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>
37 #include <asm/uaccess.h>
39 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
40 static struct kmem_cache *page_cgroup_cache __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 * We use the lower bit of the page->page_cgroup pointer as a bit spin
140 * lock. We need to ensure that page->page_cgroup is at least two
141 * byte aligned (based on comments from Nick Piggin). But since
142 * bit_spin_lock doesn't actually set that lock bit in a non-debug
143 * uniprocessor kernel, we should avoid setting it here too.
145 #define PAGE_CGROUP_LOCK_BIT 0x0
146 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
147 #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT)
149 #define PAGE_CGROUP_LOCK 0x0
153 * A page_cgroup page is associated with every page descriptor. The
154 * page_cgroup helps us identify information about the cgroup
157 struct list_head lru; /* per cgroup LRU list */
159 struct mem_cgroup *mem_cgroup;
162 #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */
163 #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */
164 #define PAGE_CGROUP_FLAG_FILE (0x4) /* page is file system backed */
165 #define PAGE_CGROUP_FLAG_UNEVICTABLE (0x8) /* page is unevictableable */
167 static int page_cgroup_nid(struct page_cgroup *pc)
169 return page_to_nid(pc->page);
172 static enum zone_type page_cgroup_zid(struct page_cgroup *pc)
174 return page_zonenum(pc->page);
178 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
179 MEM_CGROUP_CHARGE_TYPE_MAPPED,
180 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
181 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
185 * Always modified under lru lock. Then, not necessary to preempt_disable()
187 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags,
190 int val = (charge)? 1 : -1;
191 struct mem_cgroup_stat *stat = &mem->stat;
192 struct mem_cgroup_stat_cpu *cpustat;
194 VM_BUG_ON(!irqs_disabled());
196 cpustat = &stat->cpustat[smp_processor_id()];
197 if (flags & PAGE_CGROUP_FLAG_CACHE)
198 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
200 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
203 __mem_cgroup_stat_add_safe(cpustat,
204 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
206 __mem_cgroup_stat_add_safe(cpustat,
207 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
210 static struct mem_cgroup_per_zone *
211 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
213 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
216 static struct mem_cgroup_per_zone *
217 page_cgroup_zoneinfo(struct page_cgroup *pc)
219 struct mem_cgroup *mem = pc->mem_cgroup;
220 int nid = page_cgroup_nid(pc);
221 int zid = page_cgroup_zid(pc);
223 return mem_cgroup_zoneinfo(mem, nid, zid);
226 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
230 struct mem_cgroup_per_zone *mz;
233 for_each_online_node(nid)
234 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
235 mz = mem_cgroup_zoneinfo(mem, nid, zid);
236 total += MEM_CGROUP_ZSTAT(mz, idx);
241 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
243 return container_of(cgroup_subsys_state(cont,
244 mem_cgroup_subsys_id), struct mem_cgroup,
248 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
251 * mm_update_next_owner() may clear mm->owner to NULL
252 * if it races with swapoff, page migration, etc.
253 * So this can be called with p == NULL.
