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/mutex.h>
31 #include <linux/slab.h>
32 #include <linux/swap.h>
33 #include <linux/spinlock.h>
35 #include <linux/seq_file.h>
36 #include <linux/vmalloc.h>
37 #include <linux/mm_inline.h>
38 #include <linux/page_cgroup.h>
41 #include <asm/uaccess.h>
43 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
44 #define MEM_CGROUP_RECLAIM_RETRIES 5
46 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
47 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 0 */
48 int do_swap_account __read_mostly;
49 static int really_do_swap_account __initdata = 1; /* for remember boot option*/
51 #define do_swap_account (0)
56 * Statistics for memory cgroup.
58 enum mem_cgroup_stat_index {
60 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
62 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
63 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
64 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
65 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
67 MEM_CGROUP_STAT_NSTATS,
70 struct mem_cgroup_stat_cpu {
71 s64 count[MEM_CGROUP_STAT_NSTATS];
72 } ____cacheline_aligned_in_smp;
74 struct mem_cgroup_stat {
75 struct mem_cgroup_stat_cpu cpustat[0];
79 * For accounting under irq disable, no need for increment preempt count.
81 static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
82 enum mem_cgroup_stat_index idx, int val)
84 stat->count[idx] += val;
87 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
88 enum mem_cgroup_stat_index idx)
92 for_each_possible_cpu(cpu)
93 ret += stat->cpustat[cpu].count[idx];
98 * per-zone information in memory controller.
100 struct mem_cgroup_per_zone {
102 * spin_lock to protect the per cgroup LRU
104 struct list_head lists[NR_LRU_LISTS];
105 unsigned long count[NR_LRU_LISTS];
107 struct zone_reclaim_stat reclaim_stat;
109 /* Macro for accessing counter */
110 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
112 struct mem_cgroup_per_node {
113 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
116 struct mem_cgroup_lru_info {
117 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
121 * The memory controller data structure. The memory controller controls both
122 * page cache and RSS per cgroup. We would eventually like to provide
123 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
124 * to help the administrator determine what knobs to tune.
126 * TODO: Add a water mark for the memory controller. Reclaim will begin when
127 * we hit the water mark. May be even add a low water mark, such that
128 * no reclaim occurs from a cgroup at it's low water mark, this is
129 * a feature that will be implemented much later in the future.
132 struct cgroup_subsys_state css;
134 * the counter to account for memory usage
136 struct res_counter res;
138 * the counter to account for mem+swap usage.
140 struct res_counter memsw;
142 * Per cgroup active and inactive list, similar to the
143 * per zone LRU lists.
145 struct mem_cgroup_lru_info info;
147 int prev_priority; /* for recording reclaim priority */
150 * While reclaiming in a hiearchy, we cache the last child we
151 * reclaimed from. Protected by cgroup_lock()
153 struct mem_cgroup *last_scanned_child;
155 * Should the accounting and control be hierarchical, per subtree?
158 unsigned long last_oom_jiffies;
162 unsigned int inactive_ratio;
165 * statistics. This must be placed at the end of memcg.
167 struct mem_cgroup_stat stat;
171 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
172 MEM_CGROUP_CHARGE_TYPE_MAPPED,
173 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
174 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
175 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
179 /* only for here (for easy reading.) */
180 #define PCGF_CACHE (1UL << PCG_CACHE)
181 #define PCGF_USED (1UL << PCG_USED)
182 #define PCGF_LOCK (1UL << PCG_LOCK)
183 static const unsigned long
184 pcg_default_flags[NR_CHARGE_TYPE] = {
185 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
186 PCGF_USED | PCGF_LOCK, /* Anon */
187 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
191 /* for encoding cft->private value on file */
194 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
195 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
196 #define MEMFILE_ATTR(val) ((val) & 0xffff)
198 static void mem_cgroup_get(struct mem_cgroup *mem);
199 static void mem_cgroup_put(struct mem_cgroup *mem);
201 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
202 struct page_cgroup *pc,
205 int val = (charge)? 1 : -1;
206 struct mem_cgroup_stat *stat = &mem->stat;
207 struct mem_cgroup_stat_cpu *cpustat;
210 cpustat = &stat->cpustat[cpu];
211 if (PageCgroupCache(pc))
212 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
214 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
217 __mem_cgroup_stat_add_safe(cpustat,
218 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
220 __mem_cgroup_stat_add_safe(cpustat,
221 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
225 static struct mem_cgroup_per_zone *
226 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
228 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
231 static struct mem_cgroup_per_zone *
232 page_cgroup_zoneinfo(struct page_cgroup *pc)
234 struct mem_cgroup *mem = pc->mem_cgroup;
235 int nid = page_cgroup_nid(pc);
236 int zid = page_cgroup_zid(pc);
241 return mem_cgroup_zoneinfo(mem, nid, zid);
244 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
248 struct mem_cgroup_per_zone *mz;
251 for_each_online_node(nid)
252 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
253 mz = mem_cgroup_zoneinfo(mem, nid, zid);
254 total += MEM_CGROUP_ZSTAT(mz, idx);
259 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
261 return container_of(cgroup_subsys_state(cont,
262 mem_cgroup_subsys_id), struct mem_cgroup,
266 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
269 * mm_update_next_owner() may clear mm->owner to NULL
270 * if it races with swapoff, page migration, etc.
271 * So this can be called with p == NULL.
276 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
277 struct mem_cgroup, css);
281 * Following LRU functions are allowed to be used without PCG_LOCK.
