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>
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
24 #include <linux/res_counter.h>
25 #include <linux/memcontrol.h>
26 #include <linux/cgroup.h>
28 #include <linux/hugetlb.h>
29 #include <linux/pagemap.h>
30 #include <linux/smp.h>
31 #include <linux/page-flags.h>
32 #include <linux/backing-dev.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/rcupdate.h>
35 #include <linux/limits.h>
36 #include <linux/mutex.h>
37 #include <linux/rbtree.h>
38 #include <linux/slab.h>
39 #include <linux/swap.h>
40 #include <linux/swapops.h>
41 #include <linux/spinlock.h>
42 #include <linux/eventfd.h>
43 #include <linux/sort.h>
45 #include <linux/seq_file.h>
46 #include <linux/vmalloc.h>
47 #include <linux/mm_inline.h>
48 #include <linux/page_cgroup.h>
49 #include <linux/cpu.h>
50 #include <linux/oom.h>
53 #include <asm/uaccess.h>
55 #include <trace/events/vmscan.h>
57 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
58 #define MEM_CGROUP_RECLAIM_RETRIES 5
59 struct mem_cgroup *root_mem_cgroup __read_mostly;
61 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
62 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
63 int do_swap_account __read_mostly;
65 /* for remember boot option*/
66 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
67 static int really_do_swap_account __initdata = 1;
69 static int really_do_swap_account __initdata = 0;
73 #define do_swap_account (0)
77 * Per memcg event counter is incremented at every pagein/pageout. This counter
78 * is used for trigger some periodic events. This is straightforward and better
79 * than using jiffies etc. to handle periodic memcg event.
81 * These values will be used as !((event) & ((1 <<(thresh)) - 1))
83 #define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */
84 #define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */
87 * Statistics for memory cgroup.
89 enum mem_cgroup_stat_index {
91 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
93 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
94 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
95 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
96 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
97 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
98 MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
99 MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */
100 /* incremented at every pagein/pageout */
101 MEM_CGROUP_EVENTS = MEM_CGROUP_STAT_DATA,
102 MEM_CGROUP_ON_MOVE, /* someone is moving account between groups */
104 MEM_CGROUP_STAT_NSTATS,
107 struct mem_cgroup_stat_cpu {
108 s64 count[MEM_CGROUP_STAT_NSTATS];
112 * per-zone information in memory controller.
114 struct mem_cgroup_per_zone {
116 * spin_lock to protect the per cgroup LRU
118 struct list_head lists[NR_LRU_LISTS];
119 unsigned long count[NR_LRU_LISTS];
121 struct zone_reclaim_stat reclaim_stat;
122 struct rb_node tree_node; /* RB tree node */
123 unsigned long long usage_in_excess;/* Set to the value by which */
124 /* the soft limit is exceeded*/
126 struct mem_cgroup *mem; /* Back pointer, we cannot */
127 /* use container_of */
129 /* Macro for accessing counter */
130 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
132 struct mem_cgroup_per_node {
133 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
136 struct mem_cgroup_lru_info {
137 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
141 * Cgroups above their limits are maintained in a RB-Tree, independent of
142 * their hierarchy representation
145 struct mem_cgroup_tree_per_zone {
146 struct rb_root rb_root;
150 struct mem_cgroup_tree_per_node {
151 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
154 struct mem_cgroup_tree {
155 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
158 static struct mem_cgroup_tree soft_limit_tree __read_mostly;
160 struct mem_cgroup_threshold {
161 struct eventfd_ctx *eventfd;
166 struct mem_cgroup_threshold_ary {
167 /* An array index points to threshold just below usage. */
168 int current_threshold;
169 /* Size of entries[] */
171 /* Array of thresholds */
172 struct mem_cgroup_threshold entries[0];
175 struct mem_cgroup_thresholds {
176 /* Primary thresholds array */
177 struct mem_cgroup_threshold_ary *primary;
179 * Spare threshold array.
180 * This is needed to make mem_cgroup_unregister_event() "never fail".
181 * It must be able to store at least primary->size - 1 entries.
183 struct mem_cgroup_threshold_ary *spare;
187 struct mem_cgroup_eventfd_list {
188 struct list_head list;
189 struct eventfd_ctx *eventfd;
192 static void mem_cgroup_threshold(struct mem_cgroup *mem);
193 static void mem_cgroup_oom_notify(struct mem_cgroup *mem);
196 * The memory controller data structure. The memory controller controls both
197 * page cache and RSS per cgroup. We would eventually like to provide
198 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
199 * to help the administrator determine what knobs to tune.
201 * TODO: Add a water mark for the memory controller. Reclaim will begin when
202 * we hit the water mark. May be even add a low water mark, such that
203 * no reclaim occurs from a cgroup at it's low water mark, this is
204 * a feature that will be implemented much later in the future.
207 struct cgroup_subsys_state css;
209 * the counter to account for memory usage
211 struct res_counter res;
213 * the counter to account for mem+swap usage.
215 struct res_counter memsw;
217 * Per cgroup active and inactive list, similar to the
218 * per zone LRU lists.
220 struct mem_cgroup_lru_info info;
223 protect against reclaim related member.
225 spinlock_t reclaim_param_lock;
228 * While reclaiming in a hierarchy, we cache the last child we
231 int last_scanned_child;
233 * Should the accounting and control be hierarchical, per subtree?
239 unsigned int swappiness;
240 /* OOM-Killer disable */
241 int oom_kill_disable;
243 /* set when res.limit == memsw.limit */
244 bool memsw_is_minimum;
246 /* protect arrays of thresholds */
247 struct mutex thresholds_lock;
249 /* thresholds for memory usage. RCU-protected */
250 struct mem_cgroup_thresholds thresholds;
252 /* thresholds for mem+swap usage. RCU-protected */
253 struct mem_cgroup_thresholds memsw_thresholds;
255 /* For oom notifier event fd */
256 struct list_head oom_notify;
259 * Should we move charges of a task when a task is moved into this
260 * mem_cgroup ? And what type of charges should we move ?
262 unsigned long move_charge_at_immigrate;
266 struct mem_cgroup_stat_cpu *stat;
268 * used when a cpu is offlined or other synchronizations
269 * See mem_cgroup_read_stat().
271 struct mem_cgroup_stat_cpu nocpu_base;
272 spinlock_t pcp_counter_lock;
275 /* Stuffs for move charges at task migration. */
277 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
278 * left-shifted bitmap of these types.
281 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
282 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
286 /* "mc" and its members are protected by cgroup_mutex */
287 static struct move_charge_struct {
288 spinlock_t lock; /* for from, to */
289 struct mem_cgroup *from;
290 struct mem_cgroup *to;
291 unsigned long precharge;
292 unsigned long moved_charge;
293 unsigned long moved_swap;
294 struct task_struct *moving_task; /* a task moving charges */
295 wait_queue_head_t waitq; /* a waitq for other context */
297 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
298 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
301 static bool move_anon(void)
303 return test_bit(MOVE_CHARGE_TYPE_ANON,
304 &mc.to->move_charge_at_immigrate);
307 static bool move_file(void)
309 return test_bit(MOVE_CHARGE_TYPE_FILE,
310 &mc.to->move_charge_at_immigrate);
314 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
315 * limit reclaim to prevent infinite loops, if they ever occur.
317 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
318 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
321 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
322 MEM_CGROUP_CHARGE_TYPE_MAPPED,
323 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
324 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
325 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
326 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
330 /* only for here (for easy reading.) */
331 #define PCGF_CACHE (1UL << PCG_CACHE)
332 #define PCGF_USED (1UL << PCG_USED)
333 #define PCGF_LOCK (1UL << PCG_LOCK)
334 /* Not used, but added here for completeness */
335 #define PCGF_ACCT (1UL << PCG_ACCT)
337 /* for encoding cft->private value on file */
340 #define _OOM_TYPE (2)
341 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
342 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
343 #define MEMFILE_ATTR(val) ((val) & 0xffff)
344 /* Used for OOM nofiier */
345 #define OOM_CONTROL (0)
348 * Reclaim flags for mem_cgroup_hierarchical_reclaim
350 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
351 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
352 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
353 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
354 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
355 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
357 static void mem_cgroup_get(struct mem_cgroup *mem);
358 static void mem_cgroup_put(struct mem_cgroup *mem);
359 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
360 static void drain_all_stock_async(void);
362 static struct mem_cgroup_per_zone *
363 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
365 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
368 struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem)
373 static struct mem_cgroup_per_zone *
374 page_cgroup_zoneinfo(struct page_cgroup *pc)
376 struct mem_cgroup *mem = pc->mem_cgroup;
377 int nid = page_cgroup_nid(pc);
378 int zid = page_cgroup_zid(pc);
383 return mem_cgroup_zoneinfo(mem, nid, zid);
386 static struct mem_cgroup_tree_per_zone *
387 soft_limit_tree_node_zone(int nid, int zid)
389 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
392 static struct mem_cgroup_tree_per_zone *
393 soft_limit_tree_from_page(struct page *page)
395 int nid = page_to_nid(page);
396 int zid = page_zonenum(page);
398 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
402 __mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
403 struct mem_cgroup_per_zone *mz,
404 struct mem_cgroup_tree_per_zone *mctz,
405 unsigned long long new_usage_in_excess)
407 struct rb_node **p = &mctz->rb_root.rb_node;
408 struct rb_node *parent = NULL;
409 struct mem_cgroup_per_zone *mz_node;
414 mz->usage_in_excess = new_usage_in_excess;
415 if (!mz->usage_in_excess)
419 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
421 if (mz->usage_in_excess < mz_node->usage_in_excess)
424 * We can't avoid mem cgroups that are over their soft
425 * limit by the same amount
427 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
430 rb_link_node(&mz->tree_node, parent, p);
431 rb_insert_color(&mz->tree_node, &mctz->rb_root);
436 __mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
437 struct mem_cgroup_per_zone *mz,
438 struct mem_cgroup_tree_per_zone *mctz)
442 rb_erase(&mz->tree_node, &mctz->rb_root);
447 mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
448 struct mem_cgroup_per_zone *mz,
449 struct mem_cgroup_tree_per_zone *mctz)
451 spin_lock(&mctz->lock);
452 __mem_cgroup_remove_exceeded(mem, mz, mctz);
453 spin_unlock(&mctz->lock);
457 static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page)
459 unsigned long long excess;
460 struct mem_cgroup_per_zone *mz;
461 struct mem_cgroup_tree_per_zone *mctz;
462 int nid = page_to_nid(page);
463 int zid = page_zonenum(page);
464 mctz = soft_limit_tree_from_page(page);
467 * Necessary to update all ancestors when hierarchy is used.
468 * because their event counter is not touched.
470 for (; mem; mem = parent_mem_cgroup(mem)) {
471 mz = mem_cgroup_zoneinfo(mem, nid, zid);
472 excess = res_counter_soft_limit_excess(&mem->res);
474 * We have to update the tree if mz is on RB-tree or
475 * mem is over its softlimit.
477 if (excess || mz->on_tree) {
478 spin_lock(&mctz->lock);
479 /* if on-tree, remove it */
481 __mem_cgroup_remove_exceeded(mem, mz, mctz);
483 * Insert again. mz->usage_in_excess will be updated.
484 * If excess is 0, no tree ops.
486 __mem_cgroup_insert_exceeded(mem, mz, mctz, excess);
487 spin_unlock(&mctz->lock);
492 static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem)
495 struct mem_cgroup_per_zone *mz;
496 struct mem_cgroup_tree_per_zone *mctz;
498 for_each_node_state(node, N_POSSIBLE) {
499 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
500 mz = mem_cgroup_zoneinfo(mem, node, zone);
501 mctz = soft_limit_tree_node_zone(node, zone);
502 mem_cgroup_remove_exceeded(mem, mz, mctz);
507 static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem)
509 return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT;
512 static struct mem_cgroup_per_zone *
513 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
515 struct rb_node *rightmost = NULL;
516 struct mem_cgroup_per_zone *mz;
520 rightmost = rb_last(&mctz->rb_root);
522 goto done; /* Nothing to reclaim from */
524 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
526 * Remove the node now but someone else can add it back,
527 * we will to add it back at the end of reclaim to its correct
528 * position in the tree.
530 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
531 if (!res_counter_soft_limit_excess(&mz->mem->res) ||
532 !css_tryget(&mz->mem->css))
538 static struct mem_cgroup_per_zone *
539 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
541 struct mem_cgroup_per_zone *mz;
543 spin_lock(&mctz->lock);
544 mz = __mem_cgroup_largest_soft_limit_node(mctz);
545 spin_unlock(&mctz->lock);
550 * Implementation Note: reading percpu statistics for memcg.
552 * Both of vmstat[] and percpu_counter has threshold and do periodic
553 * synchronization to implement "quick" read. There are trade-off between
554 * reading cost and precision of value. Then, we may have a chance to implement
555 * a periodic synchronizion of counter in memcg's counter.
557 * But this _read() function is used for user interface now. The user accounts
558 * memory usage by memory cgroup and he _always_ requires exact value because
559 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
560 * have to visit all online cpus and make sum. So, for now, unnecessary
561 * synchronization is not implemented. (just implemented for cpu hotplug)
563 * If there are kernel internal actions which can make use of some not-exact
564 * value, and reading all cpu value can be performance bottleneck in some
565 * common workload, threashold and synchonization as vmstat[] should be
568 static s64 mem_cgroup_read_stat(struct mem_cgroup *mem,
569 enum mem_cgroup_stat_index idx)
575 for_each_online_cpu(cpu)
576 val += per_cpu(mem->stat->count[idx], cpu);
577 #ifdef CONFIG_HOTPLUG_CPU
578 spin_lock(&mem->pcp_counter_lock);
579 val += mem->nocpu_base.count[idx];
580 spin_unlock(&mem->pcp_counter_lock);
586 static s64 mem_cgroup_local_usage(struct mem_cgroup *mem)
590 ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
591 ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
595 static void mem_cgroup_swap_statistics(struct mem_cgroup *mem,
598 int val = (charge) ? 1 : -1;
599 this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
602 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
603 bool file, int nr_pages)
608 __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], nr_pages);
610 __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], nr_pages);
612 /* pagein of a big page is an event. So, ignore page size */
614 __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]);
616 __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]);
618 __this_cpu_add(mem->stat->count[MEM_CGROUP_EVENTS], nr_pages);
623 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
627 struct mem_cgroup_per_zone *mz;
630 for_each_online_node(nid)
631 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
632 mz = mem_cgroup_zoneinfo(mem, nid, zid);
633 total += MEM_CGROUP_ZSTAT(mz, idx);
638 static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift)
642 val = this_cpu_read(mem->stat->count[MEM_CGROUP_EVENTS]);
644 return !(val & ((1 << event_mask_shift) - 1));
648 * Check events in order.
