~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/vmscan.c

Version: ~ [ linux-5.14-rc3 ] ~ [ linux-5.13.5 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.53 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.135 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.198 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.240 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.276 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.276 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp