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, bool hot)
264 if (PageCgroupUnevictable(pc))
265 lru = LRU_UNEVICTABLE;
267 if (PageCgroupActive(pc))
269 if (PageCgroupFile(pc))
273 MEM_CGROUP_ZSTAT(mz, lru) += 1;
275 list_add(&pc->lru, &mz->lists[lru]);
277 list_add_tail(&pc->lru, &mz->lists[lru]);
279 mem_cgroup_charge_statistics(pc->mem_cgroup, pc, true);
282 static void __mem_cgroup_move_lists(struct page_cgroup *pc, enum lru_list lru)
284 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
285 int active = PageCgroupActive(pc);
286 int file = PageCgroupFile(pc);
287 int unevictable = PageCgroupUnevictable(pc);
288 enum lru_list from = unevictable ? LRU_UNEVICTABLE :
289 (LRU_FILE * !!file + !!active);
294 MEM_CGROUP_ZSTAT(mz, from) -= 1;
296 * However this is done under mz->lru_lock, another flags, which
297 * are not related to LRU, will be modified from out-of-lock.
298 * We have to use atomic set/clear flags.
300 if (is_unevictable_lru(lru)) {
301 ClearPageCgroupActive(pc);
302 SetPageCgroupUnevictable(pc);
304 if (is_active_lru(lru))
305 SetPageCgroupActive(pc);
307 ClearPageCgroupActive(pc);
308 ClearPageCgroupUnevictable(pc);
311 MEM_CGROUP_ZSTAT(mz, lru) += 1;
312 list_move(&pc->lru, &mz->lists[lru]);
315 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
320 ret = task->mm && mm_match_cgroup(task->mm, mem);
326 * This routine assumes that the appropriate zone's lru lock is already held
328 void mem_cgroup_move_lists(struct page *page, enum lru_list lru)
330 struct page_cgroup *pc;
331 struct mem_cgroup_per_zone *mz;
334 if (mem_cgroup_subsys.disabled)
338 * We cannot lock_page_cgroup while holding zone's lru_lock,
339 * because other holders of lock_page_cgroup can be interrupted
340 * with an attempt to rotate_reclaimable_page. But we cannot
341 * safely get to page_cgroup without it, so just try_lock it:
342 * mem_cgroup_isolate_pages allows for page left on wrong list.
344 pc = lookup_page_cgroup(page);
345 if (!trylock_page_cgroup(pc))
347 if (pc && PageCgroupUsed(pc)) {
348 mz = page_cgroup_zoneinfo(pc);
349 spin_lock_irqsave(&mz->lru_lock, flags);
350 __mem_cgroup_move_lists(pc, lru);
351 spin_unlock_irqrestore(&mz->lru_lock, flags);
353 unlock_page_cgroup(pc);
357 * Calculate mapped_ratio under memory controller. This will be used in
358 * vmscan.c for deteremining we have to reclaim mapped pages.
360 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
365 * usage is recorded in bytes. But, here, we assume the number of
366 * physical pages can be represented by "long" on any arch.
368 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
369 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
370 return (int)((rss * 100L) / total);
374 * prev_priority control...this will be used in memory reclaim path.
376 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
378 return mem->prev_priority;
381 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
383 if (priority < mem->prev_priority)
384 mem->prev_priority = priority;
387 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
389 mem->prev_priority = priority;
393 * Calculate # of pages to be scanned in this priority/zone.
396 * priority starts from "DEF_PRIORITY" and decremented in each loop.
397 * (see include/linux/mmzone.h)
400 long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone,
401 int priority, enum lru_list lru)
404 int nid = zone->zone_pgdat->node_id;
405 int zid = zone_idx(zone);
406 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
408 nr_pages = MEM_CGROUP_ZSTAT(mz, lru);
410 return (nr_pages >> priority);
413 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
414 struct list_head *dst,
415 unsigned long *scanned, int order,
416 int mode, struct zone *z,
417 struct mem_cgroup *mem_cont,
418 int active, int file)
420 unsigned long nr_taken = 0;
424 struct list_head *src;
425 struct page_cgroup *pc, *tmp;
426 int nid = z->zone_pgdat->node_id;
427 int zid = zone_idx(z);
428 struct mem_cgroup_per_zone *mz;
429 int lru = LRU_FILE * !!file + !!active;
432 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
433 src = &mz->lists[lru];
435 spin_lock(&mz->lru_lock);
437 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
438 if (scan >= nr_to_scan)
440 if (unlikely(!PageCgroupUsed(pc)))
444 if (unlikely(!PageLRU(page)))
448 * TODO: play better with lumpy reclaim, grabbing anything.
450 if (PageUnevictable(page) ||
451 (PageActive(page) && !active) ||
452 (!PageActive(page) && active)) {
453 __mem_cgroup_move_lists(pc, page_lru(page));
458 list_move(&pc->lru, &pc_list);
460 if (__isolate_lru_page(page, mode, file) == 0) {
461 list_move(&page->lru, dst);
466 list_splice(&pc_list, src);
467 spin_unlock(&mz->lru_lock);
474 * Unlike exported interface, "oom" parameter is added. if oom==true,
475 * oom-killer can be invoked.
477 static int __mem_cgroup_try_charge(struct mm_struct *mm,
478 gfp_t gfp_mask, struct mem_cgroup **memcg, bool oom)
480 struct mem_cgroup *mem;
481 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
483 * We always charge the cgroup the mm_struct belongs to.
484 * The mm_struct's mem_cgroup changes on task migration if the
485 * thread group leader migrates. It's possible that mm is not
486 * set, if so charge the init_mm (happens for pagecache usage).
488 if (likely(!*memcg)) {
490 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
491 if (unlikely(!mem)) {
496 * For every charge from the cgroup, increment reference count
507 while (unlikely(res_counter_charge(&mem->res, PAGE_SIZE))) {
508 if (!(gfp_mask & __GFP_WAIT))
511 if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
515 * try_to_free_mem_cgroup_pages() might not give us a full
516 * picture of reclaim. Some pages are reclaimed and might be
517 * moved to swap cache or just unmapped from the cgroup.
518 * Check the limit again to see if the reclaim reduced the
519 * current usage of the cgroup before giving up
521 if (res_counter_check_under_limit(&mem->res))
526 mem_cgroup_out_of_memory(mem, gfp_mask);
537 * mem_cgroup_try_charge - get charge of PAGE_SIZE.
538 * @mm: an mm_struct which is charged against. (when *memcg is NULL)
539 * @gfp_mask: gfp_mask for reclaim.
540 * @memcg: a pointer to memory cgroup which is charged against.
542 * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated
543 * memory cgroup from @mm is got and stored in *memcg.
545 * Returns 0 if success. -ENOMEM at failure.
546 * This call can invoke OOM-Killer.
549 int mem_cgroup_try_charge(struct mm_struct *mm,
550 gfp_t mask, struct mem_cgroup **memcg)
552 return __mem_cgroup_try_charge(mm, mask, memcg, true);
556 * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be
557 * USED state. If already USED, uncharge and return.
560 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
561 struct page_cgroup *pc,
562 enum charge_type ctype)
564 struct mem_cgroup_per_zone *mz;
567 /* try_charge() can return NULL to *memcg, taking care of it. */
571 lock_page_cgroup(pc);
572 if (unlikely(PageCgroupUsed(pc))) {
573 unlock_page_cgroup(pc);
574 res_counter_uncharge(&mem->res, PAGE_SIZE);
578 pc->mem_cgroup = mem;
580 * If a page is accounted as a page cache, insert to inactive list.
581 * If anon, insert to active list.
583 pc->flags = pcg_default_flags[ctype];
585 mz = page_cgroup_zoneinfo(pc);
587 spin_lock_irqsave(&mz->lru_lock, flags);
588 __mem_cgroup_add_list(mz, pc, true);
589 spin_unlock_irqrestore(&mz->lru_lock, flags);
590 unlock_page_cgroup(pc);
594 * mem_cgroup_move_account - move account of the page
595 * @pc: page_cgroup of the page.
596 * @from: mem_cgroup which the page is moved from.
597 * @to: mem_cgroup which the page is moved to. @from != @to.
599 * The caller must confirm following.
601 * 2. lru_lock of old mem_cgroup(@from) should be held.
603 * returns 0 at success,
604 * returns -EBUSY when lock is busy or "pc" is unstable.
