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>
52 #include <asm/uaccess.h>
54 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
55 #define MEM_CGROUP_RECLAIM_RETRIES 5
56 struct mem_cgroup *root_mem_cgroup __read_mostly;
58 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
59 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
60 int do_swap_account __read_mostly;
61 static int really_do_swap_account __initdata = 1; /* for remember boot option*/
63 #define do_swap_account (0)
66 #define SOFTLIMIT_EVENTS_THRESH (1000)
67 #define THRESHOLDS_EVENTS_THRESH (100)
70 * Statistics for memory cgroup.
72 enum mem_cgroup_stat_index {
74 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
76 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
77 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
78 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
79 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
80 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
81 MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
82 MEM_CGROUP_STAT_SOFTLIMIT, /* decrements on each page in/out.
83 used by soft limit implementation */
84 MEM_CGROUP_STAT_THRESHOLDS, /* decrements on each page in/out.
85 used by threshold implementation */
87 MEM_CGROUP_STAT_NSTATS,
90 struct mem_cgroup_stat_cpu {
91 s64 count[MEM_CGROUP_STAT_NSTATS];
95 * per-zone information in memory controller.
97 struct mem_cgroup_per_zone {
99 * spin_lock to protect the per cgroup LRU
101 struct list_head lists[NR_LRU_LISTS];
102 unsigned long count[NR_LRU_LISTS];
104 struct zone_reclaim_stat reclaim_stat;
105 struct rb_node tree_node; /* RB tree node */
106 unsigned long long usage_in_excess;/* Set to the value by which */
107 /* the soft limit is exceeded*/
109 struct mem_cgroup *mem; /* Back pointer, we cannot */
110 /* use container_of */
112 /* Macro for accessing counter */
113 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
115 struct mem_cgroup_per_node {
116 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
119 struct mem_cgroup_lru_info {
120 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
124 * Cgroups above their limits are maintained in a RB-Tree, independent of
125 * their hierarchy representation
128 struct mem_cgroup_tree_per_zone {
129 struct rb_root rb_root;
133 struct mem_cgroup_tree_per_node {
134 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
137 struct mem_cgroup_tree {
138 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
141 static struct mem_cgroup_tree soft_limit_tree __read_mostly;
143 struct mem_cgroup_threshold {
144 struct eventfd_ctx *eventfd;
148 struct mem_cgroup_threshold_ary {
149 /* An array index points to threshold just below usage. */
150 atomic_t current_threshold;
151 /* Size of entries[] */
153 /* Array of thresholds */
154 struct mem_cgroup_threshold entries[0];
157 static bool mem_cgroup_threshold_check(struct mem_cgroup *mem);
158 static void mem_cgroup_threshold(struct mem_cgroup *mem);
161 * The memory controller data structure. The memory controller controls both
162 * page cache and RSS per cgroup. We would eventually like to provide
163 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
164 * to help the administrator determine what knobs to tune.
166 * TODO: Add a water mark for the memory controller. Reclaim will begin when
167 * we hit the water mark. May be even add a low water mark, such that
168 * no reclaim occurs from a cgroup at it's low water mark, this is
169 * a feature that will be implemented much later in the future.
172 struct cgroup_subsys_state css;
174 * the counter to account for memory usage
176 struct res_counter res;
178 * the counter to account for mem+swap usage.
180 struct res_counter memsw;
182 * Per cgroup active and inactive list, similar to the
183 * per zone LRU lists.
185 struct mem_cgroup_lru_info info;
188 protect against reclaim related member.
190 spinlock_t reclaim_param_lock;
192 int prev_priority; /* for recording reclaim priority */
195 * While reclaiming in a hierarchy, we cache the last child we
198 int last_scanned_child;
200 * Should the accounting and control be hierarchical, per subtree?
203 unsigned long last_oom_jiffies;
206 unsigned int swappiness;
208 /* set when res.limit == memsw.limit */
209 bool memsw_is_minimum;
211 /* protect arrays of thresholds */
212 struct mutex thresholds_lock;
214 /* thresholds for memory usage. RCU-protected */
215 struct mem_cgroup_threshold_ary *thresholds;
217 /* thresholds for mem+swap usage. RCU-protected */
218 struct mem_cgroup_threshold_ary *memsw_thresholds;
221 * Should we move charges of a task when a task is moved into this
222 * mem_cgroup ? And what type of charges should we move ?
224 unsigned long move_charge_at_immigrate;
229 struct mem_cgroup_stat_cpu *stat;
232 /* Stuffs for move charges at task migration. */
234 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
235 * left-shifted bitmap of these types.
238 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
242 /* "mc" and its members are protected by cgroup_mutex */
243 static struct move_charge_struct {
244 struct mem_cgroup *from;
245 struct mem_cgroup *to;
246 unsigned long precharge;
247 unsigned long moved_charge;
248 unsigned long moved_swap;
249 struct task_struct *moving_task; /* a task moving charges */
250 wait_queue_head_t waitq; /* a waitq for other context */
252 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
256 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
257 * limit reclaim to prevent infinite loops, if they ever occur.
259 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
260 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
263 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
264 MEM_CGROUP_CHARGE_TYPE_MAPPED,
265 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
266 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
267 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
268 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
272 /* only for here (for easy reading.) */
273 #define PCGF_CACHE (1UL << PCG_CACHE)
274 #define PCGF_USED (1UL << PCG_USED)
275 #define PCGF_LOCK (1UL << PCG_LOCK)
276 /* Not used, but added here for completeness */
277 #define PCGF_ACCT (1UL << PCG_ACCT)
279 /* for encoding cft->private value on file */
282 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
283 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
284 #define MEMFILE_ATTR(val) ((val) & 0xffff)
287 * Reclaim flags for mem_cgroup_hierarchical_reclaim
289 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
290 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
291 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
292 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
293 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
294 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
296 static void mem_cgroup_get(struct mem_cgroup *mem);
297 static void mem_cgroup_put(struct mem_cgroup *mem);
298 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
299 static void drain_all_stock_async(void);
301 static struct mem_cgroup_per_zone *
302 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
304 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
307 struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem)
312 static struct mem_cgroup_per_zone *
313 page_cgroup_zoneinfo(struct page_cgroup *pc)
315 struct mem_cgroup *mem = pc->mem_cgroup;
316 int nid = page_cgroup_nid(pc);
317 int zid = page_cgroup_zid(pc);
322 return mem_cgroup_zoneinfo(mem, nid, zid);
325 static struct mem_cgroup_tree_per_zone *
326 soft_limit_tree_node_zone(int nid, int zid)
328 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
331 static struct mem_cgroup_tree_per_zone *
332 soft_limit_tree_from_page(struct page *page)
334 int nid = page_to_nid(page);
335 int zid = page_zonenum(page);
337 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
341 __mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
342 struct mem_cgroup_per_zone *mz,
343 struct mem_cgroup_tree_per_zone *mctz,
344 unsigned long long new_usage_in_excess)
346 struct rb_node **p = &mctz->rb_root.rb_node;
347 struct rb_node *parent = NULL;
348 struct mem_cgroup_per_zone *mz_node;
353 mz->usage_in_excess = new_usage_in_excess;
354 if (!mz->usage_in_excess)
358 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
360 if (mz->usage_in_excess < mz_node->usage_in_excess)
363 * We can't avoid mem cgroups that are over their soft
364 * limit by the same amount
366 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
369 rb_link_node(&mz->tree_node, parent, p);
370 rb_insert_color(&mz->tree_node, &mctz->rb_root);
375 __mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
376 struct mem_cgroup_per_zone *mz,
377 struct mem_cgroup_tree_per_zone *mctz)
381 rb_erase(&mz->tree_node, &mctz->rb_root);
386 mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
387 struct mem_cgroup_per_zone *mz,
388 struct mem_cgroup_tree_per_zone *mctz)
390 spin_lock(&mctz->lock);
391 __mem_cgroup_remove_exceeded(mem, mz, mctz);
392 spin_unlock(&mctz->lock);
395 static bool mem_cgroup_soft_limit_check(struct mem_cgroup *mem)
400 val = this_cpu_read(mem->stat->count[MEM_CGROUP_STAT_SOFTLIMIT]);
401 if (unlikely(val < 0)) {
402 this_cpu_write(mem->stat->count[MEM_CGROUP_STAT_SOFTLIMIT],
403 SOFTLIMIT_EVENTS_THRESH);
409 static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page)
411 unsigned long long excess;
412 struct mem_cgroup_per_zone *mz;
413 struct mem_cgroup_tree_per_zone *mctz;
414 int nid = page_to_nid(page);
415 int zid = page_zonenum(page);
416 mctz = soft_limit_tree_from_page(page);
419 * Necessary to update all ancestors when hierarchy is used.
420 * because their event counter is not touched.
422 for (; mem; mem = parent_mem_cgroup(mem)) {
423 mz = mem_cgroup_zoneinfo(mem, nid, zid);
424 excess = res_counter_soft_limit_excess(&mem->res);
426 * We have to update the tree if mz is on RB-tree or
427 * mem is over its softlimit.
429 if (excess || mz->on_tree) {
430 spin_lock(&mctz->lock);
431 /* if on-tree, remove it */
433 __mem_cgroup_remove_exceeded(mem, mz, mctz);
435 * Insert again. mz->usage_in_excess will be updated.
436 * If excess is 0, no tree ops.
438 __mem_cgroup_insert_exceeded(mem, mz, mctz, excess);
439 spin_unlock(&mctz->lock);
444 static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem)
447 struct mem_cgroup_per_zone *mz;
448 struct mem_cgroup_tree_per_zone *mctz;
450 for_each_node_state(node, N_POSSIBLE) {
451 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
452 mz = mem_cgroup_zoneinfo(mem, node, zone);
453 mctz = soft_limit_tree_node_zone(node, zone);
454 mem_cgroup_remove_exceeded(mem, mz, mctz);
459 static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem)
461 return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT;
464 static struct mem_cgroup_per_zone *
465 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
467 struct rb_node *rightmost = NULL;
468 struct mem_cgroup_per_zone *mz;
472 rightmost = rb_last(&mctz->rb_root);
474 goto done; /* Nothing to reclaim from */
476 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
478 * Remove the node now but someone else can add it back,
479 * we will to add it back at the end of reclaim to its correct
480 * position in the tree.
482 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
483 if (!res_counter_soft_limit_excess(&mz->mem->res) ||
484 !css_tryget(&mz->mem->css))
490 static struct mem_cgroup_per_zone *
491 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
493 struct mem_cgroup_per_zone *mz;
495 spin_lock(&mctz->lock);
496 mz = __mem_cgroup_largest_soft_limit_node(mctz);
497 spin_unlock(&mctz->lock);
501 static s64 mem_cgroup_read_stat(struct mem_cgroup *mem,
502 enum mem_cgroup_stat_index idx)
507 for_each_possible_cpu(cpu)
508 val += per_cpu(mem->stat->count[idx], cpu);
512 static s64 mem_cgroup_local_usage(struct mem_cgroup *mem)
516 ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
517 ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
521 static void mem_cgroup_swap_statistics(struct mem_cgroup *mem,
524 int val = (charge) ? 1 : -1;
525 this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
528 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
529 struct page_cgroup *pc,
532 int val = (charge) ? 1 : -1;
536 if (PageCgroupCache(pc))
537 __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], val);
539 __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], val);
542 __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]);
544 __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]);
545 __this_cpu_dec(mem->stat->count[MEM_CGROUP_STAT_SOFTLIMIT]);
546 __this_cpu_dec(mem->stat->count[MEM_CGROUP_STAT_THRESHOLDS]);
551 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
555 struct mem_cgroup_per_zone *mz;
558 for_each_online_node(nid)
559 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
560 mz = mem_cgroup_zoneinfo(mem, nid, zid);
561 total += MEM_CGROUP_ZSTAT(mz, idx);
566 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
568 return container_of(cgroup_subsys_state(cont,
569 mem_cgroup_subsys_id), struct mem_cgroup,
573 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
576 * mm_update_next_owner() may clear mm->owner to NULL
577 * if it races with swapoff, page migration, etc.
