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/limits.h>
31 #include <linux/mutex.h>
32 #include <linux/rbtree.h>
33 #include <linux/slab.h>
34 #include <linux/swap.h>
35 #include <linux/spinlock.h>
37 #include <linux/seq_file.h>
38 #include <linux/vmalloc.h>
39 #include <linux/mm_inline.h>
40 #include <linux/page_cgroup.h>
41 #include <linux/cpu.h>
44 #include <asm/uaccess.h>
46 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
47 #define MEM_CGROUP_RECLAIM_RETRIES 5
48 struct mem_cgroup *root_mem_cgroup __read_mostly;
50 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
51 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
52 int do_swap_account __read_mostly;
53 static int really_do_swap_account __initdata = 1; /* for remember boot option*/
55 #define do_swap_account (0)
58 static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */
59 #define SOFTLIMIT_EVENTS_THRESH (1000)
62 * Statistics for memory cgroup.
64 enum mem_cgroup_stat_index {
66 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
68 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
69 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
70 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
71 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
72 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
73 MEM_CGROUP_STAT_EVENTS, /* sum of pagein + pageout for internal use */
74 MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
76 MEM_CGROUP_STAT_NSTATS,
79 struct mem_cgroup_stat_cpu {
80 s64 count[MEM_CGROUP_STAT_NSTATS];
81 } ____cacheline_aligned_in_smp;
83 struct mem_cgroup_stat {
84 struct mem_cgroup_stat_cpu cpustat[0];
88 __mem_cgroup_stat_reset_safe(struct mem_cgroup_stat_cpu *stat,
89 enum mem_cgroup_stat_index idx)
95 __mem_cgroup_stat_read_local(struct mem_cgroup_stat_cpu *stat,
96 enum mem_cgroup_stat_index idx)
98 return stat->count[idx];
102 * For accounting under irq disable, no need for increment preempt count.
104 static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
105 enum mem_cgroup_stat_index idx, int val)
107 stat->count[idx] += val;
110 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
111 enum mem_cgroup_stat_index idx)
115 for_each_possible_cpu(cpu)
116 ret += stat->cpustat[cpu].count[idx];
120 static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat)
124 ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE);
125 ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS);
130 * per-zone information in memory controller.
132 struct mem_cgroup_per_zone {
134 * spin_lock to protect the per cgroup LRU
136 struct list_head lists[NR_LRU_LISTS];
137 unsigned long count[NR_LRU_LISTS];
139 struct zone_reclaim_stat reclaim_stat;
140 struct rb_node tree_node; /* RB tree node */
141 unsigned long long usage_in_excess;/* Set to the value by which */
142 /* the soft limit is exceeded*/
144 struct mem_cgroup *mem; /* Back pointer, we cannot */
145 /* use container_of */
147 /* Macro for accessing counter */
148 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
150 struct mem_cgroup_per_node {
151 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
154 struct mem_cgroup_lru_info {
155 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
159 * Cgroups above their limits are maintained in a RB-Tree, independent of
160 * their hierarchy representation
163 struct mem_cgroup_tree_per_zone {
164 struct rb_root rb_root;
168 struct mem_cgroup_tree_per_node {
169 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
172 struct mem_cgroup_tree {
173 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
176 static struct mem_cgroup_tree soft_limit_tree __read_mostly;
179 * The memory controller data structure. The memory controller controls both
180 * page cache and RSS per cgroup. We would eventually like to provide
181 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
182 * to help the administrator determine what knobs to tune.
184 * TODO: Add a water mark for the memory controller. Reclaim will begin when
185 * we hit the water mark. May be even add a low water mark, such that
186 * no reclaim occurs from a cgroup at it's low water mark, this is
187 * a feature that will be implemented much later in the future.
190 struct cgroup_subsys_state css;
192 * the counter to account for memory usage
194 struct res_counter res;
196 * the counter to account for mem+swap usage.
198 struct res_counter memsw;
200 * Per cgroup active and inactive list, similar to the
201 * per zone LRU lists.
203 struct mem_cgroup_lru_info info;
206 protect against reclaim related member.
208 spinlock_t reclaim_param_lock;
210 int prev_priority; /* for recording reclaim priority */
213 * While reclaiming in a hierarchy, we cache the last child we
216 int last_scanned_child;
218 * Should the accounting and control be hierarchical, per subtree?
221 unsigned long last_oom_jiffies;
224 unsigned int swappiness;
226 /* set when res.limit == memsw.limit */
227 bool memsw_is_minimum;
230 * statistics. This must be placed at the end of memcg.
232 struct mem_cgroup_stat stat;
236 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
237 * limit reclaim to prevent infinite loops, if they ever occur.
239 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
240 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
243 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
244 MEM_CGROUP_CHARGE_TYPE_MAPPED,
245 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
246 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
247 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
248 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
252 /* only for here (for easy reading.) */
253 #define PCGF_CACHE (1UL << PCG_CACHE)
254 #define PCGF_USED (1UL << PCG_USED)
255 #define PCGF_LOCK (1UL << PCG_LOCK)
256 /* Not used, but added here for completeness */
257 #define PCGF_ACCT (1UL << PCG_ACCT)
259 /* for encoding cft->private value on file */
262 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
263 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
264 #define MEMFILE_ATTR(val) ((val) & 0xffff)
267 * Reclaim flags for mem_cgroup_hierarchical_reclaim
269 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
270 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
271 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
272 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
273 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
274 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
276 static void mem_cgroup_get(struct mem_cgroup *mem);
277 static void mem_cgroup_put(struct mem_cgroup *mem);
278 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
279 static void drain_all_stock_async(void);
281 static struct mem_cgroup_per_zone *
282 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
284 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
287 static struct mem_cgroup_per_zone *
288 page_cgroup_zoneinfo(struct page_cgroup *pc)
290 struct mem_cgroup *mem = pc->mem_cgroup;
291 int nid = page_cgroup_nid(pc);
292 int zid = page_cgroup_zid(pc);
297 return mem_cgroup_zoneinfo(mem, nid, zid);
300 static struct mem_cgroup_tree_per_zone *
301 soft_limit_tree_node_zone(int nid, int zid)
303 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
306 static struct mem_cgroup_tree_per_zone *
307 soft_limit_tree_from_page(struct page *page)
309 int nid = page_to_nid(page);
310 int zid = page_zonenum(page);
312 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
316 __mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
317 struct mem_cgroup_per_zone *mz,
318 struct mem_cgroup_tree_per_zone *mctz,
319 unsigned long long new_usage_in_excess)
321 struct rb_node **p = &mctz->rb_root.rb_node;
322 struct rb_node *parent = NULL;
323 struct mem_cgroup_per_zone *mz_node;
328 mz->usage_in_excess = new_usage_in_excess;
329 if (!mz->usage_in_excess)
333 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
335 if (mz->usage_in_excess < mz_node->usage_in_excess)
338 * We can't avoid mem cgroups that are over their soft
339 * limit by the same amount
341 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
344 rb_link_node(&mz->tree_node, parent, p);
345 rb_insert_color(&mz->tree_node, &mctz->rb_root);
350 __mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
351 struct mem_cgroup_per_zone *mz,
352 struct mem_cgroup_tree_per_zone *mctz)
356 rb_erase(&mz->tree_node, &mctz->rb_root);
361 mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
362 struct mem_cgroup_per_zone *mz,
363 struct mem_cgroup_tree_per_zone *mctz)
365 spin_lock(&mctz->lock);
366 __mem_cgroup_remove_exceeded(mem, mz, mctz);
367 spin_unlock(&mctz->lock);
370 static bool mem_cgroup_soft_limit_check(struct mem_cgroup *mem)
375 struct mem_cgroup_stat_cpu *cpustat;
378 cpustat = &mem->stat.cpustat[cpu];
379 val = __mem_cgroup_stat_read_local(cpustat, MEM_CGROUP_STAT_EVENTS);
380 if (unlikely(val > SOFTLIMIT_EVENTS_THRESH)) {
381 __mem_cgroup_stat_reset_safe(cpustat, MEM_CGROUP_STAT_EVENTS);
388 static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page)
390 unsigned long long excess;
391 struct mem_cgroup_per_zone *mz;
392 struct mem_cgroup_tree_per_zone *mctz;
393 int nid = page_to_nid(page);
394 int zid = page_zonenum(page);
395 mctz = soft_limit_tree_from_page(page);
398 * Necessary to update all ancestors when hierarchy is used.
399 * because their event counter is not touched.
401 for (; mem; mem = parent_mem_cgroup(mem)) {
402 mz = mem_cgroup_zoneinfo(mem, nid, zid);
403 excess = res_counter_soft_limit_excess(&mem->res);
405 * We have to update the tree if mz is on RB-tree or
406 * mem is over its softlimit.
408 if (excess || mz->on_tree) {
409 spin_lock(&mctz->lock);
410 /* if on-tree, remove it */
412 __mem_cgroup_remove_exceeded(mem, mz, mctz);
414 * Insert again. mz->usage_in_excess will be updated.
415 * If excess is 0, no tree ops.
417 __mem_cgroup_insert_exceeded(mem, mz, mctz, excess);
418 spin_unlock(&mctz->lock);
423 static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem)
426 struct mem_cgroup_per_zone *mz;
427 struct mem_cgroup_tree_per_zone *mctz;
429 for_each_node_state(node, N_POSSIBLE) {
430 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
431 mz = mem_cgroup_zoneinfo(mem, node, zone);
432 mctz = soft_limit_tree_node_zone(node, zone);
433 mem_cgroup_remove_exceeded(mem, mz, mctz);
438 static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem)
440 return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT;
443 static struct mem_cgroup_per_zone *
444 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
446 struct rb_node *rightmost = NULL;
447 struct mem_cgroup_per_zone *mz;
451 rightmost = rb_last(&mctz->rb_root);
453 goto done; /* Nothing to reclaim from */
455 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
457 * Remove the node now but someone else can add it back,
458 * we will to add it back at the end of reclaim to its correct
459 * position in the tree.
461 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
462 if (!res_counter_soft_limit_excess(&mz->mem->res) ||
463 !css_tryget(&mz->mem->css))
469 static struct mem_cgroup_per_zone *
470 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
472 struct mem_cgroup_per_zone *mz;
474 spin_lock(&mctz->lock);
475 mz = __mem_cgroup_largest_soft_limit_node(mctz);
476 spin_unlock(&mctz->lock);
480 static void mem_cgroup_swap_statistics(struct mem_cgroup *mem,
483 int val = (charge) ? 1 : -1;
484 struct mem_cgroup_stat *stat = &mem->stat;
485 struct mem_cgroup_stat_cpu *cpustat;
488 cpustat = &stat->cpustat[cpu];
489 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_SWAPOUT, val);
493 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
494 struct page_cgroup *pc,
497 int val = (charge) ? 1 : -1;
498 struct mem_cgroup_stat *stat = &mem->stat;
499 struct mem_cgroup_stat_cpu *cpustat;
502 cpustat = &stat->cpustat[cpu];
503 if (PageCgroupCache(pc))
504 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
506 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
509 __mem_cgroup_stat_add_safe(cpustat,
510 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
512 __mem_cgroup_stat_add_safe(cpustat,
513 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
514 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_EVENTS, 1);
518 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
522 struct mem_cgroup_per_zone *mz;
525 for_each_online_node(nid)
526 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
527 mz = mem_cgroup_zoneinfo(mem, nid, zid);
528 total += MEM_CGROUP_ZSTAT(mz, idx);
533 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
535 return container_of(cgroup_subsys_state(cont,
536 mem_cgroup_subsys_id), struct mem_cgroup,
540 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
543 * mm_update_next_owner() may clear mm->owner to NULL
544 * if it races with swapoff, page migration, etc.
545 * So this can be called with p == NULL.
550 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
551 struct mem_cgroup, css);
554 static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
556 struct mem_cgroup *mem = NULL;
561 * Because we have no locks, mm->owner's may be being moved to other
562 * cgroup. We use css_tryget() here even if this looks
563 * pessimistic (rather than adding locks here).
567 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
570 } while (!css_tryget(&mem->css));
576 * Call callback function against all cgroup under hierarchy tree.
578 static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
579 int (*func)(struct mem_cgroup *, void *))
581 int found, ret, nextid;
582 struct cgroup_subsys_state *css;
583 struct mem_cgroup *mem;
585 if (!root->use_hierarchy)
586 return (*func)(root, data);
594 css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
596 if (css && css_tryget(css))
597 mem = container_of(css, struct mem_cgroup, css);
601 ret = (*func)(mem, data);
605 } while (!ret && css);
610 static inline bool mem_cgroup_is_root(struct mem_cgroup *mem)
612 return (mem == root_mem_cgroup);
616 * Following LRU functions are allowed to be used without PCG_LOCK.
617 * Operations are called by routine of global LRU independently from memcg.
618 * What we have to take care of here is validness of pc->mem_cgroup.
620 * Changes to pc->mem_cgroup happens when
623 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
624 * It is added to LRU before charge.
625 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
626 * When moving account, the page is not on LRU. It's isolated.
629 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
631 struct page_cgroup *pc;
632 struct mem_cgroup_per_zone *mz;
634 if (mem_cgroup_disabled())
636 pc = lookup_page_cgroup(page);
637 /* can happen while we handle swapcache. */
638 if (!TestClearPageCgroupAcctLRU(pc))
640 VM_BUG_ON(!pc->mem_cgroup);
642 * We don't check PCG_USED bit. It's cleared when the "page" is finally
643 * removed from global LRU.
645 mz = page_cgroup_zoneinfo(pc);
646 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
647 if (mem_cgroup_is_root(pc->mem_cgroup))
649 VM_BUG_ON(list_empty(&pc->lru));
650 list_del_init(&pc->lru);
654 void mem_cgroup_del_lru(struct page *page)
656 mem_cgroup_del_lru_list(page, page_lru(page));
659 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
661 struct mem_cgroup_per_zone *mz;
662 struct page_cgroup *pc;
664 if (mem_cgroup_disabled())
667 pc = lookup_page_cgroup(page);
669 * Used bit is set without atomic ops but after smp_wmb().
670 * For making pc->mem_cgroup visible, insert smp_rmb() here.
673 /* unused or root page is not rotated. */
674 if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup))
676 mz = page_cgroup_zoneinfo(pc);
677 list_move(&pc->lru, &mz->lists[lru]);
680 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
682 struct page_cgroup *pc;
683 struct mem_cgroup_per_zone *mz;
685 if (mem_cgroup_disabled())
687 pc = lookup_page_cgroup(page);
688 VM_BUG_ON(PageCgroupAcctLRU(pc));
690 * Used bit is set without atomic ops but after smp_wmb().
691 * For making pc->mem_cgroup visible, insert smp_rmb() here.
694 if (!PageCgroupUsed(pc))
697 mz = page_cgroup_zoneinfo(pc);
698 MEM_CGROUP_ZSTAT(mz, lru) += 1;
699 SetPageCgroupAcctLRU(pc);
700 if (mem_cgroup_is_root(pc->mem_cgroup))
702 list_add(&pc->lru, &mz->lists[lru]);
706 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
707 * lru because the page may.be reused after it's fully uncharged (because of
708 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
709 * it again. This function is only used to charge SwapCache. It's done under
710 * lock_page and expected that zone->lru_lock is never held.
712 static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
715 struct zone *zone = page_zone(page);
716 struct page_cgroup *pc = lookup_page_cgroup(page);
718 spin_lock_irqsave(&zone->lru_lock, flags);
720 * Forget old LRU when this page_cgroup is *not* used. This Used bit
721 * is guarded by lock_page() because the page is SwapCache.
723 if (!PageCgroupUsed(pc))
724 mem_cgroup_del_lru_list(page, page_lru(page));
725 spin_unlock_irqrestore(&zone->lru_lock, flags);
728 static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
731 struct zone *zone = page_zone(page);
732 struct page_cgroup *pc = lookup_page_cgroup(page);
734 spin_lock_irqsave(&zone->lru_lock, flags);
735 /* link when the page is linked to LRU but page_cgroup isn't */
736 if (PageLRU(page) && !PageCgroupAcctLRU(pc))
737 mem_cgroup_add_lru_list(page, page_lru(page));
738 spin_unlock_irqrestore(&zone->lru_lock, flags);
742 void mem_cgroup_move_lists(struct page *page,
743 enum lru_list from, enum lru_list to)
745 if (mem_cgroup_disabled())
747 mem_cgroup_del_lru_list(page, from);
748 mem_cgroup_add_lru_list(page, to);
751 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
754 struct mem_cgroup *curr = NULL;
758 curr = try_get_mem_cgroup_from_mm(task->mm);
764 * We should check use_hierarchy of "mem" not "curr". Because checking
765 * use_hierarchy of "curr" here make this function true if hierarchy is
766 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
767 * hierarchy(even if use_hierarchy is disabled in "mem").
769 if (mem->use_hierarchy)
770 ret = css_is_ancestor(&curr->css, &mem->css);
778 * prev_priority control...this will be used in memory reclaim path.
780 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
784 spin_lock(&mem->reclaim_param_lock);
785 prev_priority = mem->prev_priority;
786 spin_unlock(&mem->reclaim_param_lock);
788 return prev_priority;
791 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
793 spin_lock(&mem->reclaim_param_lock);
794 if (priority < mem->prev_priority)
795 mem->prev_priority = priority;
796 spin_unlock(&mem->reclaim_param_lock);
799 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
801 spin_lock(&mem->reclaim_param_lock);
802 mem->prev_priority = priority;
803 spin_unlock(&mem->reclaim_param_lock);
806 static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
808 unsigned long active;
809 unsigned long inactive;
811 unsigned long inactive_ratio;
813 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
814 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
816 gb = (inactive + active) >> (30 - PAGE_SHIFT);
818 inactive_ratio = int_sqrt(10 * gb);
823 present_pages[0] = inactive;
824 present_pages[1] = active;
827 return inactive_ratio;
830 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
832 unsigned long active;
833 unsigned long inactive;
834 unsigned long present_pages[2];
835 unsigned long inactive_ratio;
837 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
839 inactive = present_pages[0];
840 active = present_pages[1];
842 if (inactive * inactive_ratio < active)
848 int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
850 unsigned long active;
851 unsigned long inactive;
853 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
854 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
856 return (active > inactive);
859 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
863 int nid = zone->zone_pgdat->node_id;
864 int zid = zone_idx(zone);
865 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
867 return MEM_CGROUP_ZSTAT(mz, lru);
870 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
873 int nid = zone->zone_pgdat->node_id;
874 int zid = zone_idx(zone);
875 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
877 return &mz->reclaim_stat;
880 struct zone_reclaim_stat *
881 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
883 struct page_cgroup *pc;
884 struct mem_cgroup_per_zone *mz;
886 if (mem_cgroup_disabled())
889 pc = lookup_page_cgroup(page);
891 * Used bit is set without atomic ops but after smp_wmb().