606 * This function does "uncharge" from old cgroup but doesn't do "charge" to
607 * new cgroup. It should be done by a caller.
610 static int mem_cgroup_move_account(struct page_cgroup *pc,
611 struct mem_cgroup *from, struct mem_cgroup *to)
613 struct mem_cgroup_per_zone *from_mz, *to_mz;
617 VM_BUG_ON(!irqs_disabled());
618 VM_BUG_ON(from == to);
620 nid = page_cgroup_nid(pc);
621 zid = page_cgroup_zid(pc);
622 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
623 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
626 if (!trylock_page_cgroup(pc))
629 if (!PageCgroupUsed(pc))
632 if (pc->mem_cgroup != from)
635 if (spin_trylock(&to_mz->lru_lock)) {
636 __mem_cgroup_remove_list(from_mz, pc);
638 res_counter_uncharge(&from->res, PAGE_SIZE);
641 __mem_cgroup_add_list(to_mz, pc, false);
643 spin_unlock(&to_mz->lru_lock);
646 unlock_page_cgroup(pc);
651 * move charges to its parent.
654 static int mem_cgroup_move_parent(struct page_cgroup *pc,
655 struct mem_cgroup *child,
658 struct cgroup *cg = child->css.cgroup;
659 struct cgroup *pcg = cg->parent;
660 struct mem_cgroup *parent;
661 struct mem_cgroup_per_zone *mz;
669 parent = mem_cgroup_from_cont(pcg);
671 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
675 mz = mem_cgroup_zoneinfo(child,
676 page_cgroup_nid(pc), page_cgroup_zid(pc));
678 spin_lock_irqsave(&mz->lru_lock, flags);
679 ret = mem_cgroup_move_account(pc, child, parent);
680 spin_unlock_irqrestore(&mz->lru_lock, flags);
682 /* drop extra refcnt */
683 css_put(&parent->css);
684 /* uncharge if move fails */
686 res_counter_uncharge(&parent->res, PAGE_SIZE);
692 * Charge the memory controller for page usage.
694 * 0 if the charge was successful
695 * < 0 if the cgroup is over its limit
697 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
698 gfp_t gfp_mask, enum charge_type ctype,
699 struct mem_cgroup *memcg)
701 struct mem_cgroup *mem;
702 struct page_cgroup *pc;
705 pc = lookup_page_cgroup(page);
706 /* can happen at boot */
712 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
716 __mem_cgroup_commit_charge(mem, pc, ctype);
720 int mem_cgroup_newpage_charge(struct page *page,
721 struct mm_struct *mm, gfp_t gfp_mask)
723 if (mem_cgroup_subsys.disabled)
725 if (PageCompound(page))
728 * If already mapped, we don't have to account.
729 * If page cache, page->mapping has address_space.
730 * But page->mapping may have out-of-use anon_vma pointer,
731 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
734 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
738 return mem_cgroup_charge_common(page, mm, gfp_mask,
739 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
742 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
745 if (mem_cgroup_subsys.disabled)
747 if (PageCompound(page))
750 * Corner case handling. This is called from add_to_page_cache()
751 * in usual. But some FS (shmem) precharges this page before calling it
752 * and call add_to_page_cache() with GFP_NOWAIT.
754 * For GFP_NOWAIT case, the page may be pre-charged before calling
755 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
756 * charge twice. (It works but has to pay a bit larger cost.)
