: 99912 : if (PageDirty(page)) {
1016 : : /*
1017 : : * Only kswapd can writeback filesystem pages to
1018 : : * avoid risk of stack overflow but only writeback
1019 : : * if many dirty pages have been encountered.
1020 : : */
1021 [ + - ][ + - ]: 20 : if (page_is_file_cache(page) &&
1022 [ + + ]: 20 : (!current_is_kswapd() ||
1023 : : !zone_is_reclaim_dirty(zone))) {
1024 : : /*
1025 : : * Immediately reclaim when written back.
1026 : : * Similar in principal to deactivate_page()
1027 : : * except we already have the page isolated
1028 : : * and know it's dirty
1029 : : */
1030 : 13 : inc_zone_page_state(page, NR_VMSCAN_IMMEDIATE);
1031 : : SetPageReclaim(page);
1032 : :
1033 : : goto keep_locked;
1034 : : }
1035 : :
1036 [ + + ]: 7 : if (references == PAGEREF_RECLAIM_CLEAN)
1037 : : goto keep_locked;
1038 [ + - ]: 4 : if (!may_enter_fs)
1039 : : goto keep_locked;
1040 [ + - ]: 4 : if (!sc->may_writepage)
1041 : : goto keep_locked;
1042 : :
1043 : : /* Page is dirty, try to write it out here */
1044 [ + - - - ]: 4 : switch (pageout(page, mapping, sc)) {
1045 : : case PAGE_KEEP:
1046 : : goto keep_locked;
1047 : : case PAGE_ACTIVATE:
1048 : : goto activate_locked;
1049 : : case PAGE_SUCCESS:
1050 [ + - ]: 4 : if (PageWriteback(page))
1051 : : goto keep;
1052 [ - + ]: 4 : if (PageDirty(page))
1053 : : goto keep;
1054 : :
1055 : : /*
1056 : : * A synchronous write - probably a ramdisk. Go
1057 : : * ahead and try to reclaim the page.
1058 : : */
1059 [ # # ]: 0 : if (!trylock_page(page))
1060 : : goto keep;
1061 [ # # ][ # # ]: 0 : if (PageDirty(page) || PageWriteback(page))
1062 : : goto keep_locked;
1063 : 0 : mapping = page_mapping(page);
1064 : : case PAGE_CLEAN:
1065 : : ; /* try to free the page below */
1066 : : }
1067 : : }
1068 : :
1069 : : /*
1070 : : * If the page has buffers, try to free the buffer mappings
1071 : : * associated with this page. If we succeed we try to free
1072 : : * the page as well.
1073 : : *
1074 : : * We do this even if the page is PageDirty().
1075 : : * try_to_release_page() does not perform I/O, but it is
1076 : : * possible for a page to have PageDirty set, but it is actually
1077 : : * clean (all its buffers are clean). This happens if the
1078 : : * buffers were written out directly, with submit_bh(). ext3
1079 : : * will do this, as well as the blockdev mapping.
1080 : : * try_to_release_page() will discover that cleanness and will
1081 : : * drop the buffers and mark the page clean - it can be freed.
1082 : : *
1083 : : * Rarely, pages can have buffers and no ->mapping. These are
1084 : : * the pages which were not successfully invalidated in
1085 : : * truncate_complete_page(). We try to drop those buffers here
1086 : : * and if that worked, and the page is no longer mapped into
1087 : : * process address space (page_count == 1) it can be freed.
1088 : : * Otherwise, leave the page on the LRU so it is swappable.
1089 : : */
1090 [ + + ]: 99892 : if (page_has_private(page)) {
1091 [ + + ]: 84471 : if (!try_to_release_page(page, sc->gfp_mask))
1092 : : goto activate_locked;
1093 [ + + ][ + - ]: 67564 : if (!mapping && page_count(page) == 1) {
1094 : 1 : unlock_page(page);
1095 [ - + ]: 1 : if (put_page_testzero(page))
1096 : : goto free_it;
1097 : : else {
1098 : : /*
1099 : : * rare race with speculative reference.
1100 : : * the speculative reference will free
1101 : : * this page shortly, so we may
1102 : : * increment nr_reclaimed here (and
1103 : : * leave it off the LRU).
1104 : : */
1105 : 0 : nr_reclaimed++;
1106 : 105671 : continue;
1107 : : }
1108 : : }
1109 : : }
1110 : :
1111 [ + - ][ + + ]: 82983 : if (!mapping || !__remove_mapping(mapping, page))
1112 : : goto keep_locked;
1113 : :
1114 : : /*
1115 : : * At this point, we have no other references and there is
1116 : : * no way to pick any more up (removed from LRU, removed
1117 : : * from pagecache). Can use non-atomic bitops now (and
1118 : : * we obviously don't have to worry about waking up a process
1119 : : * waiting on the page lock, because there are no references.
1120 : : */
1121 : : __clear_page_locked(page);
1122 : : free_it:
1123 : 82884 : nr_reclaimed++;
1124 : :
1125 : : /*
1126 : : * Is there need to periodically free_page_list? It would
1127 : : * appear not as the counts should be low
1128 : : */
1129 : 82884 : list_add(&page->lru, &free_pages);
1130 : 82884 : continue;
1131 : :
1132 : : cull_mlocked:
1133 [ # # ]: 0 : if (PageSwapCache(page))
1134 : 0 : try_to_free_swap(page);
1135 : 0 : unlock_page(page);
1136 : 0 : putback_lru_page(page);
1137 : 0 : continue;
1138 : :
1139 : : activate_locked:
1140 : : /* Not a candidate for swapping, so reclaim swap space. */
1141 [ - + ][ # # ]: 17664 : if (PageSwapCache(page) && vm_swap_full())
1142 : 0 : try_to_free_swap(page);
1143 : : VM_BUG_ON_PAGE(PageActive(page), page);
1144 : : SetPageActive(page);
1145 : 17664 : pgactivate++;
1146 : : keep_locked:
1147 : 18940 : unlock_page(page);
1148 : : keep:
1149 : 22787 : list_add(&page->lru, &ret_pages);
1150 : : VM_BUG_ON_PAGE(PageLRU(page) || PageUnevictable(page), page);
1151 : : }
1152 : :
1153 : 6921 : free_hot_cold_page_list(&free_pages, 1);
1154 : :
1155 : : list_splice(&ret_pages, page_list);
1156 : : count_vm_events(PGACTIVATE, pgactivate);
1157 : : mem_cgroup_uncharge_end();
1158 : 6921 : *ret_nr_dirty += nr_dirty;
1159 : 6921 : *ret_nr_congested += nr_congested;
1160 : 6921 : *ret_nr_unqueued_dirty += nr_unqueued_dirty;
1161 : 6921 : *ret_nr_writeback += nr_writeback;
1162 : 6921 : *ret_nr_immediate += nr_immediate;
1163 : 6921 : return nr_reclaimed;
1164 : : }
1165 : :
1166 : 0 : unsigned long reclaim_clean_pages_from_list(struct zone *zone,
1167 : : struct list_head *page_list)
1168 : : {
1169 : 0 : struct scan_control sc = {
1170 : : .gfp_mask = GFP_KERNEL,
1171 : : .priority = DEF_PRIORITY,
1172 : : .may_unmap = 1,
1173 : : };
1174 : : unsigned long ret, dummy1, dummy2, dummy3, dummy4, dummy5;
1175 : : struct page *page, *next;
1176 : 0 : LIST_HEAD(clean_pages);
1177 : :
1178 [ # # ]: 0 : list_for_each_entry_safe(page, next, page_list, lru) {
1179 [ # # ][ # # ]: 0 : if (page_is_file_cache(page) && !PageDirty(page) &&
1180 : : !isolated_balloon_page(page)) {
1181 : : ClearPageActive(page);
1182 : : list_move(&page->lru, &clean_pages);
1183 : : }
1184 : : }
1185 : :
1186 : 0 : ret = shrink_page_list(&clean_pages, zone, &sc,
1187 : : TTU_UNMAP|TTU_IGNORE_ACCESS,
1188 : : &dummy1, &dummy2, &dummy3, &dummy4, &dummy5, true);
1189 : : list_splice(&clean_pages, page_list);
1190 : 0 : __mod_zone_page_state(zone, NR_ISOLATED_FILE, -ret);
1191 : 0 : return ret;
1192 : : }
1193 : :
1194 : : /*
1195 : : * Attempt to remove the specified page from its LRU. Only take this page
1196 : : * if it is of the appropriate PageActive status. Pages which are being
1197 : : * freed elsewhere are also ignored.
1198 : : *
1199 : : * page: page to consider
1200 : : * mode: one of the LRU isolation modes defined above
1201 : : *
1202 : : * returns 0 on success, -ve errno on failure.
1203 : : */
1204 : 0 : int __isolate_lru_page(struct page *page, isolate_mode_t mode)
1205 : : {
1206 : : int ret = -EINVAL;
1207 : :
1208 : : /* Only take pages on the LRU. */
1209 [ + - ]: 152290 : if (!PageLRU(page))
1210 : : return ret;
1211 : :
1212 : : /* Compaction should not handle unevictable pages but CMA can do so */
1213 [ - + ][ # # ]: 152290 : if (PageUnevictable(page) && !(mode & ISOLATE_UNEVICTABLE))
1214 : : return ret;
1215 : :
1216 : : ret = -EBUSY;
1217 : :
1218 : : /*
1219 : : * To minimise LRU disruption, the caller can indicate that it only
1220 : : * wants to isolate pages it will be able to operate on without
1221 : : * blocking - clean pages for the most part.
1222 : : *
1223 : : * ISOLATE_CLEAN means that only clean pages should be isolated. This
1224 : : * is used by reclaim when it is cannot write to backing storage
1225 : : *
1226 : : * ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages
1227 : : * that it is possible to migrate without blocking
1228 : : */
1229 [ - + ]: 152290 : if (mode & (ISOLATE_CLEAN|ISOLATE_ASYNC_MIGRATE)) {
1230 : : /* All the caller can do on PageWriteback is block */
1231 [ # # ]: 0 : if (PageWriteback(page))
1232 : : return ret;
1233 : :
1234 [ # # ]: 0 : if (PageDirty(page)) {
1235 : : struct address_space *mapping;
1236 : :
1237 : : /* ISOLATE_CLEAN means only clean pages */
1238 [ # # ]: 0 : if (mode & ISOLATE_CLEAN)
1239 : : return ret;
1240 : :
1241 : : /*
1242 : : * Only pages without mappings or that have a
1243 : : * ->migratepage callback are possible to migrate
1244 : : * without blocking
1245 : : */
1246 : 0 : mapping = page_mapping(page);
1247 [ # # ][ # # ]: 0 : if (mapping && !mapping->a_ops->migratepage)
1248 : : return ret;
1249 : : }
1250 : : }
1251 : :
1252 [ - + ][ # # ]: 152290 : if ((mode & ISOLATE_UNMAPPED) && page_mapped(page))
1253 : : return ret;
1254 : :
1255 [ + - ]: 152290 : if (likely(get_page_unless_zero(page))) {
1256 : : /*
1257 : : * Be careful not to clear PageLRU until after we're
1258 : : * sure the page is not being freed elsewhere -- the
1259 : : * page release code relies on it.
1260 : : */
1261 : : ClearPageLRU(page);
1262 : : ret = 0;
1263 : : }
1264 : :
1265 : 152290 : return ret;
1266 : : }
1267 : :
1268 : : /*
1269 : : * zone->lru_lock is heavily contended. Some of the functions that
1270 : : * shrink the lists perform better by taking out a batch of pages
1271 : : * and working on them outside the LRU lock.
1272 : : *
1273 : : * For pagecache intensive workloads, this function is the hottest
1274 : : * spot in the kernel (apart from copy_*_user functions).
1275 : : *
1276 : : * Appropriate locks must be held before calling this function.
1277 : : *
1278 : : * @nr_to_scan: The number of pages to look through on the list.
1279 : : * @lruvec: The LRU vector to pull pages from.
1280 : : * @dst: The temp list to put pages on to.
1281 : : * @nr_scanned: The number of pages that were scanned.
1282 : : * @sc: The scan_control struct for this reclaim session
1283 : : * @mode: One of the LRU isolation modes
1284 : : * @lru: LRU list id for isolating
1285 : : *
1286 : : * returns how many pages were moved onto *@dst.
1287 : : */
1288 : 0 : static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
1289 : : struct lruvec *lruvec, struct list_head *dst,
1290 : : unsigned long *nr_scanned, struct scan_control *sc,
1291 : : isolate_mode_t mode, enum lru_list lru)
1292 : : {
1293 : 52613 : struct list_head *src = &lruvec->lists[lru];
1294 : : unsigned long nr_taken = 0;
1295 : : unsigned long scan;
1296 : :
1297 [ + + ][ + + ]: 204903 : for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) {
1298 : : struct page *page;
1299 : : int nr_pages;
1300 : :
1301 : 152290 : page = lru_to_page(src);
1302 [ + + ]: 152290 : prefetchw_prev_lru_page(page, src, flags);
1303 : :
1304 : : VM_BUG_ON_PAGE(!PageLRU(page), page);
1305 : :
1306 [ + - - ]: 152290 : switch (__isolate_lru_page(page, mode)) {
1307 : : case 0:
1308 : : nr_pages = hpage_nr_pages(page);
1309 : : mem_cgroup_update_lru_size(lruvec, lru, -nr_pages);
1310 : 152290 : list_move(&page->lru, dst);
1311 : 152290 : nr_taken += nr_pages;
1312 : : break;
1313 : :
1314 : : case -EBUSY:
1315 : : /* else it is being freed elsewhere */
1316 : 0 : list_move(&page->lru, src);
1317 : 0 : continue;
1318 : :
1319 : : default:
1320 : 0 : BUG();
1321 : : }
1322 : : }
1323 : :
1324 : 52613 : *nr_scanned = scan;
1325 : 52613 : trace_mm_vmscan_lru_isolate(sc->order, nr_to_scan, scan,
1326 : : nr_taken, mode, is_file_lru(lru));
1327 : 0 : return nr_taken;
1328 : : }
1329 : :
1330 : : /**
1331 : : * isolate_lru_page - tries to isolate a page from its LRU list
1332 : : * @page: page to isolate from its LRU list
1333 : : *
1334 : : * Isolates a @page from an LRU list, clears PageLRU and adjusts the
1335 : : * vmstat statistic corresponding to whatever LRU list the page was on.
1336 : : *
1337 : : * Returns 0 if the page was removed from an LRU list.
1338 : : * Returns -EBUSY if the page was not on an LRU list.
1339 : : *
1340 : : * The returned page will have PageLRU() cleared. If it was found on
1341 : : * the active list, it will have PageActive set. If it was found on
1342 : : * the unevictable list, it will have the PageUnevictable bit set. That flag
1343 : : * may need to be cleared by the caller before letting the page go.
1344 : : *
1345 : : * The vmstat statistic corresponding to the list on which the page was
1346 : : * found will be decremented.
1347 : : *
1348 : : * Restrictions:
1349 : : * (1) Must be called with an elevated refcount on the page. This is a
1350 : : * fundamentnal difference from isolate_lru_pages (which is called
1351 : : * without a stable reference).
1352 : : * (2) the lru_lock must not be held.
1353 : : * (3) interrupts must be enabled.