282 * Operations are called by routine of global LRU independently from memcg.
283 * What we have to take care of here is validness of pc->mem_cgroup.
285 * Changes to pc->mem_cgroup happens when
288 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
289 * It is added to LRU before charge.
290 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
291 * When moving account, the page is not on LRU. It's isolated.
294 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
296 struct page_cgroup *pc;
297 struct mem_cgroup *mem;
298 struct mem_cgroup_per_zone *mz;
300 if (mem_cgroup_disabled())
302 pc = lookup_page_cgroup(page);
303 /* can happen while we handle swapcache. */
304 if (list_empty(&pc->lru))
306 mz = page_cgroup_zoneinfo(pc);
307 mem = pc->mem_cgroup;
308 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
309 list_del_init(&pc->lru);
313 void mem_cgroup_del_lru(struct page *page)
315 mem_cgroup_del_lru_list(page, page_lru(page));
318 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
320 struct mem_cgroup_per_zone *mz;
321 struct page_cgroup *pc;
323 if (mem_cgroup_disabled())
326 pc = lookup_page_cgroup(page);
328 /* unused page is not rotated. */
329 if (!PageCgroupUsed(pc))
331 mz = page_cgroup_zoneinfo(pc);
332 list_move(&pc->lru, &mz->lists[lru]);
335 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
337 struct page_cgroup *pc;
338 struct mem_cgroup_per_zone *mz;
340 if (mem_cgroup_disabled())
342 pc = lookup_page_cgroup(page);
343 /* barrier to sync with "charge" */
345 if (!PageCgroupUsed(pc))
348 mz = page_cgroup_zoneinfo(pc);
349 MEM_CGROUP_ZSTAT(mz, lru) += 1;
350 list_add(&pc->lru, &mz->lists[lru]);
353 * To add swapcache into LRU. Be careful to all this function.
354 * zone->lru_lock shouldn't be held and irq must not be disabled.
356 static void mem_cgroup_lru_fixup(struct page *page)
358 if (!isolate_lru_page(page))
359 putback_lru_page(page);
362 void mem_cgroup_move_lists(struct page *page,
363 enum lru_list from, enum lru_list to)
365 if (mem_cgroup_disabled())
367 mem_cgroup_del_lru_list(page, from);
368 mem_cgroup_add_lru_list(page, to);
371 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
376 ret = task->mm && mm_match_cgroup(task->mm, mem);
382 * Calculate mapped_ratio under memory controller. This will be used in
383 * vmscan.c for deteremining we have to reclaim mapped pages.
385 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
390 * usage is recorded in bytes. But, here, we assume the number of
391 * physical pages can be represented by "long" on any arch.
393 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
394 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
395 return (int)((rss * 100L) / total);
399 * prev_priority control...this will be used in memory reclaim path.
401 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
403 return mem->prev_priority;
406 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
408 if (priority < mem->prev_priority)
409 mem->prev_priority = priority;
412 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
414 mem->prev_priority = priority;
418 * Calculate # of pages to be scanned in this priority/zone.
421 * priority starts from "DEF_PRIORITY" and decremented in each loop.
422 * (see include/linux/mmzone.h)
425 long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone,
426 int priority, enum lru_list lru)
429 int nid = zone->zone_pgdat->node_id;
430 int zid = zone_idx(zone);
431 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
433 nr_pages = MEM_CGROUP_ZSTAT(mz, lru);
435 return (nr_pages >> priority);
438 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone)
440 unsigned long active;
441 unsigned long inactive;
443 inactive = mem_cgroup_get_all_zonestat(memcg, LRU_INACTIVE_ANON);
444 active = mem_cgroup_get_all_zonestat(memcg, LRU_ACTIVE_ANON);
446 if (inactive * memcg->inactive_ratio < active)
452 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
456 int nid = zone->zone_pgdat->node_id;
457 int zid = zone_idx(zone);
458 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
460 return MEM_CGROUP_ZSTAT(mz, lru);
463 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
466 int nid = zone->zone_pgdat->node_id;
467 int zid = zone_idx(zone);
468 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
470 return &mz->reclaim_stat;
473 struct zone_reclaim_stat *
474 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
476 struct page_cgroup *pc;
477 struct mem_cgroup_per_zone *mz;
479 if (mem_cgroup_disabled())
482 pc = lookup_page_cgroup(page);
483 mz = page_cgroup_zoneinfo(pc);
487 return &mz->reclaim_stat;
490 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
491 struct list_head *dst,
492 unsigned long *scanned, int order,
493 int mode, struct zone *z,
494 struct mem_cgroup *mem_cont,
495 int active, int file)
497 unsigned long nr_taken = 0;
501 struct list_head *src;
502 struct page_cgroup *pc, *tmp;
503 int nid = z->zone_pgdat->node_id;
504 int zid = zone_idx(z);
505 struct mem_cgroup_per_zone *mz;
506 int lru = LRU_FILE * !!file + !!active;
509 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
510 src = &mz->lists[lru];
513 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
514 if (scan >= nr_to_scan)
518 if (unlikely(!PageCgroupUsed(pc)))
520 if (unlikely(!PageLRU(page)))
524 if (__isolate_lru_page(page, mode, file) == 0) {
525 list_move(&page->lru, dst);
534 #define mem_cgroup_from_res_counter(counter, member) \
535 container_of(counter, struct mem_cgroup, member)
538 * This routine finds the DFS walk successor. This routine should be
539 * called with cgroup_mutex held
541 static struct mem_cgroup *
542 mem_cgroup_get_next_node(struct mem_cgroup *curr, struct mem_cgroup *root_mem)
544 struct cgroup *cgroup, *curr_cgroup, *root_cgroup;
546 curr_cgroup = curr->css.cgroup;
547 root_cgroup = root_mem->css.cgroup;
549 if (!list_empty(&curr_cgroup->children)) {
551 * Walk down to children
553 mem_cgroup_put(curr);
554 cgroup = list_entry(curr_cgroup->children.next,
555 struct cgroup, sibling);
556 curr = mem_cgroup_from_cont(cgroup);
557 mem_cgroup_get(curr);
562 if (curr_cgroup == root_cgroup) {
563 mem_cgroup_put(curr);
565 mem_cgroup_get(curr);
572 if (curr_cgroup->sibling.next != &curr_cgroup->parent->children) {
573 mem_cgroup_put(curr);
574 cgroup = list_entry(curr_cgroup->sibling.next, struct cgroup,
576 curr = mem_cgroup_from_cont(cgroup);
577 mem_cgroup_get(curr);
582 * Go up to next parent and next parent's sibling if need be
584 curr_cgroup = curr_cgroup->parent;
588 root_mem->last_scanned_child = curr;
593 * Visit the first child (need not be the first child as per the ordering
594 * of the cgroup list, since we track last_scanned_child) of @mem and use
595 * that to reclaim free pages from.
597 static struct mem_cgroup *
598 mem_cgroup_get_first_node(struct mem_cgroup *root_mem)
600 struct cgroup *cgroup;
601 struct mem_cgroup *ret;
602 bool obsolete = (root_mem->last_scanned_child &&
603 root_mem->last_scanned_child->obsolete);
606 * Scan all children under the mem_cgroup mem
609 if (list_empty(&root_mem->css.cgroup->children)) {
614 if (!root_mem->last_scanned_child || obsolete) {
617 mem_cgroup_put(root_mem->last_scanned_child);
619 cgroup = list_first_entry(&root_mem->css.cgroup->children,
620 struct cgroup, sibling);
621 ret = mem_cgroup_from_cont(cgroup);
624 ret = mem_cgroup_get_next_node(root_mem->last_scanned_child,
628 root_mem->last_scanned_child = ret;
633 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
635 if (do_swap_account) {
636 if (res_counter_check_under_limit(&mem->res) &&
637 res_counter_check_under_limit(&mem->memsw))
640 if (res_counter_check_under_limit(&mem->res))
646 * Dance down the hierarchy if needed to reclaim memory. We remember the
647 * last child we reclaimed from, so that we don't end up penalizing
648 * one child extensively based on its position in the children list.
650 * root_mem is the original ancestor that we've been reclaim from.
652 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
653 gfp_t gfp_mask, bool noswap)
655 struct mem_cgroup *next_mem;
659 * Reclaim unconditionally and don't check for return value.
660 * We need to reclaim in the current group and down the tree.
661 * One might think about checking for children before reclaiming,
662 * but there might be left over accounting, even after children
665 ret = try_to_free_mem_cgroup_pages(root_mem, gfp_mask, noswap);
666 if (mem_cgroup_check_under_limit(root_mem))
668 if (!root_mem->use_hierarchy)
671 next_mem = mem_cgroup_get_first_node(root_mem);
673 while (next_mem != root_mem) {
674 if (next_mem->obsolete) {
675 mem_cgroup_put(next_mem);
677 next_mem = mem_cgroup_get_first_node(root_mem);
681 ret = try_to_free_mem_cgroup_pages(next_mem, gfp_mask, noswap);
682 if (mem_cgroup_check_under_limit(root_mem))
685 next_mem = mem_cgroup_get_next_node(next_mem, root_mem);
691 bool mem_cgroup_oom_called(struct task_struct *task)
694 struct mem_cgroup *mem;
695 struct mm_struct *mm;
701 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
702 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
708 * Unlike exported interface, "oom" parameter is added. if oom==true,
709 * oom-killer can be invoked.
711 static int __mem_cgroup_try_charge(struct mm_struct *mm,
712 gfp_t gfp_mask, struct mem_cgroup **memcg,
715 struct mem_cgroup *mem, *mem_over_limit;
716 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
717 struct res_counter *fail_res;
719 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
720 /* Don't account this! */
726 * We always charge the cgroup the mm_struct belongs to.
727 * The mm_struct's mem_cgroup changes on task migration if the
728 * thread group leader migrates. It's possible that mm is not
729 * set, if so charge the init_mm (happens for pagecache usage).
731 if (likely(!*memcg)) {
733 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
734 if (unlikely(!mem)) {
739 * For every charge from the cgroup, increment reference count
753 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res);
755 if (!do_swap_account)
757 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
761 /* mem+swap counter fails */
762 res_counter_uncharge(&mem->res, PAGE_SIZE);
764 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
767 /* mem counter fails */
768 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
771 if (!(gfp_mask & __GFP_WAIT))
774 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
778 * try_to_free_mem_cgroup_pages() might not give us a full
779 * picture of reclaim. Some pages are reclaimed and might be
780 * moved to swap cache or just unmapped from the cgroup.
781 * Check the limit again to see if the reclaim reduced the
782 * current usage of the cgroup before giving up
785 if (mem_cgroup_check_under_limit(mem_over_limit))
790 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
791 mem_over_limit->last_oom_jiffies = jiffies;
803 * mem_cgroup_try_charge - get charge of PAGE_SIZE.
804 * @mm: an mm_struct which is charged against. (when *memcg is NULL)
805 * @gfp_mask: gfp_mask for reclaim.