651 static void memcg_check_events(struct mem_cgroup *mem, struct page *page)
653 /* threshold event is triggered in finer grain than soft limit */
654 if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) {
655 mem_cgroup_threshold(mem);
656 if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH)))
657 mem_cgroup_update_tree(mem, page);
661 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
663 return container_of(cgroup_subsys_state(cont,
664 mem_cgroup_subsys_id), struct mem_cgroup,
668 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
671 * mm_update_next_owner() may clear mm->owner to NULL
672 * if it races with swapoff, page migration, etc.
673 * So this can be called with p == NULL.
678 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
679 struct mem_cgroup, css);
682 static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
684 struct mem_cgroup *mem = NULL;
689 * Because we have no locks, mm->owner's may be being moved to other
690 * cgroup. We use css_tryget() here even if this looks
691 * pessimistic (rather than adding locks here).
695 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
698 } while (!css_tryget(&mem->css));
703 /* The caller has to guarantee "mem" exists before calling this */
704 static struct mem_cgroup *mem_cgroup_start_loop(struct mem_cgroup *mem)
706 struct cgroup_subsys_state *css;
709 if (!mem) /* ROOT cgroup has the smallest ID */
710 return root_mem_cgroup; /*css_put/get against root is ignored*/
711 if (!mem->use_hierarchy) {
712 if (css_tryget(&mem->css))
718 * searching a memory cgroup which has the smallest ID under given
719 * ROOT cgroup. (ID >= 1)
721 css = css_get_next(&mem_cgroup_subsys, 1, &mem->css, &found);
722 if (css && css_tryget(css))
723 mem = container_of(css, struct mem_cgroup, css);
730 static struct mem_cgroup *mem_cgroup_get_next(struct mem_cgroup *iter,
731 struct mem_cgroup *root,
734 int nextid = css_id(&iter->css) + 1;
737 struct cgroup_subsys_state *css;
739 hierarchy_used = iter->use_hierarchy;
742 /* If no ROOT, walk all, ignore hierarchy */
743 if (!cond || (root && !hierarchy_used))
747 root = root_mem_cgroup;
753 css = css_get_next(&mem_cgroup_subsys, nextid,
755 if (css && css_tryget(css))
756 iter = container_of(css, struct mem_cgroup, css);
758 /* If css is NULL, no more cgroups will be found */
760 } while (css && !iter);
765 * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please
766 * be careful that "break" loop is not allowed. We have reference count.
767 * Instead of that modify "cond" to be false and "continue" to exit the loop.
769 #define for_each_mem_cgroup_tree_cond(iter, root, cond) \
770 for (iter = mem_cgroup_start_loop(root);\
772 iter = mem_cgroup_get_next(iter, root, cond))
774 #define for_each_mem_cgroup_tree(iter, root) \
775 for_each_mem_cgroup_tree_cond(iter, root, true)
777 #define for_each_mem_cgroup_all(iter) \
778 for_each_mem_cgroup_tree_cond(iter, NULL, true)
781 static inline bool mem_cgroup_is_root(struct mem_cgroup *mem)
783 return (mem == root_mem_cgroup);
787 * Following LRU functions are allowed to be used without PCG_LOCK.
788 * Operations are called by routine of global LRU independently from memcg.
789 * What we have to take care of here is validness of pc->mem_cgroup.
791 * Changes to pc->mem_cgroup happens when
794 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
795 * It is added to LRU before charge.
796 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
797 * When moving account, the page is not on LRU. It's isolated.
800 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
802 struct page_cgroup *pc;
803 struct mem_cgroup_per_zone *mz;
805 if (mem_cgroup_disabled())
807 pc = lookup_page_cgroup(page);
808 /* can happen while we handle swapcache. */
809 if (!TestClearPageCgroupAcctLRU(pc))
811 VM_BUG_ON(!pc->mem_cgroup);
813 * We don't check PCG_USED bit. It's cleared when the "page" is finally
814 * removed from global LRU.
816 mz = page_cgroup_zoneinfo(pc);
817 /* huge page split is done under lru_lock. so, we have no races. */
818 MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page);
819 if (mem_cgroup_is_root(pc->mem_cgroup))
821 VM_BUG_ON(list_empty(&pc->lru));
822 list_del_init(&pc->lru);
825 void mem_cgroup_del_lru(struct page *page)
827 mem_cgroup_del_lru_list(page, page_lru(page));
830 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
832 struct mem_cgroup_per_zone *mz;
833 struct page_cgroup *pc;
835 if (mem_cgroup_disabled())
838 pc = lookup_page_cgroup(page);
840 * Used bit is set without atomic ops but after smp_wmb().
841 * For making pc->mem_cgroup visible, insert smp_rmb() here.
844 /* unused or root page is not rotated. */
845 if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup))
847 mz = page_cgroup_zoneinfo(pc);
848 list_move(&pc->lru, &mz->lists[lru]);
851 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
853 struct page_cgroup *pc;
854 struct mem_cgroup_per_zone *mz;
856 if (mem_cgroup_disabled())
858 pc = lookup_page_cgroup(page);
859 VM_BUG_ON(PageCgroupAcctLRU(pc));
861 * Used bit is set without atomic ops but after smp_wmb().
862 * For making pc->mem_cgroup visible, insert smp_rmb() here.
865 if (!PageCgroupUsed(pc))
868 mz = page_cgroup_zoneinfo(pc);
869 /* huge page split is done under lru_lock. so, we have no races. */
870 MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page);
871 SetPageCgroupAcctLRU(pc);
872 if (mem_cgroup_is_root(pc->mem_cgroup))
874 list_add(&pc->lru, &mz->lists[lru]);
878 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
879 * lru because the page may.be reused after it's fully uncharged (because of
880 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
881 * it again. This function is only used to charge SwapCache. It's done under
882 * lock_page and expected that zone->lru_lock is never held.
884 static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
887 struct zone *zone = page_zone(page);
888 struct page_cgroup *pc = lookup_page_cgroup(page);
890 spin_lock_irqsave(&zone->lru_lock, flags);
892 * Forget old LRU when this page_cgroup is *not* used. This Used bit
893 * is guarded by lock_page() because the page is SwapCache.
895 if (!PageCgroupUsed(pc))
896 mem_cgroup_del_lru_list(page, page_lru(page));
897 spin_unlock_irqrestore(&zone->lru_lock, flags);
900 static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
903 struct zone *zone = page_zone(page);
904 struct page_cgroup *pc = lookup_page_cgroup(page);
906 spin_lock_irqsave(&zone->lru_lock, flags);
907 /* link when the page is linked to LRU but page_cgroup isn't */
908 if (PageLRU(page) && !PageCgroupAcctLRU(pc))
909 mem_cgroup_add_lru_list(page, page_lru(page));
910 spin_unlock_irqrestore(&zone->lru_lock, flags);
914 void mem_cgroup_move_lists(struct page *page,
915 enum lru_list from, enum lru_list to)
917 if (mem_cgroup_disabled())
919 mem_cgroup_del_lru_list(page, from);
920 mem_cgroup_add_lru_list(page, to);
923 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
926 struct mem_cgroup *curr = NULL;
927 struct task_struct *p;
929 p = find_lock_task_mm(task);
932 curr = try_get_mem_cgroup_from_mm(p->mm);
937 * We should check use_hierarchy of "mem" not "curr". Because checking
938 * use_hierarchy of "curr" here make this function true if hierarchy is
939 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
940 * hierarchy(even if use_hierarchy is disabled in "mem").
942 if (mem->use_hierarchy)
943 ret = css_is_ancestor(&curr->css, &mem->css);
950 static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
952 unsigned long active;
953 unsigned long inactive;
955 unsigned long inactive_ratio;
957 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
958 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
960 gb = (inactive + active) >> (30 - PAGE_SHIFT);
962 inactive_ratio = int_sqrt(10 * gb);
967 present_pages[0] = inactive;
968 present_pages[1] = active;
971 return inactive_ratio;
974 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
976 unsigned long active;
977 unsigned long inactive;
978 unsigned long present_pages[2];
979 unsigned long inactive_ratio;
981 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
983 inactive = present_pages[0];
984 active = present_pages[1];
986 if (inactive * inactive_ratio < active)
992 int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
994 unsigned long active;
995 unsigned long inactive;
997 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
998 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
1000 return (active > inactive);
1003 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
1007 int nid = zone_to_nid(zone);
1008 int zid = zone_idx(zone);
1009 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
1011 return MEM_CGROUP_ZSTAT(mz, lru);
1014 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
1017 int nid = zone_to_nid(zone);
1018 int zid = zone_idx(zone);
1019 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
1021 return &mz->reclaim_stat;
1024 struct zone_reclaim_stat *
1025 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
1027 struct page_cgroup *pc;
1028 struct mem_cgroup_per_zone *mz;
1030 if (mem_cgroup_disabled())
1033 pc = lookup_page_cgroup(page);
1035 * Used bit is set without atomic ops but after smp_wmb().
1036 * For making pc->mem_cgroup visible, insert smp_rmb() here.
1039 if (!PageCgroupUsed(pc))
1042 mz = page_cgroup_zoneinfo(pc);
1046 return &mz->reclaim_stat;
1049 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
1050 struct list_head *dst,
1051 unsigned long *scanned, int order,
1052 int mode, struct zone *z,
1053 struct mem_cgroup *mem_cont,
1054 int active, int file)
1056 unsigned long nr_taken = 0;
1060 struct list_head *src;
1061 struct page_cgroup *pc, *tmp;
1062 int nid = zone_to_nid(z);
1063 int zid = zone_idx(z);
1064 struct mem_cgroup_per_zone *mz;
1065 int lru = LRU_FILE * file + active;
1069 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
1070 src = &mz->lists[lru];
1073 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
1074 if (scan >= nr_to_scan)
1078 if (unlikely(!PageCgroupUsed(pc)))
1080 if (unlikely(!PageLRU(page)))
1084 ret = __isolate_lru_page(page, mode, file);
1087 list_move(&page->lru, dst);
1088 mem_cgroup_del_lru(page);
1089 nr_taken += hpage_nr_pages(page);
1092 /* we don't affect global LRU but rotate in our LRU */
1093 mem_cgroup_rotate_lru_list(page, page_lru(page));
1102 trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken,
1108 #define mem_cgroup_from_res_counter(counter, member) \
1109 container_of(counter, struct mem_cgroup, member)
1111 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
1113 if (do_swap_account) {
1114 if (res_counter_check_under_limit(&mem->res) &&
1115 res_counter_check_under_limit(&mem->memsw))
1118 if (res_counter_check_under_limit(&mem->res))
1123 static unsigned int get_swappiness(struct mem_cgroup *memcg)
1125 struct cgroup *cgrp = memcg->css.cgroup;
1126 unsigned int swappiness;
1129 if (cgrp->parent == NULL)
1130 return vm_swappiness;
1132 spin_lock(&memcg->reclaim_param_lock);
1133 swappiness = memcg->swappiness;
1134 spin_unlock(&memcg->reclaim_param_lock);
1139 static void mem_cgroup_start_move(struct mem_cgroup *mem)
1144 spin_lock(&mem->pcp_counter_lock);
1145 for_each_online_cpu(cpu)
1146 per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1;
1147 mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1;
1148 spin_unlock(&mem->pcp_counter_lock);
1154 static void mem_cgroup_end_move(struct mem_cgroup *mem)
1161 spin_lock(&mem->pcp_counter_lock);
1162 for_each_online_cpu(cpu)
1163 per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1;
1164 mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1;
1165 spin_unlock(&mem->pcp_counter_lock);
1169 * 2 routines for checking "mem" is under move_account() or not.
1171 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1172 * for avoiding race in accounting. If true,
1173 * pc->mem_cgroup may be overwritten.
1175 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1176 * under hierarchy of moving cgroups. This is for
1177 * waiting at hith-memory prressure caused by "move".
1180 static bool mem_cgroup_stealed(struct mem_cgroup *mem)
1182 VM_BUG_ON(!rcu_read_lock_held());
1183 return this_cpu_read(mem->stat->count[MEM_CGROUP_ON_MOVE]) > 0;
1186 static bool mem_cgroup_under_move(struct mem_cgroup *mem)
1188 struct mem_cgroup *from;
1189 struct mem_cgroup *to;
1192 * Unlike task_move routines, we access mc.to, mc.from not under
1193 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1195 spin_lock(&mc.lock);
1200 if (from == mem || to == mem
1201 || (mem->use_hierarchy && css_is_ancestor(&from->css, &mem->css))
1202 || (mem->use_hierarchy && css_is_ancestor(&to->css, &mem->css)))
1205 spin_unlock(&mc.lock);
1209 static bool mem_cgroup_wait_acct_move(struct mem_cgroup *mem)
1211 if (mc.moving_task && current != mc.moving_task) {
1212 if (mem_cgroup_under_move(mem)) {
1214 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1215 /* moving charge context might have finished. */
1218 finish_wait(&mc.waitq, &wait);
1226 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1227 * @memcg: The memory cgroup that went over limit
1228 * @p: Task that is going to be killed
1230 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1233 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1235 struct cgroup *task_cgrp;
1236 struct cgroup *mem_cgrp;
1238 * Need a buffer in BSS, can't rely on allocations. The code relies
1239 * on the assumption that OOM is serialized for memory controller.
1240 * If this assumption is broken, revisit this code.
1242 static char memcg_name[PATH_MAX];
1251 mem_cgrp = memcg->css.cgroup;
1252 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1254 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1257 * Unfortunately, we are unable to convert to a useful name
1258 * But we'll still print out the usage information
1265 printk(KERN_INFO "Task in %s killed", memcg_name);
1268 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1276 * Continues from above, so we don't need an KERN_ level
1278 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1281 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1282 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1283 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1284 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1285 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1287 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1288 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1289 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1293 * This function returns the number of memcg under hierarchy tree. Returns
1294 * 1(self count) if no children.
1296 static int mem_cgroup_count_children(struct mem_cgroup *mem)
1299 struct mem_cgroup *iter;
1301 for_each_mem_cgroup_tree(iter, mem)
1307 * Return the memory (and swap, if configured) limit for a memcg.
