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/pagemap.h>
25 #include <linux/smp.h>
26 #include <linux/page-flags.h>
27 #include <linux/backing-dev.h>
28 #include <linux/bit_spinlock.h>
29 #include <linux/rcupdate.h>
30 #include <linux/slab.h>
31 #include <linux/swap.h>
32 #include <linux/spinlock.h>
34 #include <linux/seq_file.h>
35 #include <linux/vmalloc.h>
36 #include <linux/mm_inline.h>
37 #include <linux/page_cgroup.h>
39 #include <asm/uaccess.h>
41 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
42 #define MEM_CGROUP_RECLAIM_RETRIES 5
44 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
45 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 0 */
46 int do_swap_account __read_mostly;
47 static int really_do_swap_account __initdata = 1; /* for remember boot option*/
49 #define do_swap_account (0)
54 * Statistics for memory cgroup.
56 enum mem_cgroup_stat_index {
58 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
60 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
61 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
62 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
63 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
65 MEM_CGROUP_STAT_NSTATS,
68 struct mem_cgroup_stat_cpu {
69 s64 count[MEM_CGROUP_STAT_NSTATS];
70 } ____cacheline_aligned_in_smp;
72 struct mem_cgroup_stat {
73 struct mem_cgroup_stat_cpu cpustat[0];
77 * For accounting under irq disable, no need for increment preempt count.
79 static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
80 enum mem_cgroup_stat_index idx, int val)
82 stat->count[idx] += val;
85 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
86 enum mem_cgroup_stat_index idx)
90 for_each_possible_cpu(cpu)
91 ret += stat->cpustat[cpu].count[idx];
96 * per-zone information in memory controller.
98 struct mem_cgroup_per_zone {
100 * spin_lock to protect the per cgroup LRU
103 struct list_head lists[NR_LRU_LISTS];
104 unsigned long count[NR_LRU_LISTS];
106 /* Macro for accessing counter */
107 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
109 struct mem_cgroup_per_node {
110 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
113 struct mem_cgroup_lru_info {
114 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
118 * The memory controller data structure. The memory controller controls both
119 * page cache and RSS per cgroup. We would eventually like to provide
120 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
121 * to help the administrator determine what knobs to tune.
123 * TODO: Add a water mark for the memory controller. Reclaim will begin when
124 * we hit the water mark. May be even add a low water mark, such that
125 * no reclaim occurs from a cgroup at it's low water mark, this is
126 * a feature that will be implemented much later in the future.
129 struct cgroup_subsys_state css;
131 * the counter to account for memory usage
133 struct res_counter res;
135 * Per cgroup active and inactive list, similar to the
136 * per zone LRU lists.
138 struct mem_cgroup_lru_info info;
140 int prev_priority; /* for recording reclaim priority */
142 * statistics. This must be placed at the end of memcg.
144 struct mem_cgroup_stat stat;
148 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
149 MEM_CGROUP_CHARGE_TYPE_MAPPED,
150 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
151 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
152 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
156 /* only for here (for easy reading.) */
157 #define PCGF_CACHE (1UL << PCG_CACHE)
158 #define PCGF_USED (1UL << PCG_USED)
159 #define PCGF_ACTIVE (1UL << PCG_ACTIVE)
160 #define PCGF_LOCK (1UL << PCG_LOCK)
161 #define PCGF_FILE (1UL << PCG_FILE)
162 static const unsigned long
163 pcg_default_flags[NR_CHARGE_TYPE] = {
164 PCGF_CACHE | PCGF_FILE | PCGF_USED | PCGF_LOCK, /* File Cache */
165 PCGF_ACTIVE | PCGF_USED | PCGF_LOCK, /* Anon */
166 PCGF_ACTIVE | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
171 * Always modified under lru lock. Then, not necessary to preempt_disable()
173 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
174 struct page_cgroup *pc,
177 int val = (charge)? 1 : -1;
178 struct mem_cgroup_stat *stat = &mem->stat;
179 struct mem_cgroup_stat_cpu *cpustat;
181 VM_BUG_ON(!irqs_disabled());
183 cpustat = &stat->cpustat[smp_processor_id()];
184 if (PageCgroupCache(pc))
185 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
187 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
190 __mem_cgroup_stat_add_safe(cpustat,
191 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
193 __mem_cgroup_stat_add_safe(cpustat,
194 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
197 static struct mem_cgroup_per_zone *
198 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
200 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
203 static struct mem_cgroup_per_zone *
204 page_cgroup_zoneinfo(struct page_cgroup *pc)
206 struct mem_cgroup *mem = pc->mem_cgroup;
207 int nid = page_cgroup_nid(pc);
208 int zid = page_cgroup_zid(pc);
210 return mem_cgroup_zoneinfo(mem, nid, zid);
213 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
217 struct mem_cgroup_per_zone *mz;
220 for_each_online_node(nid)
221 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
222 mz = mem_cgroup_zoneinfo(mem, nid, zid);
223 total += MEM_CGROUP_ZSTAT(mz, idx);
228 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
230 return container_of(cgroup_subsys_state(cont,
231 mem_cgroup_subsys_id), struct mem_cgroup,
235 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
238 * mm_update_next_owner() may clear mm->owner to NULL
239 * if it races with swapoff, page migration, etc.
240 * So this can be called with p == NULL.
245 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
246 struct mem_cgroup, css);
249 static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz,
250 struct page_cgroup *pc)
254 if (PageCgroupUnevictable(pc))
255 lru = LRU_UNEVICTABLE;
257 if (PageCgroupActive(pc))
259 if (PageCgroupFile(pc))
263 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
265 mem_cgroup_charge_statistics(pc->mem_cgroup, pc, false);
269 static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz,
270 struct page_cgroup *pc, bool hot)
274 if (PageCgroupUnevictable(pc))
275 lru = LRU_UNEVICTABLE;
277 if (PageCgroupActive(pc))
279 if (PageCgroupFile(pc))
283 MEM_CGROUP_ZSTAT(mz, lru) += 1;
285 list_add(&pc->lru, &mz->lists[lru]);
287 list_add_tail(&pc->lru, &mz->lists[lru]);
289 mem_cgroup_charge_statistics(pc->mem_cgroup, pc, true);
292 static void __mem_cgroup_move_lists(struct page_cgroup *pc, enum lru_list lru)
294 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
295 int active = PageCgroupActive(pc);
296 int file = PageCgroupFile(pc);
297 int unevictable = PageCgroupUnevictable(pc);
298 enum lru_list from = unevictable ? LRU_UNEVICTABLE :
299 (LRU_FILE * !!file + !!active);
304 MEM_CGROUP_ZSTAT(mz, from) -= 1;
306 * However this is done under mz->lru_lock, another flags, which
307 * are not related to LRU, will be modified from out-of-lock.
308 * We have to use atomic set/clear flags.
310 if (is_unevictable_lru(lru)) {
311 ClearPageCgroupActive(pc);
312 SetPageCgroupUnevictable(pc);
314 if (is_active_lru(lru))
315 SetPageCgroupActive(pc);
317 ClearPageCgroupActive(pc);
318 ClearPageCgroupUnevictable(pc);
321 MEM_CGROUP_ZSTAT(mz, lru) += 1;
322 list_move(&pc->lru, &mz->lists[lru]);
325 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
330 ret = task->mm && mm_match_cgroup(task->mm, mem);
336 * This routine assumes that the appropriate zone's lru lock is already held
338 void mem_cgroup_move_lists(struct page *page, enum lru_list lru)
340 struct page_cgroup *pc;
341 struct mem_cgroup_per_zone *mz;
344 if (mem_cgroup_subsys.disabled)
348 * We cannot lock_page_cgroup while holding zone's lru_lock,
349 * because other holders of lock_page_cgroup can be interrupted
350 * with an attempt to rotate_reclaimable_page. But we cannot
351 * safely get to page_cgroup without it, so just try_lock it:
352 * mem_cgroup_isolate_pages allows for page left on wrong list.
354 pc = lookup_page_cgroup(page);
355 if (!trylock_page_cgroup(pc))
357 if (pc && PageCgroupUsed(pc)) {
358 mz = page_cgroup_zoneinfo(pc);
359 spin_lock_irqsave(&mz->lru_lock, flags);
360 __mem_cgroup_move_lists(pc, lru);
361 spin_unlock_irqrestore(&mz->lru_lock, flags);
363 unlock_page_cgroup(pc);
367 * Calculate mapped_ratio under memory controller. This will be used in
368 * vmscan.c for deteremining we have to reclaim mapped pages.
370 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
375 * usage is recorded in bytes. But, here, we assume the number of
376 * physical pages can be represented by "long" on any arch.
378 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
379 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
380 return (int)((rss * 100L) / total);
384 * prev_priority control...this will be used in memory reclaim path.
386 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
388 return mem->prev_priority;
391 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
393 if (priority < mem->prev_priority)
394 mem->prev_priority = priority;
397 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
399 mem->prev_priority = priority;
403 * Calculate # of pages to be scanned in this priority/zone.
406 * priority starts from "DEF_PRIORITY" and decremented in each loop.
407 * (see include/linux/mmzone.h)
410 long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone,
411 int priority, enum lru_list lru)
414 int nid = zone->zone_pgdat->node_id;
415 int zid = zone_idx(zone);
416 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
418 nr_pages = MEM_CGROUP_ZSTAT(mz, lru);
420 return (nr_pages >> priority);
423 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
424 struct list_head *dst,
425 unsigned long *scanned, int order,
426 int mode, struct zone *z,
427 struct mem_cgroup *mem_cont,
428 int active, int file)
430 unsigned long nr_taken = 0;
434 struct list_head *src;
435 struct page_cgroup *pc, *tmp;
436 int nid = z->zone_pgdat->node_id;
437 int zid = zone_idx(z);
438 struct mem_cgroup_per_zone *mz;
439 int lru = LRU_FILE * !!file + !!active;
442 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
443 src = &mz->lists[lru];
445 spin_lock(&mz->lru_lock);
447 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
448 if (scan >= nr_to_scan)
450 if (unlikely(!PageCgroupUsed(pc)))
454 if (unlikely(!PageLRU(page)))
458 * TODO: play better with lumpy reclaim, grabbing anything.
460 if (PageUnevictable(page) ||
461 (PageActive(page) && !active) ||
462 (!PageActive(page) && active)) {
463 __mem_cgroup_move_lists(pc, page_lru(page));
468 list_move(&pc->lru, &pc_list);
470 if (__isolate_lru_page(page, mode, file) == 0) {
471 list_move(&page->lru, dst);
476 list_splice(&pc_list, src);
477 spin_unlock(&mz->lru_lock);
484 * Unlike exported interface, "oom" parameter is added. if oom==true,
485 * oom-killer can be invoked.
487 static int __mem_cgroup_try_charge(struct mm_struct *mm,
488 gfp_t gfp_mask, struct mem_cgroup **memcg, bool oom)
490 struct mem_cgroup *mem;
491 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
493 * We always charge the cgroup the mm_struct belongs to.
494 * The mm_struct's mem_cgroup changes on task migration if the
495 * thread group leader migrates. It's possible that mm is not
496 * 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
517 while (unlikely(res_counter_charge(&mem->res, PAGE_SIZE))) {
518 if (!(gfp_mask & __GFP_WAIT))
521 if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
525 * try_to_free_mem_cgroup_pages() might not give us a full
526 * picture of reclaim. Some pages are reclaimed and might be
527 * moved to swap cache or just unmapped from the cgroup.
528 * Check the limit again to see if the reclaim reduced the
529 * current usage of the cgroup before giving up
531 if (res_counter_check_under_limit(&mem->res))
536 mem_cgroup_out_of_memory(mem, gfp_mask);
547 * mem_cgroup_try_charge - get charge of PAGE_SIZE.
548 * @mm: an mm_struct which is charged against. (when *memcg is NULL)
549 * @gfp_mask: gfp_mask for reclaim.
550 * @memcg: a pointer to memory cgroup which is charged against.
552 * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated
553 * memory cgroup from @mm is got and stored in *memcg.
555 * Returns 0 if success. -ENOMEM at failure.
556 * This call can invoke OOM-Killer.
559 int mem_cgroup_try_charge(struct mm_struct *mm,
560 gfp_t mask, struct mem_cgroup **memcg)
562 return __mem_cgroup_try_charge(mm, mask, memcg, true);
566 * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be
567 * USED state. If already USED, uncharge and return.
570 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
571 struct page_cgroup *pc,
572 enum charge_type ctype)
574 struct mem_cgroup_per_zone *mz;
577 /* try_charge() can return NULL to *memcg, taking care of it. */
581 lock_page_cgroup(pc);
582 if (unlikely(PageCgroupUsed(pc))) {
583 unlock_page_cgroup(pc);
584 res_counter_uncharge(&mem->res, PAGE_SIZE);
588 pc->mem_cgroup = mem;
590 * If a page is accounted as a page cache, insert to inactive list.
591 * If anon, insert to active list.
593 pc->flags = pcg_default_flags[ctype];
595 mz = page_cgroup_zoneinfo(pc);
597 spin_lock_irqsave(&mz->lru_lock, flags);
598 __mem_cgroup_add_list(mz, pc, true);
599 spin_unlock_irqrestore(&mz->lru_lock, flags);
600 unlock_page_cgroup(pc);
604 * mem_cgroup_move_account - move account of the page
605 * @pc: page_cgroup of the page.
606 * @from: mem_cgroup which the page is moved from.
607 * @to: mem_cgroup which the page is moved to. @from != @to.
609 * The caller must confirm following.
611 * 2. lru_lock of old mem_cgroup(@from) should be held.
613 * returns 0 at success,
614 * returns -EBUSY when lock is busy or "pc" is unstable.
616 * This function does "uncharge" from old cgroup but doesn't do "charge" to
617 * new cgroup. It should be done by a caller.
620 static int mem_cgroup_move_account(struct page_cgroup *pc,
621 struct mem_cgroup *from, struct mem_cgroup *to)
623 struct mem_cgroup_per_zone *from_mz, *to_mz;
627 VM_BUG_ON(!irqs_disabled());
628 VM_BUG_ON(from == to);
630 nid = page_cgroup_nid(pc);
631 zid = page_cgroup_zid(pc);
632 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
633 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
636 if (!trylock_page_cgroup(pc))
639 if (!PageCgroupUsed(pc))
642 if (pc->mem_cgroup != from)
645 if (spin_trylock(&to_mz->lru_lock)) {
646 __mem_cgroup_remove_list(from_mz, pc);
648 res_counter_uncharge(&from->res, PAGE_SIZE);
651 __mem_cgroup_add_list(to_mz, pc, false);
653 spin_unlock(&to_mz->lru_lock);
656 unlock_page_cgroup(pc);
661 * move charges to its parent.
664 static int mem_cgroup_move_parent(struct page_cgroup *pc,
665 struct mem_cgroup *child,
668 struct cgroup *cg = child->css.cgroup;
669 struct cgroup *pcg = cg->parent;
670 struct mem_cgroup *parent;
671 struct mem_cgroup_per_zone *mz;
679 parent = mem_cgroup_from_cont(pcg);
681 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
685 mz = mem_cgroup_zoneinfo(child,
686 page_cgroup_nid(pc), page_cgroup_zid(pc));
688 spin_lock_irqsave(&mz->lru_lock, flags);
689 ret = mem_cgroup_move_account(pc, child, parent);
690 spin_unlock_irqrestore(&mz->lru_lock, flags);
692 /* drop extra refcnt */
693 css_put(&parent->css);
694 /* uncharge if move fails */
696 res_counter_uncharge(&parent->res, PAGE_SIZE);
702 * Charge the memory controller for page usage.
704 * 0 if the charge was successful
705 * < 0 if the cgroup is over its limit
707 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
708 gfp_t gfp_mask, enum charge_type ctype,
709 struct mem_cgroup *memcg)
711 struct mem_cgroup *mem;
712 struct page_cgroup *pc;
715 pc = lookup_page_cgroup(page);
716 /* can happen at boot */
722 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
726 __mem_cgroup_commit_charge(mem, pc, ctype);
730 int mem_cgroup_newpage_charge(struct page *page,
731 struct mm_struct *mm, gfp_t gfp_mask)
733 if (mem_cgroup_subsys.disabled)
735 if (PageCompound(page))
738 * If already mapped, we don't have to account.
739 * If page cache, page->mapping has address_space.
740 * But page->mapping may have out-of-use anon_vma pointer,
741 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
744 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
748 return mem_cgroup_charge_common(page, mm, gfp_mask,
749 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
752 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
755 if (mem_cgroup_subsys.disabled)
757 if (PageCompound(page))
760 * Corner case handling. This is called from add_to_page_cache()
761 * in usual. But some FS (shmem) precharges this page before calling it
762 * and call add_to_page_cache() with GFP_NOWAIT.
764 * For GFP_NOWAIT case, the page may be pre-charged before calling
765 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
766 * charge twice. (It works but has to pay a bit larger cost.)
768 if (!(gfp_mask & __GFP_WAIT)) {
769 struct page_cgroup *pc;
772 pc = lookup_page_cgroup(page);
775 lock_page_cgroup(pc);
776 if (PageCgroupUsed(pc)) {
777 unlock_page_cgroup(pc);
780 unlock_page_cgroup(pc);
786 if (page_is_file_cache(page))
787 return mem_cgroup_charge_common(page, mm, gfp_mask,
788 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
790 return mem_cgroup_charge_common(page, mm, gfp_mask,
791 MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL);
795 int mem_cgroup_cache_charge_swapin(struct page *page,
796 struct mm_struct *mm, gfp_t mask, bool locked)
800 if (mem_cgroup_subsys.disabled)
807 * If not locked, the page can be dropped from SwapCache until
810 if (PageSwapCache(page)) {
811 ret = mem_cgroup_charge_common(page, mm, mask,
812 MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL);
821 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
823 struct page_cgroup *pc;
825 if (mem_cgroup_subsys.disabled)
829 pc = lookup_page_cgroup(page);
830 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED);
833 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
835 if (mem_cgroup_subsys.disabled)
839 res_counter_uncharge(&mem->res, PAGE_SIZE);
845 * uncharge if !page_mapped(page)
848 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
850 struct page_cgroup *pc;
851 struct mem_cgroup *mem;
852 struct mem_cgroup_per_zone *mz;
855 if (mem_cgroup_subsys.disabled)
858 if (PageSwapCache(page))
862 * Check if our page_cgroup is valid
864 pc = lookup_page_cgroup(page);
865 if (unlikely(!pc || !PageCgroupUsed(pc)))
868 lock_page_cgroup(pc);
870 if (!PageCgroupUsed(pc))
874 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
875 if (page_mapped(page))
878 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
879 if (!PageAnon(page)) { /* Shared memory */
880 if (page->mapping && !page_is_file_cache(page))
882 } else if (page_mapped(page)) /* Anon */
889 ClearPageCgroupUsed(pc);
890 mem = pc->mem_cgroup;
892 mz = page_cgroup_zoneinfo(pc);
893 spin_lock_irqsave(&mz->lru_lock, flags);
894 __mem_cgroup_remove_list(mz, pc);
895 spin_unlock_irqrestore(&mz->lru_lock, flags);
896 unlock_page_cgroup(pc);
898 res_counter_uncharge(&mem->res, PAGE_SIZE);
904 unlock_page_cgroup(pc);
908 void mem_cgroup_uncharge_page(struct page *page)
911 if (page_mapped(page))
913 if (page->mapping && !PageAnon(page))
915 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
918 void mem_cgroup_uncharge_cache_page(struct page *page)
920 VM_BUG_ON(page_mapped(page));
921 VM_BUG_ON(page->mapping);