892 * For making pc->mem_cgroup visible, insert smp_rmb() here.
895 if (!PageCgroupUsed(pc))
898 mz = page_cgroup_zoneinfo(pc);
902 return &mz->reclaim_stat;
905 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
906 struct list_head *dst,
907 unsigned long *scanned, int order,
908 int mode, struct zone *z,
909 struct mem_cgroup *mem_cont,
910 int active, int file)
912 unsigned long nr_taken = 0;
916 struct list_head *src;
917 struct page_cgroup *pc, *tmp;
918 int nid = z->zone_pgdat->node_id;
919 int zid = zone_idx(z);
920 struct mem_cgroup_per_zone *mz;
921 int lru = LRU_FILE * file + active;
925 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
926 src = &mz->lists[lru];
929 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
930 if (scan >= nr_to_scan)
934 if (unlikely(!PageCgroupUsed(pc)))
936 if (unlikely(!PageLRU(page)))
940 ret = __isolate_lru_page(page, mode, file);
943 list_move(&page->lru, dst);
944 mem_cgroup_del_lru(page);
948 /* we don't affect global LRU but rotate in our LRU */
949 mem_cgroup_rotate_lru_list(page, page_lru(page));
960 #define mem_cgroup_from_res_counter(counter, member) \
961 container_of(counter, struct mem_cgroup, member)
963 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
965 if (do_swap_account) {
966 if (res_counter_check_under_limit(&mem->res) &&
967 res_counter_check_under_limit(&mem->memsw))
970 if (res_counter_check_under_limit(&mem->res))
975 static unsigned int get_swappiness(struct mem_cgroup *memcg)
977 struct cgroup *cgrp = memcg->css.cgroup;
978 unsigned int swappiness;
981 if (cgrp->parent == NULL)
982 return vm_swappiness;
984 spin_lock(&memcg->reclaim_param_lock);
985 swappiness = memcg->swappiness;
986 spin_unlock(&memcg->reclaim_param_lock);
991 static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
999 * mem_cgroup_print_mem_info: Called from OOM with tasklist_lock held in read mode.
1000 * @memcg: The memory cgroup that went over limit
1001 * @p: Task that is going to be killed
1003 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1006 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1008 struct cgroup *task_cgrp;
1009 struct cgroup *mem_cgrp;
1011 * Need a buffer in BSS, can't rely on allocations. The code relies
1012 * on the assumption that OOM is serialized for memory controller.
1013 * If this assumption is broken, revisit this code.
1015 static char memcg_name[PATH_MAX];
1024 mem_cgrp = memcg->css.cgroup;
1025 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1027 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1030 * Unfortunately, we are unable to convert to a useful name
1031 * But we'll still print out the usage information
1038 printk(KERN_INFO "Task in %s killed", memcg_name);
1041 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1049 * Continues from above, so we don't need an KERN_ level
1051 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1054 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1055 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1056 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1057 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1058 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1060 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1061 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1062 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1066 * This function returns the number of memcg under hierarchy tree. Returns
1067 * 1(self count) if no children.
1069 static int mem_cgroup_count_children(struct mem_cgroup *mem)
1072 mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
1077 * Visit the first child (need not be the first child as per the ordering
1078 * of the cgroup list, since we track last_scanned_child) of @mem and use
1079 * that to reclaim free pages from.
1081 static struct mem_cgroup *
1082 mem_cgroup_select_victim(struct mem_cgroup *root_mem)
1084 struct mem_cgroup *ret = NULL;
1085 struct cgroup_subsys_state *css;
1088 if (!root_mem->use_hierarchy) {
1089 css_get(&root_mem->css);
1095 nextid = root_mem->last_scanned_child + 1;
1096 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
1098 if (css && css_tryget(css))
1099 ret = container_of(css, struct mem_cgroup, css);
1102 /* Updates scanning parameter */
1103 spin_lock(&root_mem->reclaim_param_lock);
1105 /* this means start scan from ID:1 */
1106 root_mem->last_scanned_child = 0;
1108 root_mem->last_scanned_child = found;
1109 spin_unlock(&root_mem->reclaim_param_lock);
1116 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1117 * we reclaimed from, so that we don't end up penalizing one child extensively
1118 * based on its position in the children list.
1120 * root_mem is the original ancestor that we've been reclaim from.
1122 * We give up and return to the caller when we visit root_mem twice.
1123 * (other groups can be removed while we're walking....)
1125 * If shrink==true, for avoiding to free too much, this returns immedieately.
1127 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
1130 unsigned long reclaim_options)
1132 struct mem_cgroup *victim;
1135 bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
1136 bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
1137 bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT;
1138 unsigned long excess = mem_cgroup_get_excess(root_mem);
1140 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1141 if (root_mem->memsw_is_minimum)
1145 victim = mem_cgroup_select_victim(root_mem);
1146 if (victim == root_mem) {
1149 drain_all_stock_async();
1152 * If we have not been able to reclaim
1153 * anything, it might because there are
1154 * no reclaimable pages under this hierarchy
1156 if (!check_soft || !total) {
1157 css_put(&victim->css);
1161 * We want to do more targetted reclaim.
1162 * excess >> 2 is not to excessive so as to
1163 * reclaim too much, nor too less that we keep
1164 * coming back to reclaim from this cgroup
1166 if (total >= (excess >> 2) ||
1167 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) {
1168 css_put(&victim->css);
1173 if (!mem_cgroup_local_usage(&victim->stat)) {
1174 /* this cgroup's local usage == 0 */
1175 css_put(&victim->css);
1178 /* we use swappiness of local cgroup */
1180 ret = mem_cgroup_shrink_node_zone(victim, gfp_mask,
1181 noswap, get_swappiness(victim), zone,
1182 zone->zone_pgdat->node_id);
1184 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask,
1185 noswap, get_swappiness(victim));
1186 css_put(&victim->css);
1188 * At shrinking usage, we can't check we should stop here or
1189 * reclaim more. It's depends on callers. last_scanned_child
1190 * will work enough for keeping fairness under tree.
1196 if (res_counter_check_under_soft_limit(&root_mem->res))
1198 } else if (mem_cgroup_check_under_limit(root_mem))
1204 bool mem_cgroup_oom_called(struct task_struct *task)
1207 struct mem_cgroup *mem;
1208 struct mm_struct *mm;
1214 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
1215 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
1221 static int record_last_oom_cb(struct mem_cgroup *mem, void *data)
1223 mem->last_oom_jiffies = jiffies;
1227 static void record_last_oom(struct mem_cgroup *mem)
1229 mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb);
1233 * Currently used to update mapped file statistics, but the routine can be
1234 * generalized to update other statistics as well.
1236 void mem_cgroup_update_file_mapped(struct page *page, int val)
1238 struct mem_cgroup *mem;
1239 struct mem_cgroup_stat *stat;
1240 struct mem_cgroup_stat_cpu *cpustat;
1242 struct page_cgroup *pc;
1244 pc = lookup_page_cgroup(page);
1248 lock_page_cgroup(pc);
1249 mem = pc->mem_cgroup;
1253 if (!PageCgroupUsed(pc))
1257 * Preemption is already disabled, we don't need get_cpu()
1259 cpu = smp_processor_id();
1261 cpustat = &stat->cpustat[cpu];
1263 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, val);
1265 unlock_page_cgroup(pc);
1269 * size of first charge trial. "32" comes from vmscan.c's magic value.
1270 * TODO: maybe necessary to use big numbers in big irons.
1272 #define CHARGE_SIZE (32 * PAGE_SIZE)
1273 struct memcg_stock_pcp {
1274 struct mem_cgroup *cached; /* this never be root cgroup */
1276 struct work_struct work;
1278 static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
1279 static atomic_t memcg_drain_count;
1282 * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
1283 * from local stock and true is returned. If the stock is 0 or charges from a
1284 * cgroup which is not current target, returns false. This stock will be
1287 static bool consume_stock(struct mem_cgroup *mem)
1289 struct memcg_stock_pcp *stock;
1292 stock = &get_cpu_var(memcg_stock);
1293 if (mem == stock->cached && stock->charge)
1294 stock->charge -= PAGE_SIZE;
1295 else /* need to call res_counter_charge */
1297 put_cpu_var(memcg_stock);
1302 * Returns stocks cached in percpu to res_counter and reset cached information.
1304 static void drain_stock(struct memcg_stock_pcp *stock)
1306 struct mem_cgroup *old = stock->cached;
1308 if (stock->charge) {
1309 res_counter_uncharge(&old->res, stock->charge);
1310 if (do_swap_account)
1311 res_counter_uncharge(&old->memsw, stock->charge);
1313 stock->cached = NULL;
1318 * This must be called under preempt disabled or must be called by
1319 * a thread which is pinned to local cpu.
1321 static void drain_local_stock(struct work_struct *dummy)
1323 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
1328 * Cache charges(val) which is from res_counter, to local per_cpu area.
1329 * This will be consumed by consumt_stock() function, later.
