1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
20 #include <linux/res_counter.h>
21 #include <linux/memcontrol.h>
22 #include <linux/cgroup.h>
24 #include <linux/smp.h>
25 #include <linux/page-flags.h>
26 #include <linux/backing-dev.h>
27 #include <linux/bit_spinlock.h>
28 #include <linux/rcupdate.h>
29 #include <linux/slab.h>
30 #include <linux/swap.h>
31 #include <linux/spinlock.h>
33 #include <linux/seq_file.h>
34 #include <linux/vmalloc.h>
35 #include <linux/mm_inline.h>
36 #include <linux/page_cgroup.h>
38 #include <asm/uaccess.h>
40 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
41 #define MEM_CGROUP_RECLAIM_RETRIES 5
44 * Statistics for memory cgroup.
46 enum mem_cgroup_stat_index {
48 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
50 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
51 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
52 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
53 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
55 MEM_CGROUP_STAT_NSTATS,
58 struct mem_cgroup_stat_cpu {
59 s64 count[MEM_CGROUP_STAT_NSTATS];
60 } ____cacheline_aligned_in_smp;
62 struct mem_cgroup_stat {
63 struct mem_cgroup_stat_cpu cpustat[NR_CPUS];
67 * For accounting under irq disable, no need for increment preempt count.
69 static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
70 enum mem_cgroup_stat_index idx, int val)
72 stat->count[idx] += val;
75 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
76 enum mem_cgroup_stat_index idx)
80 for_each_possible_cpu(cpu)
81 ret += stat->cpustat[cpu].count[idx];
86 * per-zone information in memory controller.
88 struct mem_cgroup_per_zone {
90 * spin_lock to protect the per cgroup LRU
93 struct list_head lists[NR_LRU_LISTS];
94 unsigned long count[NR_LRU_LISTS];
96 /* Macro for accessing counter */
97 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
99 struct mem_cgroup_per_node {
100 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
103 struct mem_cgroup_lru_info {
104 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
108 * The memory controller data structure. The memory controller controls both
109 * page cache and RSS per cgroup. We would eventually like to provide
110 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
111 * to help the administrator determine what knobs to tune.
113 * TODO: Add a water mark for the memory controller. Reclaim will begin when
114 * we hit the water mark. May be even add a low water mark, such that
115 * no reclaim occurs from a cgroup at it's low water mark, this is
116 * a feature that will be implemented much later in the future.
119 struct cgroup_subsys_state css;
121 * the counter to account for memory usage
123 struct res_counter res;
125 * Per cgroup active and inactive list, similar to the
126 * per zone LRU lists.
128 struct mem_cgroup_lru_info info;
130 int prev_priority; /* for recording reclaim priority */
134 struct mem_cgroup_stat stat;
136 static struct mem_cgroup init_mem_cgroup;
139 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
140 MEM_CGROUP_CHARGE_TYPE_MAPPED,
141 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
142 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
146 /* only for here (for easy reading.) */
147 #define PCGF_CACHE (1UL << PCG_CACHE)
148 #define PCGF_USED (1UL << PCG_USED)
149 #define PCGF_ACTIVE (1UL << PCG_ACTIVE)
150 #define PCGF_LOCK (1UL << PCG_LOCK)
151 #define PCGF_FILE (1UL << PCG_FILE)
152 static const unsigned long
153 pcg_default_flags[NR_CHARGE_TYPE] = {
154 PCGF_CACHE | PCGF_FILE | PCGF_USED | PCGF_LOCK, /* File Cache */
155 PCGF_ACTIVE | PCGF_USED | PCGF_LOCK, /* Anon */
156 PCGF_ACTIVE | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
161 * Always modified under lru lock. Then, not necessary to preempt_disable()
163 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
164 struct page_cgroup *pc,
167 int val = (charge)? 1 : -1;
168 struct mem_cgroup_stat *stat = &mem->stat;
169 struct mem_cgroup_stat_cpu *cpustat;
171 VM_BUG_ON(!irqs_disabled());
173 cpustat = &stat->cpustat[smp_processor_id()];
174 if (PageCgroupCache(pc))
175 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
177 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
180 __mem_cgroup_stat_add_safe(cpustat,
181 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
183 __mem_cgroup_stat_add_safe(cpustat,
184 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
187 static struct mem_cgroup_per_zone *
188 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
190 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
193 static struct mem_cgroup_per_zone *
194 page_cgroup_zoneinfo(struct page_cgroup *pc)
196 struct mem_cgroup *mem = pc->mem_cgroup;
197 int nid = page_cgroup_nid(pc);
198 int zid = page_cgroup_zid(pc);
200 return mem_cgroup_zoneinfo(mem, nid, zid);
203 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
207 struct mem_cgroup_per_zone *mz;
210 for_each_online_node(nid)
211 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
212 mz = mem_cgroup_zoneinfo(mem, nid, zid);
213 total += MEM_CGROUP_ZSTAT(mz, idx);
218 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
220 return container_of(cgroup_subsys_state(cont,
221 mem_cgroup_subsys_id), struct mem_cgroup,
225 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
228 * mm_update_next_owner() may clear mm->owner to NULL
229 * if it races with swapoff, page migration, etc.