1354 : : */
1355 : 0 : int isolate_lru_page(struct page *page)
1356 : : {
1357 : : int ret = -EBUSY;
1358 : :
1359 : : VM_BUG_ON_PAGE(!page_count(page), page);
1360 : :
1361 [ + - ]: 3783 : if (PageLRU(page)) {
1362 : 3783 : struct zone *zone = page_zone(page);
1363 : : struct lruvec *lruvec;
1364 : :
1365 : : spin_lock_irq(&zone->lru_lock);
1366 : : lruvec = mem_cgroup_page_lruvec(page, zone);
1367 [ + - ]: 3783 : if (PageLRU(page)) {
1368 : : int lru = page_lru(page);
1369 : : get_page(page);
1370 : : ClearPageLRU(page);
1371 : : del_page_from_lru_list(page, lruvec, lru);
1372 : : ret = 0;
1373 : : }
1374 : : spin_unlock_irq(&zone->lru_lock);
1375 : : }
1376 : 3783 : return ret;
1377 : : }
1378 : :
1379 : : /*
1380 : : * A direct reclaimer may isolate SWAP_CLUSTER_MAX pages from the LRU list and
1381 : : * then get resheduled. When there are massive number of tasks doing page
1382 : : * allocation, such sleeping direct reclaimers may keep piling up on each CPU,
1383 : : * the LRU list will go small and be scanned faster than necessary, leading to
1384 : : * unnecessary swapping, thrashing and OOM.
1385 : : */
1386 : 36028 : static int too_many_isolated(struct zone *zone, int file,
1387 : : struct scan_control *sc)
1388 : : {
1389 : : unsigned long inactive, isolated;
1390 : :
1391 [ + ]: 36028 : if (current_is_kswapd())
1392 : : return 0;
1393 : :
1394 : : if (!global_reclaim(sc))
1395 : : return 0;
1396 : :
1397 [ + - ]: 59294 : if (file) {
1398 : : inactive = zone_page_state(zone, NR_INACTIVE_FILE);
1399 : : isolated = zone_page_state(zone, NR_ISOLATED_FILE);
1400 : : } else {
1401 : : inactive = zone_page_state(zone, NR_INACTIVE_ANON);
1402 : : isolated = zone_page_state(zone, NR_ISOLATED_ANON);
1403 : : }
1404 : :
1405 : : /*
1406 : : * GFP_NOIO/GFP_NOFS callers are allowed to isolate more pages, so they
1407 : : * won't get blocked by normal direct-reclaimers, forming a circular
1408 : : * deadlock.
1409 : : */
1410 [ + - ]: 23266 : if ((sc->gfp_mask & GFP_IOFS) == GFP_IOFS)
1411 : 23266 : inactive >>= 3;
1412 : :
1413 : 23266 : return isolated > inactive;
1414 : : }
1415 : :
1416 : : static noinline_for_stack void
1417 : 0 : putback_inactive_pages(struct lruvec *lruvec, struct list_head *page_list)
1418 : : {
1419 : : struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
1420 : : struct zone *zone = lruvec_zone(lruvec);
1421 : 6922 : LIST_HEAD(pages_to_free);
1422 : :
1423 : : /*
1424 : : * Put back any unfreeable pages.
1425 : : */
1426 [ + + ]: 29712 : while (!list_empty(page_list)) {
1427 : 22790 : struct page *page = lru_to_page(page_list);
1428 : : int lru;
1429 : :
1430 : : VM_BUG_ON_PAGE(PageLRU(page), page);
1431 : : list_del(&page->lru);
1432 [ - + ]: 22790 : if (unlikely(!page_evictable(page))) {
1433 : : spin_unlock_irq(&zone->lru_lock);
1434 : 0 : putback_lru_page(page);
1435 : : spin_lock_irq(&zone->lru_lock);
1436 : 0 : continue;
1437 : : }
1438 : :
1439 : : lruvec = mem_cgroup_page_lruvec(page, zone);
1440 : :
1441 : : SetPageLRU(page);
1442 : : lru = page_lru(page);
1443 : : add_page_to_lru_list(page, lruvec, lru);
1444 : :
1445 [ + + ]: 22790 : if (is_active_lru(lru)) {
1446 : : int file = is_file_lru(lru);
1447 : : int numpages = hpage_nr_pages(page);
1448 : 17664 : reclaim_stat->recent_rotated[file] += numpages;
1449 : : }
1450 [ - + ]: 22790 : if (put_page_testzero(page)) {
1451 : : __ClearPageLRU(page);
1452 : : __ClearPageActive(page);
1453 : : del_page_from_lru_list(page, lruvec, lru);
1454 : :
1455 [ # # ]: 0 : if (unlikely(PageCompound(page))) {
1456 : : spin_unlock_irq(&zone->lru_lock);
1457 : 0 : (*get_compound_page_dtor(page))(page);
1458 : : spin_lock_irq(&zone->lru_lock);
1459 : : } else
1460 : : list_add(&page->lru, &pages_to_free);
1461 : : }
1462 : : }
1463 : :
1464 : : /*
1465 : : * To save our caller's stack, now use input list for pages to free.
1466 : : */
1467 : : list_splice(&pages_to_free, page_list);
1468 : 6922 : }
1469 : :
1470 : : /*
1471 : : * shrink_inactive_list() is a helper for shrink_zone(). It returns the number
1472 : : * of reclaimed pages
1473 : : */
1474 : : static noinline_for_stack unsigned long
1475 : 0 : shrink_inactive_list(unsigned long nr_to_scan, struct lruvec *lruvec,
1476 : : struct scan_control *sc, enum lru_list lru)
1477 : : {
1478 : 35839 : LIST_HEAD(page_list);
1479 : : unsigned long nr_scanned;
1480 : : unsigned long nr_reclaimed = 0;
1481 : : unsigned long nr_taken;
1482 : 35839 : unsigned long nr_dirty = 0;
1483 : 35839 : unsigned long nr_congested = 0;
1484 : 35839 : unsigned long nr_unqueued_dirty = 0;
1485 : 35839 : unsigned long nr_writeback = 0;
1486 : 35839 : unsigned long nr_immediate = 0;
1487 : : isolate_mode_t isolate_mode = 0;
1488 : : int file = is_file_lru(lru);
1489 : 35839 : struct zone *zone = lruvec_zone(lruvec);
1490 : : struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
1491 : :
1492 [ + + ]: 36029 : while (unlikely(too_many_isolated(zone, file, sc))) {
1493 : 190 : congestion_wait(BLK_RW_ASYNC, HZ/10);
1494 : :
1495 : : /* We are about to die and free our memory. Return now. */
1496 [ + - ]: 190 : if (fatal_signal_pending(current))
1497 : : return SWAP_CLUSTER_MAX;
1498 : : }
1499 : :
1500 : 35840 : lru_add_drain();
1501 : :
1502 [ - + ]: 35840 : if (!sc->may_unmap)
1503 : : isolate_mode |= ISOLATE_UNMAPPED;
1504 [ - + ]: 35840 : if (!sc->may_writepage)
1505 : 0 : isolate_mode |= ISOLATE_CLEAN;
1506 : :
1507 : : spin_lock_irq(&zone->lru_lock);
1508 : :
1509 : 35840 : nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &page_list,
1510 : : &nr_scanned, sc, isolate_mode, lru);
1511 : :
1512 : 35840 : __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken);
1513 : 35840 : __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken);
1514 : :
1515 : : if (global_reclaim(sc)) {
1516 : 35840 : zone->pages_scanned += nr_scanned;
1517 [ + + ]: 35840 : if (current_is_kswapd())
1518 : 12762 : __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scanned);
1519 : : else
1520 : 23078 : __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scanned);
1521 : : }
1522 : : spin_unlock_irq(&zone->lru_lock);
1523 : :
1524 [ + + ]: 35840 : if (nr_taken == 0)
1525 : : return 0;
1526 : :
1527 : 6922 : nr_reclaimed = shrink_page_list(&page_list, zone, sc, TTU_UNMAP,
1528 : : &nr_dirty, &nr_unqueued_dirty, &nr_congested,
1529 : : &nr_writeback, &nr_immediate,
1530 : : false);
1531 : :
1532 : : spin_lock_irq(&zone->lru_lock);
1533 : :
1534 : 6922 : reclaim_stat->recent_scanned[file] += nr_taken;
1535 : :
1536 : : if (global_reclaim(sc)) {
1537 [ + + ]: 6922 : if (current_is_kswapd())
1538 : 5376 : __count_zone_vm_events(PGSTEAL_KSWAPD, zone,
1539 : : nr_reclaimed);
1540 : : else
1541 : 1546 : __count_zone_vm_events(PGSTEAL_DIRECT, zone,
1542 : : nr_reclaimed);
1543 : : }
1544 : :
1545 : 6922 : putback_inactive_pages(lruvec, &page_list);
1546 : :
1547 : 6922 : __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken);
1548 : :
1549 : : spin_unlock_irq(&zone->lru_lock);
1550 : :
1551 : 6922 : free_hot_cold_page_list(&page_list, 1);
1552 : :
1553 : : /*
1554 : : * If reclaim is isolating dirty pages under writeback, it implies
1555 : : * that the long-lived page allocation rate is exceeding the page
1556 : : * laundering rate. Either the global limits are not being effective
1557 : : * at throttling processes due to the page distribution throughout
1558 : : * zones or there is heavy usage of a slow backing device. The
1559 : : * only option is to throttle from reclaim context which is not ideal
1560 : : * as there is no guarantee the dirtying process is throttled in the
1561 : : * same way balance_dirty_pages() manages.
1562 : : *
1563 : : * Once a zone is flagged ZONE_WRITEBACK, kswapd will count the number
1564 : : * of pages under pages flagged for immediate reclaim and stall if any
1565 : : * are encountered in the nr_immediate check below.
1566 : : */
1567 [ + + ][ + + ]: 6922 : if (nr_writeback && nr_writeback == nr_taken)
1568 : : zone_set_flag(zone, ZONE_WRITEBACK);
1569 : :
1570 : : /*
1571 : : * memcg will stall in page writeback so only consider forcibly
1572 : : * stalling for global reclaim
1573 : : */
1574 : : if (global_reclaim(sc)) {
1575 : : /*
1576 : : * Tag a zone as congested if all the dirty pages scanned were
1577 : : * backed by a congested BDI and wait_iff_congested will stall.
1578 : : */
1579 [ + + ][ + + ]: 6922 : if (nr_dirty && nr_dirty == nr_congested)
1580 : : zone_set_flag(zone, ZONE_CONGESTED);
1581 : :
1582 : : /*
1583 : : * If dirty pages are scanned that are not queued for IO, it
1584 : : * implies that flushers are not keeping up. In this case, flag
1585 : : * the zone ZONE_TAIL_LRU_DIRTY and kswapd will start writing
1586 : : * pages from reclaim context. It will forcibly stall in the
1587 : : * next check.
1588 : : */
1589 [ + + ]: 6922 : if (nr_unqueued_dirty == nr_taken)
1590 : : zone_set_flag(zone, ZONE_TAIL_LRU_DIRTY);
1591 : :
1592 : : /*
1593 : : * In addition, if kswapd scans pages marked marked for
1594 : : * immediate reclaim and under writeback (nr_immediate), it
1595 : : * implies that pages are cycling through the LRU faster than
1596 : : * they are written so also forcibly stall.
1597 : : */
1598 [ + + ][ + ]: 6922 : if (nr_unqueued_dirty == nr_taken || nr_immediate)
1599 : 0 : congestion_wait(BLK_RW_ASYNC, HZ/10);
1600 : : }
1601 : :
1602 : : /*
1603 : : * Stall direct reclaim for IO completions if underlying BDIs or zone
1604 : : * is congested. Allow kswapd to continue until it starts encountering
1605 : : * unqueued dirty pages or cycling through the LRU too quickly.
1606 : : */
1607 [ + - ][ + + ]: 42761 : if (!sc->hibernation_mode && !current_is_kswapd())
1608 : 1546 : wait_iff_congested(zone, BLK_RW_ASYNC, HZ/10);
1609 : :
1610 [ - + ]: 6922 : trace_mm_vmscan_lru_shrink_inactive(zone->zone_pgdat->node_id,
1611 : 6922 : zone_idx(zone),
1612 : : nr_scanned, nr_reclaimed,
1613 : : sc->priority,
1614 : : trace_shrink_flags(file));
1615 : 6922 : return nr_reclaimed;
1616 : : }
1617 : :
1618 : : /*
1619 : : * This moves pages from the active list to the inactive list.
1620 : : *
1621 : : * We move them the other way if the page is referenced by one or more
1622 : : * processes, from rmap.
1623 : : *
1624 : : * If the pages are mostly unmapped, the processing is fast and it is
1625 : : * appropriate to hold zone->lru_lock across the whole operation. But if
1626 : : * the pages are mapped, the processing is slow (page_referenced()) so we
1627 : : * should drop zone->lru_lock around each page. It's impossible to balance
1628 : : * this, so instead we remove the pages from the LRU while processing them.
1629 : : * It is safe to rely on PG_active against the non-LRU pages in here because
1630 : : * nobody will play with that bit on a non-LRU page.
1631 : : *
1632 : : * The downside is that we have to touch page->_count against each page.
1633 : : * But we had to alter page->flags anyway.
1634 : : */
1635 : :
1636 : 0 : static void move_active_pages_to_lru(struct lruvec *lruvec,
1637 : : struct list_head *list,
1638 : : struct list_head *pages_to_free,
1639 : : enum lru_list lru)
1640 : : {
1641 : 33546 : struct zone *zone = lruvec_zone(lruvec);
1642 : : unsigned long pgmoved = 0;
1643 : 0 : struct page *page;
1644 : : int nr_pages;
1645 : :
1646 [ + + ]: 80162 : while (!list_empty(list)) {
1647 : 46616 : page = lru_to_page(list);
1648 : : lruvec = mem_cgroup_page_lruvec(page, zone);
1649 : :
1650 : : VM_BUG_ON_PAGE(PageLRU(page), page);
1651 : : SetPageLRU(page);
1652 : :
1653 : : nr_pages = hpage_nr_pages(page);
1654 : : mem_cgroup_update_lru_size(lruvec, lru, nr_pages);
1655 : 46616 : list_move(&page->lru, &lruvec->lists[lru]);
1656 : 46616 : pgmoved += nr_pages;
1657 : :
1658 [ - + ]: 46616 : if (put_page_testzero(page)) {
1659 : : __ClearPageLRU(page);
1660 : : __ClearPageActive(page);
1661 : : del_page_from_lru_list(page, lruvec, lru);
1662 : :
1663 [ # # ]: 0 : if (unlikely(PageCompound(page))) {
1664 : : spin_unlock_irq(&zone->lru_lock);
1665 : 0 : (*get_compound_page_dtor(page))(page);
1666 : : spin_lock_irq(&zone->lru_lock);
1667 : : } else
1668 : : list_add(&page->lru, pages_to_free);
1669 : : }
1670 : : }
1671 : 33546 : __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
1672 [ + + ]: 33546 : if (!is_active_lru(lru))
1673 : : __count_vm_events(PGDEACTIVATE, pgmoved);
1674 : 33546 : }
1675 : :
1676 : 0 : static void shrink_active_list(unsigned long nr_to_scan,
1677 : : struct lruvec *lruvec,
1678 : : struct scan_control *sc,
1679 : : enum lru_list lru)
1680 : : {
1681 : : unsigned long nr_taken;
1682 : : unsigned long nr_scanned;
1683 : : unsigned long vm_flags;
1684 : 16773 : LIST_HEAD(l_hold); /* The pages which were snipped off */
1685 : 16773 : LIST_HEAD(l_active);
1686 : 16773 : LIST_HEAD(l_inactive);
1687 : 0 : struct page *page;
1688 : : struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
1689 : : unsigned long nr_rotated = 0;
1690 : : isolate_mode_t isolate_mode = 0;
1691 : : int file = is_file_lru(lru);
1692 : 16773 : struct zone *zone = lruvec_zone(lruvec);
1693 : :
1694 : 16773 : lru_add_drain();
1695 : :
1696 [ - + ]: 16772 : if (!sc->may_unmap)
1697 : : isolate_mode |= ISOLATE_UNMAPPED;
1698 [ - + ]: 16772 : if (!sc->may_writepage)
1699 : 0 : isolate_mode |= ISOLATE_CLEAN;
1700 : :
1701 : : spin_lock_irq(&zone->lru_lock);
1702 : :
1703 : 16773 : nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &l_hold,
1704 : : &nr_scanned, sc, isolate_mode, lru);
1705 : : if (global_reclaim(sc))
1706 : 16773 : zone->pages_scanned += nr_scanned;
1707 : :
1708 : 16773 : reclaim_stat->recent_scanned[file] += nr_taken;
1709 : :
1710 : 16773 : __count_zone_vm_events(PGREFILL, zone, nr_scanned);
1711 : 16773 : __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken);
1712 : 16773 : __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken);
1713 : : spin_unlock_irq(&zone->lru_lock);
1714 : :
1715 [ + + ]: 63386 : while (!list_empty(&l_hold)) {
1716 : 46613 : cond_resched();
1717 : 46615 : page = lru_to_page(&l_hold);
1718 : : list_del(&page->lru);
1719 : :
1720 [ - + ]: 46615 : if (unlikely(!page_evictable(page))) {
1721 : 0 : putback_lru_page(page);
1722 : 0 : continue;
1723 : : }
1724 : :
1725 [ - + ]: 46615 : if (unlikely(buffer_heads_over_limit)) {
1726 [ # # # # ]: 0 : if (page_has_private(page) && trylock_page(page)) {
1727 [ # # ]: 0 : if (page_has_private(page))
1728 : 0 : try_to_release_page(page, 0);
1729 : 0 : unlock_page(page);
1730 : : }
1731 : : }
1732 : :
1733 [ + + ]: 46615 : if (page_referenced(page, 0, sc->target_mem_cgroup,
1734 : : &vm_flags)) {
1735 : 896 : nr_rotated += hpage_nr_pages(page);
1736 : : /*
1737 : : * Identify referenced, file-backed active pages and
1738 : : * give them one more trip around the active list. So
1739 : : * that executable code get better chances to stay in
1740 : : * memory under moderate memory pressure. Anon pages
1741 : : * are not likely to be evicted by use-once streaming
1742 : : * IO, plus JVM can create lots of anon VM_EXEC pages,
1743 : : * so we ignore them here.