1331 static void refill_stock(struct mem_cgroup *mem, int val)
1333 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1335 if (stock->cached != mem) { /* reset if necessary */
1337 stock->cached = mem;
1339 stock->charge += val;
1340 put_cpu_var(memcg_stock);
1344 * Tries to drain stocked charges in other cpus. This function is asynchronous
1345 * and just put a work per cpu for draining localy on each cpu. Caller can
1346 * expects some charges will be back to res_counter later but cannot wait for
1349 static void drain_all_stock_async(void)
1352 /* This function is for scheduling "drain" in asynchronous way.
1353 * The result of "drain" is not directly handled by callers. Then,
1354 * if someone is calling drain, we don't have to call drain more.
1355 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1356 * there is a race. We just do loose check here.
1358 if (atomic_read(&memcg_drain_count))
1360 /* Notify other cpus that system-wide "drain" is running */
1361 atomic_inc(&memcg_drain_count);
1363 for_each_online_cpu(cpu) {
1364 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
1365 schedule_work_on(cpu, &stock->work);
1368 atomic_dec(&memcg_drain_count);
1369 /* We don't wait for flush_work */
1372 /* This is a synchronous drain interface. */
1373 static void drain_all_stock_sync(void)
1375 /* called when force_empty is called */
1376 atomic_inc(&memcg_drain_count);
1377 schedule_on_each_cpu(drain_local_stock);
1378 atomic_dec(&memcg_drain_count);
1381 static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb,
1382 unsigned long action,
1385 int cpu = (unsigned long)hcpu;
1386 struct memcg_stock_pcp *stock;
1388 if (action != CPU_DEAD)
1390 stock = &per_cpu(memcg_stock, cpu);
1396 * Unlike exported interface, "oom" parameter is added. if oom==true,
1397 * oom-killer can be invoked.
1399 static int __mem_cgroup_try_charge(struct mm_struct *mm,
1400 gfp_t gfp_mask, struct mem_cgroup **memcg,
1401 bool oom, struct page *page)
1403 struct mem_cgroup *mem, *mem_over_limit;
1404 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1405 struct res_counter *fail_res;
1406 int csize = CHARGE_SIZE;
1408 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
1409 /* Don't account this! */
1415 * We always charge the cgroup the mm_struct belongs to.
1416 * The mm_struct's mem_cgroup changes on task migration if the
1417 * thread group leader migrates. It's possible that mm is not
1418 * set, if so charge the init_mm (happens for pagecache usage).
1422 mem = try_get_mem_cgroup_from_mm(mm);
1430 VM_BUG_ON(css_is_removed(&mem->css));
1431 if (mem_cgroup_is_root(mem))
1436 unsigned long flags = 0;
1438 if (consume_stock(mem))
1441 ret = res_counter_charge(&mem->res, csize, &fail_res);
1443 if (!do_swap_account)
1445 ret = res_counter_charge(&mem->memsw, csize, &fail_res);
1448 /* mem+swap counter fails */
1449 res_counter_uncharge(&mem->res, csize);
1450 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
1451 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1454 /* mem counter fails */
1455 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1458 /* reduce request size and retry */
1459 if (csize > PAGE_SIZE) {
1463 if (!(gfp_mask & __GFP_WAIT))
1466 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
1472 * try_to_free_mem_cgroup_pages() might not give us a full
1473 * picture of reclaim. Some pages are reclaimed and might be
1474 * moved to swap cache or just unmapped from the cgroup.
1475 * Check the limit again to see if the reclaim reduced the
1476 * current usage of the cgroup before giving up
1479 if (mem_cgroup_check_under_limit(mem_over_limit))
1482 if (!nr_retries--) {
1484 mutex_lock(&memcg_tasklist);
1485 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
1486 mutex_unlock(&memcg_tasklist);
1487 record_last_oom(mem_over_limit);
1492 if (csize > PAGE_SIZE)
1493 refill_stock(mem, csize - PAGE_SIZE);
1496 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
1497 * if they exceeds softlimit.
1499 if (mem_cgroup_soft_limit_check(mem))
1500 mem_cgroup_update_tree(mem, page);
1509 * Somemtimes we have to undo a charge we got by try_charge().
1510 * This function is for that and do uncharge, put css's refcnt.
1511 * gotten by try_charge().
1513 static void mem_cgroup_cancel_charge(struct mem_cgroup *mem)
1515 if (!mem_cgroup_is_root(mem)) {
1516 res_counter_uncharge(&mem->res, PAGE_SIZE);
1517 if (do_swap_account)
1518 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1524 * A helper function to get mem_cgroup from ID. must be called under
1525 * rcu_read_lock(). The caller must check css_is_removed() or some if
1526 * it's concern. (dropping refcnt from swap can be called against removed
1529 static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
1531 struct cgroup_subsys_state *css;
1533 /* ID 0 is unused ID */
1536 css = css_lookup(&mem_cgroup_subsys, id);
1539 return container_of(css, struct mem_cgroup, css);
1542 static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
1544 struct mem_cgroup *mem;
1545 struct page_cgroup *pc;
1549 VM_BUG_ON(!PageLocked(page));
1551 if (!PageSwapCache(page))
1554 pc = lookup_page_cgroup(page);
1555 lock_page_cgroup(pc);
1556 if (PageCgroupUsed(pc)) {
1557 mem = pc->mem_cgroup;
1558 if (mem && !css_tryget(&mem->css))
1561 ent.val = page_private(page);
1562 id = lookup_swap_cgroup(ent);
1564 mem = mem_cgroup_lookup(id);
1565 if (mem && !css_tryget(&mem->css))
1569 unlock_page_cgroup(pc);
1574 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
1575 * USED state. If already USED, uncharge and return.
1578 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
1579 struct page_cgroup *pc,
1580 enum charge_type ctype)
1582 /* try_charge() can return NULL to *memcg, taking care of it. */
1586 lock_page_cgroup(pc);
1587 if (unlikely(PageCgroupUsed(pc))) {
1588 unlock_page_cgroup(pc);
1589 mem_cgroup_cancel_charge(mem);
1593 pc->mem_cgroup = mem;
1595 * We access a page_cgroup asynchronously without lock_page_cgroup().
1596 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
1597 * is accessed after testing USED bit. To make pc->mem_cgroup visible
1598 * before USED bit, we need memory barrier here.
1599 * See mem_cgroup_add_lru_list(), etc.
1603 case MEM_CGROUP_CHARGE_TYPE_CACHE:
1604 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
1605 SetPageCgroupCache(pc);
1606 SetPageCgroupUsed(pc);
1608 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1609 ClearPageCgroupCache(pc);
1610 SetPageCgroupUsed(pc);
1616 mem_cgroup_charge_statistics(mem, pc, true);
1618 unlock_page_cgroup(pc);
1622 * __mem_cgroup_move_account - move account of the page
1623 * @pc: page_cgroup of the page.
1624 * @from: mem_cgroup which the page is moved from.
1625 * @to: mem_cgroup which the page is moved to. @from != @to.
1627 * The caller must confirm following.
1628 * - page is not on LRU (isolate_page() is useful.)
1629 * - the pc is locked, used, and ->mem_cgroup points to @from.
1631 * This function does "uncharge" from old cgroup but doesn't do "charge" to
1632 * new cgroup. It should be done by a caller.
1635 static void __mem_cgroup_move_account(struct page_cgroup *pc,
1636 struct mem_cgroup *from, struct mem_cgroup *to)
1640 struct mem_cgroup_stat *stat;
1641 struct mem_cgroup_stat_cpu *cpustat;
1643 VM_BUG_ON(from == to);
1644 VM_BUG_ON(PageLRU(pc->page));
1645 VM_BUG_ON(!PageCgroupLocked(pc));
1646 VM_BUG_ON(!PageCgroupUsed(pc));
1647 VM_BUG_ON(pc->mem_cgroup != from);
1649 if (!mem_cgroup_is_root(from))
1650 res_counter_uncharge(&from->res, PAGE_SIZE);
1651 mem_cgroup_charge_statistics(from, pc, false);
1654 if (page_mapped(page) && !PageAnon(page)) {
1655 cpu = smp_processor_id();
1656 /* Update mapped_file data for mem_cgroup "from" */
1658 cpustat = &stat->cpustat[cpu];
1659 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED,
1662 /* Update mapped_file data for mem_cgroup "to" */
1664 cpustat = &stat->cpustat[cpu];
1665 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED,
1669 if (do_swap_account && !mem_cgroup_is_root(from))
1670 res_counter_uncharge(&from->memsw, PAGE_SIZE);
1671 css_put(&from->css);
1674 pc->mem_cgroup = to;
1675 mem_cgroup_charge_statistics(to, pc, true);
1677 * We charges against "to" which may not have any tasks. Then, "to"
1678 * can be under rmdir(). But in current implementation, caller of
1679 * this function is just force_empty() and it's garanteed that
1680 * "to" is never removed. So, we don't check rmdir status here.
1685 * check whether the @pc is valid for moving account and call
1686 * __mem_cgroup_move_account()
1688 static int mem_cgroup_move_account(struct page_cgroup *pc,
1689 struct mem_cgroup *from, struct mem_cgroup *to)
1692 lock_page_cgroup(pc);
1693 if (PageCgroupUsed(pc) && pc->mem_cgroup == from) {
1694 __mem_cgroup_move_account(pc, from, to);
1697 unlock_page_cgroup(pc);
1702 * move charges to its parent.
