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)
67 * Per memcg event counter is incremented at every pagein/pageout. This counter
68 * is used for trigger some periodic events. This is straightforward and better
69 * than using jiffies etc. to handle periodic memcg event.
71 * These values will be used as !((event) & ((1 <<(thresh)) - 1))
73 #define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */
74 #define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */
77 * Statistics for memory cgroup.
79 enum mem_cgroup_stat_index {
81 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
83 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
84 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
85 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
86 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
87 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
88 MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
89 MEM_CGROUP_EVENTS, /* incremented at every pagein/pageout */
91 MEM_CGROUP_STAT_NSTATS,
94 struct mem_cgroup_stat_cpu {
95 s64 count[MEM_CGROUP_STAT_NSTATS];
99 * per-zone information in memory controller.
101 struct mem_cgroup_per_zone {
103 * spin_lock to protect the per cgroup LRU
105 struct list_head lists[NR_LRU_LISTS];
106 unsigned long count[NR_LRU_LISTS];
108 struct zone_reclaim_stat reclaim_stat;
109 struct rb_node tree_node; /* RB tree node */
110 unsigned long long usage_in_excess;/* Set to the value by which */
111 /* the soft limit is exceeded*/
113 struct mem_cgroup *mem; /* Back pointer, we cannot */
114 /* use container_of */
116 /* Macro for accessing counter */
117 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
119 struct mem_cgroup_per_node {
120 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
123 struct mem_cgroup_lru_info {
124 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
128 * Cgroups above their limits are maintained in a RB-Tree, independent of
129 * their hierarchy representation
132 struct mem_cgroup_tree_per_zone {
133 struct rb_root rb_root;
137 struct mem_cgroup_tree_per_node {
138 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
141 struct mem_cgroup_tree {
142 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
145 static struct mem_cgroup_tree soft_limit_tree __read_mostly;
147 struct mem_cgroup_threshold {
148 struct eventfd_ctx *eventfd;
152 struct mem_cgroup_threshold_ary {
153 /* An array index points to threshold just below usage. */
154 atomic_t current_threshold;
155 /* Size of entries[] */
157 /* Array of thresholds */
158 struct mem_cgroup_threshold entries[0];
161 static void mem_cgroup_threshold(struct mem_cgroup *mem);
164 * The memory controller data structure. The memory controller controls both
165 * page cache and RSS per cgroup. We would eventually like to provide
166 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
167 * to help the administrator determine what knobs to tune.
169 * TODO: Add a water mark for the memory controller. Reclaim will begin when
170 * we hit the water mark. May be even add a low water mark, such that
171 * no reclaim occurs from a cgroup at it's low water mark, this is
172 * a feature that will be implemented much later in the future.
175 struct cgroup_subsys_state css;
177 * the counter to account for memory usage
179 struct res_counter res;
181 * the counter to account for mem+swap usage.
183 struct res_counter memsw;
185 * Per cgroup active and inactive list, similar to the
186 * per zone LRU lists.
188 struct mem_cgroup_lru_info info;
191 protect against reclaim related member.
193 spinlock_t reclaim_param_lock;
195 int prev_priority; /* for recording reclaim priority */
198 * While reclaiming in a hierarchy, we cache the last child we
201 int last_scanned_child;
203 * Should the accounting and control be hierarchical, per subtree?
209 unsigned int swappiness;
211 /* set when res.limit == memsw.limit */
212 bool memsw_is_minimum;
214 /* protect arrays of thresholds */
215 struct mutex thresholds_lock;
217 /* thresholds for memory usage. RCU-protected */
218 struct mem_cgroup_threshold_ary *thresholds;
220 /* thresholds for mem+swap usage. RCU-protected */
221 struct mem_cgroup_threshold_ary *memsw_thresholds;
224 * Should we move charges of a task when a task is moved into this
225 * mem_cgroup ? And what type of charges should we move ?
227 unsigned long move_charge_at_immigrate;
232 struct mem_cgroup_stat_cpu *stat;
235 /* Stuffs for move charges at task migration. */
237 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
238 * left-shifted bitmap of these types.
241 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
245 /* "mc" and its members are protected by cgroup_mutex */
246 static struct move_charge_struct {
247 struct mem_cgroup *from;
248 struct mem_cgroup *to;
249 unsigned long precharge;
250 unsigned long moved_charge;
251 unsigned long moved_swap;
252 struct task_struct *moving_task; /* a task moving charges */
253 wait_queue_head_t waitq; /* a waitq for other context */
255 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
259 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
260 * limit reclaim to prevent infinite loops, if they ever occur.