578 * So this can be called with p == NULL.
583 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
584 struct mem_cgroup, css);
587 static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
589 struct mem_cgroup *mem = NULL;
594 * Because we have no locks, mm->owner's may be being moved to other
595 * cgroup. We use css_tryget() here even if this looks
596 * pessimistic (rather than adding locks here).
600 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
603 } while (!css_tryget(&mem->css));
609 * Call callback function against all cgroup under hierarchy tree.
611 static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
612 int (*func)(struct mem_cgroup *, void *))
614 int found, ret, nextid;
615 struct cgroup_subsys_state *css;
616 struct mem_cgroup *mem;
618 if (!root->use_hierarchy)
619 return (*func)(root, data);
627 css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
629 if (css && css_tryget(css))
630 mem = container_of(css, struct mem_cgroup, css);
634 ret = (*func)(mem, data);
638 } while (!ret && css);
643 static inline bool mem_cgroup_is_root(struct mem_cgroup *mem)
645 return (mem == root_mem_cgroup);
649 * Following LRU functions are allowed to be used without PCG_LOCK.
650 * Operations are called by routine of global LRU independently from memcg.
651 * What we have to take care of here is validness of pc->mem_cgroup.
653 * Changes to pc->mem_cgroup happens when
656 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
657 * It is added to LRU before charge.
658 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
659 * When moving account, the page is not on LRU. It's isolated.
662 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
664 struct page_cgroup *pc;
665 struct mem_cgroup_per_zone *mz;
667 if (mem_cgroup_disabled())
669 pc = lookup_page_cgroup(page);
670 /* can happen while we handle swapcache. */
671 if (!TestClearPageCgroupAcctLRU(pc))
673 VM_BUG_ON(!pc->mem_cgroup);
675 * We don't check PCG_USED bit. It's cleared when the "page" is finally
676 * removed from global LRU.
678 mz = page_cgroup_zoneinfo(pc);
679 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
680 if (mem_cgroup_is_root(pc->mem_cgroup))
682 VM_BUG_ON(list_empty(&pc->lru));
683 list_del_init(&pc->lru);
687 void mem_cgroup_del_lru(struct page *page)
689 mem_cgroup_del_lru_list(page, page_lru(page));
692 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
694 struct mem_cgroup_per_zone *mz;
695 struct page_cgroup *pc;
697 if (mem_cgroup_disabled())
700 pc = lookup_page_cgroup(page);
702 * Used bit is set without atomic ops but after smp_wmb().
703 * For making pc->mem_cgroup visible, insert smp_rmb() here.
706 /* unused or root page is not rotated. */
707 if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup))
709 mz = page_cgroup_zoneinfo(pc);
710 list_move(&pc->lru, &mz->lists[lru]);
713 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
715 struct page_cgroup *pc;
716 struct mem_cgroup_per_zone *mz;
718 if (mem_cgroup_disabled())
720 pc = lookup_page_cgroup(page);
721 VM_BUG_ON(PageCgroupAcctLRU(pc));
723 * Used bit is set without atomic ops but after smp_wmb().
724 * For making pc->mem_cgroup visible, insert smp_rmb() here.
727 if (!PageCgroupUsed(pc))
730 mz = page_cgroup_zoneinfo(pc);
731 MEM_CGROUP_ZSTAT(mz, lru) += 1;
732 SetPageCgroupAcctLRU(pc);
733 if (mem_cgroup_is_root(pc->mem_cgroup))
735 list_add(&pc->lru, &mz->lists[lru]);
739 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
740 * lru because the page may.be reused after it's fully uncharged (because of
741 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
742 * it again. This function is only used to charge SwapCache. It's done under
743 * lock_page and expected that zone->lru_lock is never held.
745 static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
748 struct zone *zone = page_zone(page);
749 struct page_cgroup *pc = lookup_page_cgroup(page);
751 spin_lock_irqsave(&zone->lru_lock, flags);
753 * Forget old LRU when this page_cgroup is *not* used. This Used bit
754 * is guarded by lock_page() because the page is SwapCache.
756 if (!PageCgroupUsed(pc))
757 mem_cgroup_del_lru_list(page, page_lru(page));
758 spin_unlock_irqrestore(&zone->lru_lock, flags);
761 static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
764 struct zone *zone = page_zone(page);
765 struct page_cgroup *pc = lookup_page_cgroup(page);
767 spin_lock_irqsave(&zone->lru_lock, flags);
768 /* link when the page is linked to LRU but page_cgroup isn't */
769 if (PageLRU(page) && !PageCgroupAcctLRU(pc))
770 mem_cgroup_add_lru_list(page, page_lru(page));
771 spin_unlock_irqrestore(&zone->lru_lock, flags);
775 void mem_cgroup_move_lists(struct page *page,
776 enum lru_list from, enum lru_list to)
778 if (mem_cgroup_disabled())
780 mem_cgroup_del_lru_list(page, from);
781 mem_cgroup_add_lru_list(page, to);
784 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
787 struct mem_cgroup *curr = NULL;
791 curr = try_get_mem_cgroup_from_mm(task->mm);
797 * We should check use_hierarchy of "mem" not "curr". Because checking
798 * use_hierarchy of "curr" here make this function true if hierarchy is
799 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
800 * hierarchy(even if use_hierarchy is disabled in "mem").
802 if (mem->use_hierarchy)
803 ret = css_is_ancestor(&curr->css, &mem->css);
811 * prev_priority control...this will be used in memory reclaim path.
813 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
817 spin_lock(&mem->reclaim_param_lock);
818 prev_priority = mem->prev_priority;
819 spin_unlock(&mem->reclaim_param_lock);
821 return prev_priority;
824 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
826 spin_lock(&mem->reclaim_param_lock);
827 if (priority < mem->prev_priority)
828 mem->prev_priority = priority;
829 spin_unlock(&mem->reclaim_param_lock);
832 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
834 spin_lock(&mem->reclaim_param_lock);
835 mem->prev_priority = priority;
836 spin_unlock(&mem->reclaim_param_lock);
839 static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
841 unsigned long active;
842 unsigned long inactive;
844 unsigned long inactive_ratio;
846 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
847 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
849 gb = (inactive + active) >> (30 - PAGE_SHIFT);
851 inactive_ratio = int_sqrt(10 * gb);
856 present_pages[0] = inactive;
857 present_pages[1] = active;
860 return inactive_ratio;
863 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
865 unsigned long active;
866 unsigned long inactive;
867 unsigned long present_pages[2];
868 unsigned long inactive_ratio;
870 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
872 inactive = present_pages[0];
873 active = present_pages[1];
875 if (inactive * inactive_ratio < active)
881 int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
883 unsigned long active;
884 unsigned long inactive;
886 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
887 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
889 return (active > inactive);
892 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
896 int nid = zone->zone_pgdat->node_id;
897 int zid = zone_idx(zone);
898 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
900 return MEM_CGROUP_ZSTAT(mz, lru);
903 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
906 int nid = zone->zone_pgdat->node_id;
907 int zid = zone_idx(zone);
908 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
910 return &mz->reclaim_stat;
913 struct zone_reclaim_stat *
914 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
916 struct page_cgroup *pc;
917 struct mem_cgroup_per_zone *mz;
919 if (mem_cgroup_disabled())
922 pc = lookup_page_cgroup(page);
924 * Used bit is set without atomic ops but after smp_wmb().
925 * For making pc->mem_cgroup visible, insert smp_rmb() here.
928 if (!PageCgroupUsed(pc))
931 mz = page_cgroup_zoneinfo(pc);
935 return &mz->reclaim_stat;
938 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
939 struct list_head *dst,
940 unsigned long *scanned, int order,
941 int mode, struct zone *z,
942 struct mem_cgroup *mem_cont,
943 int active, int file)
945 unsigned long nr_taken = 0;
949 struct list_head *src;
950 struct page_cgroup *pc, *tmp;
951 int nid = z->zone_pgdat->node_id;
952 int zid = zone_idx(z);
953 struct mem_cgroup_per_zone *mz;
954 int lru = LRU_FILE * file + active;
958 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
959 src = &mz->lists[lru];
962 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
963 if (scan >= nr_to_scan)
967 if (unlikely(!PageCgroupUsed(pc)))
969 if (unlikely(!PageLRU(page)))
973 ret = __isolate_lru_page(page, mode, file);
976 list_move(&page->lru, dst);
977 mem_cgroup_del_lru(page);
981 /* we don't affect global LRU but rotate in our LRU */
982 mem_cgroup_rotate_lru_list(page, page_lru(page));
993 #define mem_cgroup_from_res_counter(counter, member) \
994 container_of(counter, struct mem_cgroup, member)
996 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
998 if (do_swap_account) {
999 if (res_counter_check_under_limit(&mem->res) &&
1000 res_counter_check_under_limit(&mem->memsw))
1003 if (res_counter_check_under_limit(&mem->res))
1008 static unsigned int get_swappiness(struct mem_cgroup *memcg)
1010 struct cgroup *cgrp = memcg->css.cgroup;
1011 unsigned int swappiness;
1014 if (cgrp->parent == NULL)
1015 return vm_swappiness;
1017 spin_lock(&memcg->reclaim_param_lock);
1018 swappiness = memcg->swappiness;
1019 spin_unlock(&memcg->reclaim_param_lock);
1024 static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
1032 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1033 * @memcg: The memory cgroup that went over limit
1034 * @p: Task that is going to be killed
1036 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1039 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1041 struct cgroup *task_cgrp;
1042 struct cgroup *mem_cgrp;
1044 * Need a buffer in BSS, can't rely on allocations. The code relies
1045 * on the assumption that OOM is serialized for memory controller.
1046 * If this assumption is broken, revisit this code.
1048 static char memcg_name[PATH_MAX];
1057 mem_cgrp = memcg->css.cgroup;
1058 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1060 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1063 * Unfortunately, we are unable to convert to a useful name
1064 * But we'll still print out the usage information
1071 printk(KERN_INFO "Task in %s killed", memcg_name);
1074 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1082 * Continues from above, so we don't need an KERN_ level
1084 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1087 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1088 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1089 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1090 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1091 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1093 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1094 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1095 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1099 * This function returns the number of memcg under hierarchy tree. Returns
1100 * 1(self count) if no children.
1102 static int mem_cgroup_count_children(struct mem_cgroup *mem)
1105 mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
1110 * Visit the first child (need not be the first child as per the ordering
1111 * of the cgroup list, since we track last_scanned_child) of @mem and use
1112 * that to reclaim free pages from.
1114 static struct mem_cgroup *
1115 mem_cgroup_select_victim(struct mem_cgroup *root_mem)
1117 struct mem_cgroup *ret = NULL;
1118 struct cgroup_subsys_state *css;
1121 if (!root_mem->use_hierarchy) {
1122 css_get(&root_mem->css);
1128 nextid = root_mem->last_scanned_child + 1;
1129 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
1131 if (css && css_tryget(css))
1132 ret = container_of(css, struct mem_cgroup, css);
1135 /* Updates scanning parameter */
1136 spin_lock(&root_mem->reclaim_param_lock);
1138 /* this means start scan from ID:1 */
1139 root_mem->last_scanned_child = 0;
1141 root_mem->last_scanned_child = found;
1142 spin_unlock(&root_mem->reclaim_param_lock);
1149 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1150 * we reclaimed from, so that we don't end up penalizing one child extensively
1151 * based on its position in the children list.
1153 * root_mem is the original ancestor that we've been reclaim from.
1155 * We give up and return to the caller when we visit root_mem twice.
1156 * (other groups can be removed while we're walking....)
1158 * If shrink==true, for avoiding to free too much, this returns immedieately.