1705 static int mem_cgroup_move_parent(struct page_cgroup *pc,
1706 struct mem_cgroup *child,
1709 struct page *page = pc->page;
1710 struct cgroup *cg = child->css.cgroup;
1711 struct cgroup *pcg = cg->parent;
1712 struct mem_cgroup *parent;
1720 if (!get_page_unless_zero(page))
1722 if (isolate_lru_page(page))
1725 parent = mem_cgroup_from_cont(pcg);
1726 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false, page);
1730 ret = mem_cgroup_move_account(pc, child, parent);
1732 css_put(&parent->css); /* drop extra refcnt by try_charge() */
1734 mem_cgroup_cancel_charge(parent); /* does css_put */
1736 putback_lru_page(page);
1744 * Charge the memory controller for page usage.
1746 * 0 if the charge was successful
1747 * < 0 if the cgroup is over its limit
1749 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1750 gfp_t gfp_mask, enum charge_type ctype,
1751 struct mem_cgroup *memcg)
1753 struct mem_cgroup *mem;
1754 struct page_cgroup *pc;
1757 pc = lookup_page_cgroup(page);
1758 /* can happen at boot */
1764 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true, page);
1768 __mem_cgroup_commit_charge(mem, pc, ctype);
1772 int mem_cgroup_newpage_charge(struct page *page,
1773 struct mm_struct *mm, gfp_t gfp_mask)
1775 if (mem_cgroup_disabled())
1777 if (PageCompound(page))
1780 * If already mapped, we don't have to account.
1781 * If page cache, page->mapping has address_space.
1782 * But page->mapping may have out-of-use anon_vma pointer,
1783 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1786 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1790 return mem_cgroup_charge_common(page, mm, gfp_mask,
1791 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1795 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1796 enum charge_type ctype);
1798 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1801 struct mem_cgroup *mem = NULL;
1804 if (mem_cgroup_disabled())
1806 if (PageCompound(page))
1809 * Corner case handling. This is called from add_to_page_cache()
1810 * in usual. But some FS (shmem) precharges this page before calling it
1811 * and call add_to_page_cache() with GFP_NOWAIT.
1813 * For GFP_NOWAIT case, the page may be pre-charged before calling
1814 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1815 * charge twice. (It works but has to pay a bit larger cost.)
1816 * And when the page is SwapCache, it should take swap information
1817 * into account. This is under lock_page() now.
1819 if (!(gfp_mask & __GFP_WAIT)) {
1820 struct page_cgroup *pc;
1823 pc = lookup_page_cgroup(page);
1826 lock_page_cgroup(pc);
1827 if (PageCgroupUsed(pc)) {
1828 unlock_page_cgroup(pc);
1831 unlock_page_cgroup(pc);
1834 if (unlikely(!mm && !mem))
1837 if (page_is_file_cache(page))
1838 return mem_cgroup_charge_common(page, mm, gfp_mask,
1839 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1842 if (PageSwapCache(page)) {
1843 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
1845 __mem_cgroup_commit_charge_swapin(page, mem,
1846 MEM_CGROUP_CHARGE_TYPE_SHMEM);
1848 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1849 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1855 * While swap-in, try_charge -> commit or cancel, the page is locked.
1856 * And when try_charge() successfully returns, one refcnt to memcg without
1857 * struct page_cgroup is acquired. This refcnt will be consumed by
1858 * "commit()" or removed by "cancel()"
1860 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1862 gfp_t mask, struct mem_cgroup **ptr)
1864 struct mem_cgroup *mem;
1867 if (mem_cgroup_disabled())
1870 if (!do_swap_account)
1873 * A racing thread's fault, or swapoff, may have already updated
1874 * the pte, and even removed page from swap cache: in those cases
1875 * do_swap_page()'s pte_same() test will fail; but there's also a
1876 * KSM case which does need to charge the page.
1878 if (!PageSwapCache(page))
1880 mem = try_get_mem_cgroup_from_swapcache(page);
1884 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, page);
1885 /* drop extra refcnt from tryget */
1891 return __mem_cgroup_try_charge(mm, mask, ptr, true, page);
1895 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1896 enum charge_type ctype)
1898 struct page_cgroup *pc;
1900 if (mem_cgroup_disabled())
1904 cgroup_exclude_rmdir(&ptr->css);
1905 pc = lookup_page_cgroup(page);
1906 mem_cgroup_lru_del_before_commit_swapcache(page);
1907 __mem_cgroup_commit_charge(ptr, pc, ctype);
1908 mem_cgroup_lru_add_after_commit_swapcache(page);
1910 * Now swap is on-memory. This means this page may be
1911 * counted both as mem and swap....double count.
1912 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1913 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1914 * may call delete_from_swap_cache() before reach here.
1916 if (do_swap_account && PageSwapCache(page)) {
1917 swp_entry_t ent = {.val = page_private(page)};
1919 struct mem_cgroup *memcg;
1921 id = swap_cgroup_record(ent, 0);
1923 memcg = mem_cgroup_lookup(id);
1926 * This recorded memcg can be obsolete one. So, avoid
1927 * calling css_tryget
1929 if (!mem_cgroup_is_root(memcg))
1930 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1931 mem_cgroup_swap_statistics(memcg, false);
1932 mem_cgroup_put(memcg);
1937 * At swapin, we may charge account against cgroup which has no tasks.
1938 * So, rmdir()->pre_destroy() can be called while we do this charge.
1939 * In that case, we need to call pre_destroy() again. check it here.
1941 cgroup_release_and_wakeup_rmdir(&ptr->css);
1944 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1946 __mem_cgroup_commit_charge_swapin(page, ptr,
1947 MEM_CGROUP_CHARGE_TYPE_MAPPED);
1950 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1952 if (mem_cgroup_disabled())
1956 mem_cgroup_cancel_charge(mem);
1960 __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype)
1962 struct memcg_batch_info *batch = NULL;
1963 bool uncharge_memsw = true;
1964 /* If swapout, usage of swap doesn't decrease */
1965 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1966 uncharge_memsw = false;
1968 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
1969 * In those cases, all pages freed continously can be expected to be in
1970 * the same cgroup and we have chance to coalesce uncharges.
1971 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
1972 * because we want to do uncharge as soon as possible.
1974 if (!current->memcg_batch.do_batch || test_thread_flag(TIF_MEMDIE))
1975 goto direct_uncharge;
1977 batch = ¤t->memcg_batch;
1979 * In usual, we do css_get() when we remember memcg pointer.
1980 * But in this case, we keep res->usage until end of a series of
1981 * uncharges. Then, it's ok to ignore memcg's refcnt.
1986 * In typical case, batch->memcg == mem. This means we can
1987 * merge a series of uncharges to an uncharge of res_counter.
1988 * If not, we uncharge res_counter ony by one.
1990 if (batch->memcg != mem)
1991 goto direct_uncharge;
1992 /* remember freed charge and uncharge it later */
1993 batch->bytes += PAGE_SIZE;
1995 batch->memsw_bytes += PAGE_SIZE;
1998 res_counter_uncharge(&mem->res, PAGE_SIZE);
2000 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
2005 * uncharge if !page_mapped(page)
2007 static struct mem_cgroup *
2008 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
2010 struct page_cgroup *pc;
2011 struct mem_cgroup *mem = NULL;
2012 struct mem_cgroup_per_zone *mz;
2014 if (mem_cgroup_disabled())
2017 if (PageSwapCache(page))
2021 * Check if our page_cgroup is valid
2023 pc = lookup_page_cgroup(page);
2024 if (unlikely(!pc || !PageCgroupUsed(pc)))
2027 lock_page_cgroup(pc);
2029 mem = pc->mem_cgroup;
2031 if (!PageCgroupUsed(pc))
2035 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2036 case MEM_CGROUP_CHARGE_TYPE_DROP:
2037 if (page_mapped(page))
2040 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
2041 if (!PageAnon(page)) { /* Shared memory */
2042 if (page->mapping && !page_is_file_cache(page))
2044 } else if (page_mapped(page)) /* Anon */
2051 if (!mem_cgroup_is_root(mem))
2052 __do_uncharge(mem, ctype);
2053 if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2054 mem_cgroup_swap_statistics(mem, true);
2055 mem_cgroup_charge_statistics(mem, pc, false);
2057 ClearPageCgroupUsed(pc);
2059 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2060 * freed from LRU. This is safe because uncharged page is expected not
2061 * to be reused (freed soon). Exception is SwapCache, it's handled by
2062 * special functions.
2065 mz = page_cgroup_zoneinfo(pc);
2066 unlock_page_cgroup(pc);
2068 if (mem_cgroup_soft_limit_check(mem))
2069 mem_cgroup_update_tree(mem, page);
2070 /* at swapout, this memcg will be accessed to record to swap */
2071 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2077 unlock_page_cgroup(pc);
2081 void mem_cgroup_uncharge_page(struct page *page)
2084 if (page_mapped(page))
2086 if (page->mapping && !PageAnon(page))
2088 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
2091 void mem_cgroup_uncharge_cache_page(struct page *page)
2093 VM_BUG_ON(page_mapped(page));
2094 VM_BUG_ON(page->mapping);
2095 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
2099 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2100 * In that cases, pages are freed continuously and we can expect pages
2101 * are in the same memcg. All these calls itself limits the number of
2102 * pages freed at once, then uncharge_start/end() is called properly.