1309 u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1314 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1315 limit += total_swap_pages << PAGE_SHIFT;
1317 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1319 * If memsw is finite and limits the amount of swap space available
1320 * to this memcg, return that limit.
1322 return min(limit, memsw);
1326 * Visit the first child (need not be the first child as per the ordering
1327 * of the cgroup list, since we track last_scanned_child) of @mem and use
1328 * that to reclaim free pages from.
1330 static struct mem_cgroup *
1331 mem_cgroup_select_victim(struct mem_cgroup *root_mem)
1333 struct mem_cgroup *ret = NULL;
1334 struct cgroup_subsys_state *css;
1337 if (!root_mem->use_hierarchy) {
1338 css_get(&root_mem->css);
1344 nextid = root_mem->last_scanned_child + 1;
1345 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
1347 if (css && css_tryget(css))
1348 ret = container_of(css, struct mem_cgroup, css);
1351 /* Updates scanning parameter */
1352 spin_lock(&root_mem->reclaim_param_lock);
1354 /* this means start scan from ID:1 */
1355 root_mem->last_scanned_child = 0;
1357 root_mem->last_scanned_child = found;
1358 spin_unlock(&root_mem->reclaim_param_lock);
1365 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1366 * we reclaimed from, so that we don't end up penalizing one child extensively
1367 * based on its position in the children list.
1369 * root_mem is the original ancestor that we've been reclaim from.
1371 * We give up and return to the caller when we visit root_mem twice.
1372 * (other groups can be removed while we're walking....)
1374 * If shrink==true, for avoiding to free too much, this returns immedieately.
1376 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
1379 unsigned long reclaim_options)
1381 struct mem_cgroup *victim;
1384 bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
1385 bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
1386 bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT;
1387 unsigned long excess = mem_cgroup_get_excess(root_mem);
1389 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1390 if (root_mem->memsw_is_minimum)
1394 victim = mem_cgroup_select_victim(root_mem);
1395 if (victim == root_mem) {
1398 drain_all_stock_async();
1401 * If we have not been able to reclaim
1402 * anything, it might because there are
1403 * no reclaimable pages under this hierarchy
1405 if (!check_soft || !total) {
1406 css_put(&victim->css);
1410 * We want to do more targetted reclaim.
1411 * excess >> 2 is not to excessive so as to
1412 * reclaim too much, nor too less that we keep
1413 * coming back to reclaim from this cgroup
1415 if (total >= (excess >> 2) ||
1416 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) {
1417 css_put(&victim->css);
1422 if (!mem_cgroup_local_usage(victim)) {
1423 /* this cgroup's local usage == 0 */
1424 css_put(&victim->css);
1427 /* we use swappiness of local cgroup */
1429 ret = mem_cgroup_shrink_node_zone(victim, gfp_mask,
1430 noswap, get_swappiness(victim), zone);
1432 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask,
1433 noswap, get_swappiness(victim));
1434 css_put(&victim->css);
1436 * At shrinking usage, we can't check we should stop here or
1437 * reclaim more. It's depends on callers. last_scanned_child
1438 * will work enough for keeping fairness under tree.
1444 if (res_counter_check_under_soft_limit(&root_mem->res))
1446 } else if (mem_cgroup_check_under_limit(root_mem))
1453 * Check OOM-Killer is already running under our hierarchy.
1454 * If someone is running, return false.
1456 static bool mem_cgroup_oom_lock(struct mem_cgroup *mem)
1458 int x, lock_count = 0;
1459 struct mem_cgroup *iter;
1461 for_each_mem_cgroup_tree(iter, mem) {
1462 x = atomic_inc_return(&iter->oom_lock);
1463 lock_count = max(x, lock_count);
1466 if (lock_count == 1)
1471 static int mem_cgroup_oom_unlock(struct mem_cgroup *mem)
1473 struct mem_cgroup *iter;
1476 * When a new child is created while the hierarchy is under oom,
1477 * mem_cgroup_oom_lock() may not be called. We have to use
1478 * atomic_add_unless() here.
1480 for_each_mem_cgroup_tree(iter, mem)
1481 atomic_add_unless(&iter->oom_lock, -1, 0);
1486 static DEFINE_MUTEX(memcg_oom_mutex);
1487 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1489 struct oom_wait_info {
1490 struct mem_cgroup *mem;
1494 static int memcg_oom_wake_function(wait_queue_t *wait,
1495 unsigned mode, int sync, void *arg)
1497 struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg;
1498 struct oom_wait_info *oom_wait_info;
1500 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
1502 if (oom_wait_info->mem == wake_mem)
1504 /* if no hierarchy, no match */
1505 if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy)
1508 * Both of oom_wait_info->mem and wake_mem are stable under us.
1509 * Then we can use css_is_ancestor without taking care of RCU.
1511 if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) &&
1512 !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css))
1516 return autoremove_wake_function(wait, mode, sync, arg);
1519 static void memcg_wakeup_oom(struct mem_cgroup *mem)
1521 /* for filtering, pass "mem" as argument. */
1522 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem);
1525 static void memcg_oom_recover(struct mem_cgroup *mem)
1527 if (mem && atomic_read(&mem->oom_lock))
1528 memcg_wakeup_oom(mem);
1532 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1534 bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask)
1536 struct oom_wait_info owait;
1537 bool locked, need_to_kill;
1540 owait.wait.flags = 0;
1541 owait.wait.func = memcg_oom_wake_function;
1542 owait.wait.private = current;
1543 INIT_LIST_HEAD(&owait.wait.task_list);
1544 need_to_kill = true;
1545 /* At first, try to OOM lock hierarchy under mem.*/
1546 mutex_lock(&memcg_oom_mutex);
1547 locked = mem_cgroup_oom_lock(mem);
1549 * Even if signal_pending(), we can't quit charge() loop without
1550 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1551 * under OOM is always welcomed, use TASK_KILLABLE here.
1553 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
1554 if (!locked || mem->oom_kill_disable)
1555 need_to_kill = false;
1557 mem_cgroup_oom_notify(mem);
1558 mutex_unlock(&memcg_oom_mutex);
1561 finish_wait(&memcg_oom_waitq, &owait.wait);
1562 mem_cgroup_out_of_memory(mem, mask);
1565 finish_wait(&memcg_oom_waitq, &owait.wait);
1567 mutex_lock(&memcg_oom_mutex);
1568 mem_cgroup_oom_unlock(mem);
1569 memcg_wakeup_oom(mem);
1570 mutex_unlock(&memcg_oom_mutex);
1572 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1574 /* Give chance to dying process */
1575 schedule_timeout(1);
1580 * Currently used to update mapped file statistics, but the routine can be
1581 * generalized to update other statistics as well.
1583 * Notes: Race condition
1585 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1586 * it tends to be costly. But considering some conditions, we doesn't need
1587 * to do so _always_.
1589 * Considering "charge", lock_page_cgroup() is not required because all
1590 * file-stat operations happen after a page is attached to radix-tree. There
1591 * are no race with "charge".
1593 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1594 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1595 * if there are race with "uncharge". Statistics itself is properly handled
1598 * Considering "move", this is an only case we see a race. To make the race
1599 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1600 * possibility of race condition. If there is, we take a lock.
1603 void mem_cgroup_update_page_stat(struct page *page,
1604 enum mem_cgroup_page_stat_item idx, int val)
1606 struct mem_cgroup *mem;
1607 struct page_cgroup *pc = lookup_page_cgroup(page);
1608 bool need_unlock = false;
1609 unsigned long uninitialized_var(flags);
1615 mem = pc->mem_cgroup;
1616 if (unlikely(!mem || !PageCgroupUsed(pc)))
1618 /* pc->mem_cgroup is unstable ? */
1619 if (unlikely(mem_cgroup_stealed(mem)) || PageTransHuge(page)) {
1620 /* take a lock against to access pc->mem_cgroup */
1621 move_lock_page_cgroup(pc, &flags);
1623 mem = pc->mem_cgroup;
1624 if (!mem || !PageCgroupUsed(pc))
1629 case MEMCG_NR_FILE_MAPPED:
1631 SetPageCgroupFileMapped(pc);
1632 else if (!page_mapped(page))
1633 ClearPageCgroupFileMapped(pc);
1634 idx = MEM_CGROUP_STAT_FILE_MAPPED;
1640 this_cpu_add(mem->stat->count[idx], val);
1643 if (unlikely(need_unlock))
1644 move_unlock_page_cgroup(pc, &flags);
1648 EXPORT_SYMBOL(mem_cgroup_update_page_stat);
1651 * size of first charge trial. "32" comes from vmscan.c's magic value.
1652 * TODO: maybe necessary to use big numbers in big irons.
1654 #define CHARGE_SIZE (32 * PAGE_SIZE)
1655 struct memcg_stock_pcp {
1656 struct mem_cgroup *cached; /* this never be root cgroup */
1658 struct work_struct work;
1660 static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
1661 static atomic_t memcg_drain_count;
1664 * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
1665 * from local stock and true is returned. If the stock is 0 or charges from a
1666 * cgroup which is not current target, returns false. This stock will be
1669 static bool consume_stock(struct mem_cgroup *mem)
1671 struct memcg_stock_pcp *stock;
1674 stock = &get_cpu_var(memcg_stock);
1675 if (mem == stock->cached && stock->charge)
1676 stock->charge -= PAGE_SIZE;
1677 else /* need to call res_counter_charge */
1679 put_cpu_var(memcg_stock);
1684 * Returns stocks cached in percpu to res_counter and reset cached information.
1686 static void drain_stock(struct memcg_stock_pcp *stock)
1688 struct mem_cgroup *old = stock->cached;
1690 if (stock->charge) {
1691 res_counter_uncharge(&old->res, stock->charge);
1692 if (do_swap_account)
1693 res_counter_uncharge(&old->memsw, stock->charge);
1695 stock->cached = NULL;
1700 * This must be called under preempt disabled or must be called by
1701 * a thread which is pinned to local cpu.
1703 static void drain_local_stock(struct work_struct *dummy)
1705 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
1710 * Cache charges(val) which is from res_counter, to local per_cpu area.
1711 * This will be consumed by consume_stock() function, later.
1713 static void refill_stock(struct mem_cgroup *mem, int val)
1715 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1717 if (stock->cached != mem) { /* reset if necessary */
1719 stock->cached = mem;
1721 stock->charge += val;
1722 put_cpu_var(memcg_stock);
1726 * Tries to drain stocked charges in other cpus. This function is asynchronous
1727 * and just put a work per cpu for draining localy on each cpu. Caller can
1728 * expects some charges will be back to res_counter later but cannot wait for
1731 static void drain_all_stock_async(void)
1734 /* This function is for scheduling "drain" in asynchronous way.
1735 * The result of "drain" is not directly handled by callers. Then,
1736 * if someone is calling drain, we don't have to call drain more.
1737 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1738 * there is a race. We just do loose check here.
1740 if (atomic_read(&memcg_drain_count))
1742 /* Notify other cpus that system-wide "drain" is running */
1743 atomic_inc(&memcg_drain_count);
1745 for_each_online_cpu(cpu) {
1746 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
1747 schedule_work_on(cpu, &stock->work);
1750 atomic_dec(&memcg_drain_count);
1751 /* We don't wait for flush_work */
1754 /* This is a synchronous drain interface. */
1755 static void drain_all_stock_sync(void)
1757 /* called when force_empty is called */
1758 atomic_inc(&memcg_drain_count);
1759 schedule_on_each_cpu(drain_local_stock);
1760 atomic_dec(&memcg_drain_count);
1764 * This function drains percpu counter value from DEAD cpu and
1765 * move it to local cpu. Note that this function can be preempted.
1767 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *mem, int cpu)
1771 spin_lock(&mem->pcp_counter_lock);
1772 for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) {
1773 s64 x = per_cpu(mem->stat->count[i], cpu);
1775 per_cpu(mem->stat->count[i], cpu) = 0;
1776 mem->nocpu_base.count[i] += x;
1778 /* need to clear ON_MOVE value, works as a kind of lock. */
1779 per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0;
1780 spin_unlock(&mem->pcp_counter_lock);
1783 static void synchronize_mem_cgroup_on_move(struct mem_cgroup *mem, int cpu)
1785 int idx = MEM_CGROUP_ON_MOVE;
1787 spin_lock(&mem->pcp_counter_lock);
1788 per_cpu(mem->stat->count[idx], cpu) = mem->nocpu_base.count[idx];
1789 spin_unlock(&mem->pcp_counter_lock);
1792 static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
1793 unsigned long action,
1796 int cpu = (unsigned long)hcpu;
1797 struct memcg_stock_pcp *stock;
1798 struct mem_cgroup *iter;
1800 if ((action == CPU_ONLINE)) {
1801 for_each_mem_cgroup_all(iter)
1802 synchronize_mem_cgroup_on_move(iter, cpu);
1806 if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN)
1809 for_each_mem_cgroup_all(iter)
1810 mem_cgroup_drain_pcp_counter(iter, cpu);
1812 stock = &per_cpu(memcg_stock, cpu);
1818 /* See __mem_cgroup_try_charge() for details */
1820 CHARGE_OK, /* success */
1821 CHARGE_RETRY, /* need to retry but retry is not bad */
1822 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
1823 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
1824 CHARGE_OOM_DIE, /* the current is killed because of OOM */
1827 static int __mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask,
1828 int csize, bool oom_check)
1830 struct mem_cgroup *mem_over_limit;
1831 struct res_counter *fail_res;
1832 unsigned long flags = 0;
1835 ret = res_counter_charge(&mem->res, csize, &fail_res);
1838 if (!do_swap_account)
1840 ret = res_counter_charge(&mem->memsw, csize, &fail_res);
1844 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
1845 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
1847 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
1849 if (csize > PAGE_SIZE) /* change csize and retry */
1850 return CHARGE_RETRY;
1852 if (!(gfp_mask & __GFP_WAIT))
1853 return CHARGE_WOULDBLOCK;
1855 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
1858 * try_to_free_mem_cgroup_pages() might not give us a full
1859 * picture of reclaim. Some pages are reclaimed and might be
1860 * moved to swap cache or just unmapped from the cgroup.
1861 * Check the limit again to see if the reclaim reduced the
1862 * current usage of the cgroup before giving up
1864 if (ret || mem_cgroup_check_under_limit(mem_over_limit))
1865 return CHARGE_RETRY;
1868 * At task move, charge accounts can be doubly counted. So, it's
1869 * better to wait until the end of task_move if something is going on.