758 if (!(gfp_mask & __GFP_WAIT)) {
759 struct page_cgroup *pc;
762 pc = lookup_page_cgroup(page);
765 lock_page_cgroup(pc);
766 if (PageCgroupUsed(pc)) {
767 unlock_page_cgroup(pc);
770 unlock_page_cgroup(pc);
776 if (page_is_file_cache(page))
777 return mem_cgroup_charge_common(page, mm, gfp_mask,
778 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
780 return mem_cgroup_charge_common(page, mm, gfp_mask,
781 MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL);
784 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
786 struct page_cgroup *pc;
788 if (mem_cgroup_subsys.disabled)
792 pc = lookup_page_cgroup(page);
793 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED);
796 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
798 if (mem_cgroup_subsys.disabled)
802 res_counter_uncharge(&mem->res, PAGE_SIZE);
808 * uncharge if !page_mapped(page)
811 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
813 struct page_cgroup *pc;
814 struct mem_cgroup *mem;
815 struct mem_cgroup_per_zone *mz;
818 if (mem_cgroup_subsys.disabled)
822 * Check if our page_cgroup is valid
824 pc = lookup_page_cgroup(page);
825 if (unlikely(!pc || !PageCgroupUsed(pc)))
828 lock_page_cgroup(pc);
829 if ((ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED && page_mapped(page))
830 || !PageCgroupUsed(pc)) {
831 /* This happens at race in zap_pte_range() and do_swap_page()*/
832 unlock_page_cgroup(pc);
835 ClearPageCgroupUsed(pc);
836 mem = pc->mem_cgroup;
838 mz = page_cgroup_zoneinfo(pc);
839 spin_lock_irqsave(&mz->lru_lock, flags);
840 __mem_cgroup_remove_list(mz, pc);
841 spin_unlock_irqrestore(&mz->lru_lock, flags);
842 unlock_page_cgroup(pc);
844 res_counter_uncharge(&mem->res, PAGE_SIZE);
850 void mem_cgroup_uncharge_page(struct page *page)
853 if (page_mapped(page))
855 if (page->mapping && !PageAnon(page))
857 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
860 void mem_cgroup_uncharge_cache_page(struct page *page)
862 VM_BUG_ON(page_mapped(page));
863 VM_BUG_ON(page->mapping);
864 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
868 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
871 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
873 struct page_cgroup *pc;
874 struct mem_cgroup *mem = NULL;
877 if (mem_cgroup_subsys.disabled)
880 pc = lookup_page_cgroup(page);
881 lock_page_cgroup(pc);
882 if (PageCgroupUsed(pc)) {
883 mem = pc->mem_cgroup;
886 unlock_page_cgroup(pc);
889 ret = mem_cgroup_try_charge(NULL, GFP_HIGHUSER_MOVABLE, &mem);
896 /* remove redundant charge if migration failed*/
897 void mem_cgroup_end_migration(struct mem_cgroup *mem,
898 struct page *oldpage, struct page *newpage)
900 struct page *target, *unused;
901 struct page_cgroup *pc;
902 enum charge_type ctype;
907 /* at migration success, oldpage->mapping is NULL. */
908 if (oldpage->mapping) {
916 if (PageAnon(target))
917 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
918 else if (page_is_file_cache(target))
919 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
921 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
923 /* unused page is not on radix-tree now. */
924 if (unused && ctype != MEM_CGROUP_CHARGE_TYPE_MAPPED)
925 __mem_cgroup_uncharge_common(unused, ctype);
927 pc = lookup_page_cgroup(target);
929 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
930 * So, double-counting is effectively avoided.
932 __mem_cgroup_commit_charge(mem, pc, ctype);
935 * Both of oldpage and newpage are still under lock_page().
936 * Then, we don't have to care about race in radix-tree.
937 * But we have to be careful that this page is unmapped or not.
939 * There is a case for !page_mapped(). At the start of
940 * migration, oldpage was mapped. But now, it's zapped.
941 * But we know *target* page is not freed/reused under us.
942 * mem_cgroup_uncharge_page() does all necessary checks.
944 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
945 mem_cgroup_uncharge_page(target);
949 * A call to try to shrink memory usage under specified resource controller.
950 * This is typically used for page reclaiming for shmem for reducing side
951 * effect of page allocation from shmem, which is used by some mem_cgroup.
953 int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
955 struct mem_cgroup *mem;
957 int retry = MEM_CGROUP_RECLAIM_RETRIES;
959 if (mem_cgroup_subsys.disabled)
965 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
966 if (unlikely(!mem)) {
974 progress = try_to_free_mem_cgroup_pages(mem, gfp_mask);
975 progress += res_counter_check_under_limit(&mem->res);
976 } while (!progress && --retry);
984 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
985 unsigned long long val)
988 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
992 while (res_counter_set_limit(&memcg->res, val)) {
993 if (signal_pending(current)) {
1001 progress = try_to_free_mem_cgroup_pages(memcg,
1002 GFP_HIGHUSER_MOVABLE);
1011 * This routine traverse page_cgroup in given list and drop them all.