2103 * This may be called prural(2) times in a context,
2106 void mem_cgroup_uncharge_start(void)
2108 current->memcg_batch.do_batch++;
2109 /* We can do nest. */
2110 if (current->memcg_batch.do_batch == 1) {
2111 current->memcg_batch.memcg = NULL;
2112 current->memcg_batch.bytes = 0;
2113 current->memcg_batch.memsw_bytes = 0;
2117 void mem_cgroup_uncharge_end(void)
2119 struct memcg_batch_info *batch = ¤t->memcg_batch;
2121 if (!batch->do_batch)
2125 if (batch->do_batch) /* If stacked, do nothing. */
2131 * This "batch->memcg" is valid without any css_get/put etc...
2132 * bacause we hide charges behind us.
2135 res_counter_uncharge(&batch->memcg->res, batch->bytes);
2136 if (batch->memsw_bytes)
2137 res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes);
2138 /* forget this pointer (for sanity check) */
2139 batch->memcg = NULL;
2144 * called after __delete_from_swap_cache() and drop "page" account.
2145 * memcg information is recorded to swap_cgroup of "ent"
2148 mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
2150 struct mem_cgroup *memcg;
2151 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
2153 if (!swapout) /* this was a swap cache but the swap is unused ! */
2154 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
2156 memcg = __mem_cgroup_uncharge_common(page, ctype);
2158 /* record memcg information */
2159 if (do_swap_account && swapout && memcg) {
2160 swap_cgroup_record(ent, css_id(&memcg->css));
2161 mem_cgroup_get(memcg);
2163 if (swapout && memcg)
2164 css_put(&memcg->css);
2168 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2170 * called from swap_entry_free(). remove record in swap_cgroup and
2171 * uncharge "memsw" account.
2173 void mem_cgroup_uncharge_swap(swp_entry_t ent)
2175 struct mem_cgroup *memcg;
2178 if (!do_swap_account)
2181 id = swap_cgroup_record(ent, 0);
2183 memcg = mem_cgroup_lookup(id);
2186 * We uncharge this because swap is freed.
2187 * This memcg can be obsolete one. We avoid calling css_tryget
2189 if (!mem_cgroup_is_root(memcg))
2190 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
2191 mem_cgroup_swap_statistics(memcg, false);
2192 mem_cgroup_put(memcg);
2199 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2202 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
2204 struct page_cgroup *pc;
2205 struct mem_cgroup *mem = NULL;
2208 if (mem_cgroup_disabled())
2211 pc = lookup_page_cgroup(page);
2212 lock_page_cgroup(pc);
2213 if (PageCgroupUsed(pc)) {
2214 mem = pc->mem_cgroup;
2217 unlock_page_cgroup(pc);
2220 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false,
2228 /* remove redundant charge if migration failed*/
2229 void mem_cgroup_end_migration(struct mem_cgroup *mem,
2230 struct page *oldpage, struct page *newpage)
2232 struct page *target, *unused;
2233 struct page_cgroup *pc;
2234 enum charge_type ctype;
2238 cgroup_exclude_rmdir(&mem->css);
2239 /* at migration success, oldpage->mapping is NULL. */
2240 if (oldpage->mapping) {
2248 if (PageAnon(target))
2249 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
2250 else if (page_is_file_cache(target))
2251 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
2253 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
2255 /* unused page is not on radix-tree now. */
2257 __mem_cgroup_uncharge_common(unused, ctype);
2259 pc = lookup_page_cgroup(target);
2261 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
2262 * So, double-counting is effectively avoided.
2264 __mem_cgroup_commit_charge(mem, pc, ctype);
2267 * Both of oldpage and newpage are still under lock_page().
2268 * Then, we don't have to care about race in radix-tree.
2269 * But we have to be careful that this page is unmapped or not.
2271 * There is a case for !page_mapped(). At the start of
2272 * migration, oldpage was mapped. But now, it's zapped.
2273 * But we know *target* page is not freed/reused under us.
2274 * mem_cgroup_uncharge_page() does all necessary checks.
2276 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
2277 mem_cgroup_uncharge_page(target);
2279 * At migration, we may charge account against cgroup which has no tasks
2280 * So, rmdir()->pre_destroy() can be called while we do this charge.
2281 * In that case, we need to call pre_destroy() again. check it here.
2283 cgroup_release_and_wakeup_rmdir(&mem->css);
2287 * A call to try to shrink memory usage on charge failure at shmem's swapin.
2288 * Calling hierarchical_reclaim is not enough because we should update
2289 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
2290 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
2291 * not from the memcg which this page would be charged to.
2292 * try_charge_swapin does all of these works properly.
2294 int mem_cgroup_shmem_charge_fallback(struct page *page,
2295 struct mm_struct *mm,
2298 struct mem_cgroup *mem = NULL;
2301 if (mem_cgroup_disabled())
2304 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2306 mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
2311 static DEFINE_MUTEX(set_limit_mutex);
2313 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
2314 unsigned long long val)
2320 int children = mem_cgroup_count_children(memcg);
2321 u64 curusage, oldusage;
2324 * For keeping hierarchical_reclaim simple, how long we should retry
2325 * is depends on callers. We set our retry-count to be function
2326 * of # of children which we should visit in this loop.
2328 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
2330 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2332 while (retry_count) {
2333 if (signal_pending(current)) {
2338 * Rather than hide all in some function, I do this in
2339 * open coded manner. You see what this really does.
2340 * We have to guarantee mem->res.limit < mem->memsw.limit.
2342 mutex_lock(&set_limit_mutex);
2343 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2344 if (memswlimit < val) {
2346 mutex_unlock(&set_limit_mutex);
2349 ret = res_counter_set_limit(&memcg->res, val);
2351 if (memswlimit == val)
2352 memcg->memsw_is_minimum = true;
2354 memcg->memsw_is_minimum = false;
2356 mutex_unlock(&set_limit_mutex);
2361 progress = mem_cgroup_hierarchical_reclaim(memcg, NULL,
2363 MEM_CGROUP_RECLAIM_SHRINK);
2364 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2365 /* Usage is reduced ? */
2366 if (curusage >= oldusage)
2369 oldusage = curusage;
2375 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2376 unsigned long long val)
2379 u64 memlimit, oldusage, curusage;
2380 int children = mem_cgroup_count_children(memcg);
2383 /* see mem_cgroup_resize_res_limit */
2384 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
2385 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2386 while (retry_count) {
2387 if (signal_pending(current)) {
2392 * Rather than hide all in some function, I do this in
2393 * open coded manner. You see what this really does.
2394 * We have to guarantee mem->res.limit < mem->memsw.limit.
2396 mutex_lock(&set_limit_mutex);
2397 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2398 if (memlimit > val) {
2400 mutex_unlock(&set_limit_mutex);
2403 ret = res_counter_set_limit(&memcg->memsw, val);
2405 if (memlimit == val)
2406 memcg->memsw_is_minimum = true;
2408 memcg->memsw_is_minimum = false;
2410 mutex_unlock(&set_limit_mutex);
2415 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
2416 MEM_CGROUP_RECLAIM_NOSWAP |
2417 MEM_CGROUP_RECLAIM_SHRINK);
2418 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2419 /* Usage is reduced ? */
2420 if (curusage >= oldusage)
2423 oldusage = curusage;
2428 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2429 gfp_t gfp_mask, int nid,
2432 unsigned long nr_reclaimed = 0;
2433 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2434 unsigned long reclaimed;
2436 struct mem_cgroup_tree_per_zone *mctz;
2437 unsigned long long excess;
2442 mctz = soft_limit_tree_node_zone(nid, zid);
2444 * This loop can run a while, specially if mem_cgroup's continuously
2445 * keep exceeding their soft limit and putting the system under
2452 mz = mem_cgroup_largest_soft_limit_node(mctz);
2456 reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone,
2458 MEM_CGROUP_RECLAIM_SOFT);
2459 nr_reclaimed += reclaimed;
2460 spin_lock(&mctz->lock);
2463 * If we failed to reclaim anything from this memory cgroup
2464 * it is time to move on to the next cgroup
2470 * Loop until we find yet another one.
2472 * By the time we get the soft_limit lock
2473 * again, someone might have aded the
2474 * group back on the RB tree. Iterate to
2475 * make sure we get a different mem.
2476 * mem_cgroup_largest_soft_limit_node returns
2477 * NULL if no other cgroup is present on
2481 __mem_cgroup_largest_soft_limit_node(mctz);
2482 if (next_mz == mz) {
2483 css_put(&next_mz->mem->css);
2485 } else /* next_mz == NULL or other memcg */
2489 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
2490 excess = res_counter_soft_limit_excess(&mz->mem->res);
2492 * One school of thought says that we should not add
2493 * back the node to the tree if reclaim returns 0.
2494 * But our reclaim could return 0, simply because due
2495 * to priority we are exposing a smaller subset of
2496 * memory to reclaim from. Consider this as a longer
2499 /* If excess == 0, no tree ops */
2500 __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
2501 spin_unlock(&mctz->lock);
2502 css_put(&mz->mem->css);
2505 * Could not reclaim anything and there are no more
2506 * mem cgroups to try or we seem to be looping without
2507 * reclaiming anything.
2509 if (!nr_reclaimed &&
2511 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2513 } while (!nr_reclaimed);
2515 css_put(&next_mz->mem->css);
2516 return nr_reclaimed;
2520 * This routine traverse page_cgroup in given list and drop them all.