1871 if (mem_cgroup_wait_acct_move(mem_over_limit))
1872 return CHARGE_RETRY;
1874 /* If we don't need to call oom-killer at el, return immediately */
1876 return CHARGE_NOMEM;
1878 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask))
1879 return CHARGE_OOM_DIE;
1881 return CHARGE_RETRY;
1885 * Unlike exported interface, "oom" parameter is added. if oom==true,
1886 * oom-killer can be invoked.
1888 static int __mem_cgroup_try_charge(struct mm_struct *mm,
1890 struct mem_cgroup **memcg, bool oom,
1893 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
1894 struct mem_cgroup *mem = NULL;
1896 int csize = max(CHARGE_SIZE, (unsigned long) page_size);
1899 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
1900 * in system level. So, allow to go ahead dying process in addition to
1903 if (unlikely(test_thread_flag(TIF_MEMDIE)
1904 || fatal_signal_pending(current)))
1908 * We always charge the cgroup the mm_struct belongs to.
1909 * The mm_struct's mem_cgroup changes on task migration if the
1910 * thread group leader migrates. It's possible that mm is not
1911 * set, if so charge the init_mm (happens for pagecache usage).
1916 if (*memcg) { /* css should be a valid one */
1918 VM_BUG_ON(css_is_removed(&mem->css));
1919 if (mem_cgroup_is_root(mem))
1921 if (page_size == PAGE_SIZE && consume_stock(mem))
1925 struct task_struct *p;
1928 p = rcu_dereference(mm->owner);
1930 * Because we don't have task_lock(), "p" can exit.
1931 * In that case, "mem" can point to root or p can be NULL with
1932 * race with swapoff. Then, we have small risk of mis-accouning.
1933 * But such kind of mis-account by race always happens because
1934 * we don't have cgroup_mutex(). It's overkill and we allo that
1936 * (*) swapoff at el will charge against mm-struct not against
1937 * task-struct. So, mm->owner can be NULL.
1939 mem = mem_cgroup_from_task(p);
1940 if (!mem || mem_cgroup_is_root(mem)) {
1944 if (page_size == PAGE_SIZE && consume_stock(mem)) {
1946 * It seems dagerous to access memcg without css_get().
1947 * But considering how consume_stok works, it's not
1948 * necessary. If consume_stock success, some charges
1949 * from this memcg are cached on this cpu. So, we
1950 * don't need to call css_get()/css_tryget() before
1951 * calling consume_stock().
1956 /* after here, we may be blocked. we need to get refcnt */
1957 if (!css_tryget(&mem->css)) {
1967 /* If killed, bypass charge */
1968 if (fatal_signal_pending(current)) {
1974 if (oom && !nr_oom_retries) {
1976 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
1979 ret = __mem_cgroup_do_charge(mem, gfp_mask, csize, oom_check);
1984 case CHARGE_RETRY: /* not in OOM situation but retry */
1989 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
1992 case CHARGE_NOMEM: /* OOM routine works */
1997 /* If oom, we never return -ENOMEM */
2000 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
2004 } while (ret != CHARGE_OK);
2006 if (csize > page_size)
2007 refill_stock(mem, csize - page_size);
2021 * Somemtimes we have to undo a charge we got by try_charge().
2022 * This function is for that and do uncharge, put css's refcnt.
2023 * gotten by try_charge().
2025 static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem,
2026 unsigned long count)
2028 if (!mem_cgroup_is_root(mem)) {
2029 res_counter_uncharge(&mem->res, PAGE_SIZE * count);
2030 if (do_swap_account)
2031 res_counter_uncharge(&mem->memsw, PAGE_SIZE * count);
2035 static void mem_cgroup_cancel_charge(struct mem_cgroup *mem,
2038 __mem_cgroup_cancel_charge(mem, page_size >> PAGE_SHIFT);
2042 * A helper function to get mem_cgroup from ID. must be called under
2043 * rcu_read_lock(). The caller must check css_is_removed() or some if
2044 * it's concern. (dropping refcnt from swap can be called against removed
2047 static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2049 struct cgroup_subsys_state *css;
2051 /* ID 0 is unused ID */
2054 css = css_lookup(&mem_cgroup_subsys, id);
2057 return container_of(css, struct mem_cgroup, css);
2060 struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
2062 struct mem_cgroup *mem = NULL;
2063 struct page_cgroup *pc;
2067 VM_BUG_ON(!PageLocked(page));
2069 pc = lookup_page_cgroup(page);
2070 lock_page_cgroup(pc);
2071 if (PageCgroupUsed(pc)) {
2072 mem = pc->mem_cgroup;
2073 if (mem && !css_tryget(&mem->css))
2075 } else if (PageSwapCache(page)) {
2076 ent.val = page_private(page);
2077 id = lookup_swap_cgroup(ent);
2079 mem = mem_cgroup_lookup(id);
2080 if (mem && !css_tryget(&mem->css))
2084 unlock_page_cgroup(pc);
2088 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
2089 struct page_cgroup *pc,
2090 enum charge_type ctype,
2093 int nr_pages = page_size >> PAGE_SHIFT;
2095 /* try_charge() can return NULL to *memcg, taking care of it. */
2099 lock_page_cgroup(pc);
2100 if (unlikely(PageCgroupUsed(pc))) {
2101 unlock_page_cgroup(pc);
2102 mem_cgroup_cancel_charge(mem, page_size);
2106 * we don't need page_cgroup_lock about tail pages, becase they are not
2107 * accessed by any other context at this point.
2109 pc->mem_cgroup = mem;
2111 * We access a page_cgroup asynchronously without lock_page_cgroup().
2112 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2113 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2114 * before USED bit, we need memory barrier here.
2115 * See mem_cgroup_add_lru_list(), etc.
2119 case MEM_CGROUP_CHARGE_TYPE_CACHE:
2120 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
2121 SetPageCgroupCache(pc);
2122 SetPageCgroupUsed(pc);
2124 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2125 ClearPageCgroupCache(pc);
2126 SetPageCgroupUsed(pc);
2132 mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), nr_pages);
2133 unlock_page_cgroup(pc);
2135 * "charge_statistics" updated event counter. Then, check it.
2136 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2137 * if they exceeds softlimit.
2139 memcg_check_events(mem, pc->page);
2142 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2144 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2145 (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION))
2147 * Because tail pages are not marked as "used", set it. We're under
2148 * zone->lru_lock, 'splitting on pmd' and compund_lock.
2150 void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail)
2152 struct page_cgroup *head_pc = lookup_page_cgroup(head);
2153 struct page_cgroup *tail_pc = lookup_page_cgroup(tail);
2154 unsigned long flags;
2157 * We have no races with charge/uncharge but will have races with
2158 * page state accounting.
2160 move_lock_page_cgroup(head_pc, &flags);
2162 tail_pc->mem_cgroup = head_pc->mem_cgroup;
2163 smp_wmb(); /* see __commit_charge() */
2164 if (PageCgroupAcctLRU(head_pc)) {
2166 struct mem_cgroup_per_zone *mz;
2169 * LRU flags cannot be copied because we need to add tail
2170 *.page to LRU by generic call and our hook will be called.
2171 * We hold lru_lock, then, reduce counter directly.
2173 lru = page_lru(head);
2174 mz = page_cgroup_zoneinfo(head_pc);
2175 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
2177 tail_pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2178 move_unlock_page_cgroup(head_pc, &flags);
2183 * __mem_cgroup_move_account - move account of the page
2184 * @pc: page_cgroup of the page.
2185 * @from: mem_cgroup which the page is moved from.
2186 * @to: mem_cgroup which the page is moved to. @from != @to.
2187 * @uncharge: whether we should call uncharge and css_put against @from.
2189 * The caller must confirm following.
2190 * - page is not on LRU (isolate_page() is useful.)
2191 * - the pc is locked, used, and ->mem_cgroup points to @from.
2193 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2194 * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
2195 * true, this function does "uncharge" from old cgroup, but it doesn't if
2196 * @uncharge is false, so a caller should do "uncharge".
2199 static void __mem_cgroup_move_account(struct page_cgroup *pc,
2200 struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge)
2202 VM_BUG_ON(from == to);
2203 VM_BUG_ON(PageLRU(pc->page));
2204 VM_BUG_ON(!page_is_cgroup_locked(pc));
2205 VM_BUG_ON(!PageCgroupUsed(pc));
2206 VM_BUG_ON(pc->mem_cgroup != from);
2208 if (PageCgroupFileMapped(pc)) {
2209 /* Update mapped_file data for mem_cgroup */
2211 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2212 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2215 mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -1);
2217 /* This is not "cancel", but cancel_charge does all we need. */
2218 mem_cgroup_cancel_charge(from, PAGE_SIZE);
2220 /* caller should have done css_get */
2221 pc->mem_cgroup = to;
2222 mem_cgroup_charge_statistics(to, PageCgroupCache(pc), 1);
2224 * We charges against "to" which may not have any tasks. Then, "to"
2225 * can be under rmdir(). But in current implementation, caller of
2226 * this function is just force_empty() and move charge, so it's
2227 * garanteed that "to" is never removed. So, we don't check rmdir
2233 * check whether the @pc is valid for moving account and call
2234 * __mem_cgroup_move_account()
2236 static int mem_cgroup_move_account(struct page_cgroup *pc,
2237 struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge)
2240 unsigned long flags;
2242 lock_page_cgroup(pc);
2243 if (PageCgroupUsed(pc) && pc->mem_cgroup == from) {
2244 move_lock_page_cgroup(pc, &flags);
2245 __mem_cgroup_move_account(pc, from, to, uncharge);
2246 move_unlock_page_cgroup(pc, &flags);
2249 unlock_page_cgroup(pc);
2253 memcg_check_events(to, pc->page);
2254 memcg_check_events(from, pc->page);
2259 * move charges to its parent.
2262 static int mem_cgroup_move_parent(struct page_cgroup *pc,
2263 struct mem_cgroup *child,
2266 struct page *page = pc->page;
2267 struct cgroup *cg = child->css.cgroup;
2268 struct cgroup *pcg = cg->parent;
2269 struct mem_cgroup *parent;
2277 if (!get_page_unless_zero(page))
2279 if (isolate_lru_page(page))
2282 parent = mem_cgroup_from_cont(pcg);
2283 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false,
2288 ret = mem_cgroup_move_account(pc, child, parent, true);
2290 mem_cgroup_cancel_charge(parent, PAGE_SIZE);
2292 putback_lru_page(page);
2300 * Charge the memory controller for page usage.
2302 * 0 if the charge was successful
2303 * < 0 if the cgroup is over its limit
2305 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
2306 gfp_t gfp_mask, enum charge_type ctype)
2308 struct mem_cgroup *mem = NULL;
2309 struct page_cgroup *pc;
2311 int page_size = PAGE_SIZE;
2313 if (PageTransHuge(page)) {
2314 page_size <<= compound_order(page);
2315 VM_BUG_ON(!PageTransHuge(page));
2318 pc = lookup_page_cgroup(page);
2319 /* can happen at boot */
2324 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true, page_size);
2328 __mem_cgroup_commit_charge(mem, pc, ctype, page_size);
2332 int mem_cgroup_newpage_charge(struct page *page,
2333 struct mm_struct *mm, gfp_t gfp_mask)
2335 if (mem_cgroup_disabled())
2338 * If already mapped, we don't have to account.
2339 * If page cache, page->mapping has address_space.
2340 * But page->mapping may have out-of-use anon_vma pointer,
2341 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2344 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
2348 return mem_cgroup_charge_common(page, mm, gfp_mask,
2349 MEM_CGROUP_CHARGE_TYPE_MAPPED);
2353 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2354 enum charge_type ctype);
2356 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2361 if (mem_cgroup_disabled())
2363 if (PageCompound(page))
2366 * Corner case handling. This is called from add_to_page_cache()
2367 * in usual. But some FS (shmem) precharges this page before calling it
2368 * and call add_to_page_cache() with GFP_NOWAIT.
2370 * For GFP_NOWAIT case, the page may be pre-charged before calling
2371 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
2372 * charge twice. (It works but has to pay a bit larger cost.)
2373 * And when the page is SwapCache, it should take swap information
2374 * into account. This is under lock_page() now.
2376 if (!(gfp_mask & __GFP_WAIT)) {
2377 struct page_cgroup *pc;
2379 pc = lookup_page_cgroup(page);
2382 lock_page_cgroup(pc);
2383 if (PageCgroupUsed(pc)) {
2384 unlock_page_cgroup(pc);
2387 unlock_page_cgroup(pc);
2393 if (page_is_file_cache(page))
2394 return mem_cgroup_charge_common(page, mm, gfp_mask,
2395 MEM_CGROUP_CHARGE_TYPE_CACHE);
2398 if (PageSwapCache(page)) {
2399 struct mem_cgroup *mem = NULL;
2401 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2403 __mem_cgroup_commit_charge_swapin(page, mem,
2404 MEM_CGROUP_CHARGE_TYPE_SHMEM);
2406 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
2407 MEM_CGROUP_CHARGE_TYPE_SHMEM);
2413 * While swap-in, try_charge -> commit or cancel, the page is locked.
2414 * And when try_charge() successfully returns, one refcnt to memcg without
2415 * struct page_cgroup is acquired. This refcnt will be consumed by
2416 * "commit()" or removed by "cancel()"
2418 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2420 gfp_t mask, struct mem_cgroup **ptr)
2422 struct mem_cgroup *mem;
2425 if (mem_cgroup_disabled())
2428 if (!do_swap_account)
2431 * A racing thread's fault, or swapoff, may have already updated
2432 * the pte, and even removed page from swap cache: in those cases
2433 * do_swap_page()'s pte_same() test will fail; but there's also a
2434 * KSM case which does need to charge the page.
2436 if (!PageSwapCache(page))
2438 mem = try_get_mem_cgroup_from_page(page);
2442 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, PAGE_SIZE);
2448 return __mem_cgroup_try_charge(mm, mask, ptr, true, PAGE_SIZE);
2452 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2453 enum charge_type ctype)
2455 struct page_cgroup *pc;
2457 if (mem_cgroup_disabled())
2461 cgroup_exclude_rmdir(&ptr->css);
2462 pc = lookup_page_cgroup(page);
2463 mem_cgroup_lru_del_before_commit_swapcache(page);
2464 __mem_cgroup_commit_charge(ptr, pc, ctype, PAGE_SIZE);
2465 mem_cgroup_lru_add_after_commit_swapcache(page);
2467 * Now swap is on-memory. This means this page may be
2468 * counted both as mem and swap....double count.