1744 : : */
1745 [ + + ][ + - ]: 896 : if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) {
1746 : 871 : list_add(&page->lru, &l_active);
1747 : 871 : continue;
1748 : : }
1749 : : }
1750 : :
1751 : : ClearPageActive(page); /* we are de-activating */
1752 : 45743 : list_add(&page->lru, &l_inactive);
1753 : : }
1754 : :
1755 : : /*
1756 : : * Move pages back to the lru list.
1757 : : */
1758 : : spin_lock_irq(&zone->lru_lock);
1759 : : /*
1760 : : * Count referenced pages from currently used mappings as rotated,
1761 : : * even though only some of them are actually re-activated. This
1762 : : * helps balance scan pressure between file and anonymous pages in
1763 : : * get_scan_ratio.
1764 : : */
1765 : 16773 : reclaim_stat->recent_rotated[file] += nr_rotated;
1766 : :
1767 : 16773 : move_active_pages_to_lru(lruvec, &l_active, &l_hold, lru);
1768 : 16773 : move_active_pages_to_lru(lruvec, &l_inactive, &l_hold, lru - LRU_ACTIVE);
1769 : 16773 : __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken);
1770 : : spin_unlock_irq(&zone->lru_lock);
1771 : :
1772 : 16773 : free_hot_cold_page_list(&l_hold, 1);
1773 : 16773 : }
1774 : :
1775 : : #ifdef CONFIG_SWAP
1776 : : static int inactive_anon_is_low_global(struct zone *zone)
1777 : : {
1778 : : unsigned long active, inactive;
1779 : :
1780 : : active = zone_page_state(zone, NR_ACTIVE_ANON);
1781 : : inactive = zone_page_state(zone, NR_INACTIVE_ANON);
1782 : :
1783 [ # # ][ # # ]: 0 : if (inactive * zone->inactive_ratio < active)
[ # # ]
1784 : : return 1;
1785 : :
1786 : : return 0;
1787 : : }
1788 : :
1789 : : /**
1790 : : * inactive_anon_is_low - check if anonymous pages need to be deactivated
1791 : : * @lruvec: LRU vector to check
1792 : : *
1793 : : * Returns true if the zone does not have enough inactive anon pages,
1794 : : * meaning some active anon pages need to be deactivated.
1795 : : */
1796 : : static int inactive_anon_is_low(struct lruvec *lruvec)
1797 : : {
1798 : : /*
1799 : : * If we don't have swap space, anonymous page deactivation
1800 : : * is pointless.
1801 : : */
1802 [ # # - + ]: 161285 : if (!total_swap_pages)
[ # # ]
1803 : : return 0;
1804 : :
1805 : : if (!mem_cgroup_disabled())
1806 : : return mem_cgroup_inactive_anon_is_low(lruvec);
1807 : :
1808 : : return inactive_anon_is_low_global(lruvec_zone(lruvec));
1809 : : }
1810 : : #else
1811 : : static inline int inactive_anon_is_low(struct lruvec *lruvec)
1812 : : {
1813 : : return 0;
1814 : : }
1815 : : #endif
1816 : :
1817 : : /**
1818 : : * inactive_file_is_low - check if file pages need to be deactivated
1819 : : * @lruvec: LRU vector to check
1820 : : *
1821 : : * When the system is doing streaming IO, memory pressure here
1822 : : * ensures that active file pages get deactivated, until more
1823 : : * than half of the file pages are on the inactive list.
1824 : : *
1825 : : * Once we get to that situation, protect the system's working
1826 : : * set from being evicted by disabling active file page aging.
1827 : : *
1828 : : * This uses a different ratio than the anonymous pages, because
1829 : : * the page cache uses a use-once replacement algorithm.
1830 : : */
1831 : : static int inactive_file_is_low(struct lruvec *lruvec)
1832 : : {
1833 : : unsigned long inactive;
1834 : : unsigned long active;
1835 : :
1836 : : inactive = get_lru_size(lruvec, LRU_INACTIVE_FILE);
1837 : : active = get_lru_size(lruvec, LRU_ACTIVE_FILE);
1838 : :
1839 : 31814 : return active > inactive;
1840 : : }
1841 : :
1842 : 0 : static int inactive_list_is_low(struct lruvec *lruvec, enum lru_list lru)
1843 : : {
1844 [ + - ]: 31814 : if (is_file_lru(lru))
1845 : 31814 : return inactive_file_is_low(lruvec);
1846 : : else
1847 : 0 : return inactive_anon_is_low(lruvec);
1848 : : }
1849 : :
1850 : 0 : static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
1851 : : struct lruvec *lruvec, struct scan_control *sc)
1852 : : {
1853 [ + + ]: 67653 : if (is_active_lru(lru)) {
1854 [ + + ]: 31814 : if (inactive_list_is_low(lruvec, lru))
1855 : 16773 : shrink_active_list(nr_to_scan, lruvec, sc, lru);
1856 : : return 0;
1857 : : }
1858 : :
1859 : 35839 : return shrink_inactive_list(nr_to_scan, lruvec, sc, lru);
1860 : : }
1861 : :
1862 : : static int vmscan_swappiness(struct scan_control *sc)
1863 : : {
1864 : : if (global_reclaim(sc))
1865 : 0 : return vm_swappiness;
1866 : : return mem_cgroup_swappiness(sc->target_mem_cgroup);
1867 : : }
1868 : :
1869 : : enum scan_balance {
1870 : : SCAN_EQUAL,
1871 : : SCAN_FRACT,
1872 : : SCAN_ANON,
1873 : : SCAN_FILE,
1874 : : };
1875 : :
1876 : : /*
1877 : : * Determine how aggressively the anon and file LRU lists should be
1878 : : * scanned. The relative value of each set of LRU lists is determined
1879 : : * by looking at the fraction of the pages scanned we did rotate back
1880 : : * onto the active list instead of evict.
1881 : : *
1882 : : * nr[0] = anon inactive pages to scan; nr[1] = anon active pages to scan
1883 : : * nr[2] = file inactive pages to scan; nr[3] = file active pages to scan
1884 : : */
1885 : 0 : static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
1886 : : unsigned long *nr)
1887 : : {
1888 : : struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
1889 : : u64 fraction[2];
1890 : : u64 denominator = 0; /* gcc */
1891 : 161206 : struct zone *zone = lruvec_zone(lruvec);
1892 : : unsigned long anon_prio, file_prio;
1893 : : enum scan_balance scan_balance;
1894 : : unsigned long anon, file, free;
1895 : : bool force_scan = false;
1896 : : unsigned long ap, fp;
1897 : : enum lru_list lru;
1898 : :
1899 : : /*
1900 : : * If the zone or memcg is small, nr[l] can be 0. This
1901 : : * results in no scanning on this priority and a potential
1902 : : * priority drop. Global direct reclaim can go to the next
1903 : : * zone and tends to have no problems. Global kswapd is for
1904 : : * zone balancing and it needs to scan a minimum amount. When
1905 : : * reclaiming for a memcg, a priority drop can cause high
1906 : : * latencies, so it's better to scan a minimum amount there as
1907 : : * well.
1908 : : */
1909 [ + + + + ]: 210890 : if (current_is_kswapd() && !zone_reclaimable(zone))
1910 : : force_scan = true;
1911 : : if (!global_reclaim(sc))
1912 : : force_scan = true;
1913 : :
1914 : : /* If we have no swap space, do not bother scanning anon pages. */
1915 [ + ][ - + ]: 161206 : if (!sc->may_swap || (get_nr_swap_pages() <= 0)) {
1916 : : scan_balance = SCAN_FILE;
1917 : : goto out;
1918 : : }
1919 : :
1920 : : /*
1921 : : * Global reclaim will swap to prevent OOM even with no
1922 : : * swappiness, but memcg users want to use this knob to
1923 : : * disable swapping for individual groups completely when
1924 : : * using the memory controller's swap limit feature would be
1925 : : * too expensive.
1926 : : */
1927 : : if (!global_reclaim(sc) && !vmscan_swappiness(sc)) {
1928 : : scan_balance = SCAN_FILE;
1929 : : goto out;
1930 : : }
1931 : :
1932 : : /*
1933 : : * Do not apply any pressure balancing cleverness when the
1934 : : * system is close to OOM, scan both anon and file equally
1935 : : * (unless the swappiness setting disagrees with swapping).
1936 : : */
1937 [ # # ][ # # ]: 0 : if (!sc->priority && vmscan_swappiness(sc)) {
1938 : : scan_balance = SCAN_EQUAL;
1939 : : goto out;
1940 : : }
1941 : :
1942 : 0 : anon = get_lru_size(lruvec, LRU_ACTIVE_ANON) +
1943 : : get_lru_size(lruvec, LRU_INACTIVE_ANON);
1944 : 0 : file = get_lru_size(lruvec, LRU_ACTIVE_FILE) +
1945 : : get_lru_size(lruvec, LRU_INACTIVE_FILE);
1946 : :
1947 : : /*
1948 : : * If it's foreseeable that reclaiming the file cache won't be
1949 : : * enough to get the zone back into a desirable shape, we have
1950 : : * to swap. Better start now and leave the - probably heavily
1951 : : * thrashing - remaining file pages alone.
1952 : : */
1953 : : if (global_reclaim(sc)) {
1954 : : free = zone_page_state(zone, NR_FREE_PAGES);
1955 [ # # ]: 0 : if (unlikely(file + free <= high_wmark_pages(zone))) {
1956 : : scan_balance = SCAN_ANON;
1957 : : goto out;
1958 : : }
1959 : : }
1960 : :
1961 : : /*
1962 : : * There is enough inactive page cache, do not reclaim
1963 : : * anything from the anonymous working set right now.
1964 : : */
1965 [ # # ]: 0 : if (!inactive_file_is_low(lruvec)) {
1966 : : scan_balance = SCAN_FILE;
1967 : : goto out;
1968 : : }
1969 : :
1970 : : scan_balance = SCAN_FRACT;
1971 : :
1972 : : /*
1973 : : * With swappiness at 100, anonymous and file have the same priority.
1974 : : * This scanning priority is essentially the inverse of IO cost.
1975 : : */
1976 : 0 : anon_prio = vmscan_swappiness(sc);
1977 : 0 : file_prio = 200 - anon_prio;
1978 : :
1979 : : /*
1980 : : * OK, so we have swap space and a fair amount of page cache
1981 : : * pages. We use the recently rotated / recently scanned
1982 : : * ratios to determine how valuable each cache is.
1983 : : *
1984 : : * Because workloads change over time (and to avoid overflow)
1985 : : * we keep these statistics as a floating average, which ends
1986 : : * up weighing recent references more than old ones.
1987 : : *
1988 : : * anon in [0], file in [1]
1989 : : */
1990 : : spin_lock_irq(&zone->lru_lock);
1991 [ # # ]: 0 : if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) {
1992 : 0 : reclaim_stat->recent_scanned[0] /= 2;
1993 : 0 : reclaim_stat->recent_rotated[0] /= 2;
1994 : : }
1995 : :
1996 [ # # ]: 0 : if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) {
1997 : 0 : reclaim_stat->recent_scanned[1] /= 2;
1998 : 0 : reclaim_stat->recent_rotated[1] /= 2;
1999 : : }
2000 : :
2001 : : /*
2002 : : * The amount of pressure on anon vs file pages is inversely
2003 : : * proportional to the fraction of recently scanned pages on
2004 : : * each list that were recently referenced and in active use.
2005 : : */
2006 : 0 : ap = anon_prio * (reclaim_stat->recent_scanned[0] + 1);
2007 : 0 : ap /= reclaim_stat->recent_rotated[0] + 1;
2008 : :
2009 : 0 : fp = file_prio * (reclaim_stat->recent_scanned[1] + 1);
2010 : 0 : fp /= reclaim_stat->recent_rotated[1] + 1;
2011 : : spin_unlock_irq(&zone->lru_lock);
2012 : :
2013 : 0 : fraction[0] = ap;
2014 : 0 : fraction[1] = fp;
2015 : 161206 : denominator = ap + fp + 1;
2016 : : out:
2017 [ + + ]: 806168 : for_each_evictable_lru(lru) {
2018 : : int file = is_file_lru(lru);
2019 : : unsigned long size;
2020 : : unsigned long scan;
2021 : :
2022 : : size = get_lru_size(lruvec, lru);
2023 : 644896 : scan = size >> sc->priority;
2024 : :
2025 [ + + ]: 644896 : if (!scan && force_scan)
2026 : 8130 : scan = min(size, SWAP_CLUSTER_MAX);
2027 : :
2028 [ - + + ]: 644896 : switch (scan_balance) {
2029 : : case SCAN_EQUAL:
2030 : : /* Scan lists relative to size */
2031 : : break;
2032 : : case SCAN_FRACT:
2033 : : /*
2034 : : * Scan types proportional to swappiness and
2035 : : * their relative recent reclaim efficiency.
2036 : : */
2037 : 0 : scan = div64_u64(scan * fraction[file], denominator);
2038 : : break;
2039 : : case SCAN_FILE:
2040 : : case SCAN_ANON:
2041 : : /* Scan one type exclusively */
2042 [ + + ]: 644919 : if ((scan_balance == SCAN_FILE) != file)
2043 : : scan = 0;
2044 : : break;
2045 : : default:
2046 : : /* Look ma, no brain */
2047 : 0 : BUG();
2048 : : }
2049 : 644962 : nr[lru] = scan;
2050 : : }
2051 : 161272 : }
2052 : :
2053 : : /*
2054 : : * This is a basic per-zone page freer. Used by both kswapd and direct reclaim.