922 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
925 void mem_cgroup_uncharge_swapcache(struct page *page)
927 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
931 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
934 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
936 struct page_cgroup *pc;
937 struct mem_cgroup *mem = NULL;
940 if (mem_cgroup_subsys.disabled)
943 pc = lookup_page_cgroup(page);
944 lock_page_cgroup(pc);
945 if (PageCgroupUsed(pc)) {
946 mem = pc->mem_cgroup;
949 unlock_page_cgroup(pc);
952 ret = mem_cgroup_try_charge(NULL, GFP_HIGHUSER_MOVABLE, &mem);
959 /* remove redundant charge if migration failed*/
960 void mem_cgroup_end_migration(struct mem_cgroup *mem,
961 struct page *oldpage, struct page *newpage)
963 struct page *target, *unused;
964 struct page_cgroup *pc;
965 enum charge_type ctype;
970 /* at migration success, oldpage->mapping is NULL. */
971 if (oldpage->mapping) {
979 if (PageAnon(target))
980 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
981 else if (page_is_file_cache(target))
982 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
984 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
986 /* unused page is not on radix-tree now. */
988 __mem_cgroup_uncharge_common(unused, ctype);
990 pc = lookup_page_cgroup(target);
992 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
993 * So, double-counting is effectively avoided.
995 __mem_cgroup_commit_charge(mem, pc, ctype);
998 * Both of oldpage and newpage are still under lock_page().
999 * Then, we don't have to care about race in radix-tree.
1000 * But we have to be careful that this page is unmapped or not.
1002 * There is a case for !page_mapped(). At the start of
1003 * migration, oldpage was mapped. But now, it's zapped.
1004 * But we know *target* page is not freed/reused under us.
1005 * mem_cgroup_uncharge_page() does all necessary checks.
1007 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1008 mem_cgroup_uncharge_page(target);
1012 * A call to try to shrink memory usage under specified resource controller.
1013 * This is typically used for page reclaiming for shmem for reducing side
1014 * effect of page allocation from shmem, which is used by some mem_cgroup.
1016 int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
1018 struct mem_cgroup *mem;
1020 int retry = MEM_CGROUP_RECLAIM_RETRIES;
1022 if (mem_cgroup_subsys.disabled)
1028 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
1029 if (unlikely(!mem)) {
1037 progress = try_to_free_mem_cgroup_pages(mem, gfp_mask);
1038 progress += res_counter_check_under_limit(&mem->res);
1039 } while (!progress && --retry);
1047 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1048 unsigned long long val)
1051 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1055 while (res_counter_set_limit(&memcg->res, val)) {
1056 if (signal_pending(current)) {
1064 progress = try_to_free_mem_cgroup_pages(memcg,
1065 GFP_HIGHUSER_MOVABLE);
1074 * This routine traverse page_cgroup in given list and drop them all.
1075 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1077 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
1078 struct mem_cgroup_per_zone *mz,
1081 struct page_cgroup *pc, *busy;
1082 unsigned long flags;
1084 struct list_head *list;
1087 list = &mz->lists[lru];
1089 loop = MEM_CGROUP_ZSTAT(mz, lru);
1090 /* give some margin against EBUSY etc...*/
1095 spin_lock_irqsave(&mz->lru_lock, flags);
1096 if (list_empty(list)) {
1097 spin_unlock_irqrestore(&mz->lru_lock, flags);
1100 pc = list_entry(list->prev, struct page_cgroup, lru);
1102 list_move(&pc->lru, list);
1104 spin_unlock_irqrestore(&mz->lru_lock, flags);
1107 spin_unlock_irqrestore(&mz->lru_lock, flags);
1109 ret = mem_cgroup_move_parent(pc, mem, GFP_HIGHUSER_MOVABLE);
1113 if (ret == -EBUSY || ret == -EINVAL) {
1114 /* found lock contention or "pc" is obsolete. */
1120 if (!ret && !list_empty(list))
1126 * make mem_cgroup's charge to be 0 if there is no task.
1127 * This enables deleting this mem_cgroup.