1160 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
1163 unsigned long reclaim_options)
1165 struct mem_cgroup *victim;
1168 bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
1169 bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
1170 bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT;
1171 unsigned long excess = mem_cgroup_get_excess(root_mem);
1173 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1174 if (root_mem->memsw_is_minimum)
1178 victim = mem_cgroup_select_victim(root_mem);
1179 if (victim == root_mem) {
1182 drain_all_stock_async();
1185 * If we have not been able to reclaim
1186 * anything, it might because there are
1187 * no reclaimable pages under this hierarchy
1189 if (!check_soft || !total) {
1190 css_put(&victim->css);
1194 * We want to do more targetted reclaim.
1195 * excess >> 2 is not to excessive so as to
1196 * reclaim too much, nor too less that we keep
1197 * coming back to reclaim from this cgroup
1199 if (total >= (excess >> 2) ||
1200 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) {
1201 css_put(&victim->css);
1206 if (!mem_cgroup_local_usage(victim)) {
1207 /* this cgroup's local usage == 0 */
1208 css_put(&victim->css);
1211 /* we use swappiness of local cgroup */
1213 ret = mem_cgroup_shrink_node_zone(victim, gfp_mask,
1214 noswap, get_swappiness(victim), zone,
1215 zone->zone_pgdat->node_id);
1217 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask,
1218 noswap, get_swappiness(victim));
1219 css_put(&victim->css);
1221 * At shrinking usage, we can't check we should stop here or
1222 * reclaim more. It's depends on callers. last_scanned_child
1223 * will work enough for keeping fairness under tree.
1229 if (res_counter_check_under_soft_limit(&root_mem->res))
1231 } else if (mem_cgroup_check_under_limit(root_mem))
1237 bool mem_cgroup_oom_called(struct task_struct *task)
1240 struct mem_cgroup *mem;
1241 struct mm_struct *mm;
1247 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
1248 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
1254 static int record_last_oom_cb(struct mem_cgroup *mem, void *data)
1256 mem->last_oom_jiffies = jiffies;
1260 static void record_last_oom(struct mem_cgroup *mem)
1262 mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb);
1266 * Currently used to update mapped file statistics, but the routine can be
1267 * generalized to update other statistics as well.
1269 void mem_cgroup_update_file_mapped(struct page *page, int val)
1271 struct mem_cgroup *mem;
1272 struct page_cgroup *pc;
1274 pc = lookup_page_cgroup(page);
1278 lock_page_cgroup(pc);
1279 mem = pc->mem_cgroup;
1283 if (!PageCgroupUsed(pc))
1287 * Preemption is already disabled. We can use __this_cpu_xxx
1289 __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], val);
1292 unlock_page_cgroup(pc);
1296 * size of first charge trial. "32" comes from vmscan.c's magic value.
1297 * TODO: maybe necessary to use big numbers in big irons.
1299 #define CHARGE_SIZE (32 * PAGE_SIZE)
1300 struct memcg_stock_pcp {
1301 struct mem_cgroup *cached; /* this never be root cgroup */
1303 struct work_struct work;
1305 static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
1306 static atomic_t memcg_drain_count;
1309 * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
1310 * from local stock and true is returned. If the stock is 0 or charges from a
1311 * cgroup which is not current target, returns false. This stock will be
1314 static bool consume_stock(struct mem_cgroup *mem)
1316 struct memcg_stock_pcp *stock;
1319 stock = &get_cpu_var(memcg_stock);
1320 if (mem == stock->cached && stock->charge)
1321 stock->charge -= PAGE_SIZE;
1322 else /* need to call res_counter_charge */
1324 put_cpu_var(memcg_stock);
1329 * Returns stocks cached in percpu to res_counter and reset cached information.
1331 static void drain_stock(struct memcg_stock_pcp *stock)
1333 struct mem_cgroup *old = stock->cached;
1335 if (stock->charge) {
1336 res_counter_uncharge(&old->res, stock->charge);
1337 if (do_swap_account)
1338 res_counter_uncharge(&old->memsw, stock->charge);
1340 stock->cached = NULL;
1345 * This must be called under preempt disabled or must be called by
1346 * a thread which is pinned to local cpu.
1348 static void drain_local_stock(struct work_struct *dummy)
1350 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
1355 * Cache charges(val) which is from res_counter, to local per_cpu area.
1356 * This will be consumed by consumt_stock() function, later.
1358 static void refill_stock(struct mem_cgroup *mem, int val)
1360 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1362 if (stock->cached != mem) { /* reset if necessary */
1364 stock->cached = mem;
1366 stock->charge += val;
1367 put_cpu_var(memcg_stock);
1371 * Tries to drain stocked charges in other cpus. This function is asynchronous
1372 * and just put a work per cpu for draining localy on each cpu. Caller can
1373 * expects some charges will be back to res_counter later but cannot wait for
1376 static void drain_all_stock_async(void)
1379 /* This function is for scheduling "drain" in asynchronous way.
1380 * The result of "drain" is not directly handled by callers. Then,
1381 * if someone is calling drain, we don't have to call drain more.
1382 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1383 * there is a race. We just do loose check here.
1385 if (atomic_read(&memcg_drain_count))
1387 /* Notify other cpus that system-wide "drain" is running */
1388 atomic_inc(&memcg_drain_count);
1390 for_each_online_cpu(cpu) {
1391 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
1392 schedule_work_on(cpu, &stock->work);
1395 atomic_dec(&memcg_drain_count);
1396 /* We don't wait for flush_work */
1399 /* This is a synchronous drain interface. */
1400 static void drain_all_stock_sync(void)
1402 /* called when force_empty is called */
1403 atomic_inc(&memcg_drain_count);
1404 schedule_on_each_cpu(drain_local_stock);
1405 atomic_dec(&memcg_drain_count);
1408 static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb,
1409 unsigned long action,
1412 int cpu = (unsigned long)hcpu;
1413 struct memcg_stock_pcp *stock;
1415 if (action != CPU_DEAD)
1417 stock = &per_cpu(memcg_stock, cpu);
1423 * Unlike exported interface, "oom" parameter is added. if oom==true,
1424 * oom-killer can be invoked.
1426 static int __mem_cgroup_try_charge(struct mm_struct *mm,
1427 gfp_t gfp_mask, struct mem_cgroup **memcg,
1428 bool oom, struct page *page)
1430 struct mem_cgroup *mem, *mem_over_limit;
1431 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1432 struct res_counter *fail_res;
1433 int csize = CHARGE_SIZE;
1435 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
1436 /* Don't account this! */
1442 * We always charge the cgroup the mm_struct belongs to.
1443 * The mm_struct's mem_cgroup changes on task migration if the
1444 * thread group leader migrates. It's possible that mm is not
1445 * set, if so charge the init_mm (happens for pagecache usage).
1449 mem = try_get_mem_cgroup_from_mm(mm);
1457 VM_BUG_ON(css_is_removed(&mem->css));
1458 if (mem_cgroup_is_root(mem))
1463 unsigned long flags = 0;
1465 if (consume_stock(mem))
1468 ret = res_counter_charge(&mem->res, csize, &fail_res);
1470 if (!do_swap_account)
1472 ret = res_counter_charge(&mem->memsw, csize, &fail_res);
1475 /* mem+swap counter fails */
1476 res_counter_uncharge(&mem->res, csize);
1477 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
1478 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1481 /* mem counter fails */
1482 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1485 /* reduce request size and retry */
1486 if (csize > PAGE_SIZE) {
1490 if (!(gfp_mask & __GFP_WAIT))
1493 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
1499 * try_to_free_mem_cgroup_pages() might not give us a full
1500 * picture of reclaim. Some pages are reclaimed and might be
1501 * moved to swap cache or just unmapped from the cgroup.
1502 * Check the limit again to see if the reclaim reduced the
1503 * current usage of the cgroup before giving up
1506 if (mem_cgroup_check_under_limit(mem_over_limit))
1509 /* try to avoid oom while someone is moving charge */
1510 if (mc.moving_task && current != mc.moving_task) {
1511 struct mem_cgroup *from, *to;
1512 bool do_continue = false;
1514 * There is a small race that "from" or "to" can be
1515 * freed by rmdir, so we use css_tryget().
1520 if (from && css_tryget(&from->css)) {
1521 if (mem_over_limit->use_hierarchy)
1522 do_continue = css_is_ancestor(
1524 &mem_over_limit->css);
1526 do_continue = (from == mem_over_limit);
1527 css_put(&from->css);
1529 if (!do_continue && to && css_tryget(&to->css)) {
1530 if (mem_over_limit->use_hierarchy)
1531 do_continue = css_is_ancestor(
1533 &mem_over_limit->css);
1535 do_continue = (to == mem_over_limit);
1541 prepare_to_wait(&mc.waitq, &wait,
1542 TASK_INTERRUPTIBLE);
1543 /* moving charge context might have finished. */
1546 finish_wait(&mc.waitq, &wait);
1551 if (!nr_retries--) {
1553 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
1554 record_last_oom(mem_over_limit);
1559 if (csize > PAGE_SIZE)
1560 refill_stock(mem, csize - PAGE_SIZE);
1563 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
1564 * if they exceeds softlimit.
1566 if (page && mem_cgroup_soft_limit_check(mem))
1567 mem_cgroup_update_tree(mem, page);
1569 if (mem_cgroup_threshold_check(mem))
1570 mem_cgroup_threshold(mem);
1578 * Somemtimes we have to undo a charge we got by try_charge().
1579 * This function is for that and do uncharge, put css's refcnt.
1580 * gotten by try_charge().
1582 static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem,
1583 unsigned long count)
1585 if (!mem_cgroup_is_root(mem)) {
1586 res_counter_uncharge(&mem->res, PAGE_SIZE * count);
1587 if (do_swap_account)
1588 res_counter_uncharge(&mem->memsw, PAGE_SIZE * count);
1589 VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags));
1590 WARN_ON_ONCE(count > INT_MAX);
1591 __css_put(&mem->css, (int)count);
1593 /* we don't need css_put for root */
1596 static void mem_cgroup_cancel_charge(struct mem_cgroup *mem)
1598 __mem_cgroup_cancel_charge(mem, 1);
1602 * A helper function to get mem_cgroup from ID. must be called under
1603 * rcu_read_lock(). The caller must check css_is_removed() or some if
1604 * it's concern. (dropping refcnt from swap can be called against removed
1607 static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
1609 struct cgroup_subsys_state *css;
1611 /* ID 0 is unused ID */
1614 css = css_lookup(&mem_cgroup_subsys, id);
1617 return container_of(css, struct mem_cgroup, css);
1620 struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
1622 struct mem_cgroup *mem = NULL;
1623 struct page_cgroup *pc;
1627 VM_BUG_ON(!PageLocked(page));
1629 pc = lookup_page_cgroup(page);
1630 lock_page_cgroup(pc);
1631 if (PageCgroupUsed(pc)) {
1632 mem = pc->mem_cgroup;
1633 if (mem && !css_tryget(&mem->css))
1635 } else if (PageSwapCache(page)) {
1636 ent.val = page_private(page);
1637 id = lookup_swap_cgroup(ent);
1639 mem = mem_cgroup_lookup(id);
1640 if (mem && !css_tryget(&mem->css))
1644 unlock_page_cgroup(pc);
1649 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
1650 * USED state. If already USED, uncharge and return.
1653 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
1654 struct page_cgroup *pc,
1655 enum charge_type ctype)
1657 /* try_charge() can return NULL to *memcg, taking care of it. */
1661 lock_page_cgroup(pc);
1662 if (unlikely(PageCgroupUsed(pc))) {
1663 unlock_page_cgroup(pc);
1664 mem_cgroup_cancel_charge(mem);
1668 pc->mem_cgroup = mem;
1670 * We access a page_cgroup asynchronously without lock_page_cgroup().
1671 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
1672 * is accessed after testing USED bit. To make pc->mem_cgroup visible
1673 * before USED bit, we need memory barrier here.