806 * @memcg: a pointer to memory cgroup which is charged against.
808 * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated
809 * memory cgroup from @mm is got and stored in *memcg.
811 * Returns 0 if success. -ENOMEM at failure.
812 * This call can invoke OOM-Killer.
815 int mem_cgroup_try_charge(struct mm_struct *mm,
816 gfp_t mask, struct mem_cgroup **memcg)
818 return __mem_cgroup_try_charge(mm, mask, memcg, true);
822 * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be
823 * USED state. If already USED, uncharge and return.
826 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
827 struct page_cgroup *pc,
828 enum charge_type ctype)
830 /* try_charge() can return NULL to *memcg, taking care of it. */
834 lock_page_cgroup(pc);
835 if (unlikely(PageCgroupUsed(pc))) {
836 unlock_page_cgroup(pc);
837 res_counter_uncharge(&mem->res, PAGE_SIZE);
839 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
843 pc->mem_cgroup = mem;
845 pc->flags = pcg_default_flags[ctype];
847 mem_cgroup_charge_statistics(mem, pc, true);
849 unlock_page_cgroup(pc);
853 * mem_cgroup_move_account - move account of the page
854 * @pc: page_cgroup of the page.
855 * @from: mem_cgroup which the page is moved from.
856 * @to: mem_cgroup which the page is moved to. @from != @to.
858 * The caller must confirm following.
859 * - page is not on LRU (isolate_page() is useful.)
861 * returns 0 at success,
862 * returns -EBUSY when lock is busy or "pc" is unstable.
864 * This function does "uncharge" from old cgroup but doesn't do "charge" to
865 * new cgroup. It should be done by a caller.
868 static int mem_cgroup_move_account(struct page_cgroup *pc,
869 struct mem_cgroup *from, struct mem_cgroup *to)
871 struct mem_cgroup_per_zone *from_mz, *to_mz;
875 VM_BUG_ON(from == to);
876 VM_BUG_ON(PageLRU(pc->page));
878 nid = page_cgroup_nid(pc);
879 zid = page_cgroup_zid(pc);
880 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
881 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
883 if (!trylock_page_cgroup(pc))
886 if (!PageCgroupUsed(pc))
889 if (pc->mem_cgroup != from)
893 res_counter_uncharge(&from->res, PAGE_SIZE);
894 mem_cgroup_charge_statistics(from, pc, false);
896 res_counter_uncharge(&from->memsw, PAGE_SIZE);
898 mem_cgroup_charge_statistics(to, pc, true);
902 unlock_page_cgroup(pc);
907 * move charges to its parent.
910 static int mem_cgroup_move_parent(struct page_cgroup *pc,
911 struct mem_cgroup *child,
914 struct page *page = pc->page;
915 struct cgroup *cg = child->css.cgroup;
916 struct cgroup *pcg = cg->parent;
917 struct mem_cgroup *parent;
925 parent = mem_cgroup_from_cont(pcg);
928 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
932 if (!get_page_unless_zero(page))
935 ret = isolate_lru_page(page);
940 ret = mem_cgroup_move_account(pc, child, parent);
942 /* drop extra refcnt by try_charge() (move_account increment one) */
943 css_put(&parent->css);
944 putback_lru_page(page);
949 /* uncharge if move fails */
951 res_counter_uncharge(&parent->res, PAGE_SIZE);
953 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
959 * Charge the memory controller for page usage.
961 * 0 if the charge was successful
962 * < 0 if the cgroup is over its limit
964 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
965 gfp_t gfp_mask, enum charge_type ctype,
966 struct mem_cgroup *memcg)
968 struct mem_cgroup *mem;
969 struct page_cgroup *pc;
972 pc = lookup_page_cgroup(page);
973 /* can happen at boot */
979 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
983 __mem_cgroup_commit_charge(mem, pc, ctype);
987 int mem_cgroup_newpage_charge(struct page *page,
988 struct mm_struct *mm, gfp_t gfp_mask)
990 if (mem_cgroup_disabled())
992 if (PageCompound(page))
995 * If already mapped, we don't have to account.
996 * If page cache, page->mapping has address_space.
997 * But page->mapping may have out-of-use anon_vma pointer,
998 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1001 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1005 return mem_cgroup_charge_common(page, mm, gfp_mask,
1006 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1009 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1012 if (mem_cgroup_disabled())
1014 if (PageCompound(page))
1017 * Corner case handling. This is called from add_to_page_cache()
1018 * in usual. But some FS (shmem) precharges this page before calling it
1019 * and call add_to_page_cache() with GFP_NOWAIT.
1021 * For GFP_NOWAIT case, the page may be pre-charged before calling
1022 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1023 * charge twice. (It works but has to pay a bit larger cost.)
1025 if (!(gfp_mask & __GFP_WAIT)) {
1026 struct page_cgroup *pc;
1029 pc = lookup_page_cgroup(page);
1032 lock_page_cgroup(pc);
1033 if (PageCgroupUsed(pc)) {
1034 unlock_page_cgroup(pc);
1037 unlock_page_cgroup(pc);
1043 if (page_is_file_cache(page))
1044 return mem_cgroup_charge_common(page, mm, gfp_mask,
1045 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1047 return mem_cgroup_charge_common(page, mm, gfp_mask,
1048 MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL);
1051 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1053 gfp_t mask, struct mem_cgroup **ptr)
1055 struct mem_cgroup *mem;
1058 if (mem_cgroup_disabled())
1061 if (!do_swap_account)
1065 * A racing thread's fault, or swapoff, may have already updated
1066 * the pte, and even removed page from swap cache: return success
1067 * to go on to do_swap_page()'s pte_same() test, which should fail.