230 * So this can be called with p == NULL.
235 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
236 struct mem_cgroup, css);
239 static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz,
240 struct page_cgroup *pc)
244 if (PageCgroupUnevictable(pc))
245 lru = LRU_UNEVICTABLE;
247 if (PageCgroupActive(pc))
249 if (PageCgroupFile(pc))
253 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
255 mem_cgroup_charge_statistics(pc->mem_cgroup, pc, false);
259 static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz,
260 struct page_cgroup *pc)
264 if (PageCgroupUnevictable(pc))
265 lru = LRU_UNEVICTABLE;
267 if (PageCgroupActive(pc))
269 if (PageCgroupFile(pc))
273 MEM_CGROUP_ZSTAT(mz, lru) += 1;
274 list_add(&pc->lru, &mz->lists[lru]);
276 mem_cgroup_charge_statistics(pc->mem_cgroup, pc, true);
279 static void __mem_cgroup_move_lists(struct page_cgroup *pc, enum lru_list lru)
281 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
282 int active = PageCgroupActive(pc);
283 int file = PageCgroupFile(pc);
284 int unevictable = PageCgroupUnevictable(pc);
285 enum lru_list from = unevictable ? LRU_UNEVICTABLE :
286 (LRU_FILE * !!file + !!active);
291 MEM_CGROUP_ZSTAT(mz, from) -= 1;
293 * However this is done under mz->lru_lock, another flags, which
294 * are not related to LRU, will be modified from out-of-lock.
295 * We have to use atomic set/clear flags.
297 if (is_unevictable_lru(lru)) {
298 ClearPageCgroupActive(pc);
299 SetPageCgroupUnevictable(pc);
301 if (is_active_lru(lru))
302 SetPageCgroupActive(pc);
304 ClearPageCgroupActive(pc);
305 ClearPageCgroupUnevictable(pc);
308 MEM_CGROUP_ZSTAT(mz, lru) += 1;
309 list_move(&pc->lru, &mz->lists[lru]);
312 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
317 ret = task->mm && mm_match_cgroup(task->mm, mem);
323 * This routine assumes that the appropriate zone's lru lock is already held
325 void mem_cgroup_move_lists(struct page *page, enum lru_list lru)
327 struct page_cgroup *pc;
328 struct mem_cgroup_per_zone *mz;
331 if (mem_cgroup_subsys.disabled)
335 * We cannot lock_page_cgroup while holding zone's lru_lock,
336 * because other holders of lock_page_cgroup can be interrupted
337 * with an attempt to rotate_reclaimable_page. But we cannot
338 * safely get to page_cgroup without it, so just try_lock it:
339 * mem_cgroup_isolate_pages allows for page left on wrong list.
341 pc = lookup_page_cgroup(page);
342 if (!trylock_page_cgroup(pc))
344 if (pc && PageCgroupUsed(pc)) {
345 mz = page_cgroup_zoneinfo(pc);
346 spin_lock_irqsave(&mz->lru_lock, flags);
347 __mem_cgroup_move_lists(pc, lru);
348 spin_unlock_irqrestore(&mz->lru_lock, flags);
350 unlock_page_cgroup(pc);
354 * Calculate mapped_ratio under memory controller. This will be used in
355 * vmscan.c for deteremining we have to reclaim mapped pages.
357 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
362 * usage is recorded in bytes. But, here, we assume the number of
363 * physical pages can be represented by "long" on any arch.
365 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
366 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
367 return (int)((rss * 100L) / total);
371 * prev_priority control...this will be used in memory reclaim path.