1129 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
1132 int node, zid, shrink;
1133 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1134 struct cgroup *cgrp = mem->css.cgroup;
1139 /* should free all ? */
1143 while (mem->res.usage > 0) {
1145 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
1148 if (signal_pending(current))
1150 /* This is for making all *used* pages to be on LRU. */
1151 lru_add_drain_all();
1153 for_each_node_state(node, N_POSSIBLE) {
1154 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
1155 struct mem_cgroup_per_zone *mz;
1157 mz = mem_cgroup_zoneinfo(mem, node, zid);
1159 ret = mem_cgroup_force_empty_list(mem,
1168 /* it seems parent cgroup doesn't have enough mem */
1179 /* returns EBUSY if there is a task or if we come here twice. */
1180 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
1184 /* we call try-to-free pages for make this cgroup empty */
1185 lru_add_drain_all();
1186 /* try to free all pages in this cgroup */
1188 while (nr_retries && mem->res.usage > 0) {
1191 if (signal_pending(current)) {
1195 progress = try_to_free_mem_cgroup_pages(mem,
1196 GFP_HIGHUSER_MOVABLE);
1199 /* maybe some writeback is necessary */
1200 congestion_wait(WRITE, HZ/10);
1204 /* try move_account...there may be some *locked* pages. */
1211 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1213 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1217 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
1219 return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res,
1223 * The user of this function is...
1226 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1229 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
1230 unsigned long long val;
1233 switch (cft->private) {
1235 /* This function does all necessary parse...reuse it */
1236 ret = res_counter_memparse_write_strategy(buffer, &val);
1238 ret = mem_cgroup_resize_limit(memcg, val);
1241 ret = -EINVAL; /* should be BUG() ? */
1247 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
1249 struct mem_cgroup *mem;
1251 mem = mem_cgroup_from_cont(cont);
1254 res_counter_reset_max(&mem->res);
1257 res_counter_reset_failcnt(&mem->res);
1263 static const struct mem_cgroup_stat_desc {
1266 } mem_cgroup_stat_desc[] = {
1267 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
1268 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
1269 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
1270 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
1273 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
1274 struct cgroup_map_cb *cb)
1276 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1277 struct mem_cgroup_stat *stat = &mem_cont->stat;
1280 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1283 val = mem_cgroup_read_stat(stat, i);
1284 val *= mem_cgroup_stat_desc[i].unit;
1285 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
1287 /* showing # of active pages */
1289 unsigned long active_anon, inactive_anon;
1290 unsigned long active_file, inactive_file;
1291 unsigned long unevictable;
1293 inactive_anon = mem_cgroup_get_all_zonestat(mem_cont,
1295 active_anon = mem_cgroup_get_all_zonestat(mem_cont,
1297 inactive_file = mem_cgroup_get_all_zonestat(mem_cont,
1299 active_file = mem_cgroup_get_all_zonestat(mem_cont,
1301 unevictable = mem_cgroup_get_all_zonestat(mem_cont,
1304 cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE);
1305 cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE);
1306 cb->fill(cb, "active_file", (active_file) * PAGE_SIZE);
1307 cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE);
1308 cb->fill(cb, "unevictable", unevictable * PAGE_SIZE);
1315 static struct cftype mem_cgroup_files[] = {
1317 .name = "usage_in_bytes",
1318 .private = RES_USAGE,
1319 .read_u64 = mem_cgroup_read,
1322 .name = "max_usage_in_bytes",
1323 .private = RES_MAX_USAGE,
1324 .trigger = mem_cgroup_reset,
1325 .read_u64 = mem_cgroup_read,
1328 .name = "limit_in_bytes",
1329 .private = RES_LIMIT,
1330 .write_string = mem_cgroup_write,
1331 .read_u64 = mem_cgroup_read,
1335 .private = RES_FAILCNT,
1336 .trigger = mem_cgroup_reset,
1337 .read_u64 = mem_cgroup_read,
1341 .read_map = mem_control_stat_show,
1344 .name = "force_empty",
1345 .trigger = mem_cgroup_force_empty_write,
1349 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1351 struct mem_cgroup_per_node *pn;
1352 struct mem_cgroup_per_zone *mz;
1354 int zone, tmp = node;
1356 * This routine is called against possible nodes.