1069 if (!PageSwapCache(page))
1072 ent.val = page_private(page);
1074 mem = lookup_swap_cgroup(ent);
1075 if (!mem || mem->obsolete)
1078 return __mem_cgroup_try_charge(NULL, mask, ptr, true);
1082 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1087 int mem_cgroup_cache_charge_swapin(struct page *page,
1088 struct mm_struct *mm, gfp_t mask, bool locked)
1092 if (mem_cgroup_disabled())
1099 * If not locked, the page can be dropped from SwapCache until
1102 if (PageSwapCache(page)) {
1103 struct mem_cgroup *mem = NULL;
1106 ent.val = page_private(page);
1107 if (do_swap_account) {
1108 mem = lookup_swap_cgroup(ent);
1109 if (mem && mem->obsolete)
1114 ret = mem_cgroup_charge_common(page, mm, mask,
1115 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1117 if (!ret && do_swap_account) {
1118 /* avoid double counting */
1119 mem = swap_cgroup_record(ent, NULL);
1121 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1122 mem_cgroup_put(mem);
1128 /* add this page(page_cgroup) to the LRU we want. */
1129 mem_cgroup_lru_fixup(page);
1135 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1137 struct page_cgroup *pc;
1139 if (mem_cgroup_disabled())
1143 pc = lookup_page_cgroup(page);
1144 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1146 * Now swap is on-memory. This means this page may be
1147 * counted both as mem and swap....double count.
1148 * Fix it by uncharging from memsw. This SwapCache is stable
1149 * because we're still under lock_page().
1151 if (do_swap_account) {
1152 swp_entry_t ent = {.val = page_private(page)};
1153 struct mem_cgroup *memcg;
1154 memcg = swap_cgroup_record(ent, NULL);
1156 /* If memcg is obsolete, memcg can be != ptr */
1157 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1158 mem_cgroup_put(memcg);
1162 /* add this page(page_cgroup) to the LRU we want. */
1163 mem_cgroup_lru_fixup(page);
1166 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1168 if (mem_cgroup_disabled())
1172 res_counter_uncharge(&mem->res, PAGE_SIZE);
1173 if (do_swap_account)
1174 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1180 * uncharge if !page_mapped(page)
1182 static struct mem_cgroup *
1183 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1185 struct page_cgroup *pc;
1186 struct mem_cgroup *mem = NULL;
1187 struct mem_cgroup_per_zone *mz;
1189 if (mem_cgroup_disabled())
1192 if (PageSwapCache(page))
1196 * Check if our page_cgroup is valid
1198 pc = lookup_page_cgroup(page);
1199 if (unlikely(!pc || !PageCgroupUsed(pc)))
1202 lock_page_cgroup(pc);
1204 mem = pc->mem_cgroup;
1206 if (!PageCgroupUsed(pc))
1210 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1211 if (page_mapped(page))
1214 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1215 if (!PageAnon(page)) { /* Shared memory */
1216 if (page->mapping && !page_is_file_cache(page))
1218 } else if (page_mapped(page)) /* Anon */
1225 res_counter_uncharge(&mem->res, PAGE_SIZE);
1226 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1227 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1229 mem_cgroup_charge_statistics(mem, pc, false);
1230 ClearPageCgroupUsed(pc);
1232 mz = page_cgroup_zoneinfo(pc);
1233 unlock_page_cgroup(pc);
1235 /* at swapout, this memcg will be accessed to record to swap */
1236 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1242 unlock_page_cgroup(pc);
1246 void mem_cgroup_uncharge_page(struct page *page)
1249 if (page_mapped(page))
1251 if (page->mapping && !PageAnon(page))
1253 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1256 void mem_cgroup_uncharge_cache_page(struct page *page)
1258 VM_BUG_ON(page_mapped(page));
1259 VM_BUG_ON(page->mapping);
1260 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1264 * called from __delete_from_swap_cache() and drop "page" account.
1265 * memcg information is recorded to swap_cgroup of "ent"
1267 void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
1269 struct mem_cgroup *memcg;
1271 memcg = __mem_cgroup_uncharge_common(page,
1272 MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
1273 /* record memcg information */
1274 if (do_swap_account && memcg) {
1275 swap_cgroup_record(ent, memcg);
1276 mem_cgroup_get(memcg);
1279 css_put(&memcg->css);
1282 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1284 * called from swap_entry_free(). remove record in swap_cgroup and
1285 * uncharge "memsw" account.
1287 void mem_cgroup_uncharge_swap(swp_entry_t ent)
1289 struct mem_cgroup *memcg;
1291 if (!do_swap_account)
1294 memcg = swap_cgroup_record(ent, NULL);
1296 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1297 mem_cgroup_put(memcg);
1303 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1306 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1308 struct page_cgroup *pc;
1309 struct mem_cgroup *mem = NULL;
1312 if (mem_cgroup_disabled())
1315 pc = lookup_page_cgroup(page);
1316 lock_page_cgroup(pc);
1317 if (PageCgroupUsed(pc)) {
1318 mem = pc->mem_cgroup;
1321 unlock_page_cgroup(pc);
1324 ret = mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem);
1331 /* remove redundant charge if migration failed*/
1332 void mem_cgroup_end_migration(struct mem_cgroup *mem,
1333 struct page *oldpage, struct page *newpage)
1335 struct page *target, *unused;
1336 struct page_cgroup *pc;
1337 enum charge_type ctype;
1342 /* at migration success, oldpage->mapping is NULL. */
1343 if (oldpage->mapping) {
1351 if (PageAnon(target))
1352 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1353 else if (page_is_file_cache(target))
1354 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1356 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1358 /* unused page is not on radix-tree now. */
1360 __mem_cgroup_uncharge_common(unused, ctype);
1362 pc = lookup_page_cgroup(target);
1364 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1365 * So, double-counting is effectively avoided.