258 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
259 struct mem_cgroup, css);
262 static inline int page_cgroup_locked(struct page *page)
264 return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
267 static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc)
269 VM_BUG_ON(!page_cgroup_locked(page));
270 page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK);
273 struct page_cgroup *page_get_page_cgroup(struct page *page)
275 return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK);
278 static void lock_page_cgroup(struct page *page)
280 bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
283 static int try_lock_page_cgroup(struct page *page)
285 return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
288 static void unlock_page_cgroup(struct page *page)
290 bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
293 static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz,
294 struct page_cgroup *pc)
298 if (pc->flags & PAGE_CGROUP_FLAG_UNEVICTABLE)
299 lru = LRU_UNEVICTABLE;
301 if (pc->flags & PAGE_CGROUP_FLAG_ACTIVE)
303 if (pc->flags & PAGE_CGROUP_FLAG_FILE)
307 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
309 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false);
313 static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz,
314 struct page_cgroup *pc)
318 if (pc->flags & PAGE_CGROUP_FLAG_UNEVICTABLE)
319 lru = LRU_UNEVICTABLE;
321 if (pc->flags & PAGE_CGROUP_FLAG_ACTIVE)
323 if (pc->flags & PAGE_CGROUP_FLAG_FILE)
327 MEM_CGROUP_ZSTAT(mz, lru) += 1;
328 list_add(&pc->lru, &mz->lists[lru]);
330 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true);
333 static void __mem_cgroup_move_lists(struct page_cgroup *pc, enum lru_list lru)
335 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
336 int active = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
337 int file = pc->flags & PAGE_CGROUP_FLAG_FILE;
338 int unevictable = pc->flags & PAGE_CGROUP_FLAG_UNEVICTABLE;
339 enum lru_list from = unevictable ? LRU_UNEVICTABLE :
340 (LRU_FILE * !!file + !!active);
345 MEM_CGROUP_ZSTAT(mz, from) -= 1;
347 if (is_unevictable_lru(lru)) {
348 pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE;
349 pc->flags |= PAGE_CGROUP_FLAG_UNEVICTABLE;
351 if (is_active_lru(lru))
352 pc->flags |= PAGE_CGROUP_FLAG_ACTIVE;
354 pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE;
355 pc->flags &= ~PAGE_CGROUP_FLAG_UNEVICTABLE;
358 MEM_CGROUP_ZSTAT(mz, lru) += 1;
359 list_move(&pc->lru, &mz->lists[lru]);
362 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
367 ret = task->mm && mm_match_cgroup(task->mm, mem);
373 * This routine assumes that the appropriate zone's lru lock is already held
375 void mem_cgroup_move_lists(struct page *page, enum lru_list lru)
377 struct page_cgroup *pc;
378 struct mem_cgroup_per_zone *mz;
381 if (mem_cgroup_subsys.disabled)
385 * We cannot lock_page_cgroup while holding zone's lru_lock,
386 * because other holders of lock_page_cgroup can be interrupted
387 * with an attempt to rotate_reclaimable_page. But we cannot
388 * safely get to page_cgroup without it, so just try_lock it:
389 * mem_cgroup_isolate_pages allows for page left on wrong list.
391 if (!try_lock_page_cgroup(page))
394 pc = page_get_page_cgroup(page);
396 mz = page_cgroup_zoneinfo(pc);
397 spin_lock_irqsave(&mz->lru_lock, flags);
398 __mem_cgroup_move_lists(pc, lru);
399 spin_unlock_irqrestore(&mz->lru_lock, flags);
401 unlock_page_cgroup(page);
405 * Calculate mapped_ratio under memory controller. This will be used in
406 * vmscan.c for deteremining we have to reclaim mapped pages.
408 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
413 * usage is recorded in bytes. But, here, we assume the number of
414 * physical pages can be represented by "long" on any arch.
416 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
417 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
418 return (int)((rss * 100L) / total);
422 * prev_priority control...this will be used in memory reclaim path.
424 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
426 return mem->prev_priority;
429 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
431 if (priority < mem->prev_priority)
432 mem->prev_priority = priority;
435 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
437 mem->prev_priority = priority;
441 * Calculate # of pages to be scanned in this priority/zone.
444 * priority starts from "DEF_PRIORITY" and decremented in each loop.
445 * (see include/linux/mmzone.h)
448 long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone,
449 int priority, enum lru_list lru)
452 int nid = zone->zone_pgdat->node_id;
453 int zid = zone_idx(zone);
454 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
456 nr_pages = MEM_CGROUP_ZSTAT(mz, lru);
458 return (nr_pages >> priority);
461 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
462 struct list_head *dst,
463 unsigned long *scanned, int order,
464 int mode, struct zone *z,
465 struct mem_cgroup *mem_cont,
466 int active, int file)
468 unsigned long nr_taken = 0;
472 struct list_head *src;
473 struct page_cgroup *pc, *tmp;
474 int nid = z->zone_pgdat->node_id;
475 int zid = zone_idx(z);
476 struct mem_cgroup_per_zone *mz;
477 int lru = LRU_FILE * !!file + !!active;
480 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
481 src = &mz->lists[lru];
483 spin_lock(&mz->lru_lock);
485 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
486 if (scan >= nr_to_scan)
490 if (unlikely(!PageLRU(page)))
494 * TODO: play better with lumpy reclaim, grabbing anything.