1357 * But it's BUG to call kmalloc() against offline node.
1359 * TODO: this routine can waste much memory for nodes which will
1360 * never be onlined. It's better to use memory hotplug callback
1363 if (!node_state(node, N_NORMAL_MEMORY))
1365 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
1369 mem->info.nodeinfo[node] = pn;
1370 memset(pn, 0, sizeof(*pn));
1372 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
1373 mz = &pn->zoneinfo[zone];
1374 spin_lock_init(&mz->lru_lock);
1376 INIT_LIST_HEAD(&mz->lists[l]);
1381 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1383 kfree(mem->info.nodeinfo[node]);
1386 static int mem_cgroup_size(void)
1388 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
1389 return sizeof(struct mem_cgroup) + cpustat_size;
1392 static struct mem_cgroup *mem_cgroup_alloc(void)
1394 struct mem_cgroup *mem;
1395 int size = mem_cgroup_size();
1397 if (size < PAGE_SIZE)
1398 mem = kmalloc(size, GFP_KERNEL);
1400 mem = vmalloc(size);
1403 memset(mem, 0, size);
1407 static void mem_cgroup_free(struct mem_cgroup *mem)
1409 if (mem_cgroup_size() < PAGE_SIZE)
1416 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1417 static void __init enable_swap_cgroup(void)
1419 if (!mem_cgroup_subsys.disabled && really_do_swap_account)
1420 do_swap_account = 1;
1423 static void __init enable_swap_cgroup(void)
1428 static struct cgroup_subsys_state *
1429 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
1431 struct mem_cgroup *mem;
1434 mem = mem_cgroup_alloc();
1436 return ERR_PTR(-ENOMEM);
1438 res_counter_init(&mem->res);
1440 for_each_node_state(node, N_POSSIBLE)
1441 if (alloc_mem_cgroup_per_zone_info(mem, node))
1444 if (cont->parent == NULL)
1445 enable_swap_cgroup();
1449 for_each_node_state(node, N_POSSIBLE)
1450 free_mem_cgroup_per_zone_info(mem, node);
1451 mem_cgroup_free(mem);
1452 return ERR_PTR(-ENOMEM);
1455 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1456 struct cgroup *cont)
1458 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1459 mem_cgroup_force_empty(mem, false);
1462 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
1463 struct cgroup *cont)
1466 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1468 for_each_node_state(node, N_POSSIBLE)
1469 free_mem_cgroup_per_zone_info(mem, node);
1471 mem_cgroup_free(mem_cgroup_from_cont(cont));
1474 static int mem_cgroup_populate(struct cgroup_subsys *ss,
1475 struct cgroup *cont)
1477 return cgroup_add_files(cont, ss, mem_cgroup_files,
1478 ARRAY_SIZE(mem_cgroup_files));
1481 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
1482 struct cgroup *cont,
1483 struct cgroup *old_cont,
1484 struct task_struct *p)
1486 struct mm_struct *mm;
1487 struct mem_cgroup *mem, *old_mem;
1489 mm = get_task_mm(p);
1493 mem = mem_cgroup_from_cont(cont);
1494 old_mem = mem_cgroup_from_cont(old_cont);
1497 * Only thread group leaders are allowed to migrate, the mm_struct is
1498 * in effect owned by the leader
1500 if (!thread_group_leader(p))
1507 struct cgroup_subsys mem_cgroup_subsys = {
1509 .subsys_id = mem_cgroup_subsys_id,
1510 .create = mem_cgroup_create,
1511 .pre_destroy = mem_cgroup_pre_destroy,
1512 .destroy = mem_cgroup_destroy,
1513 .populate = mem_cgroup_populate,
1514 .attach = mem_cgroup_move_task,
1518 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1520 static int __init disable_swap_account(char *s)
1522 really_do_swap_account = 0;
1525 __setup("noswapaccount", disable_swap_account);