1367 __mem_cgroup_commit_charge(mem, pc, ctype);
1370 * Both of oldpage and newpage are still under lock_page().
1371 * Then, we don't have to care about race in radix-tree.
1372 * But we have to be careful that this page is unmapped or not.
1374 * There is a case for !page_mapped(). At the start of
1375 * migration, oldpage was mapped. But now, it's zapped.
1376 * But we know *target* page is not freed/reused under us.
1377 * mem_cgroup_uncharge_page() does all necessary checks.
1379 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1380 mem_cgroup_uncharge_page(target);
1384 * A call to try to shrink memory usage under specified resource controller.
1385 * This is typically used for page reclaiming for shmem for reducing side
1386 * effect of page allocation from shmem, which is used by some mem_cgroup.
1388 int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
1390 struct mem_cgroup *mem;
1392 int retry = MEM_CGROUP_RECLAIM_RETRIES;
1394 if (mem_cgroup_disabled())
1400 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
1401 if (unlikely(!mem)) {
1409 progress = try_to_free_mem_cgroup_pages(mem, gfp_mask, true);
1410 progress += mem_cgroup_check_under_limit(mem);
1411 } while (!progress && --retry);
1420 * The inactive anon list should be small enough that the VM never has to
1421 * do too much work, but large enough that each inactive page has a chance
1422 * to be referenced again before it is swapped out.
1424 * this calculation is straightforward porting from
1425 * page_alloc.c::setup_per_zone_inactive_ratio().
1426 * it describe more detail.
1428 static void mem_cgroup_set_inactive_ratio(struct mem_cgroup *memcg)
1430 unsigned int gb, ratio;
1432 gb = res_counter_read_u64(&memcg->res, RES_LIMIT) >> 30;
1434 ratio = int_sqrt(10 * gb);
1438 memcg->inactive_ratio = ratio;
1442 static DEFINE_MUTEX(set_limit_mutex);
1444 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
1445 unsigned long long val)
1448 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1453 while (retry_count) {
1454 if (signal_pending(current)) {
1459 * Rather than hide all in some function, I do this in
1460 * open coded manner. You see what this really does.
1461 * We have to guarantee mem->res.limit < mem->memsw.limit.
1463 mutex_lock(&set_limit_mutex);
1464 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1465 if (memswlimit < val) {
1467 mutex_unlock(&set_limit_mutex);
1470 ret = res_counter_set_limit(&memcg->res, val);
1471 mutex_unlock(&set_limit_mutex);
1476 progress = try_to_free_mem_cgroup_pages(memcg,
1478 if (!progress) retry_count--;
1482 mem_cgroup_set_inactive_ratio(memcg);
1487 int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1488 unsigned long long val)
1490 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1491 u64 memlimit, oldusage, curusage;
1494 if (!do_swap_account)
1497 while (retry_count) {
1498 if (signal_pending(current)) {
1503 * Rather than hide all in some function, I do this in
1504 * open coded manner. You see what this really does.
1505 * We have to guarantee mem->res.limit < mem->memsw.limit.
1507 mutex_lock(&set_limit_mutex);
1508 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1509 if (memlimit > val) {
1511 mutex_unlock(&set_limit_mutex);
1514 ret = res_counter_set_limit(&memcg->memsw, val);
1515 mutex_unlock(&set_limit_mutex);
1520 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1521 try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, true);
1522 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1523 if (curusage >= oldusage)
1530 * This routine traverse page_cgroup in given list and drop them all.
1531 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1533 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
1534 int node, int zid, enum lru_list lru)
1537 struct mem_cgroup_per_zone *mz;
1538 struct page_cgroup *pc, *busy;
1539 unsigned long flags, loop;
1540 struct list_head *list;
1543 zone = &NODE_DATA(node)->node_zones[zid];
1544 mz = mem_cgroup_zoneinfo(mem, node, zid);
1545 list = &mz->lists[lru];
1547 loop = MEM_CGROUP_ZSTAT(mz, lru);
1548 /* give some margin against EBUSY etc...*/
1553 spin_lock_irqsave(&zone->lru_lock, flags);
1554 if (list_empty(list)) {
1555 spin_unlock_irqrestore(&zone->lru_lock, flags);
1558 pc = list_entry(list->prev, struct page_cgroup, lru);
1560 list_move(&pc->lru, list);
1562 spin_unlock_irqrestore(&zone->lru_lock, flags);
1565 spin_unlock_irqrestore(&zone->lru_lock, flags);
1567 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
1571 if (ret == -EBUSY || ret == -EINVAL) {
1572 /* found lock contention or "pc" is obsolete. */
1579 if (!ret && !list_empty(list))
1585 * make mem_cgroup's charge to be 0 if there is no task.
1586 * This enables deleting this mem_cgroup.