496 if (PageUnevictable(page) ||
497 (PageActive(page) && !active) ||
498 (!PageActive(page) && active)) {
499 __mem_cgroup_move_lists(pc, page_lru(page));
504 list_move(&pc->lru, &pc_list);
506 if (__isolate_lru_page(page, mode, file) == 0) {
507 list_move(&page->lru, dst);
512 list_splice(&pc_list, src);
513 spin_unlock(&mz->lru_lock);
520 * Charge the memory controller for page usage.
522 * 0 if the charge was successful
523 * < 0 if the cgroup is over its limit
525 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
526 gfp_t gfp_mask, enum charge_type ctype,
527 struct mem_cgroup *memcg)
529 struct mem_cgroup *mem;
530 struct page_cgroup *pc;
532 unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
533 struct mem_cgroup_per_zone *mz;
535 pc = kmem_cache_alloc(page_cgroup_cache, gfp_mask);
536 if (unlikely(pc == NULL))
540 * We always charge the cgroup the mm_struct belongs to.
541 * The mm_struct's mem_cgroup changes on task migration if the
542 * thread group leader migrates. It's possible that mm is not
543 * set, if so charge the init_mm (happens for pagecache usage).
545 if (likely(!memcg)) {
547 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
548 if (unlikely(!mem)) {
550 kmem_cache_free(page_cgroup_cache, pc);
554 * For every charge from the cgroup, increment reference count
560 css_get(&memcg->css);
563 while (unlikely(res_counter_charge(&mem->res, PAGE_SIZE))) {
564 if (!(gfp_mask & __GFP_WAIT))
567 if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
571 * try_to_free_mem_cgroup_pages() might not give us a full
572 * picture of reclaim. Some pages are reclaimed and might be
573 * moved to swap cache or just unmapped from the cgroup.
574 * Check the limit again to see if the reclaim reduced the
575 * current usage of the cgroup before giving up
577 if (res_counter_check_under_limit(&mem->res))
581 mem_cgroup_out_of_memory(mem, gfp_mask);
586 pc->mem_cgroup = mem;
589 * If a page is accounted as a page cache, insert to inactive list.
590 * If anon, insert to active list.
592 if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE) {
593 pc->flags = PAGE_CGROUP_FLAG_CACHE;
594 if (page_is_file_cache(page))
595 pc->flags |= PAGE_CGROUP_FLAG_FILE;
597 pc->flags |= PAGE_CGROUP_FLAG_ACTIVE;
598 } else if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
599 pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
600 else /* MEM_CGROUP_CHARGE_TYPE_SHMEM */
601 pc->flags = PAGE_CGROUP_FLAG_CACHE | PAGE_CGROUP_FLAG_ACTIVE;
603 lock_page_cgroup(page);
604 if (unlikely(page_get_page_cgroup(page))) {
605 unlock_page_cgroup(page);
606 res_counter_uncharge(&mem->res, PAGE_SIZE);
608 kmem_cache_free(page_cgroup_cache, pc);
611 page_assign_page_cgroup(page, pc);
613 mz = page_cgroup_zoneinfo(pc);
614 spin_lock_irqsave(&mz->lru_lock, flags);
615 __mem_cgroup_add_list(mz, pc);
616 spin_unlock_irqrestore(&mz->lru_lock, flags);
618 unlock_page_cgroup(page);
623 kmem_cache_free(page_cgroup_cache, pc);
628 int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask)
630 if (mem_cgroup_subsys.disabled)
634 * If already mapped, we don't have to account.
635 * If page cache, page->mapping has address_space.
636 * But page->mapping may have out-of-use anon_vma pointer,
637 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
640 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
644 return mem_cgroup_charge_common(page, mm, gfp_mask,
645 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
648 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
651 if (mem_cgroup_subsys.disabled)
655 * Corner case handling. This is called from add_to_page_cache()
656 * in usual. But some FS (shmem) precharges this page before calling it
657 * and call add_to_page_cache() with GFP_NOWAIT.
659 * For GFP_NOWAIT case, the page may be pre-charged before calling
660 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
661 * charge twice. (It works but has to pay a bit larger cost.)
