1 // SPDX-License-Identifier: GPL-2.0-only
10  * Linux VM subsystem. Fine-tuning documentation can be found in
11  * Documentation/admin-guide/sysctl/vm.rst.
32 #include <linux/backing-dev.h>
78  * This path almost never happens for VM activity - pages are normally freed
79  * in batches.  But it gets used by networking - and for compound pages.
97 		zone_stat_mod_folio(folio, NR_MLOCK, -nr_pages);  in __page_cache_release()
106 	free_unref_page(&folio->page, 0);  in __folio_put_small()
125 		free_zone_device_page(&folio->page);  in __folio_put()
134  * put_pages_list() - release a list of pages
135  * @pages: list of pages threaded on page->lru
145 			list_del(&folio->lru);  in put_pages_list()
149 			list_del(&folio->lru);  in put_pages_list()
188 		 * folio->mlock_count = !!folio_test_mlocked(folio)?  in lru_add_fn()
194 		folio->mlock_count = 0;  in lru_add_fn()
210 		struct folio *folio = fbatch->folios[i];  in folio_batch_move_lru()
224 	folios_put(fbatch->folios, folio_batch_count(fbatch));  in folio_batch_move_lru()
250  * to the tail of the inactive list.
287 		 * Hold lruvec->lru_lock is safe here, since  in lru_note_cost()
289 		 * 2) From a pre-LRU page during refault (which also holds the  in lru_note_cost()
293 		spin_lock_irq(&lruvec->lru_lock);  in lru_note_cost()
296 			lruvec->file_cost += cost;  in lru_note_cost()
298 			lruvec->anon_cost += cost;  in lru_note_cost()
313 		if (lruvec->file_cost + lruvec->anon_cost > lrusize / 4) {  in lru_note_cost()
314 			lruvec->file_cost /= 2;  in lru_note_cost()
315 			lruvec->anon_cost /= 2;  in lru_note_cost()
317 		spin_unlock_irq(&lruvec->lru_lock);  in lru_note_cost()
399 	 * a folio is marked active just after it is added to the inactive  in __lru_cache_activate_folio()
402 	for (i = folio_batch_count(fbatch) - 1; i >= 0; i--) {  in __lru_cache_activate_folio()
403 		struct folio *batch_folio = fbatch->folios[i];  in __lru_cache_activate_folio()
417 	unsigned long new_flags, old_flags = READ_ONCE(folio->flags);  in folio_inc_refs()
440 	} while (!try_cmpxchg(&folio->flags, &old_flags, new_flags));  in folio_inc_refs()
451  * inactive,unreferenced	->	inactive,referenced
452  * inactive,referenced		->	active,unreferenced
453  * active,unreferenced		->	active,referenced
455  * When a newly allocated page is not yet visible, so safe for non-atomic ops,
493  * folio_add_lru - Add a folio to an LRU list.
511 	    lru_gen_in_fault() && !(current->flags & PF_MEMALLOC))  in folio_add_lru()
523  * folio_add_lru_vma() - Add a folio to the appropate LRU list for this VMA.
534 	if (unlikely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) == VM_LOCKED))  in folio_add_lru_vma()
542  * inactive list to speed up its reclaim.  It is moved to the
545  * effective than the single-page writeout from reclaim.
550  * 1. active, mapped folio -> none
551  * 2. active, dirty/writeback folio -> inactive, head, reclaim
552  * 3. inactive, mapped folio -> none
553  * 4. inactive, dirty/writeback folio -> inactive, head, reclaim
554  * 5. inactive, clean -> inactive, tail
555  * 6. Others -> none
557  * In 4, it moves to the head of the inactive list so the folio is
559  * than the single-page writeout from reclaim.
589 		 * We move that folio to the tail of the inactive list.  in lru_deactivate_file_fn()
644  * disabled; or "cpu" is being hot-unplugged, and is already dead.
649 	struct folio_batch *fbatch = &fbatches->lru_add;  in lru_add_drain_cpu()
665 	fbatch = &fbatches->lru_deactivate_file;  in lru_add_drain_cpu()
669 	fbatch = &fbatches->lru_deactivate;  in lru_add_drain_cpu()
673 	fbatch = &fbatches->lru_lazyfree;  in lru_add_drain_cpu()
681  * deactivate_file_folio() - Deactivate a file folio.
706  * folio_deactivate - deactivate a folio
709  * folio_deactivate() moves @folio to the inactive list if @folio was on the
728  * folio_mark_lazyfree - make an anon folio lazyfree
731  * folio_mark_lazyfree() moves @folio to the inactive file list.
758  * It's called from per-cpu workqueue context in SMP case so
795 	return folio_batch_count(&fbatches->lru_add) ||  in cpu_needs_drain()
797 		folio_batch_count(&fbatches->lru_deactivate_file) ||  in cpu_needs_drain()
798 		folio_batch_count(&fbatches->lru_deactivate) ||  in cpu_needs_drain()
799 		folio_batch_count(&fbatches->lru_lazyfree) ||  in cpu_needs_drain()
800 		folio_batch_count(&fbatches->activate) ||  in cpu_needs_drain()
806  * Doesn't need any cpu hotplug locking because we do rely on per-cpu
815 	 * lru_drain_gen - Global pages generation number  in __lru_add_drain_all()
820 	 * This is an optimization for the highly-contended use case where a  in __lru_add_drain_all()
874 	 * below and has already reached CPU #y's per-cpu data. CPU #x comes  in __lru_add_drain_all()
875 	 * along, adds some pages to its per-cpu vectors, then calls  in __lru_add_drain_all()
949  * release_pages - batched put_page()
976 		 * Make sure the IRQ-safe lock-holding time does not get  in release_pages()
985 		if (is_huge_zero_page(&folio->page))  in release_pages()
993 			if (put_devmap_managed_page(&folio->page))  in release_pages()
996 				free_zone_device_page(&folio->page);  in release_pages()
1036 		list_add(&folio->lru, &pages_to_free);  in release_pages()
1047  * The folios which we're about to release may be in the deferred lru-addition
1049  * OK from a correctness point of view but is inefficient - those folios may be
1050  * cache-warm and we want to give them back to the page allocator ASAP.
1058 	if (!fbatch->percpu_pvec_drained) {  in __folio_batch_release()
1060 		fbatch->percpu_pvec_drained = true;  in __folio_batch_release()
1062 	release_pages(fbatch->folios, folio_batch_count(fbatch));  in __folio_batch_release()
1068  * folio_batch_remove_exceptionals() - Prune non-folios from a batch.
1072  * entries.  This function prunes all the non-folio entries from @fbatch
1073  * without leaving holes, so that it can be passed on to folio-only batch
1081 		struct folio *folio = fbatch->folios[i];  in folio_batch_remove_exceptionals()
1083 			fbatch->folios[j++] = folio;  in folio_batch_remove_exceptionals()
1085 	fbatch->nr = j;  in folio_batch_remove_exceptionals()
1093 	unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT);  in swap_setup()
1095 	/* Use a smaller cluster for small-memory machines */  in swap_setup()