2055 : : */
2056 : 0 : static void shrink_lruvec(struct lruvec *lruvec, struct scan_control *sc)
2057 : : {
2058 : : unsigned long nr[NR_LRU_LISTS];
2059 : : unsigned long targets[NR_LRU_LISTS];
2060 : : unsigned long nr_to_scan;
2061 : : enum lru_list lru;
2062 : : unsigned long nr_reclaimed = 0;
2063 : 161260 : unsigned long nr_to_reclaim = sc->nr_to_reclaim;
2064 : : struct blk_plug plug;
2065 : : bool scan_adjusted = false;
2066 : :
2067 : 161260 : get_scan_count(lruvec, sc, nr);
2068 : :
2069 : : /* Record the original scan target for proportional adjustments later */
2070 : 161268 : memcpy(targets, nr, sizeof(nr));
2071 : :
2072 : 161268 : blk_start_plug(&plug);
2073 [ + + ][ + + ]: 209264 : while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] ||
[ + + ]
2074 : 177449 : nr[LRU_INACTIVE_FILE]) {
2075 : : unsigned long nr_anon, nr_file, percentage;
2076 : : unsigned long nr_scanned;
2077 : :
2078 [ + + ]: 239893 : for_each_evictable_lru(lru) {
2079 [ + + ]: 191907 : if (nr[lru]) {
2080 : 0 : nr_to_scan = min(nr[lru], SWAP_CLUSTER_MAX);
2081 : 0 : nr[lru] -= nr_to_scan;
2082 : :
2083 : 67651 : nr_reclaimed += shrink_list(lru, nr_to_scan,
2084 : : lruvec, sc);
2085 : : }
2086 : : }
2087 : :
2088 [ + + ]: 47986 : if (nr_reclaimed < nr_to_reclaim || scan_adjusted)
2089 : 47983 : continue;
2090 : :
2091 : : /*
2092 : : * For global direct reclaim, reclaim only the number of pages
2093 : : * requested. Less care is taken to scan proportionally as it
2094 : : * is more important to minimise direct reclaim stall latency
2095 : : * than it is to properly age the LRU lists.
2096 : : */
2097 [ - + ]: 3 : if (global_reclaim(sc) && !current_is_kswapd())
2098 : : break;
2099 : :
2100 : : /*
2101 : : * For kswapd and memcg, reclaim at least the number of pages
2102 : : * requested. Ensure that the anon and file LRUs shrink
2103 : : * proportionally what was requested by get_scan_count(). We
2104 : : * stop reclaiming one LRU and reduce the amount scanning
2105 : : * proportional to the original scan target.
2106 : : */
2107 : 0 : nr_file = nr[LRU_INACTIVE_FILE] + nr[LRU_ACTIVE_FILE];
2108 : 0 : nr_anon = nr[LRU_INACTIVE_ANON] + nr[LRU_ACTIVE_ANON];
2109 : :
2110 [ # # ]: 0 : if (nr_file > nr_anon) {
2111 : 0 : unsigned long scan_target = targets[LRU_INACTIVE_ANON] +
2112 : 0 : targets[LRU_ACTIVE_ANON] + 1;
2113 : : lru = LRU_BASE;
2114 : 0 : percentage = nr_anon * 100 / scan_target;
2115 : : } else {
2116 : 0 : unsigned long scan_target = targets[LRU_INACTIVE_FILE] +
2117 : 0 : targets[LRU_ACTIVE_FILE] + 1;
2118 : : lru = LRU_FILE;
2119 : 0 : percentage = nr_file * 100 / scan_target;
2120 : : }
2121 : :
2122 : : /* Stop scanning the smaller of the LRU */
2123 : 0 : nr[lru] = 0;
2124 : 0 : nr[lru + LRU_ACTIVE] = 0;
2125 : :
2126 : : /*
2127 : : * Recalculate the other LRU scan count based on its original
2128 : : * scan target and the percentage scanning already complete
2129 : : */
2130 [ # # ]: 0 : lru = (lru == LRU_FILE) ? LRU_BASE : LRU_FILE;
2131 : 0 : nr_scanned = targets[lru] - nr[lru];
2132 : 0 : nr[lru] = targets[lru] * (100 - percentage) / 100;
2133 : 0 : nr[lru] -= min(nr[lru], nr_scanned);
2134 : :
2135 : 0 : lru += LRU_ACTIVE;
2136 : 0 : nr_scanned = targets[lru] - nr[lru];
2137 : 0 : nr[lru] = targets[lru] * (100 - percentage) / 100;
2138 : 47983 : nr[lru] -= min(nr[lru], nr_scanned);
2139 : :
2140 : : scan_adjusted = true;
2141 : : }
2142 : 161284 : blk_finish_plug(&plug);
2143 : 161285 : sc->nr_reclaimed += nr_reclaimed;
2144 : :
2145 : : /*
2146 : : * Even if we did not try to evict anon pages at all, we want to
2147 : : * rebalance the anon lru active/inactive ratio.
2148 : : */
2149 [ - + ]: 161285 : if (inactive_anon_is_low(lruvec))
2150 : 0 : shrink_active_list(SWAP_CLUSTER_MAX, lruvec,
2151 : : sc, LRU_ACTIVE_ANON);
2152 : :
2153 : 161285 : throttle_vm_writeout(sc->gfp_mask);
2154 : 161269 : }
2155 : :
2156 : : /* Use reclaim/compaction for costly allocs or under memory pressure */
2157 : 161271 : static bool in_reclaim_compaction(struct scan_control *sc)
2158 : : {
2159 [ + + ][ + - ]: 161271 : if (IS_ENABLED(CONFIG_COMPACTION) && sc->order &&
2160 [ - + ]: 161271 : (sc->order > PAGE_ALLOC_COSTLY_ORDER ||
2161 : 2 : sc->priority < DEF_PRIORITY - 2))
2162 : : return true;
2163 : :
2164 : : return false;
2165 : : }
2166 : :
2167 : : /*
2168 : : * Reclaim/compaction is used for high-order allocation requests. It reclaims
2169 : : * order-0 pages before compacting the zone. should_continue_reclaim() returns
2170 : : * true if more pages should be reclaimed such that when the page allocator
2171 : : * calls try_to_compact_zone() that it will have enough free pages to succeed.
2172 : : * It will give up earlier than that if there is difficulty reclaiming pages.
2173 : : */
2174 : : static inline bool should_continue_reclaim(struct zone *zone,
2175 : : unsigned long nr_reclaimed,
2176 : : unsigned long nr_scanned,
2177 : 161257 : struct scan_control *sc)
2178 : : {
2179 : : unsigned long pages_for_compaction;
2180 : : unsigned long inactive_lru_pages;
2181 : :
2182 : : /* If not in reclaim/compaction mode, stop */
2183 [ - + ]: 161257 : if (!in_reclaim_compaction(sc))
2184 : : return false;
2185 : :
2186 : : /* Consider stopping depending on scan and reclaim activity */
2187 [ # # ]: 0 : if (sc->gfp_mask & __GFP_REPEAT) {
2188 : : /*
2189 : : * For __GFP_REPEAT allocations, stop reclaiming if the
2190 : : * full LRU list has been scanned and we are still failing
2191 : : * to reclaim pages. This full LRU scan is potentially
2192 : : * expensive but a __GFP_REPEAT caller really wants to succeed
2193 : : */
2194 [ # # ]: 0 : if (!nr_reclaimed && !nr_scanned)
2195 : : return false;
2196 : : } else {
2197 : : /*
2198 : : * For non-__GFP_REPEAT allocations which can presumably
2199 : : * fail without consequence, stop if we failed to reclaim
2200 : : * any pages from the last SWAP_CLUSTER_MAX number of
2201 : : * pages that were scanned. This will return to the
2202 : : * caller faster at the risk reclaim/compaction and
2203 : : * the resulting allocation attempt fails
2204 : : */
2205 [ # # ]: 0 : if (!nr_reclaimed)
2206 : : return false;
2207 : : }
2208 : :
2209 : : /*
2210 : : * If we have not reclaimed enough pages for compaction and the
2211 : : * inactive lists are large enough, continue reclaiming
2212 : : */
2213 : 19 : pages_for_compaction = (2UL << sc->order);
2214 : : inactive_lru_pages = zone_page_state(zone, NR_INACTIVE_FILE);
2215 [ - + ]: 19 : if (get_nr_swap_pages() > 0)
2216 : 0 : inactive_lru_pages += zone_page_state(zone, NR_INACTIVE_ANON);
2217 [ - ][ # # ]: 19 : if (sc->nr_reclaimed < pages_for_compaction &&
2218 : : inactive_lru_pages > pages_for_compaction)
2219 : : return true;
2220 : :
2221 : : /* If compaction would go ahead or the allocation would succeed, stop */
2222 [ # # ]: 0 : switch (compaction_suitable(zone, sc->order)) {
2223 : : case COMPACT_PARTIAL:
2224 : : case COMPACT_CONTINUE:
2225 : : return false;
2226 : : default:
2227 : : return true;
2228 : : }
2229 : : }
2230 : :
2231 : 161260 : static void shrink_zone(struct zone *zone, struct scan_control *sc)
2232 : : {
2233 : : unsigned long nr_reclaimed, nr_scanned;
2234 : :
2235 : : do {
2236 : : struct mem_cgroup *root = sc->target_mem_cgroup;
2237 : : struct mem_cgroup_reclaim_cookie reclaim = {
2238 : : .zone = zone,
2239 : : .priority = sc->priority,
2240 : : };
2241 : : struct mem_cgroup *memcg;
2242 : :
2243 : 161260 : nr_reclaimed = sc->nr_reclaimed;
2244 : 161260 : nr_scanned = sc->nr_scanned;
2245 : :
2246 : : memcg = mem_cgroup_iter(root, NULL, &reclaim);
2247 : : do {
2248 : : struct lruvec *lruvec;
2249 : :
2250 : 161260 : lruvec = mem_cgroup_zone_lruvec(zone, memcg);
2251 : :
2252 : 161260 : shrink_lruvec(lruvec, sc);
2253 : :
2254 : : /*
2255 : : * Direct reclaim and kswapd have to scan all memory
2256 : : * cgroups to fulfill the overall scan target for the
2257 : : * zone.
2258 : : *
2259 : : * Limit reclaim, on the other hand, only cares about
2260 : : * nr_to_reclaim pages to be reclaimed and it will
2261 : : * retry with decreasing priority if one round over the
2262 : : * whole hierarchy is not sufficient.
2263 : : */
2264 : : if (!global_reclaim(sc) &&
2265 : : sc->nr_reclaimed >= sc->nr_to_reclaim) {
2266 : : mem_cgroup_iter_break(root, memcg);
2267 : : break;
2268 : : }
2269 : : memcg = mem_cgroup_iter(root, memcg, &reclaim);
2270 : : } while (memcg);
2271 : :
2272 : : vmpressure(sc->gfp_mask, sc->target_mem_cgroup,
2273 : : sc->nr_scanned - nr_scanned,
2274 : : sc->nr_reclaimed - nr_reclaimed);
2275 : :
2276 : 322514 : } while (should_continue_reclaim(zone, sc->nr_reclaimed - nr_reclaimed,
2277 [ - + ]: 161242 : sc->nr_scanned - nr_scanned, sc));
2278 : 161242 : }
2279 : :
2280 : : /* Returns true if compaction should go ahead for a high-order request */
2281 : : static inline bool compaction_ready(struct zone *zone, struct scan_control *sc)
2282 : : {
2283 : : unsigned long balance_gap, watermark;
2284 : : bool watermark_ok;
2285 : :
2286 : : /* Do not consider compaction for orders reclaim is meant to satisfy */
2287 [ - + ]: 111622 : if (sc->order <= PAGE_ALLOC_COSTLY_ORDER)
2288 : : return false;
2289 : :
2290 : : /*
2291 : : * Compaction takes time to run and there are potentially other
2292 : : * callers using the pages just freed. Continue reclaiming until
2293 : : * there is a buffer of free pages available to give compaction
2294 : : * a reasonable chance of completing and allocating the page
2295 : : */
2296 : 0 : balance_gap = min(low_wmark_pages(zone),
2297 : : (zone->managed_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) /
2298 : : KSWAPD_ZONE_BALANCE_GAP_RATIO);
2299 : 0 : watermark = high_wmark_pages(zone) + balance_gap + (2UL << sc->order);
2300 : 0 : watermark_ok = zone_watermark_ok_safe(zone, 0, watermark, 0, 0);
2301 : :
2302 : : /*
2303 : : * If compaction is deferred, reclaim up to a point where
2304 : : * compaction will have a chance of success when re-enabled
2305 : : */
2306 [ # # ]: 0 : if (compaction_deferred(zone, sc->order))
2307 : : return watermark_ok;
2308 : :
2309 : : /* If compaction is not ready to start, keep reclaiming */
2310 [ # # ]: 0 : if (!compaction_suitable(zone, sc->order))
2311 : : return false;
2312 : :
2313 : : return watermark_ok;
2314 : : }
2315 : :
2316 : : /*
2317 : : * This is the direct reclaim path, for page-allocating processes. We only
2318 : : * try to reclaim pages from zones which will satisfy the caller's allocation
2319 : : * request.
2320 : : *
2321 : : * We reclaim from a zone even if that zone is over high_wmark_pages(zone).
2322 : : * Because:
2323 : : * a) The caller may be trying to free *extra* pages to satisfy a higher-order
2324 : : * allocation or
2325 : : * b) The target zone may be at high_wmark_pages(zone) but the lower zones
2326 : : * must go *over* high_wmark_pages(zone) to satisfy the `incremental min'
2327 : : * zone defense algorithm.
2328 : : *
2329 : : * If a zone is deemed to be full of pinned pages then just give it a light
2330 : : * scan then give up on it.
2331 : : *
2332 : : * This function returns true if a zone is being reclaimed for a costly
2333 : : * high-order allocation and compaction is ready to begin. This indicates to
2334 : : * the caller that it should consider retrying the allocation instead of
2335 : : * further reclaim.
2336 : : */
2337 : 0 : static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
2338 : : {
2339 : : struct zoneref *z;
2340 : : struct zone *zone;
2341 : : unsigned long nr_soft_reclaimed;
2342 : : unsigned long nr_soft_scanned;
2343 : : bool aborted_reclaim = false;
2344 : :
2345 : : /*
2346 : : * If the number of buffer_heads in the machine exceeds the maximum
2347 : : * allowed level, force direct reclaim to scan the highmem zone as
2348 : : * highmem pages could be pinning lowmem pages storing buffer_heads
2349 : : */
2350 [ - + ]: 76089 : if (buffer_heads_over_limit)
2351 : 0 : sc->gfp_mask |= __GFP_HIGHMEM;
2352 : :
2353 [ + + ]: 152193 : for_each_zone_zonelist_nodemask(zone, z, zonelist,
2354 : : gfp_zone(sc->gfp_mask), sc->nodemask) {
2355 [ + ]: 152191 : if (!populated_zone(zone))
2356 : 0 : continue;
2357 : : /*
2358 : : * Take care memory controller reclaiming has small influence
2359 : : * to global LRU.
2360 : : */
2361 : : if (global_reclaim(sc)) {
2362 : : if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
2363 : : continue;
2364 [ + + + + ]: 368780 : if (sc->priority != DEF_PRIORITY &&
2365 : : !zone_reclaimable(zone))
2366 : 40581 : continue; /* Let kswapd poll it */
2367 : : if (IS_ENABLED(CONFIG_COMPACTION)) {
2368 : : /*
2369 : : * If we already have plenty of memory free for
2370 : : * compaction in this zone, don't free any more.
2371 : : * Even though compaction is invoked for any
2372 : : * non-zero order, only frequent costly order
2373 : : * reclamation is disruptive enough to become a
2374 : : * noticeable problem, like transparent huge
2375 : : * page allocations.
2376 : : */
2377 [ - + ]: 111622 : if (compaction_ready(zone, sc)) {
2378 : : aborted_reclaim = true;
2379 : 0 : continue;
2380 : : }
2381 : : }
2382 : : /*
2383 : : * This steals pages from memory cgroups over softlimit
2384 : : * and returns the number of reclaimed pages and
2385 : : * scanned pages. This works for global memory pressure
2386 : : * and balancing, not for a memcg's limit.
2387 : : */
2388 : : nr_soft_scanned = 0;
2389 : : nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone,
2390 : : sc->order, sc->gfp_mask,
2391 : : &nr_soft_scanned);
2392 : : sc->nr_reclaimed += nr_soft_reclaimed;
2393 : : sc->nr_scanned += nr_soft_scanned;
2394 : : /* need some check for avoid more shrink_zone() */
2395 : : }
2396 : :
2397 : 111622 : shrink_zone(zone, sc);
2398 : : }
2399 : :
2400 : 76093 : return aborted_reclaim;
2401 : : }
2402 : :
2403 : : /* All zones in zonelist are unreclaimable? */
2404 : 5733 : static bool all_unreclaimable(struct zonelist *zonelist,
2405 : : struct scan_control *sc)
2406 : : {
2407 : : struct zoneref *z;
2408 : : struct zone *zone;
2409 : :
2410 [ + + ]: 7726 : for_each_zone_zonelist_nodemask(zone, z, zonelist,
2411 : : gfp_zone(sc->gfp_mask), sc->nodemask) {
2412 [ - + ]: 6817 : if (!populated_zone(zone))
2413 : 0 : continue;
2414 : : if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
2415 : : continue;
2416 [ + + ]: 6817 : if (zone_reclaimable(zone))
2417 : : return false;
2418 : : }
2419 : :
2420 : : return true;
2421 : : }
2422 : :
2423 : : /*
2424 : : * This is the main entry point to direct page reclaim.
2425 : : *
2426 : : * If a full scan of the inactive list fails to free enough memory then we
2427 : : * are "out of memory" and something needs to be killed.
2428 : : *
2429 : : * If the caller is !__GFP_FS then the probability of a failure is reasonably
2430 : : * high - the zone may be full of dirty or under-writeback pages, which this
2431 : : * caller can't do much about. We kick the writeback threads and take explicit
2432 : : * naps in the hope that some of these pages can be written. But if the
2433 : : * allocating task holds filesystem locks which prevent writeout this might not
2434 : : * work, and the allocation attempt will fail.
2435 : : *
2436 : : * returns: 0, if no pages reclaimed
2437 : : * else, the number of pages reclaimed
2438 : : */
2439 : 0 : static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
2440 : : struct scan_control *sc,
2441 : : struct shrink_control *shrink)
2442 : : {
2443 : : unsigned long total_scanned = 0;
2444 : 5856 : struct reclaim_state *reclaim_state = current->reclaim_state;
2445 : : struct zoneref *z;
2446 : : struct zone *zone;
2447 : : unsigned long writeback_threshold;
2448 : : bool aborted_reclaim;
2449 : :
2450 : : delayacct_freepages_start();
2451 : :
2452 : : if (global_reclaim(sc))
2453 : : count_vm_event(ALLOCSTALL);
2454 : :
2455 : : do {
2456 : : vmpressure_prio(sc->gfp_mask, sc->target_mem_cgroup,
2457 : : sc->priority);
2458 : 76097 : sc->nr_scanned = 0;
2459 : 76097 : aborted_reclaim = shrink_zones(zonelist, sc);
2460 : :
2461 : : /*
2462 : : * Don't shrink slabs when reclaiming memory from over limit
2463 : : * cgroups but do shrink slab at least once when aborting
2464 : : * reclaim for compaction to avoid unevenly scanning file/anon
2465 : : * LRU pages over slab pages.
2466 : : */
2467 : : if (global_reclaim(sc)) {
2468 : : unsigned long lru_pages = 0;
2469 : :
2470 : : nodes_clear(shrink->nodes_to_scan);
2471 [ + + ]: 228313 : for_each_zone_zonelist(zone, z, zonelist,
2472 : : gfp_zone(sc->gfp_mask)) {
2473 : : if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
2474 : : continue;
2475 : :
2476 : 152202 : lru_pages += zone_reclaimable_pages(zone);
2477 : : node_set(zone_to_nid(zone),
2478 : : shrink->nodes_to_scan);
2479 : : }
2480 : :
2481 : 76106 : shrink_slab(shrink, sc->nr_scanned, lru_pages);
2482 [ + - ]: 76106 : if (reclaim_state) {
2483 : 76106 : sc->nr_reclaimed += reclaim_state->reclaimed_slab;
2484 : 76106 : reclaim_state->reclaimed_slab = 0;
2485 : : }
2486 : : }
2487 : 76106 : total_scanned += sc->nr_scanned;
2488 [ + + ]: 76106 : if (sc->nr_reclaimed >= sc->nr_to_reclaim)
2489 : : goto out;
2490 : :
2491 : : /*
2492 : : * If we're getting trouble reclaiming, start doing
2493 : : * writepage even in laptop mode.
2494 : : */
2495 [ + + ]: 76081 : if (sc->priority < DEF_PRIORITY - 2)
2496 : 58520 : sc->may_writepage = 1;
2497 : :
2498 : : /*
2499 : : * Try to write back as many pages as we just scanned. This
2500 : : * tends to cause slow streaming writers to write data to the
2501 : : * disk smoothly, at the dirtying rate, which is nice. But
2502 : : * that's undesirable in laptop mode, where we *want* lumpy
2503 : : * writeout. So in laptop mode, write out the whole world.
2504 : : */
2505 : 76081 : writeback_threshold = sc->nr_to_reclaim + sc->nr_to_reclaim / 2;
2506 [ + + ]: 76081 : if (total_scanned > writeback_threshold) {
2507 [ + - ]: 21 : wakeup_flusher_threads(laptop_mode ? 0 : total_scanned,
2508 : : WB_REASON_TRY_TO_FREE_PAGES);
2509 : 21 : sc->may_writepage = 1;
2510 : : }
2511 [ + + ][ + - ]: 76081 : } while (--sc->priority >= 0 && !aborted_reclaim);
2512 : :
2513 : : out:
2514 : : delayacct_freepages_end();
2515 : :
2516 [ + ]: 5864 : if (sc->nr_reclaimed)
2517 : : return sc->nr_reclaimed;
2518 : :
2519 : : /*
2520 : : * As hibernation is going on, kswapd is freezed so that it can't mark
2521 : : * the zone into all_unreclaimable. Thus bypassing all_unreclaimable
2522 : : * check.
2523 : : */
2524 [ + - ]: 5733 : if (oom_killer_disabled)
2525 : : return 0;
2526 : :
2527 : : /* Aborted reclaim to try compaction? don't OOM, then */
2528 [ + - ]: 5733 : if (aborted_reclaim)
2529 : : return 1;
2530 : :
2531 : : /* top priority shrink_zones still had more to do? don't OOM, then */
2532 [ + + ]: 5733 : if (global_reclaim(sc) && !all_unreclaimable(zonelist, sc))
2533 : : return 1;
2534 : :
2535 : : return 0;
2536 : : }
2537 : :
2538 : 0 : static bool pfmemalloc_watermark_ok(pg_data_t *pgdat)
2539 : : {
2540 : : struct zone *zone;
2541 : : unsigned long pfmemalloc_reserve = 0;
2542 : : unsigned long free_pages = 0;
2543 : : int i;
2544 : : bool wmark_ok;
2545 : :
2546 [ + + ]: 11698 : for (i = 0; i <= ZONE_NORMAL; i++) {
2547 : 5849 : zone = &pgdat->node_zones[i];
2548 : 5849 : pfmemalloc_reserve += min_wmark_pages(zone);
2549 : 5849 : free_pages += zone_page_state(zone, NR_FREE_PAGES);
2550 : : }
2551 : :
2552 : 5849 : wmark_ok = free_pages > pfmemalloc_reserve / 2;
2553 : :
2554 : : /* kswapd must be awake if processes are being throttled */
2555 [ - + ][ # # ]: 5849 : if (!wmark_ok && waitqueue_active(&pgdat->kswapd_wait)) {
2556 : 0 : pgdat->classzone_idx = min(pgdat->classzone_idx,
2557 : : (enum zone_type)ZONE_NORMAL);
2558 : 0 : wake_up_interruptible(&pgdat->kswapd_wait);
2559 : : }
2560 : :
2561 : 0 : return wmark_ok;
2562 : : }
2563 : :
2564 : : /*
2565 : : * Throttle direct reclaimers if backing storage is backed by the network
2566 : : * and the PFMEMALLOC reserve for the preferred node is getting dangerously
2567 : : * depleted. kswapd will continue to make progress and wake the processes
2568 : : * when the low watermark is reached.
2569 : : *
2570 : : * Returns true if a fatal signal was delivered during throttling. If this
2571 : : * happens, the page allocator should not consider triggering the OOM killer.
2572 : : */
2573 : 0 : static bool throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist,
2574 : : nodemask_t *nodemask)
2575 : : {
2576 : : struct zone *zone;
2577 : : int high_zoneidx = gfp_zone(gfp_mask);
2578 : : pg_data_t *pgdat;
2579 : :
2580 : : /*
2581 : : * Kernel threads should not be throttled as they may be indirectly
2582 : : * responsible for cleaning pages necessary for reclaim to make forward
2583 : : * progress. kjournald for example may enter direct reclaim while
2584 : : * committing a transaction where throttling it could forcing other
2585 : : * processes to block on log_wait_commit().
2586 : : */
2587 [ + - ]: 5856 : if (current->flags & PF_KTHREAD)
2588 : : goto out;
2589 : :
2590 : : /*
2591 : : * If a fatal signal is pending, this process should not throttle.
2592 : : * It should return quickly so it can exit and free its memory
2593 : : */
2594 [ + + ]: 5856 : if (fatal_signal_pending(current))
2595 : : goto out;
2596 : :
2597 : : /* Check if the pfmemalloc reserves are ok */
2598 : : first_zones_zonelist(zonelist, high_zoneidx, NULL, &zone);
2599 : 5847 : pgdat = zone->zone_pgdat;
2600 [ - + ]: 5847 : if (pfmemalloc_watermark_ok(pgdat))
2601 : : goto out;
2602 : :
2603 : : /* Account for the throttling */
2604 : : count_vm_event(PGSCAN_DIRECT_THROTTLE);
2605 : :
2606 : : /*
2607 : : * If the caller cannot enter the filesystem, it's possible that it
2608 : : * is due to the caller holding an FS lock or performing a journal
2609 : : * transaction in the case of a filesystem like ext[3|4]. In this case,
2610 : : * it is not safe to block on pfmemalloc_wait as kswapd could be
2611 : : * blocked waiting on the same lock. Instead, throttle for up to a
2612 : : * second before continuing.
2613 : : */
2614 [ # # ]: 0 : if (!(gfp_mask & __GFP_FS)) {
2615 [ # # ][ # # ]: 0 : wait_event_interruptible_timeout(pgdat->pfmemalloc_wait,
[ # # ][ # # ]
2616 : : pfmemalloc_watermark_ok(pgdat), HZ);
2617 : :
2618 : : goto check_pending;
2619 : : }
2620 : :
2621 : : /* Throttle until kswapd wakes the process */
2622 [ # # ][ # # ]: 0 : wait_event_killable(zone->zone_pgdat->pfmemalloc_wait,
[ # # ]
2623 : : pfmemalloc_watermark_ok(pgdat));
2624 : :
2625 : : check_pending:
2626 [ # # ]: 0 : if (fatal_signal_pending(current))
2627 : : return true;
2628 : :
2629 : : out:
2630 : : return false;
2631 : : }
2632 : :
2633 : 0 : unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
2634 : : gfp_t gfp_mask, nodemask_t *nodemask)
2635 : : {
2636 : : unsigned long nr_reclaimed;
2637 : 17547 : struct scan_control sc = {
2638 : : .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)),
2639 : 5849 : .may_writepage = !laptop_mode,
2640 : : .nr_to_reclaim = SWAP_CLUSTER_MAX,
2641 : : .may_unmap = 1,
2642 : : .may_swap = 1,
2643 : : .order = order,
2644 : : .priority = DEF_PRIORITY,
2645 : : .target_mem_cgroup = NULL,
2646 : : .nodemask = nodemask,
2647 : : };
2648 : 5849 : struct shrink_control shrink = {
2649 : : .gfp_mask = sc.gfp_mask,
2650 : : };
2651 : :
2652 : : /*
2653 : : * Do not enter reclaim if fatal signal was delivered while throttled.
2654 : : * 1 is returned so that the page allocator does not OOM kill at this
2655 : : * point.
2656 : : */
2657 [ + + ]: 5849 : if (throttle_direct_reclaim(gfp_mask, zonelist, nodemask))
2658 : : return 1;
2659 : :
2660 : 5850 : trace_mm_vmscan_direct_reclaim_begin(order,
2661 : : sc.may_writepage,
2662 : : gfp_mask);
2663 : :
2664 : 5850 : nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink);
2665 : :
2666 : : trace_mm_vmscan_direct_reclaim_end(nr_reclaimed);
2667 : :
2668 : 5855 : return nr_reclaimed;
2669 : : }
2670 : :
2671 : : #ifdef CONFIG_MEMCG
2672 : :
2673 : : unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *memcg,
2674 : : gfp_t gfp_mask, bool noswap,
2675 : : struct zone *zone,
2676 : : unsigned long *nr_scanned)
2677 : : {
2678 : : struct scan_control sc = {
2679 : : .nr_scanned = 0,
2680 : : .nr_to_reclaim = SWAP_CLUSTER_MAX,
2681 : : .may_writepage = !laptop_mode,
2682 : : .may_unmap = 1,
2683 : : .may_swap = !noswap,
2684 : : .order = 0,
2685 : : .priority = 0,
2686 : : .target_mem_cgroup = memcg,
2687 : : };
2688 : : struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg);
2689 : :
2690 : : sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
2691 : : (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK);
2692 : :
2693 : : trace_mm_vmscan_memcg_softlimit_reclaim_begin(sc.order,
2694 : : sc.may_writepage,
2695 : : sc.gfp_mask);
2696 : :
2697 : : /*
2698 : : * NOTE: Although we can get the priority field, using it
2699 : : * here is not a good idea, since it limits the pages we can scan.
2700 : : * if we don't reclaim here, the shrink_zone from balance_pgdat
2701 : : * will pick up pages from other mem cgroup's as well. We hack
2702 : : * the priority and make it zero.
2703 : : */
2704 : : shrink_lruvec(lruvec, &sc);
2705 : :
2706 : : trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed);
2707 : :
2708 : : *nr_scanned = sc.nr_scanned;
2709 : : return sc.nr_reclaimed;
2710 : : }
2711 : :
2712 : : unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg,
2713 : : gfp_t gfp_mask,
2714 : : bool noswap)
2715 : : {
2716 : : struct zonelist *zonelist;
2717 : : unsigned long nr_reclaimed;
2718 : : int nid;
2719 : : struct scan_control sc = {
2720 : : .may_writepage = !laptop_mode,
2721 : : .may_unmap = 1,
2722 : : .may_swap = !noswap,
2723 : : .nr_to_reclaim = SWAP_CLUSTER_MAX,
2724 : : .order = 0,
2725 : : .priority = DEF_PRIORITY,
2726 : : .target_mem_cgroup = memcg,
2727 : : .nodemask = NULL, /* we don't care the placement */
2728 : : .gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
2729 : : (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK),
2730 : : };
2731 : : struct shrink_control shrink = {
2732 : : .gfp_mask = sc.gfp_mask,
2733 : : };
2734 : :
2735 : : /*
2736 : : * Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't
2737 : : * take care of from where we get pages. So the node where we start the
2738 : : * scan does not need to be the current node.