373 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
375 return mem->prev_priority;
378 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
380 if (priority < mem->prev_priority)
381 mem->prev_priority = priority;
384 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
386 mem->prev_priority = priority;
390 * Calculate # of pages to be scanned in this priority/zone.
393 * priority starts from "DEF_PRIORITY" and decremented in each loop.
394 * (see include/linux/mmzone.h)
397 long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone,
398 int priority, enum lru_list lru)
401 int nid = zone->zone_pgdat->node_id;
402 int zid = zone_idx(zone);
403 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
405 nr_pages = MEM_CGROUP_ZSTAT(mz, lru);
407 return (nr_pages >> priority);
410 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
411 struct list_head *dst,
412 unsigned long *scanned, int order,
413 int mode, struct zone *z,
414 struct mem_cgroup *mem_cont,
415 int active, int file)
417 unsigned long nr_taken = 0;
421 struct list_head *src;
422 struct page_cgroup *pc, *tmp;
423 int nid = z->zone_pgdat->node_id;
424 int zid = zone_idx(z);
425 struct mem_cgroup_per_zone *mz;
426 int lru = LRU_FILE * !!file + !!active;
429 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
430 src = &mz->lists[lru];
432 spin_lock(&mz->lru_lock);
434 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
435 if (scan >= nr_to_scan)
437 if (unlikely(!PageCgroupUsed(pc)))
441 if (unlikely(!PageLRU(page)))
445 * TODO: play better with lumpy reclaim, grabbing anything.
447 if (PageUnevictable(page) ||
448 (PageActive(page) && !active) ||
449 (!PageActive(page) && active)) {
450 __mem_cgroup_move_lists(pc, page_lru(page));
455 list_move(&pc->lru, &pc_list);
457 if (__isolate_lru_page(page, mode, file) == 0) {
458 list_move(&page->lru, dst);
463 list_splice(&pc_list, src);
464 spin_unlock(&mz->lru_lock);
471 * Charge the memory controller for page usage.
473 * 0 if the charge was successful
474 * < 0 if the cgroup is over its limit
476 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
477 gfp_t gfp_mask, enum charge_type ctype,
478 struct mem_cgroup *memcg)
480 struct mem_cgroup *mem;
481 struct page_cgroup *pc;
482 unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
483 struct mem_cgroup_per_zone *mz;
486 pc = lookup_page_cgroup(page);
487 /* can happen at boot */
492 * We always charge the cgroup the mm_struct belongs to.
493 * The mm_struct's mem_cgroup changes on task migration if the
494 * thread group leader migrates. It's possible that mm is not
495 * set, if so charge the init_mm (happens for pagecache usage).
498 if (likely(!memcg)) {
500 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
501 if (unlikely(!mem)) {
506 * For every charge from the cgroup, increment reference count
512 css_get(&memcg->css);
515 while (unlikely(res_counter_charge(&mem->res, PAGE_SIZE))) {
516 if (!(gfp_mask & __GFP_WAIT))
519 if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
523 * try_to_free_mem_cgroup_pages() might not give us a full
524 * picture of reclaim. Some pages are reclaimed and might be
525 * moved to swap cache or just unmapped from the cgroup.
526 * Check the limit again to see if the reclaim reduced the
527 * current usage of the cgroup before giving up
529 if (res_counter_check_under_limit(&mem->res))
533 mem_cgroup_out_of_memory(mem, gfp_mask);
539 lock_page_cgroup(pc);
540 if (unlikely(PageCgroupUsed(pc))) {
541 unlock_page_cgroup(pc);
542 res_counter_uncharge(&mem->res, PAGE_SIZE);
547 pc->mem_cgroup = mem;
549 * If a page is accounted as a page cache, insert to inactive list.
550 * If anon, insert to active list.
552 pc->flags = pcg_default_flags[ctype];
554 mz = page_cgroup_zoneinfo(pc);
556 spin_lock_irqsave(&mz->lru_lock, flags);
557 __mem_cgroup_add_list(mz, pc);
558 spin_unlock_irqrestore(&mz->lru_lock, flags);
559 unlock_page_cgroup(pc);
568 int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask)
570 if (mem_cgroup_subsys.disabled)
572 if (PageCompound(page))
575 * If already mapped, we don't have to account.
576 * If page cache, page->mapping has address_space.