663 if (!(gfp_mask & __GFP_WAIT)) {
664 struct page_cgroup *pc;
666 lock_page_cgroup(page);
667 pc = page_get_page_cgroup(page);
669 VM_BUG_ON(pc->page != page);
670 VM_BUG_ON(!pc->mem_cgroup);
671 unlock_page_cgroup(page);
674 unlock_page_cgroup(page);
680 return mem_cgroup_charge_common(page, mm, gfp_mask,
681 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
685 * uncharge if !page_mapped(page)
688 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
690 struct page_cgroup *pc;
691 struct mem_cgroup *mem;
692 struct mem_cgroup_per_zone *mz;
695 if (mem_cgroup_subsys.disabled)
699 * Check if our page_cgroup is valid
701 lock_page_cgroup(page);
702 pc = page_get_page_cgroup(page);
706 VM_BUG_ON(pc->page != page);
708 if ((ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
709 && ((pc->flags & PAGE_CGROUP_FLAG_CACHE)
710 || page_mapped(page)))
713 mz = page_cgroup_zoneinfo(pc);
714 spin_lock_irqsave(&mz->lru_lock, flags);
715 __mem_cgroup_remove_list(mz, pc);
716 spin_unlock_irqrestore(&mz->lru_lock, flags);
718 page_assign_page_cgroup(page, NULL);
719 unlock_page_cgroup(page);
721 mem = pc->mem_cgroup;
722 res_counter_uncharge(&mem->res, PAGE_SIZE);
725 kmem_cache_free(page_cgroup_cache, pc);
728 unlock_page_cgroup(page);
731 void mem_cgroup_uncharge_page(struct page *page)
733 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
736 void mem_cgroup_uncharge_cache_page(struct page *page)
738 VM_BUG_ON(page_mapped(page));
739 VM_BUG_ON(page->mapping);
740 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
744 * Before starting migration, account against new page.
746 int mem_cgroup_prepare_migration(struct page *page, struct page *newpage)
748 struct page_cgroup *pc;
749 struct mem_cgroup *mem = NULL;
750 enum charge_type ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
753 if (mem_cgroup_subsys.disabled)
756 lock_page_cgroup(page);
757 pc = page_get_page_cgroup(page);
759 mem = pc->mem_cgroup;
761 if (pc->flags & PAGE_CGROUP_FLAG_CACHE) {
762 if (page_is_file_cache(page))
763 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
765 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
768 unlock_page_cgroup(page);
770 ret = mem_cgroup_charge_common(newpage, NULL, GFP_KERNEL,
777 /* remove redundant charge if migration failed*/
778 void mem_cgroup_end_migration(struct page *newpage)
781 * At success, page->mapping is not NULL.
782 * special rollback care is necessary when
783 * 1. at migration failure. (newpage->mapping is cleared in this case)
784 * 2. the newpage was moved but not remapped again because the task
785 * exits and the newpage is obsolete. In this case, the new page
786 * may be a swapcache. So, we just call mem_cgroup_uncharge_page()
787 * always for avoiding mess. The page_cgroup will be removed if
788 * unnecessary. File cache pages is still on radix-tree. Don't
791 if (!newpage->mapping)
792 __mem_cgroup_uncharge_common(newpage,
793 MEM_CGROUP_CHARGE_TYPE_FORCE);
794 else if (PageAnon(newpage))
795 mem_cgroup_uncharge_page(newpage);
799 * A call to try to shrink memory usage under specified resource controller.
800 * This is typically used for page reclaiming for shmem for reducing side
801 * effect of page allocation from shmem, which is used by some mem_cgroup.
803 int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
805 struct mem_cgroup *mem;
807 int retry = MEM_CGROUP_RECLAIM_RETRIES;
809 if (mem_cgroup_subsys.disabled)
815 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
816 if (unlikely(!mem)) {
824 progress = try_to_free_mem_cgroup_pages(mem, gfp_mask);
825 progress += res_counter_check_under_limit(&mem->res);
826 } while (!progress && --retry);
834 int mem_cgroup_resize_limit(struct mem_cgroup *memcg, unsigned long long val)
837 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
841 while (res_counter_set_limit(&memcg->res, val)) {
842 if (signal_pending(current)) {
850 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL);
859 * This routine traverse page_cgroup in given list and drop them all.