2739 : : */
2740 : : nid = mem_cgroup_select_victim_node(memcg);
2741 : :
2742 : : zonelist = NODE_DATA(nid)->node_zonelists;
2743 : :
2744 : : trace_mm_vmscan_memcg_reclaim_begin(0,
2745 : : sc.may_writepage,
2746 : : sc.gfp_mask);
2747 : :
2748 : : nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink);
2749 : :
2750 : : trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed);
2751 : :
2752 : : return nr_reclaimed;
2753 : : }
2754 : : #endif
2755 : :
2756 : 0 : static void age_active_anon(struct zone *zone, struct scan_control *sc)
2757 : : {
2758 : : struct mem_cgroup *memcg;
2759 : :
2760 [ - + ]: 31014 : if (!total_swap_pages)
2761 : 0 : return;
2762 : :
2763 : : memcg = mem_cgroup_iter(NULL, NULL, NULL);
2764 : : do {
2765 : 0 : struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg);
2766 : :
2767 [ - ]: 0 : if (inactive_anon_is_low(lruvec))
2768 : 0 : shrink_active_list(SWAP_CLUSTER_MAX, lruvec,
2769 : : sc, LRU_ACTIVE_ANON);
2770 : :
2771 : : memcg = mem_cgroup_iter(NULL, memcg, NULL);
2772 : : } while (memcg);
2773 : : }
2774 : :
2775 : 0 : static bool zone_balanced(struct zone *zone, int order,
2776 : : unsigned long balance_gap, int classzone_idx)
2777 : : {
2778 [ + + ]: 211223 : if (!zone_watermark_ok_safe(zone, order, high_wmark_pages(zone) +
2779 : : balance_gap, classzone_idx, 0))
2780 : : return false;
2781 : :
2782 [ + + + - ]: 53189 : if (IS_ENABLED(CONFIG_COMPACTION) && order &&
2783 : 2788 : !compaction_suitable(zone, order))
2784 : : return false;
2785 : :
2786 : : return true;
2787 : : }
2788 : :
2789 : : /*
2790 : : * pgdat_balanced() is used when checking if a node is balanced.
2791 : : *
2792 : : * For order-0, all zones must be balanced!
2793 : : *
2794 : : * For high-order allocations only zones that meet watermarks and are in a
2795 : : * zone allowed by the callers classzone_idx are added to balanced_pages. The
2796 : : * total of balanced pages must be at least 25% of the zones allowed by
2797 : : * classzone_idx for the node to be considered balanced. Forcing all zones to
2798 : : * be balanced for high orders can cause excessive reclaim when there are
2799 : : * imbalanced zones.
2800 : : * The choice of 25% is due to
2801 : : * o a 16M DMA zone that is balanced will not balance a zone on any
2802 : : * reasonable sized machine
2803 : : * o On all other machines, the top zone must be at least a reasonable
2804 : : * percentage of the middle zones. For example, on 32-bit x86, highmem
2805 : : * would need to be at least 256M for it to be balance a whole node.
2806 : : * Similarly, on x86-64 the Normal zone would need to be at least 1G
2807 : : * to balance a node on its own. These seemed like reasonable ratios.
2808 : : */
2809 : 0 : static bool pgdat_balanced(pg_data_t *pgdat, int order, int classzone_idx)
2810 : : {
2811 : : unsigned long managed_pages = 0;
2812 : : unsigned long balanced_pages = 0;
2813 : : int i;
2814 : :
2815 : : /* Check the watermark levels */
2816 [ + + ]: 71555 : for (i = 0; i <= classzone_idx; i++) {
2817 : 69251 : struct zone *zone = pgdat->node_zones + i;
2818 : :
2819 [ - + ]: 69251 : if (!populated_zone(zone))
2820 : 0 : continue;
2821 : :
2822 : 69251 : managed_pages += zone->managed_pages;
2823 : :
2824 : : /*
2825 : : * A special case here:
2826 : : *
2827 : : * balance_pgdat() skips over all_unreclaimable after
2828 : : * DEF_PRIORITY. Effectively, it considers them balanced so
2829 : : * they must be considered balanced here as well!
2830 : : */
2831 [ + + ]: 69251 : if (!zone_reclaimable(zone)) {
2832 : 14906 : balanced_pages += zone->managed_pages;
2833 : 14906 : continue;
2834 : : }
2835 : :
2836 [ + + ]: 54345 : if (zone_balanced(zone, order, 0, i))
2837 : 22022 : balanced_pages += zone->managed_pages;
2838 [ + + ]: 32323 : else if (!order)
2839 : : return false;
2840 : : }
2841 : :
2842 [ + + ]: 2304 : if (order)
2843 : 2 : return balanced_pages >= (managed_pages >> 2);
2844 : : else
2845 : : return true;
2846 : : }
2847 : :
2848 : : /*
2849 : : * Prepare kswapd for sleeping. This verifies that there are no processes
2850 : : * waiting in throttle_direct_reclaim() and that watermarks have been met.
2851 : : *
2852 : : * Returns true if kswapd is ready to sleep
2853 : : */
2854 : 0 : static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, long remaining,
2855 : : int classzone_idx)
2856 : : {
2857 : : /* If a direct reclaimer woke kswapd within HZ/10, it's premature */
2858 [ + + ]: 7212 : if (remaining)
2859 : : return false;
2860 : :
2861 : : /*
2862 : : * There is a potential race between when kswapd checks its watermarks
2863 : : * and a process gets throttled. There is also a potential race if
2864 : : * processes get throttled, kswapd wakes, a large process exits therby
2865 : : * balancing the zones that causes kswapd to miss a wakeup. If kswapd
2866 : : * is going to sleep, no process should be sleeping on pfmemalloc_wait
2867 : : * so wake them now if necessary. If necessary, processes will wake
2868 : : * kswapd and get throttled again
2869 : : */
2870 [ - + ]: 6073 : if (waitqueue_active(&pgdat->pfmemalloc_wait)) {
2871 : 0 : wake_up(&pgdat->pfmemalloc_wait);
2872 : 0 : return false;
2873 : : }
2874 : :
2875 : 6073 : return pgdat_balanced(pgdat, order, classzone_idx);
2876 : : }
2877 : :
2878 : : /*
2879 : : * kswapd shrinks the zone by the number of pages required to reach
2880 : : * the high watermark.
2881 : : *
2882 : : * Returns true if kswapd scanned at least the requested number of pages to
2883 : : * reclaim or if the lack of progress was due to pages under writeback.
2884 : : * This is used to determine if the scanning priority needs to be raised.
2885 : : */
2886 : 0 : static bool kswapd_shrink_zone(struct zone *zone,
2887 : : int classzone_idx,
2888 : : struct scan_control *sc,
2889 : : unsigned long lru_pages,
2890 : : unsigned long *nr_attempted)
2891 : : {
2892 : 52574 : int testorder = sc->order;
2893 : : unsigned long balance_gap;
2894 : 52574 : struct reclaim_state *reclaim_state = current->reclaim_state;
2895 : 105148 : struct shrink_control shrink = {
2896 : 52574 : .gfp_mask = sc->gfp_mask,
2897 : : };
2898 : : bool lowmem_pressure;
2899 : :
2900 : : /* Reclaim above the high watermark. */
2901 : 52574 : sc->nr_to_reclaim = max(SWAP_CLUSTER_MAX, high_wmark_pages(zone));
2902 : :
2903 : : /*
2904 : : * Kswapd reclaims only single pages with compaction enabled. Trying
2905 : : * too hard to reclaim until contiguous free pages have become
2906 : : * available can hurt performance by evicting too much useful data
2907 : : * from memory. Do not reclaim more than needed for compaction.
2908 : : */
2909 [ + + + - ]: 52576 : if (IS_ENABLED(CONFIG_COMPACTION) && sc->order &&
2910 : 2 : compaction_suitable(zone, sc->order) !=
2911 : : COMPACT_SKIPPED)
2912 : : testorder = 0;
2913 : :
2914 : : /*
2915 : : * We put equal pressure on every zone, unless one zone has way too
2916 : : * many pages free already. The "too many pages" is defined as the
2917 : : * high wmark plus a "gap" where the gap is either the low
2918 : : * watermark or 1% of the zone, whichever is smaller.
2919 : : */
2920 : 52574 : balance_gap = min(low_wmark_pages(zone),
2921 : : (zone->managed_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) /
2922 : : KSWAPD_ZONE_BALANCE_GAP_RATIO);
2923 : :
2924 : : /*
2925 : : * If there is no low memory pressure or the zone is balanced then no
2926 : : * reclaim is necessary
2927 : : */
2928 [ - + ][ # # ]: 52574 : lowmem_pressure = (buffer_heads_over_limit && is_highmem(zone));
2929 [ + - ][ + + ]: 52574 : if (!lowmem_pressure && zone_balanced(zone, testorder,
2930 : : balance_gap, classzone_idx))
2931 : : return true;
2932 : :
2933 : 49684 : shrink_zone(zone, sc);
2934 : : nodes_clear(shrink.nodes_to_scan);
2935 : : node_set(zone_to_nid(zone), shrink.nodes_to_scan);
2936 : :
2937 : 49684 : reclaim_state->reclaimed_slab = 0;
2938 : 49684 : shrink_slab(&shrink, sc->nr_scanned, lru_pages);
2939 : 49684 : sc->nr_reclaimed += reclaim_state->reclaimed_slab;
2940 : :
2941 : : /* Account for the number of pages attempted to reclaim */
2942 : 49684 : *nr_attempted += sc->nr_to_reclaim;
2943 : :
2944 : : zone_clear_flag(zone, ZONE_WRITEBACK);
2945 : :
2946 : : /*
2947 : : * If a zone reaches its high watermark, consider it to be no longer
2948 : : * congested. It's possible there are dirty pages backed by congested
2949 : : * BDIs but as pressure is relieved, speculatively avoid congestion
2950 : : * waits.
2951 : : */
2952 [ + + + + ]: 96641 : if (zone_reclaimable(zone) &&
2953 : 46957 : zone_balanced(zone, testorder, 0, classzone_idx)) {
2954 : : zone_clear_flag(zone, ZONE_CONGESTED);
2955 : : zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY);
2956 : : }
2957 : :
2958 : 49684 : return sc->nr_scanned >= sc->nr_to_reclaim;
2959 : : }
2960 : :
2961 : : /*
2962 : : * For kswapd, balance_pgdat() will work across all this node's zones until
2963 : : * they are all at high_wmark_pages(zone).
2964 : : *
2965 : : * Returns the final order kswapd was reclaiming at
2966 : : *
2967 : : * There is special handling here for zones which are full of pinned pages.
2968 : : * This can happen if the pages are all mlocked, or if they are all used by
2969 : : * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb.
2970 : : * What we do is to detect the case where all pages in the zone have been
2971 : : * scanned twice and there has been zero successful reclaim. Mark the zone as
2972 : : * dead and from now on, only perform a short scan. Basically we're polling
2973 : : * the zone for when the problem goes away.
2974 : : *
2975 : : * kswapd scans the zones in the highmem->normal->dma direction. It skips
2976 : : * zones which have free_pages > high_wmark_pages(zone), but once a zone is
2977 : : * found to have free_pages <= high_wmark_pages(zone), we scan that zone and the
2978 : : * lower zones regardless of the number of free pages in the lower zones. This
2979 : : * interoperates with the page allocator fallback scheme to ensure that aging
2980 : : * of pages is balanced across the zones.
2981 : : */
2982 : 0 : static unsigned long balance_pgdat(pg_data_t *pgdat, int order,
2983 : : int *classzone_idx)
2984 : : {
2985 : : int i;
2986 : : int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */
2987 : : unsigned long nr_soft_reclaimed;
2988 : : unsigned long nr_soft_scanned;
2989 : 7212 : struct scan_control sc = {
2990 : : .gfp_mask = GFP_KERNEL,
2991 : : .priority = DEF_PRIORITY,
2992 : : .may_unmap = 1,
2993 : : .may_swap = 1,
2994 : 3606 : .may_writepage = !laptop_mode,
2995 : : .order = order,
2996 : : .target_mem_cgroup = NULL,
2997 : : };
2998 : : count_vm_event(PAGEOUTRUN);
2999 : :
3000 : : do {
3001 : : unsigned long lru_pages = 0;
3002 : 31014 : unsigned long nr_attempted = 0;
3003 : : bool raise_priority = true;
3004 : 31014 : bool pgdat_needs_compaction = (order > 0);
3005 : :
3006 : 31014 : sc.nr_reclaimed = 0;
3007 : :
3008 : : /*
3009 : : * Scan in the highmem->dma direction for the highest
3010 : : * zone which needs scanning
3011 : : */
3012 [ + - ]: 31014 : for (i = pgdat->nr_zones - 1; i >= 0; i--) {
3013 : 31014 : struct zone *zone = pgdat->node_zones + i;
3014 : :
3015 [ - + ]: 31014 : if (!populated_zone(zone))
3016 : 0 : continue;
3017 : :
3018 [ + + - + ]: 58419 : if (sc.priority != DEF_PRIORITY &&
3019 : : !zone_reclaimable(zone))
3020 : 0 : continue;
3021 : :
3022 : : /*
3023 : : * Do some background aging of the anon list, to give
3024 : : * pages a chance to be referenced before reclaiming.
3025 : : */
3026 : 31014 : age_active_anon(zone, &sc);
3027 : :
3028 : : /*
3029 : : * If the number of buffer_heads in the machine
3030 : : * exceeds the maximum allowed level and this node
3031 : : * has a highmem zone, force kswapd to reclaim from
3032 : : * it to relieve lowmem pressure.
3033 : : */
3034 [ - + ][ # # ]: 31014 : if (buffer_heads_over_limit && is_highmem_idx(i)) {
3035 : : end_zone = i;
3036 : : break;
3037 : : }
3038 : :
3039 [ - + ]: 31014 : if (!zone_balanced(zone, order, 0, 0)) {
3040 : : end_zone = i;
3041 : : break;
3042 : : } else {
3043 : : /*
3044 : : * If balanced, clear the dirty and congested
3045 : : * flags
3046 : : */
3047 : : zone_clear_flag(zone, ZONE_CONGESTED);
3048 : : zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY);
3049 : : }
3050 : : }
3051 : :
3052 [ + - ]: 31014 : if (i < 0)
3053 : : goto out;
3054 : :
3055 [ + + ]: 93042 : for (i = 0; i <= end_zone; i++) {
3056 : 62028 : struct zone *zone = pgdat->node_zones + i;
3057 : :
3058 [ - + ]: 62028 : if (!populated_zone(zone))
3059 : 0 : continue;
3060 : :
3061 : 62028 : lru_pages += zone_reclaimable_pages(zone);
3062 : :
3063 : : /*
3064 : : * If any zone is currently balanced then kswapd will
3065 : : * not call compaction as it is expected that the
3066 : : * necessary pages are already available.
3067 : : */
3068 [ + + + - ]: 62029 : if (pgdat_needs_compaction &&
3069 : 1 : zone_watermark_ok(zone, order,
3070 : : low_wmark_pages(zone),
3071 : : *classzone_idx, 0))
3072 : : pgdat_needs_compaction = false;
3073 : : }
3074 : :
3075 : : /*
3076 : : * If we're getting trouble reclaiming, start doing writepage
3077 : : * even in laptop mode.
3078 : : */
3079 [ + + ]: 31014 : if (sc.priority < DEF_PRIORITY - 2)
3080 : 31014 : sc.may_writepage = 1;
3081 : :
3082 : : /*
3083 : : * Now scan the zone in the dma->highmem direction, stopping
3084 : : * at the last zone which needs scanning.