2469 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2470 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2471 * may call delete_from_swap_cache() before reach here.
2473 if (do_swap_account && PageSwapCache(page)) {
2474 swp_entry_t ent = {.val = page_private(page)};
2476 struct mem_cgroup *memcg;
2478 id = swap_cgroup_record(ent, 0);
2480 memcg = mem_cgroup_lookup(id);
2483 * This recorded memcg can be obsolete one. So, avoid
2484 * calling css_tryget
2486 if (!mem_cgroup_is_root(memcg))
2487 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
2488 mem_cgroup_swap_statistics(memcg, false);
2489 mem_cgroup_put(memcg);
2494 * At swapin, we may charge account against cgroup which has no tasks.
2495 * So, rmdir()->pre_destroy() can be called while we do this charge.
2496 * In that case, we need to call pre_destroy() again. check it here.
2498 cgroup_release_and_wakeup_rmdir(&ptr->css);
2501 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
2503 __mem_cgroup_commit_charge_swapin(page, ptr,
2504 MEM_CGROUP_CHARGE_TYPE_MAPPED);
2507 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
2509 if (mem_cgroup_disabled())
2513 mem_cgroup_cancel_charge(mem, PAGE_SIZE);
2517 __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype,
2520 struct memcg_batch_info *batch = NULL;
2521 bool uncharge_memsw = true;
2522 /* If swapout, usage of swap doesn't decrease */
2523 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2524 uncharge_memsw = false;
2526 batch = ¤t->memcg_batch;
2528 * In usual, we do css_get() when we remember memcg pointer.
2529 * But in this case, we keep res->usage until end of a series of
2530 * uncharges. Then, it's ok to ignore memcg's refcnt.
2535 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2536 * In those cases, all pages freed continously can be expected to be in
2537 * the same cgroup and we have chance to coalesce uncharges.
2538 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2539 * because we want to do uncharge as soon as possible.
2542 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2543 goto direct_uncharge;
2545 if (page_size != PAGE_SIZE)
2546 goto direct_uncharge;
2549 * In typical case, batch->memcg == mem. This means we can
2550 * merge a series of uncharges to an uncharge of res_counter.
2551 * If not, we uncharge res_counter ony by one.
2553 if (batch->memcg != mem)
2554 goto direct_uncharge;
2555 /* remember freed charge and uncharge it later */
2556 batch->bytes += PAGE_SIZE;
2558 batch->memsw_bytes += PAGE_SIZE;
2561 res_counter_uncharge(&mem->res, page_size);
2563 res_counter_uncharge(&mem->memsw, page_size);
2564 if (unlikely(batch->memcg != mem))
2565 memcg_oom_recover(mem);
2570 * uncharge if !page_mapped(page)
2572 static struct mem_cgroup *
2573 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
2576 struct page_cgroup *pc;
2577 struct mem_cgroup *mem = NULL;
2578 int page_size = PAGE_SIZE;
2580 if (mem_cgroup_disabled())
2583 if (PageSwapCache(page))
2586 if (PageTransHuge(page)) {
2587 page_size <<= compound_order(page);
2588 VM_BUG_ON(!PageTransHuge(page));
2591 count = page_size >> PAGE_SHIFT;
2593 * Check if our page_cgroup is valid
2595 pc = lookup_page_cgroup(page);
2596 if (unlikely(!pc || !PageCgroupUsed(pc)))
2599 lock_page_cgroup(pc);
2601 mem = pc->mem_cgroup;
2603 if (!PageCgroupUsed(pc))
2607 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2608 case MEM_CGROUP_CHARGE_TYPE_DROP:
2609 /* See mem_cgroup_prepare_migration() */
2610 if (page_mapped(page) || PageCgroupMigration(pc))
2613 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
2614 if (!PageAnon(page)) { /* Shared memory */
2615 if (page->mapping && !page_is_file_cache(page))
2617 } else if (page_mapped(page)) /* Anon */
2624 mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), -count);
2626 ClearPageCgroupUsed(pc);
2628 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2629 * freed from LRU. This is safe because uncharged page is expected not
2630 * to be reused (freed soon). Exception is SwapCache, it's handled by
2631 * special functions.
2634 unlock_page_cgroup(pc);
2636 * even after unlock, we have mem->res.usage here and this memcg
2637 * will never be freed.
2639 memcg_check_events(mem, page);
2640 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
2641 mem_cgroup_swap_statistics(mem, true);
2642 mem_cgroup_get(mem);
2644 if (!mem_cgroup_is_root(mem))
2645 __do_uncharge(mem, ctype, page_size);
2650 unlock_page_cgroup(pc);
2654 void mem_cgroup_uncharge_page(struct page *page)
2657 if (page_mapped(page))
2659 if (page->mapping && !PageAnon(page))
2661 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
2664 void mem_cgroup_uncharge_cache_page(struct page *page)
2666 VM_BUG_ON(page_mapped(page));
2667 VM_BUG_ON(page->mapping);
2668 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
2672 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2673 * In that cases, pages are freed continuously and we can expect pages
2674 * are in the same memcg. All these calls itself limits the number of
2675 * pages freed at once, then uncharge_start/end() is called properly.
2676 * This may be called prural(2) times in a context,
2679 void mem_cgroup_uncharge_start(void)
2681 current->memcg_batch.do_batch++;
2682 /* We can do nest. */
2683 if (current->memcg_batch.do_batch == 1) {
2684 current->memcg_batch.memcg = NULL;
2685 current->memcg_batch.bytes = 0;
2686 current->memcg_batch.memsw_bytes = 0;
2690 void mem_cgroup_uncharge_end(void)
2692 struct memcg_batch_info *batch = ¤t->memcg_batch;
2694 if (!batch->do_batch)
2698 if (batch->do_batch) /* If stacked, do nothing. */
2704 * This "batch->memcg" is valid without any css_get/put etc...
2705 * bacause we hide charges behind us.
2708 res_counter_uncharge(&batch->memcg->res, batch->bytes);
2709 if (batch->memsw_bytes)
2710 res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes);
2711 memcg_oom_recover(batch->memcg);
2712 /* forget this pointer (for sanity check) */
2713 batch->memcg = NULL;
2718 * called after __delete_from_swap_cache() and drop "page" account.
2719 * memcg information is recorded to swap_cgroup of "ent"
2722 mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
2724 struct mem_cgroup *memcg;
2725 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
2727 if (!swapout) /* this was a swap cache but the swap is unused ! */
2728 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
2730 memcg = __mem_cgroup_uncharge_common(page, ctype);
2733 * record memcg information, if swapout && memcg != NULL,
2734 * mem_cgroup_get() was called in uncharge().
2736 if (do_swap_account && swapout && memcg)
2737 swap_cgroup_record(ent, css_id(&memcg->css));
2741 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2743 * called from swap_entry_free(). remove record in swap_cgroup and
2744 * uncharge "memsw" account.
2746 void mem_cgroup_uncharge_swap(swp_entry_t ent)
2748 struct mem_cgroup *memcg;
2751 if (!do_swap_account)
2754 id = swap_cgroup_record(ent, 0);
2756 memcg = mem_cgroup_lookup(id);
2759 * We uncharge this because swap is freed.
2760 * This memcg can be obsolete one. We avoid calling css_tryget
2762 if (!mem_cgroup_is_root(memcg))
2763 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
2764 mem_cgroup_swap_statistics(memcg, false);
2765 mem_cgroup_put(memcg);
2771 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2772 * @entry: swap entry to be moved
2773 * @from: mem_cgroup which the entry is moved from
2774 * @to: mem_cgroup which the entry is moved to
2775 * @need_fixup: whether we should fixup res_counters and refcounts.
2777 * It succeeds only when the swap_cgroup's record for this entry is the same
2778 * as the mem_cgroup's id of @from.
2780 * Returns 0 on success, -EINVAL on failure.
2782 * The caller must have charged to @to, IOW, called res_counter_charge() about
2783 * both res and memsw, and called css_get().
2785 static int mem_cgroup_move_swap_account(swp_entry_t entry,
2786 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
2788 unsigned short old_id, new_id;
2790 old_id = css_id(&from->css);
2791 new_id = css_id(&to->css);
2793 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
2794 mem_cgroup_swap_statistics(from, false);
2795 mem_cgroup_swap_statistics(to, true);
2797 * This function is only called from task migration context now.
2798 * It postpones res_counter and refcount handling till the end
2799 * of task migration(mem_cgroup_clear_mc()) for performance
2800 * improvement. But we cannot postpone mem_cgroup_get(to)
2801 * because if the process that has been moved to @to does
2802 * swap-in, the refcount of @to might be decreased to 0.
2806 if (!mem_cgroup_is_root(from))
2807 res_counter_uncharge(&from->memsw, PAGE_SIZE);
2808 mem_cgroup_put(from);
2810 * we charged both to->res and to->memsw, so we should
2813 if (!mem_cgroup_is_root(to))
2814 res_counter_uncharge(&to->res, PAGE_SIZE);
2821 static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
2822 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
2829 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2832 int mem_cgroup_prepare_migration(struct page *page,
2833 struct page *newpage, struct mem_cgroup **ptr)
2835 struct page_cgroup *pc;
2836 struct mem_cgroup *mem = NULL;
2837 enum charge_type ctype;
2840 VM_BUG_ON(PageTransHuge(page));
2841 if (mem_cgroup_disabled())
2844 pc = lookup_page_cgroup(page);
2845 lock_page_cgroup(pc);
2846 if (PageCgroupUsed(pc)) {
2847 mem = pc->mem_cgroup;
2850 * At migrating an anonymous page, its mapcount goes down
2851 * to 0 and uncharge() will be called. But, even if it's fully
2852 * unmapped, migration may fail and this page has to be
2853 * charged again. We set MIGRATION flag here and delay uncharge
2854 * until end_migration() is called
2856 * Corner Case Thinking
2858 * When the old page was mapped as Anon and it's unmap-and-freed
2859 * while migration was ongoing.
2860 * If unmap finds the old page, uncharge() of it will be delayed
2861 * until end_migration(). If unmap finds a new page, it's
2862 * uncharged when it make mapcount to be 1->0. If unmap code
2863 * finds swap_migration_entry, the new page will not be mapped
2864 * and end_migration() will find it(mapcount==0).
2867 * When the old page was mapped but migraion fails, the kernel
2868 * remaps it. A charge for it is kept by MIGRATION flag even
2869 * if mapcount goes down to 0. We can do remap successfully
2870 * without charging it again.
2873 * The "old" page is under lock_page() until the end of
2874 * migration, so, the old page itself will not be swapped-out.
2875 * If the new page is swapped out before end_migraton, our
2876 * hook to usual swap-out path will catch the event.
2879 SetPageCgroupMigration(pc);
2881 unlock_page_cgroup(pc);
2883 * If the page is not charged at this point,
2890 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, ptr, false, PAGE_SIZE);
2891 css_put(&mem->css);/* drop extra refcnt */
2892 if (ret || *ptr == NULL) {
2893 if (PageAnon(page)) {
2894 lock_page_cgroup(pc);
2895 ClearPageCgroupMigration(pc);
2896 unlock_page_cgroup(pc);
2898 * The old page may be fully unmapped while we kept it.
2900 mem_cgroup_uncharge_page(page);
2905 * We charge new page before it's used/mapped. So, even if unlock_page()
2906 * is called before end_migration, we can catch all events on this new
2907 * page. In the case new page is migrated but not remapped, new page's
2908 * mapcount will be finally 0 and we call uncharge in end_migration().
2910 pc = lookup_page_cgroup(newpage);
2912 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
2913 else if (page_is_file_cache(page))
2914 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
2916 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
2917 __mem_cgroup_commit_charge(mem, pc, ctype, PAGE_SIZE);
2921 /* remove redundant charge if migration failed*/
2922 void mem_cgroup_end_migration(struct mem_cgroup *mem,
2923 struct page *oldpage, struct page *newpage, bool migration_ok)
2925 struct page *used, *unused;
2926 struct page_cgroup *pc;
2930 /* blocks rmdir() */
2931 cgroup_exclude_rmdir(&mem->css);
2932 if (!migration_ok) {
2940 * We disallowed uncharge of pages under migration because mapcount
2941 * of the page goes down to zero, temporarly.
2942 * Clear the flag and check the page should be charged.
2944 pc = lookup_page_cgroup(oldpage);
2945 lock_page_cgroup(pc);
2946 ClearPageCgroupMigration(pc);
2947 unlock_page_cgroup(pc);
2949 __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE);
2952 * If a page is a file cache, radix-tree replacement is very atomic
2953 * and we can skip this check. When it was an Anon page, its mapcount
2954 * goes down to 0. But because we added MIGRATION flage, it's not
2955 * uncharged yet. There are several case but page->mapcount check
2956 * and USED bit check in mem_cgroup_uncharge_page() will do enough
2957 * check. (see prepare_charge() also)
2960 mem_cgroup_uncharge_page(used);
2962 * At migration, we may charge account against cgroup which has no
2964 * So, rmdir()->pre_destroy() can be called while we do this charge.
2965 * In that case, we need to call pre_destroy() again. check it here.
2967 cgroup_release_and_wakeup_rmdir(&mem->css);
2971 * A call to try to shrink memory usage on charge failure at shmem's swapin.
2972 * Calling hierarchical_reclaim is not enough because we should update
2973 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
2974 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
2975 * not from the memcg which this page would be charged to.
2976 * try_charge_swapin does all of these works properly.
2978 int mem_cgroup_shmem_charge_fallback(struct page *page,
2979 struct mm_struct *mm,
2982 struct mem_cgroup *mem = NULL;
2985 if (mem_cgroup_disabled())
2988 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2990 mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
2995 static DEFINE_MUTEX(set_limit_mutex);
2997 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
2998 unsigned long long val)
3001 u64 memswlimit, memlimit;
3003 int children = mem_cgroup_count_children(memcg);
3004 u64 curusage, oldusage;
3008 * For keeping hierarchical_reclaim simple, how long we should retry
3009 * is depends on callers. We set our retry-count to be function
3010 * of # of children which we should visit in this loop.
3012 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3014 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3017 while (retry_count) {
3018 if (signal_pending(current)) {
3023 * Rather than hide all in some function, I do this in
3024 * open coded manner. You see what this really does.
3025 * We have to guarantee mem->res.limit < mem->memsw.limit.