1674 * See mem_cgroup_add_lru_list(), etc.
1678 case MEM_CGROUP_CHARGE_TYPE_CACHE:
1679 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
1680 SetPageCgroupCache(pc);
1681 SetPageCgroupUsed(pc);
1683 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1684 ClearPageCgroupCache(pc);
1685 SetPageCgroupUsed(pc);
1691 mem_cgroup_charge_statistics(mem, pc, true);
1693 unlock_page_cgroup(pc);
1697 * __mem_cgroup_move_account - move account of the page
1698 * @pc: page_cgroup of the page.
1699 * @from: mem_cgroup which the page is moved from.
1700 * @to: mem_cgroup which the page is moved to. @from != @to.
1701 * @uncharge: whether we should call uncharge and css_put against @from.
1703 * The caller must confirm following.
1704 * - page is not on LRU (isolate_page() is useful.)
1705 * - the pc is locked, used, and ->mem_cgroup points to @from.
1707 * This function doesn't do "charge" nor css_get to new cgroup. It should be
1708 * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
1709 * true, this function does "uncharge" from old cgroup, but it doesn't if
1710 * @uncharge is false, so a caller should do "uncharge".
1713 static void __mem_cgroup_move_account(struct page_cgroup *pc,
1714 struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge)
1718 VM_BUG_ON(from == to);
1719 VM_BUG_ON(PageLRU(pc->page));
1720 VM_BUG_ON(!PageCgroupLocked(pc));
1721 VM_BUG_ON(!PageCgroupUsed(pc));
1722 VM_BUG_ON(pc->mem_cgroup != from);
1725 if (page_mapped(page) && !PageAnon(page)) {
1726 /* Update mapped_file data for mem_cgroup */
1728 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
1729 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
1732 mem_cgroup_charge_statistics(from, pc, false);
1734 /* This is not "cancel", but cancel_charge does all we need. */
1735 mem_cgroup_cancel_charge(from);
1737 /* caller should have done css_get */
1738 pc->mem_cgroup = to;
1739 mem_cgroup_charge_statistics(to, pc, true);
1741 * We charges against "to" which may not have any tasks. Then, "to"
1742 * can be under rmdir(). But in current implementation, caller of
1743 * this function is just force_empty() and move charge, so it's
1744 * garanteed that "to" is never removed. So, we don't check rmdir
1750 * check whether the @pc is valid for moving account and call
1751 * __mem_cgroup_move_account()
1753 static int mem_cgroup_move_account(struct page_cgroup *pc,
1754 struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge)
1757 lock_page_cgroup(pc);
1758 if (PageCgroupUsed(pc) && pc->mem_cgroup == from) {
1759 __mem_cgroup_move_account(pc, from, to, uncharge);
1762 unlock_page_cgroup(pc);
1767 * move charges to its parent.
1770 static int mem_cgroup_move_parent(struct page_cgroup *pc,
1771 struct mem_cgroup *child,
1774 struct page *page = pc->page;
1775 struct cgroup *cg = child->css.cgroup;
1776 struct cgroup *pcg = cg->parent;
1777 struct mem_cgroup *parent;
1785 if (!get_page_unless_zero(page))
1787 if (isolate_lru_page(page))
1790 parent = mem_cgroup_from_cont(pcg);
1791 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false, page);
1795 ret = mem_cgroup_move_account(pc, child, parent, true);
1797 mem_cgroup_cancel_charge(parent);
1799 putback_lru_page(page);
1807 * Charge the memory controller for page usage.
1809 * 0 if the charge was successful
1810 * < 0 if the cgroup is over its limit
1812 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1813 gfp_t gfp_mask, enum charge_type ctype,
1814 struct mem_cgroup *memcg)
1816 struct mem_cgroup *mem;
1817 struct page_cgroup *pc;
1820 pc = lookup_page_cgroup(page);
1821 /* can happen at boot */
1827 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true, page);
1831 __mem_cgroup_commit_charge(mem, pc, ctype);
1835 int mem_cgroup_newpage_charge(struct page *page,
1836 struct mm_struct *mm, gfp_t gfp_mask)
1838 if (mem_cgroup_disabled())
1840 if (PageCompound(page))
1843 * If already mapped, we don't have to account.
1844 * If page cache, page->mapping has address_space.
1845 * But page->mapping may have out-of-use anon_vma pointer,
1846 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1849 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1853 return mem_cgroup_charge_common(page, mm, gfp_mask,
1854 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1858 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1859 enum charge_type ctype);
1861 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1864 struct mem_cgroup *mem = NULL;
1867 if (mem_cgroup_disabled())
1869 if (PageCompound(page))
1872 * Corner case handling. This is called from add_to_page_cache()
1873 * in usual. But some FS (shmem) precharges this page before calling it
1874 * and call add_to_page_cache() with GFP_NOWAIT.
1876 * For GFP_NOWAIT case, the page may be pre-charged before calling
1877 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1878 * charge twice. (It works but has to pay a bit larger cost.)
1879 * And when the page is SwapCache, it should take swap information
1880 * into account. This is under lock_page() now.
1882 if (!(gfp_mask & __GFP_WAIT)) {
1883 struct page_cgroup *pc;
1886 pc = lookup_page_cgroup(page);
1889 lock_page_cgroup(pc);
1890 if (PageCgroupUsed(pc)) {
1891 unlock_page_cgroup(pc);
1894 unlock_page_cgroup(pc);
1897 if (unlikely(!mm && !mem))
1900 if (page_is_file_cache(page))
1901 return mem_cgroup_charge_common(page, mm, gfp_mask,
1902 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1905 if (PageSwapCache(page)) {
1906 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
1908 __mem_cgroup_commit_charge_swapin(page, mem,
1909 MEM_CGROUP_CHARGE_TYPE_SHMEM);
1911 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1912 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1918 * While swap-in, try_charge -> commit or cancel, the page is locked.
1919 * And when try_charge() successfully returns, one refcnt to memcg without
1920 * struct page_cgroup is acquired. This refcnt will be consumed by
1921 * "commit()" or removed by "cancel()"
1923 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1925 gfp_t mask, struct mem_cgroup **ptr)
1927 struct mem_cgroup *mem;
1930 if (mem_cgroup_disabled())
1933 if (!do_swap_account)
1936 * A racing thread's fault, or swapoff, may have already updated
1937 * the pte, and even removed page from swap cache: in those cases
1938 * do_swap_page()'s pte_same() test will fail; but there's also a
1939 * KSM case which does need to charge the page.
1941 if (!PageSwapCache(page))
1943 mem = try_get_mem_cgroup_from_page(page);
1947 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, page);
1948 /* drop extra refcnt from tryget */
1954 return __mem_cgroup_try_charge(mm, mask, ptr, true, page);
1958 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1959 enum charge_type ctype)
1961 struct page_cgroup *pc;
1963 if (mem_cgroup_disabled())
1967 cgroup_exclude_rmdir(&ptr->css);
1968 pc = lookup_page_cgroup(page);
1969 mem_cgroup_lru_del_before_commit_swapcache(page);
1970 __mem_cgroup_commit_charge(ptr, pc, ctype);
1971 mem_cgroup_lru_add_after_commit_swapcache(page);
1973 * Now swap is on-memory. This means this page may be
1974 * counted both as mem and swap....double count.
1975 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1976 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1977 * may call delete_from_swap_cache() before reach here.
1979 if (do_swap_account && PageSwapCache(page)) {
1980 swp_entry_t ent = {.val = page_private(page)};
1982 struct mem_cgroup *memcg;
1984 id = swap_cgroup_record(ent, 0);
1986 memcg = mem_cgroup_lookup(id);
1989 * This recorded memcg can be obsolete one. So, avoid
1990 * calling css_tryget
1992 if (!mem_cgroup_is_root(memcg))
1993 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1994 mem_cgroup_swap_statistics(memcg, false);
1995 mem_cgroup_put(memcg);
2000 * At swapin, we may charge account against cgroup which has no tasks.
2001 * So, rmdir()->pre_destroy() can be called while we do this charge.
2002 * In that case, we need to call pre_destroy() again. check it here.
2004 cgroup_release_and_wakeup_rmdir(&ptr->css);
2007 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
2009 __mem_cgroup_commit_charge_swapin(page, ptr,
2010 MEM_CGROUP_CHARGE_TYPE_MAPPED);
2013 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
2015 if (mem_cgroup_disabled())
2019 mem_cgroup_cancel_charge(mem);
2023 __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype)
2025 struct memcg_batch_info *batch = NULL;
2026 bool uncharge_memsw = true;
2027 /* If swapout, usage of swap doesn't decrease */
2028 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2029 uncharge_memsw = false;
2031 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2032 * In those cases, all pages freed continously can be expected to be in
2033 * the same cgroup and we have chance to coalesce uncharges.
2034 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2035 * because we want to do uncharge as soon as possible.
2037 if (!current->memcg_batch.do_batch || test_thread_flag(TIF_MEMDIE))
2038 goto direct_uncharge;
2040 batch = ¤t->memcg_batch;
2042 * In usual, we do css_get() when we remember memcg pointer.
2043 * But in this case, we keep res->usage until end of a series of
2044 * uncharges. Then, it's ok to ignore memcg's refcnt.
2049 * In typical case, batch->memcg == mem. This means we can
2050 * merge a series of uncharges to an uncharge of res_counter.
2051 * If not, we uncharge res_counter ony by one.
2053 if (batch->memcg != mem)
2054 goto direct_uncharge;
2055 /* remember freed charge and uncharge it later */
2056 batch->bytes += PAGE_SIZE;
2058 batch->memsw_bytes += PAGE_SIZE;
2061 res_counter_uncharge(&mem->res, PAGE_SIZE);
2063 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
2068 * uncharge if !page_mapped(page)
2070 static struct mem_cgroup *
2071 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
2073 struct page_cgroup *pc;
2074 struct mem_cgroup *mem = NULL;
2075 struct mem_cgroup_per_zone *mz;
2077 if (mem_cgroup_disabled())
2080 if (PageSwapCache(page))
2084 * Check if our page_cgroup is valid
2086 pc = lookup_page_cgroup(page);
2087 if (unlikely(!pc || !PageCgroupUsed(pc)))
2090 lock_page_cgroup(pc);
2092 mem = pc->mem_cgroup;
2094 if (!PageCgroupUsed(pc))
2098 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2099 case MEM_CGROUP_CHARGE_TYPE_DROP:
2100 if (page_mapped(page))
2103 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
2104 if (!PageAnon(page)) { /* Shared memory */
2105 if (page->mapping && !page_is_file_cache(page))
2107 } else if (page_mapped(page)) /* Anon */
2114 if (!mem_cgroup_is_root(mem))
2115 __do_uncharge(mem, ctype);
2116 if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2117 mem_cgroup_swap_statistics(mem, true);
2118 mem_cgroup_charge_statistics(mem, pc, false);
2120 ClearPageCgroupUsed(pc);
2122 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2123 * freed from LRU. This is safe because uncharged page is expected not
2124 * to be reused (freed soon). Exception is SwapCache, it's handled by
2125 * special functions.
2128 mz = page_cgroup_zoneinfo(pc);
2129 unlock_page_cgroup(pc);
2131 if (mem_cgroup_soft_limit_check(mem))
2132 mem_cgroup_update_tree(mem, page);
2133 if (mem_cgroup_threshold_check(mem))
2134 mem_cgroup_threshold(mem);
2135 /* at swapout, this memcg will be accessed to record to swap */
2136 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2142 unlock_page_cgroup(pc);
2146 void mem_cgroup_uncharge_page(struct page *page)
2149 if (page_mapped(page))
2151 if (page->mapping && !PageAnon(page))
2153 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
2156 void mem_cgroup_uncharge_cache_page(struct page *page)
2158 VM_BUG_ON(page_mapped(page));
2159 VM_BUG_ON(page->mapping);
2160 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
2164 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2165 * In that cases, pages are freed continuously and we can expect pages
2166 * are in the same memcg. All these calls itself limits the number of
2167 * pages freed at once, then uncharge_start/end() is called properly.
2168 * This may be called prural(2) times in a context,
2171 void mem_cgroup_uncharge_start(void)
2173 current->memcg_batch.do_batch++;
2174 /* We can do nest. */
2175 if (current->memcg_batch.do_batch == 1) {
2176 current->memcg_batch.memcg = NULL;
2177 current->memcg_batch.bytes = 0;
2178 current->memcg_batch.memsw_bytes = 0;
2182 void mem_cgroup_uncharge_end(void)
2184 struct memcg_batch_info *batch = ¤t->memcg_batch;
2186 if (!batch->do_batch)
2190 if (batch->do_batch) /* If stacked, do nothing. */
2196 * This "batch->memcg" is valid without any css_get/put etc...
2197 * bacause we hide charges behind us.
2200 res_counter_uncharge(&batch->memcg->res, batch->bytes);
2201 if (batch->memsw_bytes)
2202 res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes);
2203 /* forget this pointer (for sanity check) */
2204 batch->memcg = NULL;
2209 * called after __delete_from_swap_cache() and drop "page" account.
2210 * memcg information is recorded to swap_cgroup of "ent"
2213 mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
2215 struct mem_cgroup *memcg;
2216 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
2218 if (!swapout) /* this was a swap cache but the swap is unused ! */
2219 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
2221 memcg = __mem_cgroup_uncharge_common(page, ctype);
2223 /* record memcg information */
2224 if (do_swap_account && swapout && memcg) {
2225 swap_cgroup_record(ent, css_id(&memcg->css));
2226 mem_cgroup_get(memcg);
2228 if (swapout && memcg)
2229 css_put(&memcg->css);
2233 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2235 * called from swap_entry_free(). remove record in swap_cgroup and
2236 * uncharge "memsw" account.
2238 void mem_cgroup_uncharge_swap(swp_entry_t ent)
2240 struct mem_cgroup *memcg;
2243 if (!do_swap_account)
2246 id = swap_cgroup_record(ent, 0);
2248 memcg = mem_cgroup_lookup(id);
2251 * We uncharge this because swap is freed.
2252 * This memcg can be obsolete one. We avoid calling css_tryget
2254 if (!mem_cgroup_is_root(memcg))
2255 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
2256 mem_cgroup_swap_statistics(memcg, false);
2257 mem_cgroup_put(memcg);
2263 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2264 * @entry: swap entry to be moved
2265 * @from: mem_cgroup which the entry is moved from
2266 * @to: mem_cgroup which the entry is moved to
2267 * @need_fixup: whether we should fixup res_counters and refcounts.
2269 * It succeeds only when the swap_cgroup's record for this entry is the same
2270 * as the mem_cgroup's id of @from.
2272 * Returns 0 on success, -EINVAL on failure.
2274 * The caller must have charged to @to, IOW, called res_counter_charge() about
2275 * both res and memsw, and called css_get().
2277 static int mem_cgroup_move_swap_account(swp_entry_t entry,
2278 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
2280 unsigned short old_id, new_id;
2282 old_id = css_id(&from->css);
2283 new_id = css_id(&to->css);
2285 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
2286 mem_cgroup_swap_statistics(from, false);
2287 mem_cgroup_swap_statistics(to, true);
2289 * This function is only called from task migration context now.
2290 * It postpones res_counter and refcount handling till the end
2291 * of task migration(mem_cgroup_clear_mc()) for performance
2292 * improvement. But we cannot postpone mem_cgroup_get(to)
2293 * because if the process that has been moved to @to does
2294 * swap-in, the refcount of @to might be decreased to 0.
2298 if (!mem_cgroup_is_root(from))
2299 res_counter_uncharge(&from->memsw, PAGE_SIZE);
2300 mem_cgroup_put(from);
2302 * we charged both to->res and to->memsw, so we should
2305 if (!mem_cgroup_is_root(to))
2306 res_counter_uncharge(&to->res, PAGE_SIZE);
2314 static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
2315 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
2322 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2325 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
2327 struct page_cgroup *pc;
2328 struct mem_cgroup *mem = NULL;
2331 if (mem_cgroup_disabled())
2334 pc = lookup_page_cgroup(page);
2335 lock_page_cgroup(pc);
2336 if (PageCgroupUsed(pc)) {
2337 mem = pc->mem_cgroup;
2340 unlock_page_cgroup(pc);
2343 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false,
2351 /* remove redundant charge if migration failed*/
2352 void mem_cgroup_end_migration(struct mem_cgroup *mem,
2353 struct page *oldpage, struct page *newpage)
2355 struct page *target, *unused;
2356 struct page_cgroup *pc;
2357 enum charge_type ctype;
2361 cgroup_exclude_rmdir(&mem->css);
2362 /* at migration success, oldpage->mapping is NULL. */
2363 if (oldpage->mapping) {
2371 if (PageAnon(target))
2372 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
2373 else if (page_is_file_cache(target))
2374 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
2376 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
2378 /* unused page is not on radix-tree now. */
2380 __mem_cgroup_uncharge_common(unused, ctype);
2382 pc = lookup_page_cgroup(target);
2384 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
2385 * So, double-counting is effectively avoided.
2387 __mem_cgroup_commit_charge(mem, pc, ctype);
2390 * Both of oldpage and newpage are still under lock_page().
2391 * Then, we don't have to care about race in radix-tree.
2392 * But we have to be careful that this page is unmapped or not.
2394 * There is a case for !page_mapped(). At the start of
2395 * migration, oldpage was mapped. But now, it's zapped.
2396 * But we know *target* page is not freed/reused under us.
2397 * mem_cgroup_uncharge_page() does all necessary checks.
2399 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
2400 mem_cgroup_uncharge_page(target);
2402 * At migration, we may charge account against cgroup which has no tasks
2403 * So, rmdir()->pre_destroy() can be called while we do this charge.
2404 * In that case, we need to call pre_destroy() again. check it here.
2406 cgroup_release_and_wakeup_rmdir(&mem->css);
2410 * A call to try to shrink memory usage on charge failure at shmem's swapin.
2411 * Calling hierarchical_reclaim is not enough because we should update
2412 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
2413 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
2414 * not from the memcg which this page would be charged to.
2415 * try_charge_swapin does all of these works properly.
2417 int mem_cgroup_shmem_charge_fallback(struct page *page,
2418 struct mm_struct *mm,
2421 struct mem_cgroup *mem = NULL;
2424 if (mem_cgroup_disabled())
2427 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2429 mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
2434 static DEFINE_MUTEX(set_limit_mutex);
2436 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
2437 unsigned long long val)
2442 int children = mem_cgroup_count_children(memcg);
2443 u64 curusage, oldusage;
2446 * For keeping hierarchical_reclaim simple, how long we should retry
2447 * is depends on callers. We set our retry-count to be function
2448 * of # of children which we should visit in this loop.
2450 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
2452 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2454 while (retry_count) {
2455 if (signal_pending(current)) {
2460 * Rather than hide all in some function, I do this in
2461 * open coded manner. You see what this really does.
2462 * We have to guarantee mem->res.limit < mem->memsw.limit.
2464 mutex_lock(&set_limit_mutex);
2465 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2466 if (memswlimit < val) {
2468 mutex_unlock(&set_limit_mutex);
2471 ret = res_counter_set_limit(&memcg->res, val);
2473 if (memswlimit == val)
2474 memcg->memsw_is_minimum = true;
2476 memcg->memsw_is_minimum = false;
2478 mutex_unlock(&set_limit_mutex);
2483 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
2484 MEM_CGROUP_RECLAIM_SHRINK);
2485 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2486 /* Usage is reduced ? */
2487 if (curusage >= oldusage)
2490 oldusage = curusage;
2496 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2497 unsigned long long val)
2500 u64 memlimit, oldusage, curusage;
2501 int children = mem_cgroup_count_children(memcg);
2504 /* see mem_cgroup_resize_res_limit */
2505 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
2506 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2507 while (retry_count) {
2508 if (signal_pending(current)) {
2513 * Rather than hide all in some function, I do this in
2514 * open coded manner. You see what this really does.
2515 * We have to guarantee mem->res.limit < mem->memsw.limit.
2517 mutex_lock(&set_limit_mutex);
2518 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2519 if (memlimit > val) {
2521 mutex_unlock(&set_limit_mutex);
2524 ret = res_counter_set_limit(&memcg->memsw, val);
2526 if (memlimit == val)
2527 memcg->memsw_is_minimum = true;
2529 memcg->memsw_is_minimum = false;
2531 mutex_unlock(&set_limit_mutex);
2536 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
2537 MEM_CGROUP_RECLAIM_NOSWAP |
2538 MEM_CGROUP_RECLAIM_SHRINK);
2539 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2540 /* Usage is reduced ? */
2541 if (curusage >= oldusage)
2544 oldusage = curusage;
2549 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2550 gfp_t gfp_mask, int nid,
2553 unsigned long nr_reclaimed = 0;
2554 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2555 unsigned long reclaimed;
2557 struct mem_cgroup_tree_per_zone *mctz;
2558 unsigned long long excess;
2563 mctz = soft_limit_tree_node_zone(nid, zid);
2565 * This loop can run a while, specially if mem_cgroup's continuously
2566 * keep exceeding their soft limit and putting the system under
2573 mz = mem_cgroup_largest_soft_limit_node(mctz);
2577 reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone,
2579 MEM_CGROUP_RECLAIM_SOFT);
2580 nr_reclaimed += reclaimed;
2581 spin_lock(&mctz->lock);
2584 * If we failed to reclaim anything from this memory cgroup
2585 * it is time to move on to the next cgroup
2591 * Loop until we find yet another one.
2593 * By the time we get the soft_limit lock
2594 * again, someone might have aded the
2595 * group back on the RB tree. Iterate to
2596 * make sure we get a different mem.
2597 * mem_cgroup_largest_soft_limit_node returns
2598 * NULL if no other cgroup is present on
2602 __mem_cgroup_largest_soft_limit_node(mctz);
2603 if (next_mz == mz) {
2604 css_put(&next_mz->mem->css);
2606 } else /* next_mz == NULL or other memcg */
2610 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
2611 excess = res_counter_soft_limit_excess(&mz->mem->res);
2613 * One school of thought says that we should not add
2614 * back the node to the tree if reclaim returns 0.
2615 * But our reclaim could return 0, simply because due
2616 * to priority we are exposing a smaller subset of
2617 * memory to reclaim from. Consider this as a longer
2620 /* If excess == 0, no tree ops */
2621 __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
2622 spin_unlock(&mctz->lock);
2623 css_put(&mz->mem->css);
2626 * Could not reclaim anything and there are no more
2627 * mem cgroups to try or we seem to be looping without
2628 * reclaiming anything.
2630 if (!nr_reclaimed &&
2632 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2634 } while (!nr_reclaimed);
2636 css_put(&next_mz->mem->css);
2637 return nr_reclaimed;
2641 * This routine traverse page_cgroup in given list and drop them all.
2642 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
2644 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
2645 int node, int zid, enum lru_list lru)
2648 struct mem_cgroup_per_zone *mz;
2649 struct page_cgroup *pc, *busy;
2650 unsigned long flags, loop;
2651 struct list_head *list;
2654 zone = &NODE_DATA(node)->node_zones[zid];
2655 mz = mem_cgroup_zoneinfo(mem, node, zid);
2656 list = &mz->lists[lru];
2658 loop = MEM_CGROUP_ZSTAT(mz, lru);
2659 /* give some margin against EBUSY etc...*/
2664 spin_lock_irqsave(&zone->lru_lock, flags);
2665 if (list_empty(list)) {
2666 spin_unlock_irqrestore(&zone->lru_lock, flags);
2669 pc = list_entry(list->prev, struct page_cgroup, lru);
2671 list_move(&pc->lru, list);
2673 spin_unlock_irqrestore(&zone->lru_lock, flags);
2676 spin_unlock_irqrestore(&zone->lru_lock, flags);
2678 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
2682 if (ret == -EBUSY || ret == -EINVAL) {
2683 /* found lock contention or "pc" is obsolete. */
2690 if (!ret && !list_empty(list))
2696 * make mem_cgroup's charge to be 0 if there is no task.