1012 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1014 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
1015 struct mem_cgroup_per_zone *mz,
1018 struct page_cgroup *pc, *busy;
1019 unsigned long flags;
1021 struct list_head *list;
1024 list = &mz->lists[lru];
1026 loop = MEM_CGROUP_ZSTAT(mz, lru);
1027 /* give some margin against EBUSY etc...*/
1032 spin_lock_irqsave(&mz->lru_lock, flags);
1033 if (list_empty(list)) {
1034 spin_unlock_irqrestore(&mz->lru_lock, flags);
1037 pc = list_entry(list->prev, struct page_cgroup, lru);
1039 list_move(&pc->lru, list);
1041 spin_unlock_irqrestore(&mz->lru_lock, flags);
1044 spin_unlock_irqrestore(&mz->lru_lock, flags);
1046 ret = mem_cgroup_move_parent(pc, mem, GFP_HIGHUSER_MOVABLE);
1050 if (ret == -EBUSY || ret == -EINVAL) {
1051 /* found lock contention or "pc" is obsolete. */
1057 if (!ret && !list_empty(list))
1063 * make mem_cgroup's charge to be 0 if there is no task.
1064 * This enables deleting this mem_cgroup.
1066 static int mem_cgroup_force_empty(struct mem_cgroup *mem)
1069 int node, zid, shrink;
1070 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1076 while (mem->res.usage > 0) {
1078 if (atomic_read(&mem->css.cgroup->count) > 0)
1081 /* This is for making all *used* pages to be on LRU. */
1082 lru_add_drain_all();
1084 for_each_node_state(node, N_POSSIBLE) {
1085 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
1086 struct mem_cgroup_per_zone *mz;
1088 mz = mem_cgroup_zoneinfo(mem, node, zid);
1090 ret = mem_cgroup_force_empty_list(mem,
1099 /* it seems parent cgroup doesn't have enough mem */
1110 /* returns EBUSY if we come here twice. */
1115 /* try to free all pages in this cgroup */
1117 while (nr_retries && mem->res.usage > 0) {
1119 progress = try_to_free_mem_cgroup_pages(mem,
1120 GFP_HIGHUSER_MOVABLE);
1125 /* try move_account...there may be some *locked* pages. */
1132 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
1134 return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res,
1138 * The user of this function is...
1141 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1144 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
1145 unsigned long long val;
1148 switch (cft->private) {
1150 /* This function does all necessary parse...reuse it */
1151 ret = res_counter_memparse_write_strategy(buffer, &val);
1153 ret = mem_cgroup_resize_limit(memcg, val);
1156 ret = -EINVAL; /* should be BUG() ? */
1162 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
1164 struct mem_cgroup *mem;
1166 mem = mem_cgroup_from_cont(cont);
1169 res_counter_reset_max(&mem->res);
1172 res_counter_reset_failcnt(&mem->res);
1178 static const struct mem_cgroup_stat_desc {
1181 } mem_cgroup_stat_desc[] = {
1182 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
1183 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
1184 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
1185 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
1188 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
1189 struct cgroup_map_cb *cb)
1191 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1192 struct mem_cgroup_stat *stat = &mem_cont->stat;
1195 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1198 val = mem_cgroup_read_stat(stat, i);
1199 val *= mem_cgroup_stat_desc[i].unit;
1200 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
1202 /* showing # of active pages */
1204 unsigned long active_anon, inactive_anon;
1205 unsigned long active_file, inactive_file;
1206 unsigned long unevictable;
1208 inactive_anon = mem_cgroup_get_all_zonestat(mem_cont,
1210 active_anon = mem_cgroup_get_all_zonestat(mem_cont,
1212 inactive_file = mem_cgroup_get_all_zonestat(mem_cont,
1214 active_file = mem_cgroup_get_all_zonestat(mem_cont,
1216 unevictable = mem_cgroup_get_all_zonestat(mem_cont,
1219 cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE);
1220 cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE);
1221 cb->fill(cb, "active_file", (active_file) * PAGE_SIZE);
1222 cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE);
1223 cb->fill(cb, "unevictable", unevictable * PAGE_SIZE);
1229 static struct cftype mem_cgroup_files[] = {
1231 .name = "usage_in_bytes",
1232 .private = RES_USAGE,
1233 .read_u64 = mem_cgroup_read,
1236 .name = "max_usage_in_bytes",
1237 .private = RES_MAX_USAGE,
1238 .trigger = mem_cgroup_reset,
1239 .read_u64 = mem_cgroup_read,
1242 .name = "limit_in_bytes",
1243 .private = RES_LIMIT,
1244 .write_string = mem_cgroup_write,
1245 .read_u64 = mem_cgroup_read,
1249 .private = RES_FAILCNT,
1250 .trigger = mem_cgroup_reset,
1251 .read_u64 = mem_cgroup_read,
1255 .read_map = mem_control_stat_show,
1259 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1261 struct mem_cgroup_per_node *pn;
1262 struct mem_cgroup_per_zone *mz;
1264 int zone, tmp = node;
1266 * This routine is called against possible nodes.