3085 : : *
3086 : : * We do this because the page allocator works in the opposite
3087 : : * direction. This prevents the page allocator from allocating
3088 : : * pages behind kswapd's direction of progress, which would
3089 : : * cause too much scanning of the lower zones.
3090 : : */
3091 [ + + ]: 93042 : for (i = 0; i <= end_zone; i++) {
3092 : 62028 : struct zone *zone = pgdat->node_zones + i;
3093 : :
3094 [ - + ]: 62028 : if (!populated_zone(zone))
3095 : 0 : continue;
3096 : :
3097 [ + + + + ]: 116838 : if (sc.priority != DEF_PRIORITY &&
3098 : : !zone_reclaimable(zone))
3099 : 9454 : continue;
3100 : :
3101 : 52574 : sc.nr_scanned = 0;
3102 : :
3103 : : nr_soft_scanned = 0;
3104 : : /*
3105 : : * Call soft limit reclaim before calling shrink_zone.
3106 : : */
3107 : : nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone,
3108 : : order, sc.gfp_mask,
3109 : : &nr_soft_scanned);
3110 : : sc.nr_reclaimed += nr_soft_reclaimed;
3111 : :
3112 : : /*
3113 : : * There should be no need to raise the scanning
3114 : : * priority if enough pages are already being scanned
3115 : : * that that high watermark would be met at 100%
3116 : : * efficiency.
3117 : : */
3118 [ + + ]: 52574 : if (kswapd_shrink_zone(zone, end_zone, &sc,
3119 : : lru_pages, &nr_attempted))
3120 : : raise_priority = false;
3121 : : }
3122 : :
3123 : : /*
3124 : : * If the low watermark is met there is no need for processes
3125 : : * to be throttled on pfmemalloc_wait as they should not be
3126 : : * able to safely make forward progress. Wake them
3127 : : */
3128 [ - + # # ]: 31014 : if (waitqueue_active(&pgdat->pfmemalloc_wait) &&
3129 : 0 : pfmemalloc_watermark_ok(pgdat))
3130 : 0 : wake_up(&pgdat->pfmemalloc_wait);
3131 : :
3132 : : /*
3133 : : * Fragmentation may mean that the system cannot be rebalanced
3134 : : * for high-order allocations in all zones. If twice the
3135 : : * allocation size has been reclaimed and the zones are still
3136 : : * not balanced then recheck the watermarks at order-0 to
3137 : : * prevent kswapd reclaiming excessively. Assume that a
3138 : : * process requested a high-order can direct reclaim/compact.
3139 : : */
3140 [ + + ][ - + ]: 31014 : if (order && sc.nr_reclaimed >= 2UL << order)
3141 : 0 : order = sc.order = 0;
3142 : :
3143 : : /* Check if kswapd should be suspending */
3144 [ + - ][ + - ]: 31014 : if (try_to_freeze() || kthread_should_stop())
3145 : : break;
3146 : :
3147 : : /*
3148 : : * Compact if necessary and kswapd is reclaiming at least the
3149 : : * high watermark number of pages as requsted
3150 : : */
3151 [ - + ][ # # ]: 31014 : if (pgdat_needs_compaction && sc.nr_reclaimed > nr_attempted)
3152 : 0 : compact_pgdat(pgdat, order);
3153 : :
3154 : : /*
3155 : : * Raise priority if scanning rate is too low or there was no
3156 : : * progress in reclaiming pages
3157 : : */
3158 [ + + ][ + + ]: 31014 : if (raise_priority || !sc.nr_reclaimed)
3159 : 30854 : sc.priority--;
3160 [ + + ]: 28553 : } while (sc.priority >= 1 &&
3161 [ + + ]: 31014 : !pgdat_balanced(pgdat, order, *classzone_idx));
3162 : :
3163 : : out:
3164 : : /*
3165 : : * Return the order we were reclaiming at so prepare_kswapd_sleep()
3166 : : * makes a decision on the order we were last reclaiming at. However,
3167 : : * if another caller entered the allocator slow path while kswapd
3168 : : * was awake, order will remain at the higher level
3169 : : */
3170 : 3606 : *classzone_idx = end_zone;
3171 : 3606 : return order;
3172 : : }
3173 : :
3174 : 0 : static void kswapd_try_to_sleep(pg_data_t *pgdat, int order, int classzone_idx)
3175 : : {
3176 : : long remaining = 0;
3177 : 7212 : DEFINE_WAIT(wait);
3178 : :
3179 [ + - ][ + - ]: 3606 : if (freezing(current) || kthread_should_stop())
3180 : 0 : return;
3181 : :
3182 : 3606 : prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE);
3183 : :
3184 : : /* Try to sleep for a short interval */
3185 [ + + ]: 3606 : if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) {
3186 : 1149 : remaining = schedule_timeout(HZ/10);
3187 : 1149 : finish_wait(&pgdat->kswapd_wait, &wait);
3188 : 1149 : prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE);
3189 : : }
3190 : :
3191 : : /*
3192 : : * After a short sleep, check if it was a premature sleep. If not, then
3193 : : * go fully to sleep until explicitly woken up.
3194 : : */
3195 [ + + ]: 3606 : if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) {
3196 : 10 : trace_mm_vmscan_kswapd_sleep(pgdat->node_id);
3197 : :
3198 : : /*
3199 : : * vmstat counters are not perfectly accurate and the estimated
3200 : : * value for counters such as NR_FREE_PAGES can deviate from the
3201 : : * true value by nr_online_cpus * threshold. To avoid the zone
3202 : : * watermarks being breached while under pressure, we reduce the
3203 : : * per-cpu vmstat threshold while kswapd is awake and restore
3204 : : * them before going back to sleep.
3205 : : */
3206 : 10 : set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold);
3207 : :
3208 : : /*
3209 : : * Compaction records what page blocks it recently failed to
3210 : : * isolate pages from and skips them in the future scanning.
3211 : : * When kswapd is going to sleep, it is reasonable to assume
3212 : : * that pages and compaction may succeed so reset the cache.
3213 : : */
3214 : 10 : reset_isolation_suitable(pgdat);
3215 : :
3216 [ + - ]: 10 : if (!kthread_should_stop())
3217 : 10 : schedule();
3218 : :
3219 : 10 : set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold);
3220 : : } else {
3221 [ + + ]: 3596 : if (remaining)
3222 : : count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY);
3223 : : else
3224 : : count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY);
3225 : : }
3226 : 3606 : finish_wait(&pgdat->kswapd_wait, &wait);
3227 : : }
3228 : :
3229 : : /*
3230 : : * The background pageout daemon, started as a kernel thread
3231 : : * from the init process.
3232 : : *
3233 : : * This basically trickles out pages so that we have _some_
3234 : : * free memory available even if there is no other activity
3235 : : * that frees anything up. This is needed for things like routing
3236 : : * etc, where we otherwise might have all activity going on in
3237 : : * asynchronous contexts that cannot page things out.
3238 : : *
3239 : : * If there are applications that are active memory-allocators
3240 : : * (most normal use), this basically shouldn't matter.
3241 : : */
3242 : 0 : static int kswapd(void *p)
3243 : : {
3244 : : unsigned long order, new_order;
3245 : : unsigned balanced_order;
3246 : : int classzone_idx, new_classzone_idx;
3247 : : int balanced_classzone_idx;
3248 : : pg_data_t *pgdat = (pg_data_t*)p;
3249 : 0 : struct task_struct *tsk = current;
3250 : :
3251 : 0 : struct reclaim_state reclaim_state = {
3252 : : .reclaimed_slab = 0,
3253 : : };
3254 : 0 : const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
3255 : :
3256 : : lockdep_set_current_reclaim_state(GFP_KERNEL);
3257 : :
3258 [ # # ]: 0 : if (!cpumask_empty(cpumask))
3259 : 0 : set_cpus_allowed_ptr(tsk, cpumask);
3260 : 0 : current->reclaim_state = &reclaim_state;
3261 : :
3262 : : /*
3263 : : * Tell the memory management that we're a "memory allocator",
3264 : : * and that if we need more memory we should get access to it
3265 : : * regardless (see "__alloc_pages()"). "kswapd" should
3266 : : * never get caught in the normal page freeing logic.
3267 : : *
3268 : : * (Kswapd normally doesn't need memory anyway, but sometimes
3269 : : * you need a small amount of memory in order to be able to
3270 : : * page out something else, and this flag essentially protects
3271 : : * us from recursively trying to free more memory as we're
3272 : : * trying to free the first piece of memory in the first place).
3273 : : */
3274 : 0 : tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
3275 : 0 : set_freezable();
3276 : :
3277 : : order = new_order = 0;
3278 : : balanced_order = 0;
3279 : 0 : classzone_idx = new_classzone_idx = pgdat->nr_zones - 1;
3280 : 3606 : balanced_classzone_idx = classzone_idx;
3281 : : for ( ; ; ) {
3282 : : bool ret;
3283 : :
3284 : : /*
3285 : : * If the last balance_pgdat was unsuccessful it's unlikely a
3286 : : * new request of a similar or harder type will succeed soon
3287 : : * so consider going to sleep on the basis we reclaimed at
3288 : : */
3289 [ + - ]: 3606 : if (balanced_classzone_idx >= new_classzone_idx &&
3290 : 3606 : balanced_order == new_order) {
3291 : 3606 : new_order = pgdat->kswapd_max_order;
3292 : 3606 : new_classzone_idx = pgdat->classzone_idx;
3293 : 3606 : pgdat->kswapd_max_order = 0;
3294 : 3606 : pgdat->classzone_idx = pgdat->nr_zones - 1;
3295 : : }
3296 : :
3297 [ + - ]: 3606 : if (order < new_order || classzone_idx > new_classzone_idx) {
3298 : : /*
3299 : : * Don't sleep if someone wants a larger 'order'
3300 : : * allocation or has tigher zone constraints
3301 : : */
3302 : : order = new_order;
3303 : : classzone_idx = new_classzone_idx;
3304 : : } else {
3305 : 3606 : kswapd_try_to_sleep(pgdat, balanced_order,
3306 : : balanced_classzone_idx);
3307 : 3606 : order = pgdat->kswapd_max_order;
3308 : 3606 : classzone_idx = pgdat->classzone_idx;
3309 : : new_order = order;
3310 : : new_classzone_idx = classzone_idx;
3311 : 3606 : pgdat->kswapd_max_order = 0;
3312 : 3606 : pgdat->classzone_idx = pgdat->nr_zones - 1;
3313 : : }
3314 : :
3315 : : ret = try_to_freeze();
3316 [ + - ]: 3606 : if (kthread_should_stop())
3317 : : break;
3318 : :
3319 : : /*
3320 : : * We can speed up thawing tasks if we don't call balance_pgdat
3321 : : * after returning from the refrigerator
3322 : : */
3323 [ - + ]: 3606 : if (!ret) {
3324 : 3606 : trace_mm_vmscan_kswapd_wake(pgdat->node_id, order);
3325 : 3606 : balanced_classzone_idx = classzone_idx;
3326 : 3606 : balanced_order = balance_pgdat(pgdat, order,
3327 : : &balanced_classzone_idx);
3328 : : }
3329 : : }
3330 : :
3331 : 0 : current->reclaim_state = NULL;
3332 : 0 : return 0;
3333 : : }
3334 : :
3335 : : /*
3336 : : * A zone is low on free memory, so wake its kswapd task to service it.
3337 : : */
3338 : 0 : void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx)
3339 : : {
3340 : : pg_data_t *pgdat;
3341 : :
3342 [ + + ]: 69302 : if (!populated_zone(zone))
3343 : : return;
3344 : :
3345 : : if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
3346 : : return;
3347 : 69289 : pgdat = zone->zone_pgdat;
3348 [ + + ]: 69289 : if (pgdat->kswapd_max_order < order) {
3349 : 2 : pgdat->kswapd_max_order = order;
3350 : 2 : pgdat->classzone_idx = min(pgdat->classzone_idx, classzone_idx);
3351 : : }
3352 [ + + ]: 69289 : if (!waitqueue_active(&pgdat->kswapd_wait))
3353 : : return;
3354 [ + + ]: 26329 : if (zone_balanced(zone, order, 0, 0))
3355 : : return;
3356 : :
3357 : 1186 : trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order);
3358 : 1186 : wake_up_interruptible(&pgdat->kswapd_wait);
3359 : : }
3360 : :
3361 : : #ifdef CONFIG_HIBERNATION
3362 : : /*
3363 : : * Try to free `nr_to_reclaim' of memory, system-wide, and return the number of
3364 : : * freed pages.
3365 : : *
3366 : : * Rather than trying to age LRUs the aim is to preserve the overall
3367 : : * LRU order by reclaiming preferentially
3368 : : * inactive > active > active referenced > active mapped
3369 : : */
3370 : : unsigned long shrink_all_memory(unsigned long nr_to_reclaim)
3371 : : {
3372 : : struct reclaim_state reclaim_state;
3373 : : struct scan_control sc = {
3374 : : .gfp_mask = GFP_HIGHUSER_MOVABLE,
3375 : : .may_swap = 1,
3376 : : .may_unmap = 1,
3377 : : .may_writepage = 1,
3378 : : .nr_to_reclaim = nr_to_reclaim,
3379 : : .hibernation_mode = 1,
3380 : : .order = 0,
3381 : : .priority = DEF_PRIORITY,
3382 : : };
3383 : : struct shrink_control shrink = {
3384 : : .gfp_mask = sc.gfp_mask,
3385 : : };
3386 : : struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask);
3387 : : struct task_struct *p = current;
3388 : : unsigned long nr_reclaimed;
3389 : :
3390 : : p->flags |= PF_MEMALLOC;
3391 : : lockdep_set_current_reclaim_state(sc.gfp_mask);
3392 : : reclaim_state.reclaimed_slab = 0;
3393 : : p->reclaim_state = &reclaim_state;
3394 : :
3395 : : nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink);
3396 : :
3397 : : p->reclaim_state = NULL;
3398 : : lockdep_clear_current_reclaim_state();
3399 : : p->flags &= ~PF_MEMALLOC;
3400 : :
3401 : : return nr_reclaimed;
3402 : : }
3403 : : #endif /* CONFIG_HIBERNATION */
3404 : :
3405 : : /* It's optimal to keep kswapds on the same CPUs as their memory, but
3406 : : not required for correctness. So if the last cpu in a node goes
3407 : : away, we get changed to run anywhere: as the first one comes back,
3408 : : restore their cpu bindings. */
3409 : 0 : static int cpu_callback(struct notifier_block *nfb, unsigned long action,
3410 : : void *hcpu)
3411 : : {
3412 : : int nid;
3413 : :
3414 [ + + ]: 555 : if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) {
3415 [ + + ]: 162 : for_each_node_state(nid, N_MEMORY) {
3416 : : pg_data_t *pgdat = NODE_DATA(nid);
3417 : : const struct cpumask *mask;
3418 : :
3419 : 81 : mask = cpumask_of_node(pgdat->node_id);
3420 : :
3421 [ - + ]: 81 : if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids)
3422 : : /* One of our CPUs online: restore mask */
3423 : 162 : set_cpus_allowed_ptr(pgdat->kswapd, mask);
3424 : : }
3425 : : }
3426 : 555 : return NOTIFY_OK;
3427 : : }
3428 : :
3429 : : /*
3430 : : * This kswapd start function will be called by init and node-hot-add.
3431 : : * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added.