2521 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
2523 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
2524 int node, int zid, enum lru_list lru)
2527 struct mem_cgroup_per_zone *mz;
2528 struct page_cgroup *pc, *busy;
2529 unsigned long flags, loop;
2530 struct list_head *list;
2533 zone = &NODE_DATA(node)->node_zones[zid];
2534 mz = mem_cgroup_zoneinfo(mem, node, zid);
2535 list = &mz->lists[lru];
2537 loop = MEM_CGROUP_ZSTAT(mz, lru);
2538 /* give some margin against EBUSY etc...*/
2543 spin_lock_irqsave(&zone->lru_lock, flags);
2544 if (list_empty(list)) {
2545 spin_unlock_irqrestore(&zone->lru_lock, flags);
2548 pc = list_entry(list->prev, struct page_cgroup, lru);
2550 list_move(&pc->lru, list);
2552 spin_unlock_irqrestore(&zone->lru_lock, flags);
2555 spin_unlock_irqrestore(&zone->lru_lock, flags);
2557 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
2561 if (ret == -EBUSY || ret == -EINVAL) {
2562 /* found lock contention or "pc" is obsolete. */
2569 if (!ret && !list_empty(list))
2575 * make mem_cgroup's charge to be 0 if there is no task.
2576 * This enables deleting this mem_cgroup.
2578 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
2581 int node, zid, shrink;
2582 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
2583 struct cgroup *cgrp = mem->css.cgroup;
2588 /* should free all ? */
2592 while (mem->res.usage > 0) {
2594 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
2597 if (signal_pending(current))
2599 /* This is for making all *used* pages to be on LRU. */
2600 lru_add_drain_all();
2601 drain_all_stock_sync();
2603 for_each_node_state(node, N_HIGH_MEMORY) {
2604 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
2607 ret = mem_cgroup_force_empty_list(mem,
2616 /* it seems parent cgroup doesn't have enough mem */
2627 /* returns EBUSY if there is a task or if we come here twice. */
2628 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
2632 /* we call try-to-free pages for make this cgroup empty */
2633 lru_add_drain_all();
2634 /* try to free all pages in this cgroup */
2636 while (nr_retries && mem->res.usage > 0) {
2639 if (signal_pending(current)) {
2643 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
2644 false, get_swappiness(mem));
2647 /* maybe some writeback is necessary */
2648 congestion_wait(BLK_RW_ASYNC, HZ/10);
2653 /* try move_account...there may be some *locked* pages. */
2660 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
2662 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
2666 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
2668 return mem_cgroup_from_cont(cont)->use_hierarchy;
2671 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
2675 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2676 struct cgroup *parent = cont->parent;
2677 struct mem_cgroup *parent_mem = NULL;
2680 parent_mem = mem_cgroup_from_cont(parent);
2684 * If parent's use_hierarchy is set, we can't make any modifications
2685 * in the child subtrees. If it is unset, then the change can
2686 * occur, provided the current cgroup has no children.
2688 * For the root cgroup, parent_mem is NULL, we allow value to be
2689 * set if there are no children.
2691 if ((!parent_mem || !parent_mem->use_hierarchy) &&
2692 (val == 1 || val == 0)) {
2693 if (list_empty(&cont->children))
2694 mem->use_hierarchy = val;
2704 struct mem_cgroup_idx_data {
2706 enum mem_cgroup_stat_index idx;
2710 mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data)
2712 struct mem_cgroup_idx_data *d = data;
2713 d->val += mem_cgroup_read_stat(&mem->stat, d->idx);
2718 mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem,
2719 enum mem_cgroup_stat_index idx, s64 *val)
2721 struct mem_cgroup_idx_data d;
2724 mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat);
2728 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
2730 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2734 type = MEMFILE_TYPE(cft->private);
2735 name = MEMFILE_ATTR(cft->private);
2738 if (name == RES_USAGE && mem_cgroup_is_root(mem)) {
2739 mem_cgroup_get_recursive_idx_stat(mem,
2740 MEM_CGROUP_STAT_CACHE, &idx_val);
2742 mem_cgroup_get_recursive_idx_stat(mem,
2743 MEM_CGROUP_STAT_RSS, &idx_val);
2747 val = res_counter_read_u64(&mem->res, name);
2750 if (name == RES_USAGE && mem_cgroup_is_root(mem)) {
2751 mem_cgroup_get_recursive_idx_stat(mem,
2752 MEM_CGROUP_STAT_CACHE, &idx_val);
2754 mem_cgroup_get_recursive_idx_stat(mem,
2755 MEM_CGROUP_STAT_RSS, &idx_val);
2757 mem_cgroup_get_recursive_idx_stat(mem,
2758 MEM_CGROUP_STAT_SWAPOUT, &idx_val);
2762 val = res_counter_read_u64(&mem->memsw, name);
2771 * The user of this function is...
2774 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
2777 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
2779 unsigned long long val;
2782 type = MEMFILE_TYPE(cft->private);
2783 name = MEMFILE_ATTR(cft->private);
2786 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2790 /* This function does all necessary parse...reuse it */
2791 ret = res_counter_memparse_write_strategy(buffer, &val);
2795 ret = mem_cgroup_resize_limit(memcg, val);
2797 ret = mem_cgroup_resize_memsw_limit(memcg, val);
2799 case RES_SOFT_LIMIT:
2800 ret = res_counter_memparse_write_strategy(buffer, &val);
2804 * For memsw, soft limits are hard to implement in terms
2805 * of semantics, for now, we support soft limits for
2806 * control without swap
2809 ret = res_counter_set_soft_limit(&memcg->res, val);
2814 ret = -EINVAL; /* should be BUG() ? */
2820 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
2821 unsigned long long *mem_limit, unsigned long long *memsw_limit)
2823 struct cgroup *cgroup;
2824 unsigned long long min_limit, min_memsw_limit, tmp;
2826 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2827 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2828 cgroup = memcg->css.cgroup;
2829 if (!memcg->use_hierarchy)
2832 while (cgroup->parent) {
2833 cgroup = cgroup->parent;
2834 memcg = mem_cgroup_from_cont(cgroup);
2835 if (!memcg->use_hierarchy)
2837 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
2838 min_limit = min(min_limit, tmp);
2839 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2840 min_memsw_limit = min(min_memsw_limit, tmp);
2843 *mem_limit = min_limit;
2844 *memsw_limit = min_memsw_limit;
2848 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
2850 struct mem_cgroup *mem;
2853 mem = mem_cgroup_from_cont(cont);
2854 type = MEMFILE_TYPE(event);
2855 name = MEMFILE_ATTR(event);
2859 res_counter_reset_max(&mem->res);
2861 res_counter_reset_max(&mem->memsw);
2865 res_counter_reset_failcnt(&mem->res);
2867 res_counter_reset_failcnt(&mem->memsw);
2875 /* For read statistics */
2891 struct mcs_total_stat {
2892 s64 stat[NR_MCS_STAT];
2898 } memcg_stat_strings[NR_MCS_STAT] = {
2899 {"cache", "total_cache"},
2900 {"rss", "total_rss"},
2901 {"mapped_file", "total_mapped_file"},
2902 {"pgpgin", "total_pgpgin"},
2903 {"pgpgout", "total_pgpgout"},
2904 {"swap", "total_swap"},
2905 {"inactive_anon", "total_inactive_anon"},
2906 {"active_anon", "total_active_anon"},
2907 {"inactive_file", "total_inactive_file"},
2908 {"active_file", "total_active_file"},
2909 {"unevictable", "total_unevictable"}
2913 static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
2915 struct mcs_total_stat *s = data;
2919 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE);
2920 s->stat[MCS_CACHE] += val * PAGE_SIZE;
2921 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
2922 s->stat[MCS_RSS] += val * PAGE_SIZE;
2923 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_FILE_MAPPED);
2924 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
2925 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT);
2926 s->stat[MCS_PGPGIN] += val;
2927 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT);
2928 s->stat[MCS_PGPGOUT] += val;
2929 if (do_swap_account) {
2930 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_SWAPOUT);
2931 s->stat[MCS_SWAP] += val * PAGE_SIZE;
2935 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
2936 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
2937 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
2938 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
2939 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
2940 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
2941 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
2942 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
2943 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
2944 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
2949 mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
2951 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
2954 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
2955 struct cgroup_map_cb *cb)
2957 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
2958 struct mcs_total_stat mystat;
2961 memset(&mystat, 0, sizeof(mystat));
2962 mem_cgroup_get_local_stat(mem_cont, &mystat);
2964 for (i = 0; i < NR_MCS_STAT; i++) {
2965 if (i == MCS_SWAP && !do_swap_account)
2967 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
2970 /* Hierarchical information */
2972 unsigned long long limit, memsw_limit;
2973 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
2974 cb->fill(cb, "hierarchical_memory_limit", limit);
2975 if (do_swap_account)
2976 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
2979 memset(&mystat, 0, sizeof(mystat));
2980 mem_cgroup_get_total_stat(mem_cont, &mystat);
2981 for (i = 0; i < NR_MCS_STAT; i++) {
2982 if (i == MCS_SWAP && !do_swap_account)
2984 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
2987 #ifdef CONFIG_DEBUG_VM