1588 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
1591 int node, zid, shrink;
1592 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1593 struct cgroup *cgrp = mem->css.cgroup;
1598 /* should free all ? */
1602 while (mem->res.usage > 0) {
1604 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
1607 if (signal_pending(current))
1609 /* This is for making all *used* pages to be on LRU. */
1610 lru_add_drain_all();
1612 for_each_node_state(node, N_POSSIBLE) {
1613 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
1616 ret = mem_cgroup_force_empty_list(mem,
1625 /* it seems parent cgroup doesn't have enough mem */
1636 /* returns EBUSY if there is a task or if we come here twice. */
1637 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
1641 /* we call try-to-free pages for make this cgroup empty */
1642 lru_add_drain_all();
1643 /* try to free all pages in this cgroup */
1645 while (nr_retries && mem->res.usage > 0) {
1648 if (signal_pending(current)) {
1652 progress = try_to_free_mem_cgroup_pages(mem,
1656 /* maybe some writeback is necessary */
1657 congestion_wait(WRITE, HZ/10);
1662 /* try move_account...there may be some *locked* pages. */
1669 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1671 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1675 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
1677 return mem_cgroup_from_cont(cont)->use_hierarchy;
1680 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
1684 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1685 struct cgroup *parent = cont->parent;
1686 struct mem_cgroup *parent_mem = NULL;
1689 parent_mem = mem_cgroup_from_cont(parent);
1693 * If parent's use_hiearchy is set, we can't make any modifications
1694 * in the child subtrees. If it is unset, then the change can
1695 * occur, provided the current cgroup has no children.
1697 * For the root cgroup, parent_mem is NULL, we allow value to be
1698 * set if there are no children.
1700 if ((!parent_mem || !parent_mem->use_hierarchy) &&
1701 (val == 1 || val == 0)) {
1702 if (list_empty(&cont->children))
1703 mem->use_hierarchy = val;
1713 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
1715 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1719 type = MEMFILE_TYPE(cft->private);
1720 name = MEMFILE_ATTR(cft->private);
1723 val = res_counter_read_u64(&mem->res, name);
1726 if (do_swap_account)
1727 val = res_counter_read_u64(&mem->memsw, name);
1736 * The user of this function is...
1739 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1742 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
1744 unsigned long long val;
1747 type = MEMFILE_TYPE(cft->private);
1748 name = MEMFILE_ATTR(cft->private);
1751 /* This function does all necessary parse...reuse it */
1752 ret = res_counter_memparse_write_strategy(buffer, &val);
1756 ret = mem_cgroup_resize_limit(memcg, val);
1758 ret = mem_cgroup_resize_memsw_limit(memcg, val);
1761 ret = -EINVAL; /* should be BUG() ? */
1767 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
1769 struct mem_cgroup *mem;
1772 mem = mem_cgroup_from_cont(cont);
1773 type = MEMFILE_TYPE(event);
1774 name = MEMFILE_ATTR(event);
1778 res_counter_reset_max(&mem->res);
1780 res_counter_reset_max(&mem->memsw);
1784 res_counter_reset_failcnt(&mem->res);
1786 res_counter_reset_failcnt(&mem->memsw);
1792 static const struct mem_cgroup_stat_desc {
1795 } mem_cgroup_stat_desc[] = {
1796 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
1797 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
1798 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
1799 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
1802 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
1803 struct cgroup_map_cb *cb)
1805 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1806 struct mem_cgroup_stat *stat = &mem_cont->stat;
1809 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1812 val = mem_cgroup_read_stat(stat, i);
1813 val *= mem_cgroup_stat_desc[i].unit;
1814 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
1816 /* showing # of active pages */
1818 unsigned long active_anon, inactive_anon;
1819 unsigned long active_file, inactive_file;
1820 unsigned long unevictable;
1822 inactive_anon = mem_cgroup_get_all_zonestat(mem_cont,
1824 active_anon = mem_cgroup_get_all_zonestat(mem_cont,
1826 inactive_file = mem_cgroup_get_all_zonestat(mem_cont,
1828 active_file = mem_cgroup_get_all_zonestat(mem_cont,
1830 unevictable = mem_cgroup_get_all_zonestat(mem_cont,
1833 cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE);
1834 cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE);
1835 cb->fill(cb, "active_file", (active_file) * PAGE_SIZE);
1836 cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE);
1837 cb->fill(cb, "unevictable", unevictable * PAGE_SIZE);
1844 static struct cftype mem_cgroup_files[] = {
1846 .name = "usage_in_bytes",
1847 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
1848 .read_u64 = mem_cgroup_read,
1851 .name = "max_usage_in_bytes",
1852 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
1853 .trigger = mem_cgroup_reset,
1854 .read_u64 = mem_cgroup_read,
1857 .name = "limit_in_bytes",
1858 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
1859 .write_string = mem_cgroup_write,
1860 .read_u64 = mem_cgroup_read,
1864 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
1865 .trigger = mem_cgroup_reset,
1866 .read_u64 = mem_cgroup_read,
1870 .read_map = mem_control_stat_show,
1873 .name = "force_empty",
1874 .trigger = mem_cgroup_force_empty_write,
1877 .name = "use_hierarchy",
1878 .write_u64 = mem_cgroup_hierarchy_write,
1879 .read_u64 = mem_cgroup_hierarchy_read,
1883 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1884 static struct cftype memsw_cgroup_files[] = {
1886 .name = "memsw.usage_in_bytes",
1887 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
1888 .read_u64 = mem_cgroup_read,
1891 .name = "memsw.max_usage_in_bytes",
1892 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
1893 .trigger = mem_cgroup_reset,
1894 .read_u64 = mem_cgroup_read,
1897 .name = "memsw.limit_in_bytes",
1898 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
1899 .write_string = mem_cgroup_write,
1900 .read_u64 = mem_cgroup_read,
1903 .name = "memsw.failcnt",
1904 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
1905 .trigger = mem_cgroup_reset,
1906 .read_u64 = mem_cgroup_read,
1910 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
1912 if (!do_swap_account)
1914 return cgroup_add_files(cont, ss, memsw_cgroup_files,
1915 ARRAY_SIZE(memsw_cgroup_files));
1918 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
1924 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1926 struct mem_cgroup_per_node *pn;
1927 struct mem_cgroup_per_zone *mz;
1929 int zone, tmp = node;
1931 * This routine is called against possible nodes.