3432 : : */
3433 : 0 : int kswapd_run(int nid)
3434 : : {
3435 : : pg_data_t *pgdat = NODE_DATA(nid);
3436 : : int ret = 0;
3437 : :
3438 [ # # ]: 0 : if (pgdat->kswapd)
3439 : : return 0;
3440 : :
3441 [ # # ]: 0 : pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid);
3442 [ # # ]: 0 : if (IS_ERR(pgdat->kswapd)) {
3443 : : /* failure at boot is fatal */
3444 [ # # ]: 0 : BUG_ON(system_state == SYSTEM_BOOTING);
3445 : 0 : pr_err("Failed to start kswapd on node %d\n", nid);
3446 : 0 : ret = PTR_ERR(pgdat->kswapd);
3447 : 0 : pgdat->kswapd = NULL;
3448 : : }
3449 : 0 : return ret;
3450 : : }
3451 : :
3452 : : /*
3453 : : * Called by memory hotplug when all memory in a node is offlined. Caller must
3454 : : * hold lock_memory_hotplug().
3455 : : */
3456 : 0 : void kswapd_stop(int nid)
3457 : : {
3458 : 0 : struct task_struct *kswapd = NODE_DATA(nid)->kswapd;
3459 : :
3460 [ # # ]: 0 : if (kswapd) {
3461 : 0 : kthread_stop(kswapd);
3462 : 0 : NODE_DATA(nid)->kswapd = NULL;
3463 : : }
3464 : 0 : }
3465 : :
3466 : 0 : static int __init kswapd_init(void)
3467 : : {
3468 : : int nid;
3469 : :
3470 : 0 : swap_setup();
3471 [ # # ]: 0 : for_each_node_state(nid, N_MEMORY)
3472 : 0 : kswapd_run(nid);
3473 : 0 : hotcpu_notifier(cpu_callback, 0);
3474 : 0 : return 0;
3475 : : }
3476 : :
3477 : : module_init(kswapd_init)
3478 : :
3479 : : #ifdef CONFIG_NUMA
3480 : : /*
3481 : : * Zone reclaim mode
3482 : : *
3483 : : * If non-zero call zone_reclaim when the number of free pages falls below
3484 : : * the watermarks.
3485 : : */
3486 : : int zone_reclaim_mode __read_mostly;
3487 : :
3488 : : #define RECLAIM_OFF 0
3489 : : #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */
3490 : : #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */
3491 : : #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */
3492 : :
3493 : : /*
3494 : : * Priority for ZONE_RECLAIM. This determines the fraction of pages
3495 : : * of a node considered for each zone_reclaim. 4 scans 1/16th of
3496 : : * a zone.
3497 : : */
3498 : : #define ZONE_RECLAIM_PRIORITY 4
3499 : :
3500 : : /*
3501 : : * Percentage of pages in a zone that must be unmapped for zone_reclaim to
3502 : : * occur.
3503 : : */
3504 : : int sysctl_min_unmapped_ratio = 1;
3505 : :
3506 : : /*
3507 : : * If the number of slab pages in a zone grows beyond this percentage then
3508 : : * slab reclaim needs to occur.
3509 : : */
3510 : : int sysctl_min_slab_ratio = 5;
3511 : :
3512 : : static inline unsigned long zone_unmapped_file_pages(struct zone *zone)
3513 : : {
3514 : : unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED);
3515 : : unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) +
3516 : : zone_page_state(zone, NR_ACTIVE_FILE);
3517 : :
3518 : : /*
3519 : : * It's possible for there to be more file mapped pages than
3520 : : * accounted for by the pages on the file LRU lists because
3521 : : * tmpfs pages accounted for as ANON can also be FILE_MAPPED
3522 : : */
3523 : : return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0;
3524 : : }
3525 : :
3526 : : /* Work out how many page cache pages we can reclaim in this reclaim_mode */
3527 : : static long zone_pagecache_reclaimable(struct zone *zone)
3528 : : {
3529 : : long nr_pagecache_reclaimable;
3530 : : long delta = 0;
3531 : :
3532 : : /*
3533 : : * If RECLAIM_SWAP is set, then all file pages are considered
3534 : : * potentially reclaimable. Otherwise, we have to worry about
3535 : : * pages like swapcache and zone_unmapped_file_pages() provides
3536 : : * a better estimate
3537 : : */
3538 : : if (zone_reclaim_mode & RECLAIM_SWAP)
3539 : : nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES);
3540 : : else
3541 : : nr_pagecache_reclaimable = zone_unmapped_file_pages(zone);
3542 : :
3543 : : /* If we can't clean pages, remove dirty pages from consideration */
3544 : : if (!(zone_reclaim_mode & RECLAIM_WRITE))
3545 : : delta += zone_page_state(zone, NR_FILE_DIRTY);
3546 : :
3547 : : /* Watch for any possible underflows due to delta */
3548 : : if (unlikely(delta > nr_pagecache_reclaimable))
3549 : : delta = nr_pagecache_reclaimable;
3550 : :
3551 : : return nr_pagecache_reclaimable - delta;
3552 : : }
3553 : :
3554 : : /*
3555 : : * Try to free up some pages from this zone through reclaim.
3556 : : */
3557 : : static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
3558 : : {
3559 : : /* Minimum pages needed in order to stay on node */
3560 : : const unsigned long nr_pages = 1 << order;
3561 : : struct task_struct *p = current;
3562 : : struct reclaim_state reclaim_state;
3563 : : struct scan_control sc = {
3564 : : .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE),
3565 : : .may_unmap = !!(zone_reclaim_mode & RECLAIM_SWAP),
3566 : : .may_swap = 1,
3567 : : .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX),
3568 : : .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)),
3569 : : .order = order,
3570 : : .priority = ZONE_RECLAIM_PRIORITY,
3571 : : };
3572 : : struct shrink_control shrink = {
3573 : : .gfp_mask = sc.gfp_mask,
3574 : : };
3575 : : unsigned long nr_slab_pages0, nr_slab_pages1;
3576 : :
3577 : : cond_resched();
3578 : : /*
3579 : : * We need to be able to allocate from the reserves for RECLAIM_SWAP
3580 : : * and we also need to be able to write out pages for RECLAIM_WRITE
3581 : : * and RECLAIM_SWAP.
3582 : : */
3583 : : p->flags |= PF_MEMALLOC | PF_SWAPWRITE;
3584 : : lockdep_set_current_reclaim_state(gfp_mask);
3585 : : reclaim_state.reclaimed_slab = 0;
3586 : : p->reclaim_state = &reclaim_state;
3587 : :
3588 : : if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) {
3589 : : /*
3590 : : * Free memory by calling shrink zone with increasing
3591 : : * priorities until we have enough memory freed.
3592 : : */
3593 : : do {
3594 : : shrink_zone(zone, &sc);
3595 : : } while (sc.nr_reclaimed < nr_pages && --sc.priority >= 0);
3596 : : }
3597 : :
3598 : : nr_slab_pages0 = zone_page_state(zone, NR_SLAB_RECLAIMABLE);
3599 : : if (nr_slab_pages0 > zone->min_slab_pages) {
3600 : : /*
3601 : : * shrink_slab() does not currently allow us to determine how
3602 : : * many pages were freed in this zone. So we take the current
3603 : : * number of slab pages and shake the slab until it is reduced
3604 : : * by the same nr_pages that we used for reclaiming unmapped
3605 : : * pages.
3606 : : */
3607 : : nodes_clear(shrink.nodes_to_scan);
3608 : : node_set(zone_to_nid(zone), shrink.nodes_to_scan);
3609 : : for (;;) {
3610 : : unsigned long lru_pages = zone_reclaimable_pages(zone);
3611 : :
3612 : : /* No reclaimable slab or very low memory pressure */
3613 : : if (!shrink_slab(&shrink, sc.nr_scanned, lru_pages))
3614 : : break;
3615 : :
3616 : : /* Freed enough memory */
3617 : : nr_slab_pages1 = zone_page_state(zone,
3618 : : NR_SLAB_RECLAIMABLE);
3619 : : if (nr_slab_pages1 + nr_pages <= nr_slab_pages0)
3620 : : break;
3621 : : }
3622 : :
3623 : : /*
3624 : : * Update nr_reclaimed by the number of slab pages we
3625 : : * reclaimed from this zone.
3626 : : */
3627 : : nr_slab_pages1 = zone_page_state(zone, NR_SLAB_RECLAIMABLE);
3628 : : if (nr_slab_pages1 < nr_slab_pages0)
3629 : : sc.nr_reclaimed += nr_slab_pages0 - nr_slab_pages1;
3630 : : }
3631 : :
3632 : : p->reclaim_state = NULL;
3633 : : current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE);
3634 : : lockdep_clear_current_reclaim_state();
3635 : : return sc.nr_reclaimed >= nr_pages;
3636 : : }
3637 : :
3638 : : int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
3639 : : {
3640 : : int node_id;
3641 : : int ret;
3642 : :
3643 : : /*
3644 : : * Zone reclaim reclaims unmapped file backed pages and
3645 : : * slab pages if we are over the defined limits.
3646 : : *
3647 : : * A small portion of unmapped file backed pages is needed for
3648 : : * file I/O otherwise pages read by file I/O will be immediately
3649 : : * thrown out if the zone is overallocated. So we do not reclaim
3650 : : * if less than a specified percentage of the zone is used by
3651 : : * unmapped file backed pages.
3652 : : */
3653 : : if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages &&
3654 : : zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages)
3655 : : return ZONE_RECLAIM_FULL;
3656 : :
3657 : : if (!zone_reclaimable(zone))
3658 : : return ZONE_RECLAIM_FULL;
3659 : :
3660 : : /*
3661 : : * Do not scan if the allocation should not be delayed.
3662 : : */
3663 : : if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC))
3664 : : return ZONE_RECLAIM_NOSCAN;
3665 : :
3666 : : /*
3667 : : * Only run zone reclaim on the local zone or on zones that do not
3668 : : * have associated processors. This will favor the local processor
3669 : : * over remote processors and spread off node memory allocations
3670 : : * as wide as possible.
3671 : : */
3672 : : node_id = zone_to_nid(zone);
3673 : : if (node_state(node_id, N_CPU) && node_id != numa_node_id())
3674 : : return ZONE_RECLAIM_NOSCAN;
3675 : :
3676 : : if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED))
3677 : : return ZONE_RECLAIM_NOSCAN;
3678 : :
3679 : : ret = __zone_reclaim(zone, gfp_mask, order);
3680 : : zone_clear_flag(zone, ZONE_RECLAIM_LOCKED);
3681 : :
3682 : : if (!ret)
3683 : : count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED);
3684 : :
3685 : : return ret;
3686 : : }
3687 : : #endif
3688 : :
3689 : : /*
3690 : : * page_evictable - test whether a page is evictable
3691 : : * @page: the page to test
3692 : : *
3693 : : * Test whether page is evictable--i.e., should be placed on active/inactive
3694 : : * lists vs unevictable list.
3695 : : *
3696 : : * Reasons page might not be evictable:
3697 : : * (1) page's mapping marked unevictable
3698 : : * (2) page is part of an mlocked VMA
3699 : : *
3700 : : */
3701 : 0 : int page_evictable(struct page *page)
3702 : : {
3703 [ + ][ + + ]: 377696 : return !mapping_unevictable(page_mapping(page)) && !PageMlocked(page);
3704 : : }
3705 : :
3706 : : #ifdef CONFIG_SHMEM
3707 : : /**
3708 : : * check_move_unevictable_pages - check pages for evictability and move to appropriate zone lru list
3709 : : * @pages: array of pages to check
3710 : : * @nr_pages: number of pages to check
3711 : : *
3712 : : * Checks pages for evictability and moves them to the appropriate lru list.
3713 : : *
3714 : : * This function is only used for SysV IPC SHM_UNLOCK.
3715 : : */
3716 : 0 : void check_move_unevictable_pages(struct page **pages, int nr_pages)
3717 : : {
3718 : : struct lruvec *lruvec;
3719 : : struct zone *zone = NULL;
3720 : : int pgscanned = 0;
3721 : : int pgrescued = 0;
3722 : : int i;
3723 : :
3724 [ + + ]: 2 : for (i = 0; i < nr_pages; i++) {
3725 : 1 : struct page *page = pages[i];
3726 : : struct zone *pagezone;
3727 : :
3728 : 1 : pgscanned++;
3729 : 1 : pagezone = page_zone(page);
3730 [ + - ]: 1 : if (pagezone != zone) {
3731 [ - + ]: 1 : if (zone)
3732 : : spin_unlock_irq(&zone->lru_lock);
3733 : : zone = pagezone;
3734 : : spin_lock_irq(&zone->lru_lock);
3735 : : }
3736 : : lruvec = mem_cgroup_page_lruvec(page, zone);
3737 : :
3738 [ + - ][ + - ]: 2 : if (!PageLRU(page) || !PageUnevictable(page))
3739 : 1 : continue;
3740 : :
3741 [ # # ]: 0 : if (page_evictable(page)) {
3742 : : enum lru_list lru = page_lru_base_type(page);
3743 : :
3744 : : VM_BUG_ON_PAGE(PageActive(page), page);
3745 : : ClearPageUnevictable(page);
3746 : : del_page_from_lru_list(page, lruvec, LRU_UNEVICTABLE);
3747 : : add_page_to_lru_list(page, lruvec, lru);
3748 : 0 : pgrescued++;
3749 : : }
3750 : : }
3751 : :
3752 [ + - ]: 1 : if (zone) {
3753 : : __count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
3754 : : __count_vm_events(UNEVICTABLE_PGSCANNED, pgscanned);
3755 : : spin_unlock_irq(&zone->lru_lock);
3756 : : }
3757 : 1 : }
3758 : : #endif /* CONFIG_SHMEM */
3759 : :
3760 : 0 : static void warn_scan_unevictable_pages(void)
3761 : : {
3762 [ + + ]: 2 : printk_once(KERN_WARNING
3763 : : "%s: The scan_unevictable_pages sysctl/node-interface has been "
3764 : : "disabled for lack of a legitimate use case. If you have "
3765 : : "one, please send an email to linux-mm@kvack.org.\n",
3766 : : current->comm);
3767 : 0 : }
3768 : :
3769 : : /*
3770 : : * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of
3771 : : * all nodes' unevictable lists for evictable pages
3772 : : */
3773 : : unsigned long scan_unevictable_pages;
3774 : :
3775 : 0 : int scan_unevictable_handler(struct ctl_table *table, int write,
3776 : : void __user *buffer,
3777 : : size_t *length, loff_t *ppos)
3778 : : {
3779 : 2 : warn_scan_unevictable_pages();
3780 : 2 : proc_doulongvec_minmax(table, write, buffer, length, ppos);
3781 : 2 : scan_unevictable_pages = 0;
3782 : 2 : return 0;
3783 : : }
3784 : :
3785 : : #ifdef CONFIG_NUMA
3786 : : /*
3787 : : * per node 'scan_unevictable_pages' attribute. On demand re-scan of
3788 : : * a specified node's per zone unevictable lists for evictable pages.
3789 : : */
3790 : :
3791 : : static ssize_t read_scan_unevictable_node(struct device *dev,
3792 : : struct device_attribute *attr,
3793 : : char *buf)
3794 : : {
3795 : : warn_scan_unevictable_pages();
3796 : : return sprintf(buf, "0\n"); /* always zero; should fit... */
3797 : : }
3798 : :
3799 : : static ssize_t write_scan_unevictable_node(struct device *dev,
3800 : : struct device_attribute *attr,
3801 : : const char *buf, size_t count)
3802 : : {
3803 : : warn_scan_unevictable_pages();
3804 : : return 1;
3805 : : }
3806 : :
3807 : :
3808 : : static DEVICE_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR,
3809 : : read_scan_unevictable_node,
3810 : : write_scan_unevictable_node);
3811 : :
3812 : : int scan_unevictable_register_node(struct node *node)
3813 : : {
3814 : : return device_create_file(&node->dev, &dev_attr_scan_unevictable_pages);
3815 : : }
3816 : :
3817 : : void scan_unevictable_unregister_node(struct node *node)
3818 : : {
3819 : : device_remove_file(&node->dev, &dev_attr_scan_unevictable_pages);
3820 : : }
3821 : : #endif
|