2988 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
2992 struct mem_cgroup_per_zone *mz;
2993 unsigned long recent_rotated[2] = {0, 0};
2994 unsigned long recent_scanned[2] = {0, 0};
2996 for_each_online_node(nid)
2997 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2998 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
3000 recent_rotated[0] +=
3001 mz->reclaim_stat.recent_rotated[0];
3002 recent_rotated[1] +=
3003 mz->reclaim_stat.recent_rotated[1];
3004 recent_scanned[0] +=
3005 mz->reclaim_stat.recent_scanned[0];
3006 recent_scanned[1] +=
3007 mz->reclaim_stat.recent_scanned[1];
3009 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
3010 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
3011 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
3012 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
3019 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
3021 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3023 return get_swappiness(memcg);
3026 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
3029 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3030 struct mem_cgroup *parent;
3035 if (cgrp->parent == NULL)
3038 parent = mem_cgroup_from_cont(cgrp->parent);
3042 /* If under hierarchy, only empty-root can set this value */
3043 if ((parent->use_hierarchy) ||
3044 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
3049 spin_lock(&memcg->reclaim_param_lock);
3050 memcg->swappiness = val;
3051 spin_unlock(&memcg->reclaim_param_lock);
3059 static struct cftype mem_cgroup_files[] = {
3061 .name = "usage_in_bytes",
3062 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
3063 .read_u64 = mem_cgroup_read,
3066 .name = "max_usage_in_bytes",
3067 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
3068 .trigger = mem_cgroup_reset,
3069 .read_u64 = mem_cgroup_read,
3072 .name = "limit_in_bytes",
3073 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
3074 .write_string = mem_cgroup_write,
3075 .read_u64 = mem_cgroup_read,
3078 .name = "soft_limit_in_bytes",
3079 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
3080 .write_string = mem_cgroup_write,
3081 .read_u64 = mem_cgroup_read,
3085 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
3086 .trigger = mem_cgroup_reset,
3087 .read_u64 = mem_cgroup_read,
3091 .read_map = mem_control_stat_show,
3094 .name = "force_empty",
3095 .trigger = mem_cgroup_force_empty_write,
3098 .name = "use_hierarchy",
3099 .write_u64 = mem_cgroup_hierarchy_write,
3100 .read_u64 = mem_cgroup_hierarchy_read,
3103 .name = "swappiness",
3104 .read_u64 = mem_cgroup_swappiness_read,
3105 .write_u64 = mem_cgroup_swappiness_write,
3109 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3110 static struct cftype memsw_cgroup_files[] = {
3112 .name = "memsw.usage_in_bytes",
3113 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
3114 .read_u64 = mem_cgroup_read,
3117 .name = "memsw.max_usage_in_bytes",
3118 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
3119 .trigger = mem_cgroup_reset,
3120 .read_u64 = mem_cgroup_read,
3123 .name = "memsw.limit_in_bytes",
3124 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
3125 .write_string = mem_cgroup_write,
3126 .read_u64 = mem_cgroup_read,
3129 .name = "memsw.failcnt",
3130 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
3131 .trigger = mem_cgroup_reset,
3132 .read_u64 = mem_cgroup_read,
3136 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
3138 if (!do_swap_account)
3140 return cgroup_add_files(cont, ss, memsw_cgroup_files,
3141 ARRAY_SIZE(memsw_cgroup_files));
3144 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
3150 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
3152 struct mem_cgroup_per_node *pn;
3153 struct mem_cgroup_per_zone *mz;
3155 int zone, tmp = node;
3157 * This routine is called against possible nodes.
3158 * But it's BUG to call kmalloc() against offline node.
3160 * TODO: this routine can waste much memory for nodes which will
3161 * never be onlined. It's better to use memory hotplug callback
3164 if (!node_state(node, N_NORMAL_MEMORY))
3166 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
3170 mem->info.nodeinfo[node] = pn;
3171 memset(pn, 0, sizeof(*pn));
3173 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
3174 mz = &pn->zoneinfo[zone];
3176 INIT_LIST_HEAD(&mz->lists[l]);
3177 mz->usage_in_excess = 0;
3178 mz->on_tree = false;
3184 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
3186 kfree(mem->info.nodeinfo[node]);
3189 static int mem_cgroup_size(void)
3191 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
3192 return sizeof(struct mem_cgroup) + cpustat_size;
3195 static struct mem_cgroup *mem_cgroup_alloc(void)
3197 struct mem_cgroup *mem;
3198 int size = mem_cgroup_size();
3200 if (size < PAGE_SIZE)
3201 mem = kmalloc(size, GFP_KERNEL);
3203 mem = vmalloc(size);
3206 memset(mem, 0, size);
3211 * At destroying mem_cgroup, references from swap_cgroup can remain.
3212 * (scanning all at force_empty is too costly...)
3214 * Instead of clearing all references at force_empty, we remember
3215 * the number of reference from swap_cgroup and free mem_cgroup when
3216 * it goes down to 0.
3218 * Removal of cgroup itself succeeds regardless of refs from swap.
3221 static void __mem_cgroup_free(struct mem_cgroup *mem)
3225 mem_cgroup_remove_from_trees(mem);
3226 free_css_id(&mem_cgroup_subsys, &mem->css);
3228 for_each_node_state(node, N_POSSIBLE)
3229 free_mem_cgroup_per_zone_info(mem, node);
3231 if (mem_cgroup_size() < PAGE_SIZE)
3237 static void mem_cgroup_get(struct mem_cgroup *mem)
3239 atomic_inc(&mem->refcnt);
3242 static void mem_cgroup_put(struct mem_cgroup *mem)
3244 if (atomic_dec_and_test(&mem->refcnt)) {
3245 struct mem_cgroup *parent = parent_mem_cgroup(mem);
3246 __mem_cgroup_free(mem);
3248 mem_cgroup_put(parent);
3253 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
3255 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
3257 if (!mem->res.parent)
3259 return mem_cgroup_from_res_counter(mem->res.parent, res);
3262 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3263 static void __init enable_swap_cgroup(void)
3265 if (!mem_cgroup_disabled() && really_do_swap_account)
3266 do_swap_account = 1;
3269 static void __init enable_swap_cgroup(void)
3274 static int mem_cgroup_soft_limit_tree_init(void)
3276 struct mem_cgroup_tree_per_node *rtpn;
3277 struct mem_cgroup_tree_per_zone *rtpz;
3278 int tmp, node, zone;
3280 for_each_node_state(node, N_POSSIBLE) {
3282 if (!node_state(node, N_NORMAL_MEMORY))
3284 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
3288 soft_limit_tree.rb_tree_per_node[node] = rtpn;
3290 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
3291 rtpz = &rtpn->rb_tree_per_zone[zone];
3292 rtpz->rb_root = RB_ROOT;
3293 spin_lock_init(&rtpz->lock);
3299 static struct cgroup_subsys_state * __ref
3300 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
3302 struct mem_cgroup *mem, *parent;
3303 long error = -ENOMEM;
3306 mem = mem_cgroup_alloc();
3308 return ERR_PTR(error);
3310 for_each_node_state(node, N_POSSIBLE)
3311 if (alloc_mem_cgroup_per_zone_info(mem, node))
3315 if (cont->parent == NULL) {
3317 enable_swap_cgroup();
3319 root_mem_cgroup = mem;
3320 if (mem_cgroup_soft_limit_tree_init())
3322 for_each_possible_cpu(cpu) {
3323 struct memcg_stock_pcp *stock =
3324 &per_cpu(memcg_stock, cpu);
3325 INIT_WORK(&stock->work, drain_local_stock);
3327 hotcpu_notifier(memcg_stock_cpu_callback, 0);
3330 parent = mem_cgroup_from_cont(cont->parent);
3331 mem->use_hierarchy = parent->use_hierarchy;
3334 if (parent && parent->use_hierarchy) {
3335 res_counter_init(&mem->res, &parent->res);
3336 res_counter_init(&mem->memsw, &parent->memsw);
3338 * We increment refcnt of the parent to ensure that we can
3339 * safely access it on res_counter_charge/uncharge.
3340 * This refcnt will be decremented when freeing this
3341 * mem_cgroup(see mem_cgroup_put).
3343 mem_cgroup_get(parent);
3345 res_counter_init(&mem->res, NULL);
3346 res_counter_init(&mem->memsw, NULL);
3348 mem->last_scanned_child = 0;
3349 spin_lock_init(&mem->reclaim_param_lock);
3352 mem->swappiness = get_swappiness(parent);
3353 atomic_set(&mem->refcnt, 1);
3356 __mem_cgroup_free(mem);
3357 root_mem_cgroup = NULL;
3358 return ERR_PTR(error);
3361 static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
3362 struct cgroup *cont)
3364 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3366 return mem_cgroup_force_empty(mem, false);
3369 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
3370 struct cgroup *cont)
3372 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3374 mem_cgroup_put(mem);
3377 static int mem_cgroup_populate(struct cgroup_subsys *ss,
3378 struct cgroup *cont)
3382 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
3383 ARRAY_SIZE(mem_cgroup_files));
3386 ret = register_memsw_files(cont, ss);
3390 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
3391 struct cgroup *cont,
3392 struct cgroup *old_cont,
3393 struct task_struct *p,
3396 mutex_lock(&memcg_tasklist);
3398 * FIXME: It's better to move charges of this process from old
3399 * memcg to new memcg. But it's just on TODO-List now.
3401 mutex_unlock(&memcg_tasklist);
3404 struct cgroup_subsys mem_cgroup_subsys = {
3406 .subsys_id = mem_cgroup_subsys_id,
3407 .create = mem_cgroup_create,
3408 .pre_destroy = mem_cgroup_pre_destroy,
3409 .destroy = mem_cgroup_destroy,
3410 .populate = mem_cgroup_populate,
3411 .attach = mem_cgroup_move_task,
3416 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3418 static int __init disable_swap_account(char *s)
3420 really_do_swap_account = 0;
3423 __setup("noswapaccount", disable_swap_account);