1932 * But it's BUG to call kmalloc() against offline node.
1934 * TODO: this routine can waste much memory for nodes which will
1935 * never be onlined. It's better to use memory hotplug callback
1938 if (!node_state(node, N_NORMAL_MEMORY))
1940 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
1944 mem->info.nodeinfo[node] = pn;
1945 memset(pn, 0, sizeof(*pn));
1947 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
1948 mz = &pn->zoneinfo[zone];
1950 INIT_LIST_HEAD(&mz->lists[l]);
1955 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1957 kfree(mem->info.nodeinfo[node]);
1960 static int mem_cgroup_size(void)
1962 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
1963 return sizeof(struct mem_cgroup) + cpustat_size;
1966 static struct mem_cgroup *mem_cgroup_alloc(void)
1968 struct mem_cgroup *mem;
1969 int size = mem_cgroup_size();
1971 if (size < PAGE_SIZE)
1972 mem = kmalloc(size, GFP_KERNEL);
1974 mem = vmalloc(size);
1977 memset(mem, 0, size);
1982 * At destroying mem_cgroup, references from swap_cgroup can remain.
1983 * (scanning all at force_empty is too costly...)
1985 * Instead of clearing all references at force_empty, we remember
1986 * the number of reference from swap_cgroup and free mem_cgroup when
1987 * it goes down to 0.
1989 * When mem_cgroup is destroyed, mem->obsolete will be set to 0 and
1990 * entry which points to this memcg will be ignore at swapin.
1992 * Removal of cgroup itself succeeds regardless of refs from swap.
1995 static void mem_cgroup_free(struct mem_cgroup *mem)
1999 if (atomic_read(&mem->refcnt) > 0)
2003 for_each_node_state(node, N_POSSIBLE)
2004 free_mem_cgroup_per_zone_info(mem, node);
2006 if (mem_cgroup_size() < PAGE_SIZE)
2012 static void mem_cgroup_get(struct mem_cgroup *mem)
2014 atomic_inc(&mem->refcnt);
2017 static void mem_cgroup_put(struct mem_cgroup *mem)
2019 if (atomic_dec_and_test(&mem->refcnt)) {
2022 mem_cgroup_free(mem);
2027 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2028 static void __init enable_swap_cgroup(void)
2030 if (!mem_cgroup_disabled() && really_do_swap_account)
2031 do_swap_account = 1;
2034 static void __init enable_swap_cgroup(void)
2039 static struct cgroup_subsys_state *
2040 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2042 struct mem_cgroup *mem, *parent;
2045 mem = mem_cgroup_alloc();
2047 return ERR_PTR(-ENOMEM);
2049 for_each_node_state(node, N_POSSIBLE)
2050 if (alloc_mem_cgroup_per_zone_info(mem, node))
2053 if (cont->parent == NULL) {
2054 enable_swap_cgroup();
2057 parent = mem_cgroup_from_cont(cont->parent);
2058 mem->use_hierarchy = parent->use_hierarchy;
2061 if (parent && parent->use_hierarchy) {
2062 res_counter_init(&mem->res, &parent->res);
2063 res_counter_init(&mem->memsw, &parent->memsw);
2065 res_counter_init(&mem->res, NULL);
2066 res_counter_init(&mem->memsw, NULL);
2068 mem_cgroup_set_inactive_ratio(mem);
2069 mem->last_scanned_child = NULL;
2073 for_each_node_state(node, N_POSSIBLE)
2074 free_mem_cgroup_per_zone_info(mem, node);
2075 mem_cgroup_free(mem);
2076 return ERR_PTR(-ENOMEM);
2079 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
2080 struct cgroup *cont)
2082 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2084 mem_cgroup_force_empty(mem, false);
2087 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2088 struct cgroup *cont)
2090 mem_cgroup_free(mem_cgroup_from_cont(cont));
2093 static int mem_cgroup_populate(struct cgroup_subsys *ss,
2094 struct cgroup *cont)
2098 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2099 ARRAY_SIZE(mem_cgroup_files));
2102 ret = register_memsw_files(cont, ss);
2106 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2107 struct cgroup *cont,
2108 struct cgroup *old_cont,
2109 struct task_struct *p)
2112 * FIXME: It's better to move charges of this process from old
2113 * memcg to new memcg. But it's just on TODO-List now.
2117 struct cgroup_subsys mem_cgroup_subsys = {
2119 .subsys_id = mem_cgroup_subsys_id,
2120 .create = mem_cgroup_create,
2121 .pre_destroy = mem_cgroup_pre_destroy,
2122 .destroy = mem_cgroup_destroy,
2123 .populate = mem_cgroup_populate,
2124 .attach = mem_cgroup_move_task,
2128 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2130 static int __init disable_swap_account(char *s)
2132 really_do_swap_account = 0;
2135 __setup("noswapaccount", disable_swap_account);