1267 * But it's BUG to call kmalloc() against offline node.
1269 * TODO: this routine can waste much memory for nodes which will
1270 * never be onlined. It's better to use memory hotplug callback
1273 if (!node_state(node, N_NORMAL_MEMORY))
1275 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
1279 mem->info.nodeinfo[node] = pn;
1280 memset(pn, 0, sizeof(*pn));
1282 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
1283 mz = &pn->zoneinfo[zone];
1284 spin_lock_init(&mz->lru_lock);
1286 INIT_LIST_HEAD(&mz->lists[l]);
1291 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1293 kfree(mem->info.nodeinfo[node]);
1296 static struct mem_cgroup *mem_cgroup_alloc(void)
1298 struct mem_cgroup *mem;
1300 if (sizeof(*mem) < PAGE_SIZE)
1301 mem = kmalloc(sizeof(*mem), GFP_KERNEL);
1303 mem = vmalloc(sizeof(*mem));
1306 memset(mem, 0, sizeof(*mem));
1310 static void mem_cgroup_free(struct mem_cgroup *mem)
1312 if (sizeof(*mem) < PAGE_SIZE)
1319 static struct cgroup_subsys_state *
1320 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
1322 struct mem_cgroup *mem;
1325 if (unlikely((cont->parent) == NULL)) {
1326 mem = &init_mem_cgroup;
1328 mem = mem_cgroup_alloc();
1330 return ERR_PTR(-ENOMEM);
1333 res_counter_init(&mem->res);
1335 for_each_node_state(node, N_POSSIBLE)
1336 if (alloc_mem_cgroup_per_zone_info(mem, node))
1341 for_each_node_state(node, N_POSSIBLE)
1342 free_mem_cgroup_per_zone_info(mem, node);
1343 if (cont->parent != NULL)
1344 mem_cgroup_free(mem);
1345 return ERR_PTR(-ENOMEM);
1348 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1349 struct cgroup *cont)
1351 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1352 mem_cgroup_force_empty(mem);
1355 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
1356 struct cgroup *cont)
1359 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1361 for_each_node_state(node, N_POSSIBLE)
1362 free_mem_cgroup_per_zone_info(mem, node);
1364 mem_cgroup_free(mem_cgroup_from_cont(cont));
1367 static int mem_cgroup_populate(struct cgroup_subsys *ss,
1368 struct cgroup *cont)
1370 return cgroup_add_files(cont, ss, mem_cgroup_files,
1371 ARRAY_SIZE(mem_cgroup_files));
1374 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
1375 struct cgroup *cont,
1376 struct cgroup *old_cont,
1377 struct task_struct *p)
1379 struct mm_struct *mm;
1380 struct mem_cgroup *mem, *old_mem;
1382 mm = get_task_mm(p);
1386 mem = mem_cgroup_from_cont(cont);
1387 old_mem = mem_cgroup_from_cont(old_cont);
1390 * Only thread group leaders are allowed to migrate, the mm_struct is
1391 * in effect owned by the leader
1393 if (!thread_group_leader(p))
1400 struct cgroup_subsys mem_cgroup_subsys = {
1402 .subsys_id = mem_cgroup_subsys_id,
1403 .create = mem_cgroup_create,
1404 .pre_destroy = mem_cgroup_pre_destroy,
1405 .destroy = mem_cgroup_destroy,
1406 .populate = mem_cgroup_populate,
1407 .attach = mem_cgroup_move_task,