860 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
862 #define FORCE_UNCHARGE_BATCH (128)
863 static void mem_cgroup_force_empty_list(struct mem_cgroup *mem,
864 struct mem_cgroup_per_zone *mz,
867 struct page_cgroup *pc;
869 int count = FORCE_UNCHARGE_BATCH;
871 struct list_head *list;
873 list = &mz->lists[lru];
875 spin_lock_irqsave(&mz->lru_lock, flags);
876 while (!list_empty(list)) {
877 pc = list_entry(list->prev, struct page_cgroup, lru);
880 spin_unlock_irqrestore(&mz->lru_lock, flags);
882 * Check if this page is on LRU. !LRU page can be found
883 * if it's under page migration.
886 __mem_cgroup_uncharge_common(page,
887 MEM_CGROUP_CHARGE_TYPE_FORCE);
890 count = FORCE_UNCHARGE_BATCH;
895 spin_lock_irqsave(&mz->lru_lock, flags);
897 spin_unlock_irqrestore(&mz->lru_lock, flags);
901 * make mem_cgroup's charge to be 0 if there is no task.
902 * This enables deleting this mem_cgroup.
904 static int mem_cgroup_force_empty(struct mem_cgroup *mem)
911 * page reclaim code (kswapd etc..) will move pages between
912 * active_list <-> inactive_list while we don't take a lock.
913 * So, we have to do loop here until all lists are empty.
915 while (mem->res.usage > 0) {
916 if (atomic_read(&mem->css.cgroup->count) > 0)
918 for_each_node_state(node, N_POSSIBLE)
919 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
920 struct mem_cgroup_per_zone *mz;
922 mz = mem_cgroup_zoneinfo(mem, node, zid);
924 mem_cgroup_force_empty_list(mem, mz, l);
933 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
935 return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res,
939 * The user of this function is...
942 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
945 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
946 unsigned long long val;
949 switch (cft->private) {
951 /* This function does all necessary parse...reuse it */
952 ret = res_counter_memparse_write_strategy(buffer, &val);
954 ret = mem_cgroup_resize_limit(memcg, val);
957 ret = -EINVAL; /* should be BUG() ? */
963 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
965 struct mem_cgroup *mem;
967 mem = mem_cgroup_from_cont(cont);
970 res_counter_reset_max(&mem->res);
973 res_counter_reset_failcnt(&mem->res);
979 static int mem_force_empty_write(struct cgroup *cont, unsigned int event)
981 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont));
984 static const struct mem_cgroup_stat_desc {
987 } mem_cgroup_stat_desc[] = {
988 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
989 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
990 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
991 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
994 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
995 struct cgroup_map_cb *cb)
997 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
998 struct mem_cgroup_stat *stat = &mem_cont->stat;
1001 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1004 val = mem_cgroup_read_stat(stat, i);
1005 val *= mem_cgroup_stat_desc[i].unit;
1006 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
1008 /* showing # of active pages */
1010 unsigned long active_anon, inactive_anon;
1011 unsigned long active_file, inactive_file;
1012 unsigned long unevictable;
1014 inactive_anon = mem_cgroup_get_all_zonestat(mem_cont,
1016 active_anon = mem_cgroup_get_all_zonestat(mem_cont,
1018 inactive_file = mem_cgroup_get_all_zonestat(mem_cont,
1020 active_file = mem_cgroup_get_all_zonestat(mem_cont,
1022 unevictable = mem_cgroup_get_all_zonestat(mem_cont,
1025 cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE);
1026 cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE);
1027 cb->fill(cb, "active_file", (active_file) * PAGE_SIZE);
1028 cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE);
1029 cb->fill(cb, "unevictable", unevictable * PAGE_SIZE);
1035 static struct cftype mem_cgroup_files[] = {
1037 .name = "usage_in_bytes",
1038 .private = RES_USAGE,
1039 .read_u64 = mem_cgroup_read,
1042 .name = "max_usage_in_bytes",
1043 .private = RES_MAX_USAGE,
1044 .trigger = mem_cgroup_reset,
1045 .read_u64 = mem_cgroup_read,
1048 .name = "limit_in_bytes",
1049 .private = RES_LIMIT,
1050 .write_string = mem_cgroup_write,
1051 .read_u64 = mem_cgroup_read,
1055 .private = RES_FAILCNT,
1056 .trigger = mem_cgroup_reset,
1057 .read_u64 = mem_cgroup_read,
1060 .name = "force_empty",
1061 .trigger = mem_force_empty_write,
1065 .read_map = mem_control_stat_show,
1069 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1071 struct mem_cgroup_per_node *pn;
1072 struct mem_cgroup_per_zone *mz;
1074 int zone, tmp = node;
1076 * This routine is called against possible nodes.