2697 * This enables deleting this mem_cgroup.
2699 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
2702 int node, zid, shrink;
2703 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
2704 struct cgroup *cgrp = mem->css.cgroup;
2709 /* should free all ? */
2715 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
2718 if (signal_pending(current))
2720 /* This is for making all *used* pages to be on LRU. */
2721 lru_add_drain_all();
2722 drain_all_stock_sync();
2724 for_each_node_state(node, N_HIGH_MEMORY) {
2725 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
2728 ret = mem_cgroup_force_empty_list(mem,
2737 /* it seems parent cgroup doesn't have enough mem */
2741 /* "ret" should also be checked to ensure all lists are empty. */
2742 } while (mem->res.usage > 0 || ret);
2748 /* returns EBUSY if there is a task or if we come here twice. */
2749 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
2753 /* we call try-to-free pages for make this cgroup empty */
2754 lru_add_drain_all();
2755 /* try to free all pages in this cgroup */
2757 while (nr_retries && mem->res.usage > 0) {
2760 if (signal_pending(current)) {
2764 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
2765 false, get_swappiness(mem));
2768 /* maybe some writeback is necessary */
2769 congestion_wait(BLK_RW_ASYNC, HZ/10);
2774 /* try move_account...there may be some *locked* pages. */
2778 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
2780 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
2784 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
2786 return mem_cgroup_from_cont(cont)->use_hierarchy;
2789 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
2793 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2794 struct cgroup *parent = cont->parent;
2795 struct mem_cgroup *parent_mem = NULL;
2798 parent_mem = mem_cgroup_from_cont(parent);
2802 * If parent's use_hierarchy is set, we can't make any modifications
2803 * in the child subtrees. If it is unset, then the change can
2804 * occur, provided the current cgroup has no children.
2806 * For the root cgroup, parent_mem is NULL, we allow value to be
2807 * set if there are no children.
2809 if ((!parent_mem || !parent_mem->use_hierarchy) &&
2810 (val == 1 || val == 0)) {
2811 if (list_empty(&cont->children))
2812 mem->use_hierarchy = val;
2822 struct mem_cgroup_idx_data {
2824 enum mem_cgroup_stat_index idx;
2828 mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data)
2830 struct mem_cgroup_idx_data *d = data;
2831 d->val += mem_cgroup_read_stat(mem, d->idx);
2836 mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem,
2837 enum mem_cgroup_stat_index idx, s64 *val)
2839 struct mem_cgroup_idx_data d;
2842 mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat);
2846 static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap)
2850 if (!mem_cgroup_is_root(mem)) {
2852 return res_counter_read_u64(&mem->res, RES_USAGE);
2854 return res_counter_read_u64(&mem->memsw, RES_USAGE);
2857 mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE, &idx_val);
2859 mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS, &idx_val);
2863 mem_cgroup_get_recursive_idx_stat(mem,
2864 MEM_CGROUP_STAT_SWAPOUT, &idx_val);
2868 return val << PAGE_SHIFT;
2871 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
2873 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2877 type = MEMFILE_TYPE(cft->private);
2878 name = MEMFILE_ATTR(cft->private);
2881 if (name == RES_USAGE)
2882 val = mem_cgroup_usage(mem, false);
2884 val = res_counter_read_u64(&mem->res, name);
2887 if (name == RES_USAGE)
2888 val = mem_cgroup_usage(mem, true);
2890 val = res_counter_read_u64(&mem->memsw, name);
2899 * The user of this function is...
2902 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
2905 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
2907 unsigned long long val;
2910 type = MEMFILE_TYPE(cft->private);
2911 name = MEMFILE_ATTR(cft->private);
2914 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2918 /* This function does all necessary parse...reuse it */
2919 ret = res_counter_memparse_write_strategy(buffer, &val);
2923 ret = mem_cgroup_resize_limit(memcg, val);
2925 ret = mem_cgroup_resize_memsw_limit(memcg, val);
2927 case RES_SOFT_LIMIT:
2928 ret = res_counter_memparse_write_strategy(buffer, &val);
2932 * For memsw, soft limits are hard to implement in terms
2933 * of semantics, for now, we support soft limits for
2934 * control without swap
2937 ret = res_counter_set_soft_limit(&memcg->res, val);
2942 ret = -EINVAL; /* should be BUG() ? */
2948 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
2949 unsigned long long *mem_limit, unsigned long long *memsw_limit)
2951 struct cgroup *cgroup;
2952 unsigned long long min_limit, min_memsw_limit, tmp;
2954 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2955 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2956 cgroup = memcg->css.cgroup;
2957 if (!memcg->use_hierarchy)
2960 while (cgroup->parent) {
2961 cgroup = cgroup->parent;
2962 memcg = mem_cgroup_from_cont(cgroup);
2963 if (!memcg->use_hierarchy)
2965 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
2966 min_limit = min(min_limit, tmp);
2967 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2968 min_memsw_limit = min(min_memsw_limit, tmp);
2971 *mem_limit = min_limit;
2972 *memsw_limit = min_memsw_limit;
2976 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
2978 struct mem_cgroup *mem;
2981 mem = mem_cgroup_from_cont(cont);
2982 type = MEMFILE_TYPE(event);
2983 name = MEMFILE_ATTR(event);
2987 res_counter_reset_max(&mem->res);
2989 res_counter_reset_max(&mem->memsw);
2993 res_counter_reset_failcnt(&mem->res);
2995 res_counter_reset_failcnt(&mem->memsw);
3002 static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
3005 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
3009 static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3010 struct cftype *cft, u64 val)
3012 struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
3014 if (val >= (1 << NR_MOVE_TYPE))
3017 * We check this value several times in both in can_attach() and
3018 * attach(), so we need cgroup lock to prevent this value from being
3022 mem->move_charge_at_immigrate = val;
3028 static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3029 struct cftype *cft, u64 val)
3036 /* For read statistics */
3052 struct mcs_total_stat {
3053 s64 stat[NR_MCS_STAT];
3059 } memcg_stat_strings[NR_MCS_STAT] = {
3060 {"cache", "total_cache"},
3061 {"rss", "total_rss"},
3062 {"mapped_file", "total_mapped_file"},
3063 {"pgpgin", "total_pgpgin"},
3064 {"pgpgout", "total_pgpgout"},
3065 {"swap", "total_swap"},
3066 {"inactive_anon", "total_inactive_anon"},
3067 {"active_anon", "total_active_anon"},
3068 {"inactive_file", "total_inactive_file"},
3069 {"active_file", "total_active_file"},
3070 {"unevictable", "total_unevictable"}
3074 static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
3076 struct mcs_total_stat *s = data;
3080 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
3081 s->stat[MCS_CACHE] += val * PAGE_SIZE;
3082 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
3083 s->stat[MCS_RSS] += val * PAGE_SIZE;
3084 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED);
3085 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
3086 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT);
3087 s->stat[MCS_PGPGIN] += val;
3088 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT);
3089 s->stat[MCS_PGPGOUT] += val;
3090 if (do_swap_account) {
3091 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT);
3092 s->stat[MCS_SWAP] += val * PAGE_SIZE;
3096 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
3097 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
3098 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
3099 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
3100 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
3101 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
3102 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
3103 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
3104 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
3105 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
3110 mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
3112 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
3115 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
3116 struct cgroup_map_cb *cb)
3118 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
3119 struct mcs_total_stat mystat;
3122 memset(&mystat, 0, sizeof(mystat));
3123 mem_cgroup_get_local_stat(mem_cont, &mystat);
3125 for (i = 0; i < NR_MCS_STAT; i++) {
3126 if (i == MCS_SWAP && !do_swap_account)
3128 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
3131 /* Hierarchical information */
3133 unsigned long long limit, memsw_limit;
3134 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
3135 cb->fill(cb, "hierarchical_memory_limit", limit);
3136 if (do_swap_account)
3137 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
3140 memset(&mystat, 0, sizeof(mystat));
3141 mem_cgroup_get_total_stat(mem_cont, &mystat);
3142 for (i = 0; i < NR_MCS_STAT; i++) {
3143 if (i == MCS_SWAP && !do_swap_account)
3145 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
3148 #ifdef CONFIG_DEBUG_VM
3149 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
3153 struct mem_cgroup_per_zone *mz;
3154 unsigned long recent_rotated[2] = {0, 0};
3155 unsigned long recent_scanned[2] = {0, 0};
3157 for_each_online_node(nid)
3158 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
3159 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
3161 recent_rotated[0] +=
3162 mz->reclaim_stat.recent_rotated[0];
3163 recent_rotated[1] +=
3164 mz->reclaim_stat.recent_rotated[1];
3165 recent_scanned[0] +=
3166 mz->reclaim_stat.recent_scanned[0];
3167 recent_scanned[1] +=
3168 mz->reclaim_stat.recent_scanned[1];
3170 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
3171 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
3172 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
3173 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
3180 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
3182 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3184 return get_swappiness(memcg);
3187 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
3190 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3191 struct mem_cgroup *parent;
3196 if (cgrp->parent == NULL)
3199 parent = mem_cgroup_from_cont(cgrp->parent);
3203 /* If under hierarchy, only empty-root can set this value */
3204 if ((parent->use_hierarchy) ||
3205 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
3210 spin_lock(&memcg->reclaim_param_lock);
3211 memcg->swappiness = val;
3212 spin_unlock(&memcg->reclaim_param_lock);
3219 static bool mem_cgroup_threshold_check(struct mem_cgroup *mem)
3224 val = this_cpu_read(mem->stat->count[MEM_CGROUP_STAT_THRESHOLDS]);
3225 if (unlikely(val < 0)) {
3226 this_cpu_write(mem->stat->count[MEM_CGROUP_STAT_THRESHOLDS],
3227 THRESHOLDS_EVENTS_THRESH);
3233 static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3235 struct mem_cgroup_threshold_ary *t;
3241 t = rcu_dereference(memcg->thresholds);
3243 t = rcu_dereference(memcg->memsw_thresholds);
3248 usage = mem_cgroup_usage(memcg, swap);
3251 * current_threshold points to threshold just below usage.
3252 * If it's not true, a threshold was crossed after last
3253 * call of __mem_cgroup_threshold().
3255 i = atomic_read(&t->current_threshold);
3258 * Iterate backward over array of thresholds starting from
3259 * current_threshold and check if a threshold is crossed.
3260 * If none of thresholds below usage is crossed, we read
3261 * only one element of the array here.
3263 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3264 eventfd_signal(t->entries[i].eventfd, 1);
3266 /* i = current_threshold + 1 */
3270 * Iterate forward over array of thresholds starting from
3271 * current_threshold+1 and check if a threshold is crossed.
3272 * If none of thresholds above usage is crossed, we read
3273 * only one element of the array here.
3275 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3276 eventfd_signal(t->entries[i].eventfd, 1);
3278 /* Update current_threshold */
3279 atomic_set(&t->current_threshold, i - 1);
3284 static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3286 __mem_cgroup_threshold(memcg, false);
3287 if (do_swap_account)
3288 __mem_cgroup_threshold(memcg, true);
3291 static int compare_thresholds(const void *a, const void *b)
3293 const struct mem_cgroup_threshold *_a = a;
3294 const struct mem_cgroup_threshold *_b = b;
3296 return _a->threshold - _b->threshold;
3299 static int mem_cgroup_register_event(struct cgroup *cgrp, struct cftype *cft,
3300 struct eventfd_ctx *eventfd, const char *args)
3302 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3303 struct mem_cgroup_threshold_ary *thresholds, *thresholds_new;
3304 int type = MEMFILE_TYPE(cft->private);
3305 u64 threshold, usage;
3309 ret = res_counter_memparse_write_strategy(args, &threshold);
3313 mutex_lock(&memcg->thresholds_lock);
3315 thresholds = memcg->thresholds;
3316 else if (type == _MEMSWAP)
3317 thresholds = memcg->memsw_thresholds;
3321 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
3323 /* Check if a threshold crossed before adding a new one */
3325 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3328 size = thresholds->size + 1;
3332 /* Allocate memory for new array of thresholds */
3333 thresholds_new = kmalloc(sizeof(*thresholds_new) +
3334 size * sizeof(struct mem_cgroup_threshold),
3336 if (!thresholds_new) {
3340 thresholds_new->size = size;
3342 /* Copy thresholds (if any) to new array */
3344 memcpy(thresholds_new->entries, thresholds->entries,
3346 sizeof(struct mem_cgroup_threshold));
3347 /* Add new threshold */
3348 thresholds_new->entries[size - 1].eventfd = eventfd;
3349 thresholds_new->entries[size - 1].threshold = threshold;
3351 /* Sort thresholds. Registering of new threshold isn't time-critical */
3352 sort(thresholds_new->entries, size,
3353 sizeof(struct mem_cgroup_threshold),
3354 compare_thresholds, NULL);
3356 /* Find current threshold */
3357 atomic_set(&thresholds_new->current_threshold, -1);
3358 for (i = 0; i < size; i++) {
3359 if (thresholds_new->entries[i].threshold < usage) {
3361 * thresholds_new->current_threshold will not be used
3362 * until rcu_assign_pointer(), so it's safe to increment
3365 atomic_inc(&thresholds_new->current_threshold);
3370 * We need to increment refcnt to be sure that all thresholds
3371 * will be unregistered before calling __mem_cgroup_free()
3373 mem_cgroup_get(memcg);
3376 rcu_assign_pointer(memcg->thresholds, thresholds_new);
3378 rcu_assign_pointer(memcg->memsw_thresholds, thresholds_new);
3380 /* To be sure that nobody uses thresholds before freeing it */
3385 mutex_unlock(&memcg->thresholds_lock);
3390 static int mem_cgroup_unregister_event(struct cgroup *cgrp, struct cftype *cft,
3391 struct eventfd_ctx *eventfd)
3393 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3394 struct mem_cgroup_threshold_ary *thresholds, *thresholds_new;
3395 int type = MEMFILE_TYPE(cft->private);
3400 mutex_lock(&memcg->thresholds_lock);
3402 thresholds = memcg->thresholds;
3403 else if (type == _MEMSWAP)
3404 thresholds = memcg->memsw_thresholds;
3409 * Something went wrong if we trying to unregister a threshold
3410 * if we don't have thresholds
3412 BUG_ON(!thresholds);
3414 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
3416 /* Check if a threshold crossed before removing */
3417 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3419 /* Calculate new number of threshold */
3420 for (i = 0; i < thresholds->size; i++) {
3421 if (thresholds->entries[i].eventfd != eventfd)
3425 /* Set thresholds array to NULL if we don't have thresholds */
3427 thresholds_new = NULL;
3431 /* Allocate memory for new array of thresholds */
3432 thresholds_new = kmalloc(sizeof(*thresholds_new) +
3433 size * sizeof(struct mem_cgroup_threshold),
3435 if (!thresholds_new) {
3439 thresholds_new->size = size;
3441 /* Copy thresholds and find current threshold */
3442 atomic_set(&thresholds_new->current_threshold, -1);
3443 for (i = 0, j = 0; i < thresholds->size; i++) {
3444 if (thresholds->entries[i].eventfd == eventfd)
3447 thresholds_new->entries[j] = thresholds->entries[i];
3448 if (thresholds_new->entries[j].threshold < usage) {
3450 * thresholds_new->current_threshold will not be used
3451 * until rcu_assign_pointer(), so it's safe to increment
3454 atomic_inc(&thresholds_new->current_threshold);
3461 rcu_assign_pointer(memcg->thresholds, thresholds_new);
3463 rcu_assign_pointer(memcg->memsw_thresholds, thresholds_new);
3465 /* To be sure that nobody uses thresholds before freeing it */
3468 for (i = 0; i < thresholds->size - size; i++)
3469 mem_cgroup_put(memcg);
3473 mutex_unlock(&memcg->thresholds_lock);
3478 static struct cftype mem_cgroup_files[] = {
3480 .name = "usage_in_bytes",
3481 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
3482 .read_u64 = mem_cgroup_read,
3483 .register_event = mem_cgroup_register_event,
3484 .unregister_event = mem_cgroup_unregister_event,
3487 .name = "max_usage_in_bytes",
3488 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
3489 .trigger = mem_cgroup_reset,
3490 .read_u64 = mem_cgroup_read,
3493 .name = "limit_in_bytes",
3494 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
3495 .write_string = mem_cgroup_write,
3496 .read_u64 = mem_cgroup_read,
3499 .name = "soft_limit_in_bytes",
3500 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
3501 .write_string = mem_cgroup_write,
3502 .read_u64 = mem_cgroup_read,
3506 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
3507 .trigger = mem_cgroup_reset,
3508 .read_u64 = mem_cgroup_read,
3512 .read_map = mem_control_stat_show,
3515 .name = "force_empty",
3516 .trigger = mem_cgroup_force_empty_write,
3519 .name = "use_hierarchy",
3520 .write_u64 = mem_cgroup_hierarchy_write,
3521 .read_u64 = mem_cgroup_hierarchy_read,
3524 .name = "swappiness",
3525 .read_u64 = mem_cgroup_swappiness_read,
3526 .write_u64 = mem_cgroup_swappiness_write,
3529 .name = "move_charge_at_immigrate",
3530 .read_u64 = mem_cgroup_move_charge_read,
3531 .write_u64 = mem_cgroup_move_charge_write,
3535 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3536 static struct cftype memsw_cgroup_files[] = {
3538 .name = "memsw.usage_in_bytes",
3539 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
3540 .read_u64 = mem_cgroup_read,
3541 .register_event = mem_cgroup_register_event,
3542 .unregister_event = mem_cgroup_unregister_event,
3545 .name = "memsw.max_usage_in_bytes",
3546 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
3547 .trigger = mem_cgroup_reset,
3548 .read_u64 = mem_cgroup_read,
3551 .name = "memsw.limit_in_bytes",
3552 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
3553 .write_string = mem_cgroup_write,
3554 .read_u64 = mem_cgroup_read,
3557 .name = "memsw.failcnt",
3558 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
3559 .trigger = mem_cgroup_reset,
3560 .read_u64 = mem_cgroup_read,
3564 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
3566 if (!do_swap_account)
3568 return cgroup_add_files(cont, ss, memsw_cgroup_files,
3569 ARRAY_SIZE(memsw_cgroup_files));
3572 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
3578 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
3580 struct mem_cgroup_per_node *pn;
3581 struct mem_cgroup_per_zone *mz;
3583 int zone, tmp = node;
3585 * This routine is called against possible nodes.
3586 * But it's BUG to call kmalloc() against offline node.
3588 * TODO: this routine can waste much memory for nodes which will
3589 * never be onlined. It's better to use memory hotplug callback
3592 if (!node_state(node, N_NORMAL_MEMORY))
3594 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
3598 mem->info.nodeinfo[node] = pn;
3599 memset(pn, 0, sizeof(*pn));
3601 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
3602 mz = &pn->zoneinfo[zone];
3604 INIT_LIST_HEAD(&mz->lists[l]);
3605 mz->usage_in_excess = 0;
3606 mz->on_tree = false;
3612 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
3614 kfree(mem->info.nodeinfo[node]);
3617 static struct mem_cgroup *mem_cgroup_alloc(void)
3619 struct mem_cgroup *mem;
3620 int size = sizeof(struct mem_cgroup);
3622 /* Can be very big if MAX_NUMNODES is very big */
3623 if (size < PAGE_SIZE)
3624 mem = kmalloc(size, GFP_KERNEL);
3626 mem = vmalloc(size);
3629 memset(mem, 0, size);
3630 mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
3632 if (size < PAGE_SIZE)
3642 * At destroying mem_cgroup, references from swap_cgroup can remain.
3643 * (scanning all at force_empty is too costly...)
3645 * Instead of clearing all references at force_empty, we remember
3646 * the number of reference from swap_cgroup and free mem_cgroup when
3647 * it goes down to 0.
3649 * Removal of cgroup itself succeeds regardless of refs from swap.
3652 static void __mem_cgroup_free(struct mem_cgroup *mem)
3656 mem_cgroup_remove_from_trees(mem);
3657 free_css_id(&mem_cgroup_subsys, &mem->css);
3659 for_each_node_state(node, N_POSSIBLE)
3660 free_mem_cgroup_per_zone_info(mem, node);
3662 free_percpu(mem->stat);
3663 if (sizeof(struct mem_cgroup) < PAGE_SIZE)
3669 static void mem_cgroup_get(struct mem_cgroup *mem)
3671 atomic_inc(&mem->refcnt);
3674 static void __mem_cgroup_put(struct mem_cgroup *mem, int count)
3676 if (atomic_sub_and_test(count, &mem->refcnt)) {
3677 struct mem_cgroup *parent = parent_mem_cgroup(mem);
3678 __mem_cgroup_free(mem);
3680 mem_cgroup_put(parent);
3684 static void mem_cgroup_put(struct mem_cgroup *mem)
3686 __mem_cgroup_put(mem, 1);
3690 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
3692 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
3694 if (!mem->res.parent)
3696 return mem_cgroup_from_res_counter(mem->res.parent, res);
3699 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3700 static void __init enable_swap_cgroup(void)
3702 if (!mem_cgroup_disabled() && really_do_swap_account)
3703 do_swap_account = 1;
3706 static void __init enable_swap_cgroup(void)
3711 static int mem_cgroup_soft_limit_tree_init(void)
3713 struct mem_cgroup_tree_per_node *rtpn;
3714 struct mem_cgroup_tree_per_zone *rtpz;
3715 int tmp, node, zone;
3717 for_each_node_state(node, N_POSSIBLE) {
3719 if (!node_state(node, N_NORMAL_MEMORY))
3721 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
3725 soft_limit_tree.rb_tree_per_node[node] = rtpn;
3727 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
3728 rtpz = &rtpn->rb_tree_per_zone[zone];
3729 rtpz->rb_root = RB_ROOT;
3730 spin_lock_init(&rtpz->lock);
3736 static struct cgroup_subsys_state * __ref
3737 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
3739 struct mem_cgroup *mem, *parent;
3740 long error = -ENOMEM;
3743 mem = mem_cgroup_alloc();
3745 return ERR_PTR(error);
3747 for_each_node_state(node, N_POSSIBLE)
3748 if (alloc_mem_cgroup_per_zone_info(mem, node))
3752 if (cont->parent == NULL) {
3754 enable_swap_cgroup();
3756 root_mem_cgroup = mem;
3757 if (mem_cgroup_soft_limit_tree_init())
3759 for_each_possible_cpu(cpu) {
3760 struct memcg_stock_pcp *stock =
3761 &per_cpu(memcg_stock, cpu);
3762 INIT_WORK(&stock->work, drain_local_stock);
3764 hotcpu_notifier(memcg_stock_cpu_callback, 0);
3766 parent = mem_cgroup_from_cont(cont->parent);
3767 mem->use_hierarchy = parent->use_hierarchy;
3770 if (parent && parent->use_hierarchy) {
3771 res_counter_init(&mem->res, &parent->res);
3772 res_counter_init(&mem->memsw, &parent->memsw);
3774 * We increment refcnt of the parent to ensure that we can
3775 * safely access it on res_counter_charge/uncharge.