3027 mutex_lock(&set_limit_mutex);
3028 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3029 if (memswlimit < val) {
3031 mutex_unlock(&set_limit_mutex);
3035 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3039 ret = res_counter_set_limit(&memcg->res, val);
3041 if (memswlimit == val)
3042 memcg->memsw_is_minimum = true;
3044 memcg->memsw_is_minimum = false;
3046 mutex_unlock(&set_limit_mutex);
3051 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
3052 MEM_CGROUP_RECLAIM_SHRINK);
3053 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3054 /* Usage is reduced ? */
3055 if (curusage >= oldusage)
3058 oldusage = curusage;
3060 if (!ret && enlarge)
3061 memcg_oom_recover(memcg);
3066 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3067 unsigned long long val)
3070 u64 memlimit, memswlimit, oldusage, curusage;
3071 int children = mem_cgroup_count_children(memcg);
3075 /* see mem_cgroup_resize_res_limit */
3076 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3077 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
3078 while (retry_count) {
3079 if (signal_pending(current)) {
3084 * Rather than hide all in some function, I do this in
3085 * open coded manner. You see what this really does.
3086 * We have to guarantee mem->res.limit < mem->memsw.limit.
3088 mutex_lock(&set_limit_mutex);
3089 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3090 if (memlimit > val) {
3092 mutex_unlock(&set_limit_mutex);
3095 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3096 if (memswlimit < val)
3098 ret = res_counter_set_limit(&memcg->memsw, val);
3100 if (memlimit == val)
3101 memcg->memsw_is_minimum = true;
3103 memcg->memsw_is_minimum = false;
3105 mutex_unlock(&set_limit_mutex);
3110 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
3111 MEM_CGROUP_RECLAIM_NOSWAP |
3112 MEM_CGROUP_RECLAIM_SHRINK);
3113 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
3114 /* Usage is reduced ? */
3115 if (curusage >= oldusage)
3118 oldusage = curusage;
3120 if (!ret && enlarge)
3121 memcg_oom_recover(memcg);
3125 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
3128 unsigned long nr_reclaimed = 0;
3129 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3130 unsigned long reclaimed;
3132 struct mem_cgroup_tree_per_zone *mctz;
3133 unsigned long long excess;
3138 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
3140 * This loop can run a while, specially if mem_cgroup's continuously
3141 * keep exceeding their soft limit and putting the system under
3148 mz = mem_cgroup_largest_soft_limit_node(mctz);
3152 reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone,
3154 MEM_CGROUP_RECLAIM_SOFT);
3155 nr_reclaimed += reclaimed;
3156 spin_lock(&mctz->lock);
3159 * If we failed to reclaim anything from this memory cgroup
3160 * it is time to move on to the next cgroup
3166 * Loop until we find yet another one.
3168 * By the time we get the soft_limit lock
3169 * again, someone might have aded the
3170 * group back on the RB tree. Iterate to
3171 * make sure we get a different mem.
3172 * mem_cgroup_largest_soft_limit_node returns
3173 * NULL if no other cgroup is present on
3177 __mem_cgroup_largest_soft_limit_node(mctz);
3178 if (next_mz == mz) {
3179 css_put(&next_mz->mem->css);
3181 } else /* next_mz == NULL or other memcg */
3185 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
3186 excess = res_counter_soft_limit_excess(&mz->mem->res);
3188 * One school of thought says that we should not add
3189 * back the node to the tree if reclaim returns 0.
3190 * But our reclaim could return 0, simply because due
3191 * to priority we are exposing a smaller subset of
3192 * memory to reclaim from. Consider this as a longer
3195 /* If excess == 0, no tree ops */
3196 __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
3197 spin_unlock(&mctz->lock);
3198 css_put(&mz->mem->css);
3201 * Could not reclaim anything and there are no more
3202 * mem cgroups to try or we seem to be looping without
3203 * reclaiming anything.
3205 if (!nr_reclaimed &&
3207 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3209 } while (!nr_reclaimed);
3211 css_put(&next_mz->mem->css);
3212 return nr_reclaimed;
3216 * This routine traverse page_cgroup in given list and drop them all.
3217 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3219 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
3220 int node, int zid, enum lru_list lru)
3223 struct mem_cgroup_per_zone *mz;
3224 struct page_cgroup *pc, *busy;
3225 unsigned long flags, loop;
3226 struct list_head *list;
3229 zone = &NODE_DATA(node)->node_zones[zid];
3230 mz = mem_cgroup_zoneinfo(mem, node, zid);
3231 list = &mz->lists[lru];
3233 loop = MEM_CGROUP_ZSTAT(mz, lru);
3234 /* give some margin against EBUSY etc...*/
3239 spin_lock_irqsave(&zone->lru_lock, flags);
3240 if (list_empty(list)) {
3241 spin_unlock_irqrestore(&zone->lru_lock, flags);
3244 pc = list_entry(list->prev, struct page_cgroup, lru);
3246 list_move(&pc->lru, list);
3248 spin_unlock_irqrestore(&zone->lru_lock, flags);
3251 spin_unlock_irqrestore(&zone->lru_lock, flags);
3253 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
3257 if (ret == -EBUSY || ret == -EINVAL) {
3258 /* found lock contention or "pc" is obsolete. */
3265 if (!ret && !list_empty(list))
3271 * make mem_cgroup's charge to be 0 if there is no task.
3272 * This enables deleting this mem_cgroup.
3274 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
3277 int node, zid, shrink;
3278 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
3279 struct cgroup *cgrp = mem->css.cgroup;
3284 /* should free all ? */
3290 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
3293 if (signal_pending(current))
3295 /* This is for making all *used* pages to be on LRU. */
3296 lru_add_drain_all();
3297 drain_all_stock_sync();
3299 mem_cgroup_start_move(mem);
3300 for_each_node_state(node, N_HIGH_MEMORY) {
3301 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
3304 ret = mem_cgroup_force_empty_list(mem,
3313 mem_cgroup_end_move(mem);
3314 memcg_oom_recover(mem);
3315 /* it seems parent cgroup doesn't have enough mem */
3319 /* "ret" should also be checked to ensure all lists are empty. */
3320 } while (mem->res.usage > 0 || ret);
3326 /* returns EBUSY if there is a task or if we come here twice. */
3327 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
3331 /* we call try-to-free pages for make this cgroup empty */
3332 lru_add_drain_all();
3333 /* try to free all pages in this cgroup */
3335 while (nr_retries && mem->res.usage > 0) {
3338 if (signal_pending(current)) {
3342 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
3343 false, get_swappiness(mem));
3346 /* maybe some writeback is necessary */
3347 congestion_wait(BLK_RW_ASYNC, HZ/10);
3352 /* try move_account...there may be some *locked* pages. */
3356 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3358 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3362 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3364 return mem_cgroup_from_cont(cont)->use_hierarchy;
3367 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3371 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3372 struct cgroup *parent = cont->parent;
3373 struct mem_cgroup *parent_mem = NULL;
3376 parent_mem = mem_cgroup_from_cont(parent);
3380 * If parent's use_hierarchy is set, we can't make any modifications
3381 * in the child subtrees. If it is unset, then the change can
3382 * occur, provided the current cgroup has no children.
3384 * For the root cgroup, parent_mem is NULL, we allow value to be
3385 * set if there are no children.
3387 if ((!parent_mem || !parent_mem->use_hierarchy) &&
3388 (val == 1 || val == 0)) {
3389 if (list_empty(&cont->children))
3390 mem->use_hierarchy = val;
3401 static u64 mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem,
3402 enum mem_cgroup_stat_index idx)
3404 struct mem_cgroup *iter;
3407 /* each per cpu's value can be minus.Then, use s64 */
3408 for_each_mem_cgroup_tree(iter, mem)
3409 val += mem_cgroup_read_stat(iter, idx);
3411 if (val < 0) /* race ? */
3416 static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap)
3420 if (!mem_cgroup_is_root(mem)) {
3422 return res_counter_read_u64(&mem->res, RES_USAGE);
3424 return res_counter_read_u64(&mem->memsw, RES_USAGE);
3427 val = mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE);
3428 val += mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS);
3431 val += mem_cgroup_get_recursive_idx_stat(mem,
3432 MEM_CGROUP_STAT_SWAPOUT);
3434 return val << PAGE_SHIFT;
3437 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
3439 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3443 type = MEMFILE_TYPE(cft->private);
3444 name = MEMFILE_ATTR(cft->private);
3447 if (name == RES_USAGE)
3448 val = mem_cgroup_usage(mem, false);
3450 val = res_counter_read_u64(&mem->res, name);
3453 if (name == RES_USAGE)
3454 val = mem_cgroup_usage(mem, true);
3456 val = res_counter_read_u64(&mem->memsw, name);
3465 * The user of this function is...
3468 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3471 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3473 unsigned long long val;
3476 type = MEMFILE_TYPE(cft->private);
3477 name = MEMFILE_ATTR(cft->private);
3480 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3484 /* This function does all necessary parse...reuse it */
3485 ret = res_counter_memparse_write_strategy(buffer, &val);
3489 ret = mem_cgroup_resize_limit(memcg, val);
3491 ret = mem_cgroup_resize_memsw_limit(memcg, val);
3493 case RES_SOFT_LIMIT:
3494 ret = res_counter_memparse_write_strategy(buffer, &val);
3498 * For memsw, soft limits are hard to implement in terms
3499 * of semantics, for now, we support soft limits for
3500 * control without swap
3503 ret = res_counter_set_soft_limit(&memcg->res, val);
3508 ret = -EINVAL; /* should be BUG() ? */
3514 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3515 unsigned long long *mem_limit, unsigned long long *memsw_limit)
3517 struct cgroup *cgroup;
3518 unsigned long long min_limit, min_memsw_limit, tmp;
3520 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3521 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3522 cgroup = memcg->css.cgroup;
3523 if (!memcg->use_hierarchy)
3526 while (cgroup->parent) {
3527 cgroup = cgroup->parent;
3528 memcg = mem_cgroup_from_cont(cgroup);
3529 if (!memcg->use_hierarchy)
3531 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
3532 min_limit = min(min_limit, tmp);
3533 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3534 min_memsw_limit = min(min_memsw_limit, tmp);
3537 *mem_limit = min_limit;
3538 *memsw_limit = min_memsw_limit;
3542 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
3544 struct mem_cgroup *mem;
3547 mem = mem_cgroup_from_cont(cont);
3548 type = MEMFILE_TYPE(event);
3549 name = MEMFILE_ATTR(event);
3553 res_counter_reset_max(&mem->res);
3555 res_counter_reset_max(&mem->memsw);
3559 res_counter_reset_failcnt(&mem->res);
3561 res_counter_reset_failcnt(&mem->memsw);
3568 static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
3571 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
3575 static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3576 struct cftype *cft, u64 val)
3578 struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
3580 if (val >= (1 << NR_MOVE_TYPE))
3583 * We check this value several times in both in can_attach() and
3584 * attach(), so we need cgroup lock to prevent this value from being
3588 mem->move_charge_at_immigrate = val;
3594 static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3595 struct cftype *cft, u64 val)
3602 /* For read statistics */
3618 struct mcs_total_stat {
3619 s64 stat[NR_MCS_STAT];
3625 } memcg_stat_strings[NR_MCS_STAT] = {
3626 {"cache", "total_cache"},
3627 {"rss", "total_rss"},
3628 {"mapped_file", "total_mapped_file"},
3629 {"pgpgin", "total_pgpgin"},
3630 {"pgpgout", "total_pgpgout"},
3631 {"swap", "total_swap"},
3632 {"inactive_anon", "total_inactive_anon"},
3633 {"active_anon", "total_active_anon"},
3634 {"inactive_file", "total_inactive_file"},
3635 {"active_file", "total_active_file"},
3636 {"unevictable", "total_unevictable"}
3641 mem_cgroup_get_local_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
3646 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
3647 s->stat[MCS_CACHE] += val * PAGE_SIZE;
3648 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
3649 s->stat[MCS_RSS] += val * PAGE_SIZE;
3650 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED);
3651 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
3652 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT);
3653 s->stat[MCS_PGPGIN] += val;
3654 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT);
3655 s->stat[MCS_PGPGOUT] += val;
3656 if (do_swap_account) {
3657 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT);
3658 s->stat[MCS_SWAP] += val * PAGE_SIZE;
3662 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
3663 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
3664 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
3665 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
3666 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
3667 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
3668 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
3669 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
3670 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
3671 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
3675 mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
3677 struct mem_cgroup *iter;
3679 for_each_mem_cgroup_tree(iter, mem)
3680 mem_cgroup_get_local_stat(iter, s);
3683 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
3684 struct cgroup_map_cb *cb)
3686 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
3687 struct mcs_total_stat mystat;
3690 memset(&mystat, 0, sizeof(mystat));
3691 mem_cgroup_get_local_stat(mem_cont, &mystat);
3693 for (i = 0; i < NR_MCS_STAT; i++) {
3694 if (i == MCS_SWAP && !do_swap_account)
3696 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
3699 /* Hierarchical information */
3701 unsigned long long limit, memsw_limit;
3702 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
3703 cb->fill(cb, "hierarchical_memory_limit", limit);
3704 if (do_swap_account)
3705 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
3708 memset(&mystat, 0, sizeof(mystat));
3709 mem_cgroup_get_total_stat(mem_cont, &mystat);
3710 for (i = 0; i < NR_MCS_STAT; i++) {
3711 if (i == MCS_SWAP && !do_swap_account)
3713 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
3716 #ifdef CONFIG_DEBUG_VM
3717 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
3721 struct mem_cgroup_per_zone *mz;
3722 unsigned long recent_rotated[2] = {0, 0};
3723 unsigned long recent_scanned[2] = {0, 0};
3725 for_each_online_node(nid)
3726 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
3727 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
3729 recent_rotated[0] +=
3730 mz->reclaim_stat.recent_rotated[0];
3731 recent_rotated[1] +=
3732 mz->reclaim_stat.recent_rotated[1];
3733 recent_scanned[0] +=
3734 mz->reclaim_stat.recent_scanned[0];
3735 recent_scanned[1] +=
3736 mz->reclaim_stat.recent_scanned[1];
3738 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
3739 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
3740 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
3741 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
3748 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
3750 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3752 return get_swappiness(memcg);
3755 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
3758 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3759 struct mem_cgroup *parent;
3764 if (cgrp->parent == NULL)
3767 parent = mem_cgroup_from_cont(cgrp->parent);
3771 /* If under hierarchy, only empty-root can set this value */
3772 if ((parent->use_hierarchy) ||
3773 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
3778 spin_lock(&memcg->reclaim_param_lock);
3779 memcg->swappiness = val;
3780 spin_unlock(&memcg->reclaim_param_lock);
3787 static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3789 struct mem_cgroup_threshold_ary *t;
3795 t = rcu_dereference(memcg->thresholds.primary);
3797 t = rcu_dereference(memcg->memsw_thresholds.primary);
3802 usage = mem_cgroup_usage(memcg, swap);
3805 * current_threshold points to threshold just below usage.