1077 * But it's BUG to call kmalloc() against offline node.
1079 * TODO: this routine can waste much memory for nodes which will
1080 * never be onlined. It's better to use memory hotplug callback
1083 if (!node_state(node, N_NORMAL_MEMORY))
1085 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
1089 mem->info.nodeinfo[node] = pn;
1090 memset(pn, 0, sizeof(*pn));
1092 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
1093 mz = &pn->zoneinfo[zone];
1094 spin_lock_init(&mz->lru_lock);
1096 INIT_LIST_HEAD(&mz->lists[l]);
1101 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1103 kfree(mem->info.nodeinfo[node]);
1106 static struct mem_cgroup *mem_cgroup_alloc(void)
1108 struct mem_cgroup *mem;
1110 if (sizeof(*mem) < PAGE_SIZE)
1111 mem = kmalloc(sizeof(*mem), GFP_KERNEL);
1113 mem = vmalloc(sizeof(*mem));
1116 memset(mem, 0, sizeof(*mem));
1120 static void mem_cgroup_free(struct mem_cgroup *mem)
1122 if (sizeof(*mem) < PAGE_SIZE)
1129 static struct cgroup_subsys_state *
1130 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
1132 struct mem_cgroup *mem;
1135 if (unlikely((cont->parent) == NULL)) {
1136 mem = &init_mem_cgroup;
1137 page_cgroup_cache = KMEM_CACHE(page_cgroup, SLAB_PANIC);
1139 mem = mem_cgroup_alloc();
1141 return ERR_PTR(-ENOMEM);
1144 res_counter_init(&mem->res);
1146 for_each_node_state(node, N_POSSIBLE)
1147 if (alloc_mem_cgroup_per_zone_info(mem, node))
1152 for_each_node_state(node, N_POSSIBLE)
1153 free_mem_cgroup_per_zone_info(mem, node);
1154 if (cont->parent != NULL)
1155 mem_cgroup_free(mem);
1156 return ERR_PTR(-ENOMEM);
1159 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1160 struct cgroup *cont)
1162 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1163 mem_cgroup_force_empty(mem);
1166 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
1167 struct cgroup *cont)
1170 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1172 for_each_node_state(node, N_POSSIBLE)
1173 free_mem_cgroup_per_zone_info(mem, node);
1175 mem_cgroup_free(mem_cgroup_from_cont(cont));
1178 static int mem_cgroup_populate(struct cgroup_subsys *ss,
1179 struct cgroup *cont)
1181 return cgroup_add_files(cont, ss, mem_cgroup_files,
1182 ARRAY_SIZE(mem_cgroup_files));
1185 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
1186 struct cgroup *cont,
1187 struct cgroup *old_cont,
1188 struct task_struct *p)
1190 struct mm_struct *mm;
1191 struct mem_cgroup *mem, *old_mem;
1193 mm = get_task_mm(p);
1197 mem = mem_cgroup_from_cont(cont);
1198 old_mem = mem_cgroup_from_cont(old_cont);
1201 * Only thread group leaders are allowed to migrate, the mm_struct is
1202 * in effect owned by the leader
1204 if (!thread_group_leader(p))
1211 struct cgroup_subsys mem_cgroup_subsys = {
1213 .subsys_id = mem_cgroup_subsys_id,
1214 .create = mem_cgroup_create,
1215 .pre_destroy = mem_cgroup_pre_destroy,
1216 .destroy = mem_cgroup_destroy,
1217 .populate = mem_cgroup_populate,
1218 .attach = mem_cgroup_move_task,