262 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
263 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
266 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
267 MEM_CGROUP_CHARGE_TYPE_MAPPED,
268 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
269 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
270 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
271 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
275 /* only for here (for easy reading.) */
276 #define PCGF_CACHE (1UL << PCG_CACHE)
277 #define PCGF_USED (1UL << PCG_USED)
278 #define PCGF_LOCK (1UL << PCG_LOCK)
279 /* Not used, but added here for completeness */
280 #define PCGF_ACCT (1UL << PCG_ACCT)
282 /* for encoding cft->private value on file */
285 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
286 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
287 #define MEMFILE_ATTR(val) ((val) & 0xffff)
290 * Reclaim flags for mem_cgroup_hierarchical_reclaim
292 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
293 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
294 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
295 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
296 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
297 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
299 static void mem_cgroup_get(struct mem_cgroup *mem);
300 static void mem_cgroup_put(struct mem_cgroup *mem);
301 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
302 static void drain_all_stock_async(void);
304 static struct mem_cgroup_per_zone *
305 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
307 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
310 struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem)
315 static struct mem_cgroup_per_zone *
316 page_cgroup_zoneinfo(struct page_cgroup *pc)
318 struct mem_cgroup *mem = pc->mem_cgroup;
319 int nid = page_cgroup_nid(pc);
320 int zid = page_cgroup_zid(pc);
325 return mem_cgroup_zoneinfo(mem, nid, zid);
328 static struct mem_cgroup_tree_per_zone *
329 soft_limit_tree_node_zone(int nid, int zid)
331 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
334 static struct mem_cgroup_tree_per_zone *
335 soft_limit_tree_from_page(struct page *page)
337 int nid = page_to_nid(page);
338 int zid = page_zonenum(page);
340 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
344 __mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
345 struct mem_cgroup_per_zone *mz,
346 struct mem_cgroup_tree_per_zone *mctz,
347 unsigned long long new_usage_in_excess)
349 struct rb_node **p = &mctz->rb_root.rb_node;
350 struct rb_node *parent = NULL;
351 struct mem_cgroup_per_zone *mz_node;
356 mz->usage_in_excess = new_usage_in_excess;
357 if (!mz->usage_in_excess)
361 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
363 if (mz->usage_in_excess < mz_node->usage_in_excess)
366 * We can't avoid mem cgroups that are over their soft
367 * limit by the same amount
369 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
372 rb_link_node(&mz->tree_node, parent, p);
373 rb_insert_color(&mz->tree_node, &mctz->rb_root);
378 __mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
379 struct mem_cgroup_per_zone *mz,
380 struct mem_cgroup_tree_per_zone *mctz)
384 rb_erase(&mz->tree_node, &mctz->rb_root);
389 mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
390 struct mem_cgroup_per_zone *mz,
391 struct mem_cgroup_tree_per_zone *mctz)
393 spin_lock(&mctz->lock);
394 __mem_cgroup_remove_exceeded(mem, mz, mctz);
395 spin_unlock(&mctz->lock);
399 static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page)
401 unsigned long long excess;
402 struct mem_cgroup_per_zone *mz;
403 struct mem_cgroup_tree_per_zone *mctz;
404 int nid = page_to_nid(page);
405 int zid = page_zonenum(page);
406 mctz = soft_limit_tree_from_page(page);
409 * Necessary to update all ancestors when hierarchy is used.
410 * because their event counter is not touched.
412 for (; mem; mem = parent_mem_cgroup(mem)) {
413 mz = mem_cgroup_zoneinfo(mem, nid, zid);
414 excess = res_counter_soft_limit_excess(&mem->res);
416 * We have to update the tree if mz is on RB-tree or
417 * mem is over its softlimit.
419 if (excess || mz->on_tree) {
420 spin_lock(&mctz->lock);
421 /* if on-tree, remove it */
423 __mem_cgroup_remove_exceeded(mem, mz, mctz);
425 * Insert again. mz->usage_in_excess will be updated.
426 * If excess is 0, no tree ops.
428 __mem_cgroup_insert_exceeded(mem, mz, mctz, excess);
429 spin_unlock(&mctz->lock);
434 static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem)
437 struct mem_cgroup_per_zone *mz;
438 struct mem_cgroup_tree_per_zone *mctz;
440 for_each_node_state(node, N_POSSIBLE) {
441 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
442 mz = mem_cgroup_zoneinfo(mem, node, zone);
443 mctz = soft_limit_tree_node_zone(node, zone);
444 mem_cgroup_remove_exceeded(mem, mz, mctz);
449 static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem)
451 return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT;
454 static struct mem_cgroup_per_zone *
455 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
457 struct rb_node *rightmost = NULL;
458 struct mem_cgroup_per_zone *mz;
462 rightmost = rb_last(&mctz->rb_root);
464 goto done; /* Nothing to reclaim from */
466 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
468 * Remove the node now but someone else can add it back,
469 * we will to add it back at the end of reclaim to its correct
470 * position in the tree.