3806 * If it's not true, a threshold was crossed after last
3807 * call of __mem_cgroup_threshold().
3809 i = t->current_threshold;
3812 * Iterate backward over array of thresholds starting from
3813 * current_threshold and check if a threshold is crossed.
3814 * If none of thresholds below usage is crossed, we read
3815 * only one element of the array here.
3817 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3818 eventfd_signal(t->entries[i].eventfd, 1);
3820 /* i = current_threshold + 1 */
3824 * Iterate forward over array of thresholds starting from
3825 * current_threshold+1 and check if a threshold is crossed.
3826 * If none of thresholds above usage is crossed, we read
3827 * only one element of the array here.
3829 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3830 eventfd_signal(t->entries[i].eventfd, 1);
3832 /* Update current_threshold */
3833 t->current_threshold = i - 1;
3838 static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3841 __mem_cgroup_threshold(memcg, false);
3842 if (do_swap_account)
3843 __mem_cgroup_threshold(memcg, true);
3845 memcg = parent_mem_cgroup(memcg);
3849 static int compare_thresholds(const void *a, const void *b)
3851 const struct mem_cgroup_threshold *_a = a;
3852 const struct mem_cgroup_threshold *_b = b;
3854 return _a->threshold - _b->threshold;
3857 static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem)
3859 struct mem_cgroup_eventfd_list *ev;
3861 list_for_each_entry(ev, &mem->oom_notify, list)
3862 eventfd_signal(ev->eventfd, 1);
3866 static void mem_cgroup_oom_notify(struct mem_cgroup *mem)
3868 struct mem_cgroup *iter;
3870 for_each_mem_cgroup_tree(iter, mem)
3871 mem_cgroup_oom_notify_cb(iter);
3874 static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
3875 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
3877 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3878 struct mem_cgroup_thresholds *thresholds;
3879 struct mem_cgroup_threshold_ary *new;
3880 int type = MEMFILE_TYPE(cft->private);
3881 u64 threshold, usage;
3884 ret = res_counter_memparse_write_strategy(args, &threshold);
3888 mutex_lock(&memcg->thresholds_lock);
3891 thresholds = &memcg->thresholds;
3892 else if (type == _MEMSWAP)
3893 thresholds = &memcg->memsw_thresholds;
3897 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
3899 /* Check if a threshold crossed before adding a new one */
3900 if (thresholds->primary)
3901 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3903 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
3905 /* Allocate memory for new array of thresholds */
3906 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
3914 /* Copy thresholds (if any) to new array */
3915 if (thresholds->primary) {
3916 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
3917 sizeof(struct mem_cgroup_threshold));
3920 /* Add new threshold */
3921 new->entries[size - 1].eventfd = eventfd;
3922 new->entries[size - 1].threshold = threshold;
3924 /* Sort thresholds. Registering of new threshold isn't time-critical */
3925 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
3926 compare_thresholds, NULL);
3928 /* Find current threshold */
3929 new->current_threshold = -1;
3930 for (i = 0; i < size; i++) {
3931 if (new->entries[i].threshold < usage) {
3933 * new->current_threshold will not be used until
3934 * rcu_assign_pointer(), so it's safe to increment
3937 ++new->current_threshold;
3941 /* Free old spare buffer and save old primary buffer as spare */
3942 kfree(thresholds->spare);
3943 thresholds->spare = thresholds->primary;
3945 rcu_assign_pointer(thresholds->primary, new);
3947 /* To be sure that nobody uses thresholds */
3951 mutex_unlock(&memcg->thresholds_lock);
3956 static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
3957 struct cftype *cft, struct eventfd_ctx *eventfd)
3959 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3960 struct mem_cgroup_thresholds *thresholds;
3961 struct mem_cgroup_threshold_ary *new;
3962 int type = MEMFILE_TYPE(cft->private);
3966 mutex_lock(&memcg->thresholds_lock);
3968 thresholds = &memcg->thresholds;
3969 else if (type == _MEMSWAP)
3970 thresholds = &memcg->memsw_thresholds;
3975 * Something went wrong if we trying to unregister a threshold
3976 * if we don't have thresholds
3978 BUG_ON(!thresholds);
3980 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
3982 /* Check if a threshold crossed before removing */
3983 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3985 /* Calculate new number of threshold */
3987 for (i = 0; i < thresholds->primary->size; i++) {
3988 if (thresholds->primary->entries[i].eventfd != eventfd)
3992 new = thresholds->spare;
3994 /* Set thresholds array to NULL if we don't have thresholds */
4003 /* Copy thresholds and find current threshold */
4004 new->current_threshold = -1;
4005 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4006 if (thresholds->primary->entries[i].eventfd == eventfd)
4009 new->entries[j] = thresholds->primary->entries[i];
4010 if (new->entries[j].threshold < usage) {
4012 * new->current_threshold will not be used
4013 * until rcu_assign_pointer(), so it's safe to increment
4016 ++new->current_threshold;
4022 /* Swap primary and spare array */
4023 thresholds->spare = thresholds->primary;
4024 rcu_assign_pointer(thresholds->primary, new);
4026 /* To be sure that nobody uses thresholds */
4029 mutex_unlock(&memcg->thresholds_lock);
4032 static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4033 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4035 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4036 struct mem_cgroup_eventfd_list *event;
4037 int type = MEMFILE_TYPE(cft->private);
4039 BUG_ON(type != _OOM_TYPE);
4040 event = kmalloc(sizeof(*event), GFP_KERNEL);
4044 mutex_lock(&memcg_oom_mutex);
4046 event->eventfd = eventfd;
4047 list_add(&event->list, &memcg->oom_notify);
4049 /* already in OOM ? */
4050 if (atomic_read(&memcg->oom_lock))
4051 eventfd_signal(eventfd, 1);
4052 mutex_unlock(&memcg_oom_mutex);
4057 static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
4058 struct cftype *cft, struct eventfd_ctx *eventfd)
4060 struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
4061 struct mem_cgroup_eventfd_list *ev, *tmp;
4062 int type = MEMFILE_TYPE(cft->private);
4064 BUG_ON(type != _OOM_TYPE);
4066 mutex_lock(&memcg_oom_mutex);
4068 list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) {
4069 if (ev->eventfd == eventfd) {
4070 list_del(&ev->list);
4075 mutex_unlock(&memcg_oom_mutex);
4078 static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4079 struct cftype *cft, struct cgroup_map_cb *cb)
4081 struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
4083 cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable);
4085 if (atomic_read(&mem->oom_lock))
4086 cb->fill(cb, "under_oom", 1);
4088 cb->fill(cb, "under_oom", 0);
4092 static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4093 struct cftype *cft, u64 val)
4095 struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
4096 struct mem_cgroup *parent;
4098 /* cannot set to root cgroup and only 0 and 1 are allowed */
4099 if (!cgrp->parent || !((val == 0) || (val == 1)))
4102 parent = mem_cgroup_from_cont(cgrp->parent);
4105 /* oom-kill-disable is a flag for subhierarchy. */
4106 if ((parent->use_hierarchy) ||
4107 (mem->use_hierarchy && !list_empty(&cgrp->children))) {
4111 mem->oom_kill_disable = val;
4113 memcg_oom_recover(mem);
4118 static struct cftype mem_cgroup_files[] = {
4120 .name = "usage_in_bytes",
4121 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
4122 .read_u64 = mem_cgroup_read,
4123 .register_event = mem_cgroup_usage_register_event,
4124 .unregister_event = mem_cgroup_usage_unregister_event,
4127 .name = "max_usage_in_bytes",
4128 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
4129 .trigger = mem_cgroup_reset,
4130 .read_u64 = mem_cgroup_read,
4133 .name = "limit_in_bytes",
4134 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
4135 .write_string = mem_cgroup_write,
4136 .read_u64 = mem_cgroup_read,
4139 .name = "soft_limit_in_bytes",
4140 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4141 .write_string = mem_cgroup_write,
4142 .read_u64 = mem_cgroup_read,
4146 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
4147 .trigger = mem_cgroup_reset,
4148 .read_u64 = mem_cgroup_read,
4152 .read_map = mem_control_stat_show,
4155 .name = "force_empty",
4156 .trigger = mem_cgroup_force_empty_write,
4159 .name = "use_hierarchy",
4160 .write_u64 = mem_cgroup_hierarchy_write,
4161 .read_u64 = mem_cgroup_hierarchy_read,
4164 .name = "swappiness",
4165 .read_u64 = mem_cgroup_swappiness_read,
4166 .write_u64 = mem_cgroup_swappiness_write,
4169 .name = "move_charge_at_immigrate",
4170 .read_u64 = mem_cgroup_move_charge_read,
4171 .write_u64 = mem_cgroup_move_charge_write,
4174 .name = "oom_control",
4175 .read_map = mem_cgroup_oom_control_read,
4176 .write_u64 = mem_cgroup_oom_control_write,
4177 .register_event = mem_cgroup_oom_register_event,
4178 .unregister_event = mem_cgroup_oom_unregister_event,
4179 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4183 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4184 static struct cftype memsw_cgroup_files[] = {
4186 .name = "memsw.usage_in_bytes",
4187 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
4188 .read_u64 = mem_cgroup_read,
4189 .register_event = mem_cgroup_usage_register_event,
4190 .unregister_event = mem_cgroup_usage_unregister_event,
4193 .name = "memsw.max_usage_in_bytes",
4194 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4195 .trigger = mem_cgroup_reset,
4196 .read_u64 = mem_cgroup_read,
4199 .name = "memsw.limit_in_bytes",
4200 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4201 .write_string = mem_cgroup_write,
4202 .read_u64 = mem_cgroup_read,
4205 .name = "memsw.failcnt",
4206 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4207 .trigger = mem_cgroup_reset,
4208 .read_u64 = mem_cgroup_read,
4212 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4214 if (!do_swap_account)
4216 return cgroup_add_files(cont, ss, memsw_cgroup_files,
4217 ARRAY_SIZE(memsw_cgroup_files));
4220 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4226 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
4228 struct mem_cgroup_per_node *pn;
4229 struct mem_cgroup_per_zone *mz;
4231 int zone, tmp = node;
4233 * This routine is called against possible nodes.
4234 * But it's BUG to call kmalloc() against offline node.
4236 * TODO: this routine can waste much memory for nodes which will
4237 * never be onlined. It's better to use memory hotplug callback
4240 if (!node_state(node, N_NORMAL_MEMORY))
4242 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
4246 mem->info.nodeinfo[node] = pn;
4247 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4248 mz = &pn->zoneinfo[zone];
4250 INIT_LIST_HEAD(&mz->lists[l]);
4251 mz->usage_in_excess = 0;
4252 mz->on_tree = false;
4258 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
4260 kfree(mem->info.nodeinfo[node]);
4263 static struct mem_cgroup *mem_cgroup_alloc(void)
4265 struct mem_cgroup *mem;
4266 int size = sizeof(struct mem_cgroup);
4268 /* Can be very big if MAX_NUMNODES is very big */
4269 if (size < PAGE_SIZE)
4270 mem = kzalloc(size, GFP_KERNEL);
4272 mem = vzalloc(size);
4277 mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4280 spin_lock_init(&mem->pcp_counter_lock);
4284 if (size < PAGE_SIZE)
4292 * At destroying mem_cgroup, references from swap_cgroup can remain.
4293 * (scanning all at force_empty is too costly...)
4295 * Instead of clearing all references at force_empty, we remember
4296 * the number of reference from swap_cgroup and free mem_cgroup when
4297 * it goes down to 0.
4299 * Removal of cgroup itself succeeds regardless of refs from swap.
4302 static void __mem_cgroup_free(struct mem_cgroup *mem)
4306 mem_cgroup_remove_from_trees(mem);
4307 free_css_id(&mem_cgroup_subsys, &mem->css);
4309 for_each_node_state(node, N_POSSIBLE)
4310 free_mem_cgroup_per_zone_info(mem, node);
4312 free_percpu(mem->stat);
4313 if (sizeof(struct mem_cgroup) < PAGE_SIZE)
4319 static void mem_cgroup_get(struct mem_cgroup *mem)
4321 atomic_inc(&mem->refcnt);
4324 static void __mem_cgroup_put(struct mem_cgroup *mem, int count)
4326 if (atomic_sub_and_test(count, &mem->refcnt)) {
4327 struct mem_cgroup *parent = parent_mem_cgroup(mem);
4328 __mem_cgroup_free(mem);
4330 mem_cgroup_put(parent);
4334 static void mem_cgroup_put(struct mem_cgroup *mem)
4336 __mem_cgroup_put(mem, 1);
4340 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4342 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
4344 if (!mem->res.parent)
4346 return mem_cgroup_from_res_counter(mem->res.parent, res);
4349 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4350 static void __init enable_swap_cgroup(void)
4352 if (!mem_cgroup_disabled() && really_do_swap_account)
4353 do_swap_account = 1;
4356 static void __init enable_swap_cgroup(void)
4361 static int mem_cgroup_soft_limit_tree_init(void)
4363 struct mem_cgroup_tree_per_node *rtpn;
4364 struct mem_cgroup_tree_per_zone *rtpz;
4365 int tmp, node, zone;
4367 for_each_node_state(node, N_POSSIBLE) {
4369 if (!node_state(node, N_NORMAL_MEMORY))
4371 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4375 soft_limit_tree.rb_tree_per_node[node] = rtpn;
4377 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4378 rtpz = &rtpn->rb_tree_per_zone[zone];
4379 rtpz->rb_root = RB_ROOT;
4380 spin_lock_init(&rtpz->lock);
4386 static struct cgroup_subsys_state * __ref
4387 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
4389 struct mem_cgroup *mem, *parent;
4390 long error = -ENOMEM;
4393 mem = mem_cgroup_alloc();
4395 return ERR_PTR(error);
4397 for_each_node_state(node, N_POSSIBLE)
4398 if (alloc_mem_cgroup_per_zone_info(mem, node))
4402 if (cont->parent == NULL) {
4404 enable_swap_cgroup();
4406 root_mem_cgroup = mem;
4407 if (mem_cgroup_soft_limit_tree_init())
4409 for_each_possible_cpu(cpu) {
4410 struct memcg_stock_pcp *stock =
4411 &per_cpu(memcg_stock, cpu);
4412 INIT_WORK(&stock->work, drain_local_stock);
4414 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
4416 parent = mem_cgroup_from_cont(cont->parent);
4417 mem->use_hierarchy = parent->use_hierarchy;
4418 mem->oom_kill_disable = parent->oom_kill_disable;
4421 if (parent && parent->use_hierarchy) {
4422 res_counter_init(&mem->res, &parent->res);
4423 res_counter_init(&mem->memsw, &parent->memsw);
4425 * We increment refcnt of the parent to ensure that we can
4426 * safely access it on res_counter_charge/uncharge.