3776 * This refcnt will be decremented when freeing this
3777 * mem_cgroup(see mem_cgroup_put).
3779 mem_cgroup_get(parent);
3781 res_counter_init(&mem->res, NULL);
3782 res_counter_init(&mem->memsw, NULL);
3784 mem->last_scanned_child = 0;
3785 spin_lock_init(&mem->reclaim_param_lock);
3788 mem->swappiness = get_swappiness(parent);
3789 atomic_set(&mem->refcnt, 1);
3790 mem->move_charge_at_immigrate = 0;
3791 mutex_init(&mem->thresholds_lock);
3794 __mem_cgroup_free(mem);
3795 root_mem_cgroup = NULL;
3796 return ERR_PTR(error);
3799 static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
3800 struct cgroup *cont)
3802 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3804 return mem_cgroup_force_empty(mem, false);
3807 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
3808 struct cgroup *cont)
3810 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3812 mem_cgroup_put(mem);
3815 static int mem_cgroup_populate(struct cgroup_subsys *ss,
3816 struct cgroup *cont)
3820 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
3821 ARRAY_SIZE(mem_cgroup_files));
3824 ret = register_memsw_files(cont, ss);
3829 /* Handlers for move charge at task migration. */
3830 #define PRECHARGE_COUNT_AT_ONCE 256
3831 static int mem_cgroup_do_precharge(unsigned long count)
3834 int batch_count = PRECHARGE_COUNT_AT_ONCE;
3835 struct mem_cgroup *mem = mc.to;
3837 if (mem_cgroup_is_root(mem)) {
3838 mc.precharge += count;
3839 /* we don't need css_get for root */
3842 /* try to charge at once */
3844 struct res_counter *dummy;
3846 * "mem" cannot be under rmdir() because we've already checked
3847 * by cgroup_lock_live_cgroup() that it is not removed and we
3848 * are still under the same cgroup_mutex. So we can postpone
3851 if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy))
3853 if (do_swap_account && res_counter_charge(&mem->memsw,
3854 PAGE_SIZE * count, &dummy)) {
3855 res_counter_uncharge(&mem->res, PAGE_SIZE * count);
3858 mc.precharge += count;
3859 VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags));
3860 WARN_ON_ONCE(count > INT_MAX);
3861 __css_get(&mem->css, (int)count);
3865 /* fall back to one by one charge */
3867 if (signal_pending(current)) {
3871 if (!batch_count--) {
3872 batch_count = PRECHARGE_COUNT_AT_ONCE;
3875 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem,
3878 /* mem_cgroup_clear_mc() will do uncharge later */
3884 #else /* !CONFIG_MMU */
3885 static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
3886 struct cgroup *cgroup,
3887 struct task_struct *p,
3892 static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
3893 struct cgroup *cgroup,
3894 struct task_struct *p,
3898 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
3899 struct cgroup *cont,
3900 struct cgroup *old_cont,
3901 struct task_struct *p,
3908 * is_target_pte_for_mc - check a pte whether it is valid for move charge
3909 * @vma: the vma the pte to be checked belongs
3910 * @addr: the address corresponding to the pte to be checked
3911 * @ptent: the pte to be checked
3912 * @target: the pointer the target page or swap ent will be stored(can be NULL)
3915 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
3916 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
3917 * move charge. if @target is not NULL, the page is stored in target->page
3918 * with extra refcnt got(Callers should handle it).
3919 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
3920 * target for charge migration. if @target is not NULL, the entry is stored
3923 * Called with pte lock held.
3930 enum mc_target_type {
3931 MC_TARGET_NONE, /* not used */
3936 static int is_target_pte_for_mc(struct vm_area_struct *vma,
3937 unsigned long addr, pte_t ptent, union mc_target *target)
3939 struct page *page = NULL;
3940 struct page_cgroup *pc;
3942 swp_entry_t ent = { .val = 0 };
3943 int usage_count = 0;
3944 bool move_anon = test_bit(MOVE_CHARGE_TYPE_ANON,
3945 &mc.to->move_charge_at_immigrate);
3947 if (!pte_present(ptent)) {
3948 /* TODO: handle swap of shmes/tmpfs */
3949 if (pte_none(ptent) || pte_file(ptent))
3951 else if (is_swap_pte(ptent)) {
3952 ent = pte_to_swp_entry(ptent);
3953 if (!move_anon || non_swap_entry(ent))
3955 usage_count = mem_cgroup_count_swap_user(ent, &page);
3958 page = vm_normal_page(vma, addr, ptent);
3959 if (!page || !page_mapped(page))
3962 * TODO: We don't move charges of file(including shmem/tmpfs)
3965 if (!move_anon || !PageAnon(page))
3967 if (!get_page_unless_zero(page))
3969 usage_count = page_mapcount(page);
3971 if (usage_count > 1) {
3973 * TODO: We don't move charges of shared(used by multiple
3974 * processes) pages for now.
3981 pc = lookup_page_cgroup(page);
3983 * Do only loose check w/o page_cgroup lock.
3984 * mem_cgroup_move_account() checks the pc is valid or not under
3987 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
3988 ret = MC_TARGET_PAGE;
3990 target->page = page;
3992 if (!ret || !target)
3996 if (ent.val && do_swap_account && !ret &&
3997 css_id(&mc.from->css) == lookup_swap_cgroup(ent)) {
3998 ret = MC_TARGET_SWAP;
4005 static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4006 unsigned long addr, unsigned long end,
4007 struct mm_walk *walk)
4009 struct vm_area_struct *vma = walk->private;
4013 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4014 for (; addr != end; pte++, addr += PAGE_SIZE)
4015 if (is_target_pte_for_mc(vma, addr, *pte, NULL))
4016 mc.precharge++; /* increment precharge temporarily */
4017 pte_unmap_unlock(pte - 1, ptl);
4023 static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4025 unsigned long precharge;
4026 struct vm_area_struct *vma;
4028 down_read(&mm->mmap_sem);
4029 for (vma = mm->mmap; vma; vma = vma->vm_next) {
4030 struct mm_walk mem_cgroup_count_precharge_walk = {
4031 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4035 if (is_vm_hugetlb_page(vma))
4037 /* TODO: We don't move charges of shmem/tmpfs pages for now. */
4038 if (vma->vm_flags & VM_SHARED)
4040 walk_page_range(vma->vm_start, vma->vm_end,
4041 &mem_cgroup_count_precharge_walk);
4043 up_read(&mm->mmap_sem);
4045 precharge = mc.precharge;
4051 static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4053 return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm));
4056 static void mem_cgroup_clear_mc(void)
4058 /* we must uncharge all the leftover precharges from mc.to */
4060 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
4064 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4065 * we must uncharge here.
4067 if (mc.moved_charge) {
4068 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
4069 mc.moved_charge = 0;
4071 /* we must fixup refcnts and charges */
4072 if (mc.moved_swap) {
4073 WARN_ON_ONCE(mc.moved_swap > INT_MAX);
4074 /* uncharge swap account from the old cgroup */
4075 if (!mem_cgroup_is_root(mc.from))
4076 res_counter_uncharge(&mc.from->memsw,
4077 PAGE_SIZE * mc.moved_swap);
4078 __mem_cgroup_put(mc.from, mc.moved_swap);
4080 if (!mem_cgroup_is_root(mc.to)) {
4082 * we charged both to->res and to->memsw, so we should
4085 res_counter_uncharge(&mc.to->res,
4086 PAGE_SIZE * mc.moved_swap);
4087 VM_BUG_ON(test_bit(CSS_ROOT, &mc.to->css.flags));
4088 __css_put(&mc.to->css, mc.moved_swap);
4090 /* we've already done mem_cgroup_get(mc.to) */
4096 mc.moving_task = NULL;
4097 wake_up_all(&mc.waitq);
4100 static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
4101 struct cgroup *cgroup,
4102 struct task_struct *p,
4106 struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup);
4108 if (mem->move_charge_at_immigrate) {
4109 struct mm_struct *mm;
4110 struct mem_cgroup *from = mem_cgroup_from_task(p);
4112 VM_BUG_ON(from == mem);
4114 mm = get_task_mm(p);
4117 /* We move charges only when we move a owner of the mm */
4118 if (mm->owner == p) {
4121 VM_BUG_ON(mc.precharge);
4122 VM_BUG_ON(mc.moved_charge);
4123 VM_BUG_ON(mc.moved_swap);
4124 VM_BUG_ON(mc.moving_task);
4128 mc.moved_charge = 0;
4130 mc.moving_task = current;
4132 ret = mem_cgroup_precharge_mc(mm);
4134 mem_cgroup_clear_mc();
4141 static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
4142 struct cgroup *cgroup,
4143 struct task_struct *p,
4146 mem_cgroup_clear_mc();
4149 static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4150 unsigned long addr, unsigned long end,
4151 struct mm_walk *walk)
4154 struct vm_area_struct *vma = walk->private;
4159 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4160 for (; addr != end; addr += PAGE_SIZE) {
4161 pte_t ptent = *(pte++);
4162 union mc_target target;
4165 struct page_cgroup *pc;
4171 type = is_target_pte_for_mc(vma, addr, ptent, &target);
4173 case MC_TARGET_PAGE:
4175 if (isolate_lru_page(page))
4177 pc = lookup_page_cgroup(page);
4178 if (!mem_cgroup_move_account(pc,
4179 mc.from, mc.to, false)) {
4181 /* we uncharge from mc.from later. */
4184 putback_lru_page(page);
4185 put: /* is_target_pte_for_mc() gets the page */
4188 case MC_TARGET_SWAP:
4190 if (!mem_cgroup_move_swap_account(ent,
4191 mc.from, mc.to, false)) {
4193 /* we fixup refcnts and charges later. */
4201 pte_unmap_unlock(pte - 1, ptl);
4206 * We have consumed all precharges we got in can_attach().
4207 * We try charge one by one, but don't do any additional
4208 * charges to mc.to if we have failed in charge once in attach()
4211 ret = mem_cgroup_do_precharge(1);
4219 static void mem_cgroup_move_charge(struct mm_struct *mm)
4221 struct vm_area_struct *vma;
4223 lru_add_drain_all();
4224 down_read(&mm->mmap_sem);
4225 for (vma = mm->mmap; vma; vma = vma->vm_next) {
4227 struct mm_walk mem_cgroup_move_charge_walk = {
4228 .pmd_entry = mem_cgroup_move_charge_pte_range,
4232 if (is_vm_hugetlb_page(vma))
4234 /* TODO: We don't move charges of shmem/tmpfs pages for now. */
4235 if (vma->vm_flags & VM_SHARED)
4237 ret = walk_page_range(vma->vm_start, vma->vm_end,
4238 &mem_cgroup_move_charge_walk);
4241 * means we have consumed all precharges and failed in
4242 * doing additional charge. Just abandon here.
4246 up_read(&mm->mmap_sem);
4249 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
4250 struct cgroup *cont,
4251 struct cgroup *old_cont,
4252 struct task_struct *p,
4255 struct mm_struct *mm;
4258 /* no need to move charge */
4261 mm = get_task_mm(p);
4263 mem_cgroup_move_charge(mm);
4266 mem_cgroup_clear_mc();
4269 struct cgroup_subsys mem_cgroup_subsys = {
4271 .subsys_id = mem_cgroup_subsys_id,
4272 .create = mem_cgroup_create,
4273 .pre_destroy = mem_cgroup_pre_destroy,
4274 .destroy = mem_cgroup_destroy,
4275 .populate = mem_cgroup_populate,
4276 .can_attach = mem_cgroup_can_attach,
4277 .cancel_attach = mem_cgroup_cancel_attach,
4278 .attach = mem_cgroup_move_task,
4283 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4285 static int __init disable_swap_account(char *s)
4287 really_do_swap_account = 0;
4290 __setup("noswapaccount", disable_swap_account);