577 * But page->mapping may have out-of-use anon_vma pointer,
578 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
581 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
585 return mem_cgroup_charge_common(page, mm, gfp_mask,
586 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
589 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
592 if (mem_cgroup_subsys.disabled)
594 if (PageCompound(page))
597 * Corner case handling. This is called from add_to_page_cache()
598 * in usual. But some FS (shmem) precharges this page before calling it
599 * and call add_to_page_cache() with GFP_NOWAIT.
601 * For GFP_NOWAIT case, the page may be pre-charged before calling
602 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
603 * charge twice. (It works but has to pay a bit larger cost.)
605 if (!(gfp_mask & __GFP_WAIT)) {
606 struct page_cgroup *pc;
609 pc = lookup_page_cgroup(page);
612 lock_page_cgroup(pc);
613 if (PageCgroupUsed(pc)) {
614 unlock_page_cgroup(pc);
617 unlock_page_cgroup(pc);
623 if (page_is_file_cache(page))
624 return mem_cgroup_charge_common(page, mm, gfp_mask,
625 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
627 return mem_cgroup_charge_common(page, mm, gfp_mask,
628 MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL);
632 * uncharge if !page_mapped(page)
635 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
637 struct page_cgroup *pc;
638 struct mem_cgroup *mem;
639 struct mem_cgroup_per_zone *mz;
642 if (mem_cgroup_subsys.disabled)
646 * Check if our page_cgroup is valid
648 pc = lookup_page_cgroup(page);
649 if (unlikely(!pc || !PageCgroupUsed(pc)))
652 lock_page_cgroup(pc);
653 if ((ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED && page_mapped(page))
654 || !PageCgroupUsed(pc)) {
655 /* This happens at race in zap_pte_range() and do_swap_page()*/
656 unlock_page_cgroup(pc);
659 ClearPageCgroupUsed(pc);
660 mem = pc->mem_cgroup;
662 mz = page_cgroup_zoneinfo(pc);
663 spin_lock_irqsave(&mz->lru_lock, flags);
664 __mem_cgroup_remove_list(mz, pc);
665 spin_unlock_irqrestore(&mz->lru_lock, flags);
666 unlock_page_cgroup(pc);
668 res_counter_uncharge(&mem->res, PAGE_SIZE);
674 void mem_cgroup_uncharge_page(struct page *page)
677 if (page_mapped(page))
679 if (page->mapping && !PageAnon(page))
681 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
684 void mem_cgroup_uncharge_cache_page(struct page *page)
686 VM_BUG_ON(page_mapped(page));
687 VM_BUG_ON(page->mapping);
688 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
692 * Before starting migration, account against new page.
694 int mem_cgroup_prepare_migration(struct page *page, struct page *newpage)
696 struct page_cgroup *pc;
697 struct mem_cgroup *mem = NULL;
698 enum charge_type ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
701 if (mem_cgroup_subsys.disabled)
704 pc = lookup_page_cgroup(page);
705 lock_page_cgroup(pc);
706 if (PageCgroupUsed(pc)) {
707 mem = pc->mem_cgroup;
709 if (PageCgroupCache(pc)) {
710 if (page_is_file_cache(page))
711 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
713 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
716 unlock_page_cgroup(pc);
718 ret = mem_cgroup_charge_common(newpage, NULL, GFP_KERNEL,
725 /* remove redundant charge if migration failed*/
726 void mem_cgroup_end_migration(struct page *newpage)
729 * At success, page->mapping is not NULL.
730 * special rollback care is necessary when
731 * 1. at migration failure. (newpage->mapping is cleared in this case)
732 * 2. the newpage was moved but not remapped again because the task
733 * exits and the newpage is obsolete. In this case, the new page
734 * may be a swapcache. So, we just call mem_cgroup_uncharge_page()
735 * always for avoiding mess. The page_cgroup will be removed if
736 * unnecessary. File cache pages is still on radix-tree. Don't
739 if (!newpage->mapping)
740 __mem_cgroup_uncharge_common(newpage,
741 MEM_CGROUP_CHARGE_TYPE_FORCE);
742 else if (PageAnon(newpage))
743 mem_cgroup_uncharge_page(newpage);
747 * A call to try to shrink memory usage under specified resource controller.
748 * This is typically used for page reclaiming for shmem for reducing side
749 * effect of page allocation from shmem, which is used by some mem_cgroup.