4427 * This refcnt will be decremented when freeing this
4428 * mem_cgroup(see mem_cgroup_put).
4430 mem_cgroup_get(parent);
4432 res_counter_init(&mem->res, NULL);
4433 res_counter_init(&mem->memsw, NULL);
4435 mem->last_scanned_child = 0;
4436 spin_lock_init(&mem->reclaim_param_lock);
4437 INIT_LIST_HEAD(&mem->oom_notify);
4440 mem->swappiness = get_swappiness(parent);
4441 atomic_set(&mem->refcnt, 1);
4442 mem->move_charge_at_immigrate = 0;
4443 mutex_init(&mem->thresholds_lock);
4446 __mem_cgroup_free(mem);
4447 root_mem_cgroup = NULL;
4448 return ERR_PTR(error);
4451 static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
4452 struct cgroup *cont)
4454 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
4456 return mem_cgroup_force_empty(mem, false);
4459 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
4460 struct cgroup *cont)
4462 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
4464 mem_cgroup_put(mem);
4467 static int mem_cgroup_populate(struct cgroup_subsys *ss,
4468 struct cgroup *cont)
4472 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
4473 ARRAY_SIZE(mem_cgroup_files));
4476 ret = register_memsw_files(cont, ss);
4481 /* Handlers for move charge at task migration. */
4482 #define PRECHARGE_COUNT_AT_ONCE 256
4483 static int mem_cgroup_do_precharge(unsigned long count)
4486 int batch_count = PRECHARGE_COUNT_AT_ONCE;
4487 struct mem_cgroup *mem = mc.to;
4489 if (mem_cgroup_is_root(mem)) {
4490 mc.precharge += count;
4491 /* we don't need css_get for root */
4494 /* try to charge at once */
4496 struct res_counter *dummy;
4498 * "mem" cannot be under rmdir() because we've already checked
4499 * by cgroup_lock_live_cgroup() that it is not removed and we
4500 * are still under the same cgroup_mutex. So we can postpone
4503 if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy))
4505 if (do_swap_account && res_counter_charge(&mem->memsw,
4506 PAGE_SIZE * count, &dummy)) {
4507 res_counter_uncharge(&mem->res, PAGE_SIZE * count);
4510 mc.precharge += count;
4514 /* fall back to one by one charge */
4516 if (signal_pending(current)) {
4520 if (!batch_count--) {
4521 batch_count = PRECHARGE_COUNT_AT_ONCE;
4524 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false,
4527 /* mem_cgroup_clear_mc() will do uncharge later */
4535 * is_target_pte_for_mc - check a pte whether it is valid for move charge
4536 * @vma: the vma the pte to be checked belongs
4537 * @addr: the address corresponding to the pte to be checked
4538 * @ptent: the pte to be checked
4539 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4542 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4543 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4544 * move charge. if @target is not NULL, the page is stored in target->page
4545 * with extra refcnt got(Callers should handle it).
4546 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4547 * target for charge migration. if @target is not NULL, the entry is stored
4550 * Called with pte lock held.
4557 enum mc_target_type {
4558 MC_TARGET_NONE, /* not used */
4563 static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4564 unsigned long addr, pte_t ptent)
4566 struct page *page = vm_normal_page(vma, addr, ptent);
4568 if (!page || !page_mapped(page))
4570 if (PageAnon(page)) {
4571 /* we don't move shared anon */
4572 if (!move_anon() || page_mapcount(page) > 2)
4574 } else if (!move_file())
4575 /* we ignore mapcount for file pages */
4577 if (!get_page_unless_zero(page))
4583 static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4584 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4587 struct page *page = NULL;
4588 swp_entry_t ent = pte_to_swp_entry(ptent);
4590 if (!move_anon() || non_swap_entry(ent))
4592 usage_count = mem_cgroup_count_swap_user(ent, &page);
4593 if (usage_count > 1) { /* we don't move shared anon */
4598 if (do_swap_account)
4599 entry->val = ent.val;
4604 static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4605 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4607 struct page *page = NULL;
4608 struct inode *inode;
4609 struct address_space *mapping;
4612 if (!vma->vm_file) /* anonymous vma */
4617 inode = vma->vm_file->f_path.dentry->d_inode;
4618 mapping = vma->vm_file->f_mapping;
4619 if (pte_none(ptent))
4620 pgoff = linear_page_index(vma, addr);
4621 else /* pte_file(ptent) is true */
4622 pgoff = pte_to_pgoff(ptent);
4624 /* page is moved even if it's not RSS of this task(page-faulted). */
4625 if (!mapping_cap_swap_backed(mapping)) { /* normal file */
4626 page = find_get_page(mapping, pgoff);
4627 } else { /* shmem/tmpfs file. we should take account of swap too. */
4629 mem_cgroup_get_shmem_target(inode, pgoff, &page, &ent);
4630 if (do_swap_account)
4631 entry->val = ent.val;
4637 static int is_target_pte_for_mc(struct vm_area_struct *vma,
4638 unsigned long addr, pte_t ptent, union mc_target *target)
4640 struct page *page = NULL;
4641 struct page_cgroup *pc;
4643 swp_entry_t ent = { .val = 0 };
4645 if (pte_present(ptent))
4646 page = mc_handle_present_pte(vma, addr, ptent);
4647 else if (is_swap_pte(ptent))
4648 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
4649 else if (pte_none(ptent) || pte_file(ptent))
4650 page = mc_handle_file_pte(vma, addr, ptent, &ent);
4652 if (!page && !ent.val)
4655 pc = lookup_page_cgroup(page);
4657 * Do only loose check w/o page_cgroup lock.
4658 * mem_cgroup_move_account() checks the pc is valid or not under
4661 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
4662 ret = MC_TARGET_PAGE;
4664 target->page = page;
4666 if (!ret || !target)
4669 /* There is a swap entry and a page doesn't exist or isn't charged */
4670 if (ent.val && !ret &&
4671 css_id(&mc.from->css) == lookup_swap_cgroup(ent)) {
4672 ret = MC_TARGET_SWAP;
4679 static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4680 unsigned long addr, unsigned long end,
4681 struct mm_walk *walk)
4683 struct vm_area_struct *vma = walk->private;
4687 VM_BUG_ON(pmd_trans_huge(*pmd));
4688 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4689 for (; addr != end; pte++, addr += PAGE_SIZE)
4690 if (is_target_pte_for_mc(vma, addr, *pte, NULL))
4691 mc.precharge++; /* increment precharge temporarily */
4692 pte_unmap_unlock(pte - 1, ptl);
4698 static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4700 unsigned long precharge;
4701 struct vm_area_struct *vma;
4703 down_read(&mm->mmap_sem);
4704 for (vma = mm->mmap; vma; vma = vma->vm_next) {
4705 struct mm_walk mem_cgroup_count_precharge_walk = {
4706 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4710 if (is_vm_hugetlb_page(vma))
4712 walk_page_range(vma->vm_start, vma->vm_end,
4713 &mem_cgroup_count_precharge_walk);
4715 up_read(&mm->mmap_sem);
4717 precharge = mc.precharge;
4723 static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4725 unsigned long precharge = mem_cgroup_count_precharge(mm);
4727 VM_BUG_ON(mc.moving_task);
4728 mc.moving_task = current;
4729 return mem_cgroup_do_precharge(precharge);
4732 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4733 static void __mem_cgroup_clear_mc(void)
4735 struct mem_cgroup *from = mc.from;
4736 struct mem_cgroup *to = mc.to;
4738 /* we must uncharge all the leftover precharges from mc.to */
4740 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
4744 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4745 * we must uncharge here.
4747 if (mc.moved_charge) {
4748 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
4749 mc.moved_charge = 0;
4751 /* we must fixup refcnts and charges */
4752 if (mc.moved_swap) {
4753 /* uncharge swap account from the old cgroup */
4754 if (!mem_cgroup_is_root(mc.from))
4755 res_counter_uncharge(&mc.from->memsw,
4756 PAGE_SIZE * mc.moved_swap);
4757 __mem_cgroup_put(mc.from, mc.moved_swap);
4759 if (!mem_cgroup_is_root(mc.to)) {
4761 * we charged both to->res and to->memsw, so we should
4764 res_counter_uncharge(&mc.to->res,
4765 PAGE_SIZE * mc.moved_swap);
4767 /* we've already done mem_cgroup_get(mc.to) */
4770 memcg_oom_recover(from);
4771 memcg_oom_recover(to);
4772 wake_up_all(&mc.waitq);
4775 static void mem_cgroup_clear_mc(void)
4777 struct mem_cgroup *from = mc.from;
4780 * we must clear moving_task before waking up waiters at the end of
4783 mc.moving_task = NULL;
4784 __mem_cgroup_clear_mc();
4785 spin_lock(&mc.lock);
4788 spin_unlock(&mc.lock);
4789 mem_cgroup_end_move(from);
4792 static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
4793 struct cgroup *cgroup,
4794 struct task_struct *p,
4798 struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup);
4800 if (mem->move_charge_at_immigrate) {
4801 struct mm_struct *mm;
4802 struct mem_cgroup *from = mem_cgroup_from_task(p);
4804 VM_BUG_ON(from == mem);
4806 mm = get_task_mm(p);
4809 /* We move charges only when we move a owner of the mm */
4810 if (mm->owner == p) {
4813 VM_BUG_ON(mc.precharge);
4814 VM_BUG_ON(mc.moved_charge);
4815 VM_BUG_ON(mc.moved_swap);
4816 mem_cgroup_start_move(from);
4817 spin_lock(&mc.lock);
4820 spin_unlock(&mc.lock);
4821 /* We set mc.moving_task later */
4823 ret = mem_cgroup_precharge_mc(mm);
4825 mem_cgroup_clear_mc();
4832 static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
4833 struct cgroup *cgroup,
4834 struct task_struct *p,
4837 mem_cgroup_clear_mc();
4840 static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4841 unsigned long addr, unsigned long end,
4842 struct mm_walk *walk)
4845 struct vm_area_struct *vma = walk->private;
4850 VM_BUG_ON(pmd_trans_huge(*pmd));
4851 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4852 for (; addr != end; addr += PAGE_SIZE) {
4853 pte_t ptent = *(pte++);
4854 union mc_target target;
4857 struct page_cgroup *pc;
4863 type = is_target_pte_for_mc(vma, addr, ptent, &target);
4865 case MC_TARGET_PAGE:
4867 if (isolate_lru_page(page))
4869 pc = lookup_page_cgroup(page);
4870 if (!mem_cgroup_move_account(pc,
4871 mc.from, mc.to, false)) {
4873 /* we uncharge from mc.from later. */
4876 putback_lru_page(page);
4877 put: /* is_target_pte_for_mc() gets the page */
4880 case MC_TARGET_SWAP:
4882 if (!mem_cgroup_move_swap_account(ent,
4883 mc.from, mc.to, false)) {
4885 /* we fixup refcnts and charges later. */
4893 pte_unmap_unlock(pte - 1, ptl);
4898 * We have consumed all precharges we got in can_attach().
4899 * We try charge one by one, but don't do any additional
4900 * charges to mc.to if we have failed in charge once in attach()
4903 ret = mem_cgroup_do_precharge(1);
4911 static void mem_cgroup_move_charge(struct mm_struct *mm)
4913 struct vm_area_struct *vma;
4915 lru_add_drain_all();
4917 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
4919 * Someone who are holding the mmap_sem might be waiting in
4920 * waitq. So we cancel all extra charges, wake up all waiters,
4921 * and retry. Because we cancel precharges, we might not be able
4922 * to move enough charges, but moving charge is a best-effort
4923 * feature anyway, so it wouldn't be a big problem.
4925 __mem_cgroup_clear_mc();
4929 for (vma = mm->mmap; vma; vma = vma->vm_next) {
4931 struct mm_walk mem_cgroup_move_charge_walk = {
4932 .pmd_entry = mem_cgroup_move_charge_pte_range,
4936 if (is_vm_hugetlb_page(vma))
4938 ret = walk_page_range(vma->vm_start, vma->vm_end,
4939 &mem_cgroup_move_charge_walk);
4942 * means we have consumed all precharges and failed in
4943 * doing additional charge. Just abandon here.
4947 up_read(&mm->mmap_sem);
4950 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
4951 struct cgroup *cont,
4952 struct cgroup *old_cont,
4953 struct task_struct *p,
4956 struct mm_struct *mm;
4959 /* no need to move charge */
4962 mm = get_task_mm(p);
4964 mem_cgroup_move_charge(mm);
4967 mem_cgroup_clear_mc();
4969 #else /* !CONFIG_MMU */
4970 static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
4971 struct cgroup *cgroup,
4972 struct task_struct *p,
4977 static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
4978 struct cgroup *cgroup,
4979 struct task_struct *p,
4983 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
4984 struct cgroup *cont,
4985 struct cgroup *old_cont,
4986 struct task_struct *p,
4992 struct cgroup_subsys mem_cgroup_subsys = {
4994 .subsys_id = mem_cgroup_subsys_id,
4995 .create = mem_cgroup_create,
4996 .pre_destroy = mem_cgroup_pre_destroy,
4997 .destroy = mem_cgroup_destroy,
4998 .populate = mem_cgroup_populate,
4999 .can_attach = mem_cgroup_can_attach,
5000 .cancel_attach = mem_cgroup_cancel_attach,
5001 .attach = mem_cgroup_move_task,
5006 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5007 static int __init enable_swap_account(char *s)
5009 /* consider enabled if no parameter or 1 is given */
5010 if (!s || !strcmp(s, "1"))
5011 really_do_swap_account = 1;
5012 else if (!strcmp(s, "0"))
5013 really_do_swap_account = 0;
5016 __setup("swapaccount", enable_swap_account);
5018 static int __init disable_swap_account(char *s)
5020 enable_swap_account("0");
5023 __setup("noswapaccount", disable_swap_account);