751 int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
753 struct mem_cgroup *mem;
755 int retry = MEM_CGROUP_RECLAIM_RETRIES;
757 if (mem_cgroup_subsys.disabled)
763 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
764 if (unlikely(!mem)) {
772 progress = try_to_free_mem_cgroup_pages(mem, gfp_mask);
773 progress += res_counter_check_under_limit(&mem->res);
774 } while (!progress && --retry);
782 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
783 unsigned long long val)
786 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
790 while (res_counter_set_limit(&memcg->res, val)) {
791 if (signal_pending(current)) {
799 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL);
808 * This routine traverse page_cgroup in given list and drop them all.
809 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
811 #define FORCE_UNCHARGE_BATCH (128)
812 static void mem_cgroup_force_empty_list(struct mem_cgroup *mem,
813 struct mem_cgroup_per_zone *mz,
816 struct page_cgroup *pc;
818 int count = FORCE_UNCHARGE_BATCH;
820 struct list_head *list;
822 list = &mz->lists[lru];
824 spin_lock_irqsave(&mz->lru_lock, flags);
825 while (!list_empty(list)) {
826 pc = list_entry(list->prev, struct page_cgroup, lru);
828 if (!PageCgroupUsed(pc))
831 spin_unlock_irqrestore(&mz->lru_lock, flags);
833 * Check if this page is on LRU. !LRU page can be found
834 * if it's under page migration.
837 __mem_cgroup_uncharge_common(page,
838 MEM_CGROUP_CHARGE_TYPE_FORCE);
841 count = FORCE_UNCHARGE_BATCH;
845 spin_lock_irqsave(&mz->lru_lock, flags);
848 spin_lock_irqsave(&mz->lru_lock, flags);
850 spin_unlock_irqrestore(&mz->lru_lock, flags);
854 * make mem_cgroup's charge to be 0 if there is no task.
855 * This enables deleting this mem_cgroup.
857 static int mem_cgroup_force_empty(struct mem_cgroup *mem)
864 * page reclaim code (kswapd etc..) will move pages between
865 * active_list <-> inactive_list while we don't take a lock.
866 * So, we have to do loop here until all lists are empty.
868 while (mem->res.usage > 0) {
869 if (atomic_read(&mem->css.cgroup->count) > 0)
871 /* This is for making all *used* pages to be on LRU. */
873 for_each_node_state(node, N_POSSIBLE)
874 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
875 struct mem_cgroup_per_zone *mz;
877 mz = mem_cgroup_zoneinfo(mem, node, zid);
879 mem_cgroup_force_empty_list(mem, mz, l);
889 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
891 return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res,
895 * The user of this function is...
898 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
901 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
902 unsigned long long val;
905 switch (cft->private) {
907 /* This function does all necessary parse...reuse it */
908 ret = res_counter_memparse_write_strategy(buffer, &val);
910 ret = mem_cgroup_resize_limit(memcg, val);
913 ret = -EINVAL; /* should be BUG() ? */
919 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
921 struct mem_cgroup *mem;
923 mem = mem_cgroup_from_cont(cont);
926 res_counter_reset_max(&mem->res);
929 res_counter_reset_failcnt(&mem->res);
935 static int mem_force_empty_write(struct cgroup *cont, unsigned int event)
937 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont));
940 static const struct mem_cgroup_stat_desc {
943 } mem_cgroup_stat_desc[] = {
944 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
945 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
946 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
947 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
950 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
951 struct cgroup_map_cb *cb)
953 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
954 struct mem_cgroup_stat *stat = &mem_cont->stat;
957 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
960 val = mem_cgroup_read_stat(stat, i);
961 val *= mem_cgroup_stat_desc[i].unit;
962 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
964 /* showing # of active pages */
966 unsigned long active_anon, inactive_anon;
967 unsigned long active_file, inactive_file;
968 unsigned long unevictable;
970 inactive_anon = mem_cgroup_get_all_zonestat(mem_cont,
972 active_anon = mem_cgroup_get_all_zonestat(mem_cont,
974 inactive_file = mem_cgroup_get_all_zonestat(mem_cont,
976 active_file = mem_cgroup_get_all_zonestat(mem_cont,
978 unevictable = mem_cgroup_get_all_zonestat(mem_cont,
981 cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE);
982 cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE);
983 cb->fill(cb, "active_file", (active_file) * PAGE_SIZE);
984 cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE);
985 cb->fill(cb, "unevictable", unevictable * PAGE_SIZE);
991 static struct cftype mem_cgroup_files[] = {
993 .name = "usage_in_bytes",
994 .private = RES_USAGE,
995 .read_u64 = mem_cgroup_read,
998 .name = "max_usage_in_bytes",
999 .private = RES_MAX_USAGE,
1000 .trigger = mem_cgroup_reset,
1001 .read_u64 = mem_cgroup_read,
1004 .name = "limit_in_bytes",
1005 .private = RES_LIMIT,
1006 .write_string = mem_cgroup_write,
1007 .read_u64 = mem_cgroup_read,
1011 .private = RES_FAILCNT,
1012 .trigger = mem_cgroup_reset,
1013 .read_u64 = mem_cgroup_read,
1016 .name = "force_empty",
1017 .trigger = mem_force_empty_write,
1021 .read_map = mem_control_stat_show,
1025 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1027 struct mem_cgroup_per_node *pn;
1028 struct mem_cgroup_per_zone *mz;
1030 int zone, tmp = node;
1032 * This routine is called against possible nodes.
1033 * But it's BUG to call kmalloc() against offline node.
1035 * TODO: this routine can waste much memory for nodes which will
1036 * never be onlined. It's better to use memory hotplug callback
1039 if (!node_state(node, N_NORMAL_MEMORY))
1041 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
1045 mem->info.nodeinfo[node] = pn;
1046 memset(pn, 0, sizeof(*pn));
1048 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
1049 mz = &pn->zoneinfo[zone];
1050 spin_lock_init(&mz->lru_lock);
1052 INIT_LIST_HEAD(&mz->lists[l]);
1057 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1059 kfree(mem->info.nodeinfo[node]);
1062 static struct mem_cgroup *mem_cgroup_alloc(void)
1064 struct mem_cgroup *mem;
1066 if (sizeof(*mem) < PAGE_SIZE)
1067 mem = kmalloc(sizeof(*mem), GFP_KERNEL);
1069 mem = vmalloc(sizeof(*mem));
1072 memset(mem, 0, sizeof(*mem));
1076 static void mem_cgroup_free(struct mem_cgroup *mem)
1078 if (sizeof(*mem) < PAGE_SIZE)
1085 static struct cgroup_subsys_state *
1086 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
1088 struct mem_cgroup *mem;
1091 if (unlikely((cont->parent) == NULL)) {
1092 mem = &init_mem_cgroup;
1094 mem = mem_cgroup_alloc();
1096 return ERR_PTR(-ENOMEM);
1099 res_counter_init(&mem->res);
1101 for_each_node_state(node, N_POSSIBLE)
1102 if (alloc_mem_cgroup_per_zone_info(mem, node))
1107 for_each_node_state(node, N_POSSIBLE)
1108 free_mem_cgroup_per_zone_info(mem, node);
1109 if (cont->parent != NULL)
1110 mem_cgroup_free(mem);
1111 return ERR_PTR(-ENOMEM);
1114 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1115 struct cgroup *cont)
1117 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1118 mem_cgroup_force_empty(mem);
1121 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
1122 struct cgroup *cont)
1125 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1127 for_each_node_state(node, N_POSSIBLE)
1128 free_mem_cgroup_per_zone_info(mem, node);
1130 mem_cgroup_free(mem_cgroup_from_cont(cont));
1133 static int mem_cgroup_populate(struct cgroup_subsys *ss,
1134 struct cgroup *cont)
1136 return cgroup_add_files(cont, ss, mem_cgroup_files,
1137 ARRAY_SIZE(mem_cgroup_files));
1140 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
1141 struct cgroup *cont,
1142 struct cgroup *old_cont,
1143 struct task_struct *p)
1145 struct mm_struct *mm;
1146 struct mem_cgroup *mem, *old_mem;
1148 mm = get_task_mm(p);
1152 mem = mem_cgroup_from_cont(cont);
1153 old_mem = mem_cgroup_from_cont(old_cont);
1156 * Only thread group leaders are allowed to migrate, the mm_struct is
1157 * in effect owned by the leader
1159 if (!thread_group_leader(p))
1166 struct cgroup_subsys mem_cgroup_subsys = {
1168 .subsys_id = mem_cgroup_subsys_id,
1169 .create = mem_cgroup_create,
1170 .pre_destroy = mem_cgroup_pre_destroy,
1171 .destroy = mem_cgroup_destroy,
1172 .populate = mem_cgroup_populate,
1173 .attach = mem_cgroup_move_task,