472 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
473 if (!res_counter_soft_limit_excess(&mz->mem->res) ||
474 !css_tryget(&mz->mem->css))
480 static struct mem_cgroup_per_zone *
481 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
483 struct mem_cgroup_per_zone *mz;
485 spin_lock(&mctz->lock);
486 mz = __mem_cgroup_largest_soft_limit_node(mctz);
487 spin_unlock(&mctz->lock);
491 static s64 mem_cgroup_read_stat(struct mem_cgroup *mem,
492 enum mem_cgroup_stat_index idx)
497 for_each_possible_cpu(cpu)
498 val += per_cpu(mem->stat->count[idx], cpu);
502 static s64 mem_cgroup_local_usage(struct mem_cgroup *mem)
506 ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
507 ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
511 static void mem_cgroup_swap_statistics(struct mem_cgroup *mem,
514 int val = (charge) ? 1 : -1;
515 this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
518 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
519 struct page_cgroup *pc,
522 int val = (charge) ? 1 : -1;
526 if (PageCgroupCache(pc))
527 __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], val);
529 __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], val);
532 __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]);
534 __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]);
535 __this_cpu_inc(mem->stat->count[MEM_CGROUP_EVENTS]);
540 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
544 struct mem_cgroup_per_zone *mz;
547 for_each_online_node(nid)
548 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
549 mz = mem_cgroup_zoneinfo(mem, nid, zid);
550 total += MEM_CGROUP_ZSTAT(mz, idx);
555 static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift)
559 val = this_cpu_read(mem->stat->count[MEM_CGROUP_EVENTS]);
561 return !(val & ((1 << event_mask_shift) - 1));
565 * Check events in order.
568 static void memcg_check_events(struct mem_cgroup *mem, struct page *page)
570 /* threshold event is triggered in finer grain than soft limit */
571 if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) {
572 mem_cgroup_threshold(mem);
573 if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH)))
574 mem_cgroup_update_tree(mem, page);
578 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
580 return container_of(cgroup_subsys_state(cont,
581 mem_cgroup_subsys_id), struct mem_cgroup,
585 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
588 * mm_update_next_owner() may clear mm->owner to NULL
589 * if it races with swapoff, page migration, etc.
590 * So this can be called with p == NULL.
595 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
596 struct mem_cgroup, css);
599 static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
601 struct mem_cgroup *mem = NULL;
606 * Because we have no locks, mm->owner's may be being moved to other
607 * cgroup. We use css_tryget() here even if this looks
608 * pessimistic (rather than adding locks here).
612 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
615 } while (!css_tryget(&mem->css));
621 * Call callback function against all cgroup under hierarchy tree.
623 static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
624 int (*func)(struct mem_cgroup *, void *))
626 int found, ret, nextid;
627 struct cgroup_subsys_state *css;
628 struct mem_cgroup *mem;
630 if (!root->use_hierarchy)
631 return (*func)(root, data);
639 css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
641 if (css && css_tryget(css))
642 mem = container_of(css, struct mem_cgroup, css);
646 ret = (*func)(mem, data);
650 } while (!ret && css);
655 static inline bool mem_cgroup_is_root(struct mem_cgroup *mem)
657 return (mem == root_mem_cgroup);
661 * Following LRU functions are allowed to be used without PCG_LOCK.
662 * Operations are called by routine of global LRU independently from memcg.
663 * What we have to take care of here is validness of pc->mem_cgroup.
665 * Changes to pc->mem_cgroup happens when
668 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
669 * It is added to LRU before charge.
670 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
671 * When moving account, the page is not on LRU. It's isolated.
674 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
676 struct page_cgroup *pc;
677 struct mem_cgroup_per_zone *mz;
679 if (mem_cgroup_disabled())
681 pc = lookup_page_cgroup(page);
682 /* can happen while we handle swapcache. */
683 if (!TestClearPageCgroupAcctLRU(pc))
685 VM_BUG_ON(!pc->mem_cgroup);
687 * We don't check PCG_USED bit. It's cleared when the "page" is finally
688 * removed from global LRU.
690 mz = page_cgroup_zoneinfo(pc);
691 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
692 if (mem_cgroup_is_root(pc->mem_cgroup))
694 VM_BUG_ON(list_empty(&pc->lru));
695 list_del_init(&pc->lru);
699 void mem_cgroup_del_lru(struct page *page)
701 mem_cgroup_del_lru_list(page, page_lru(page));
704 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
706 struct mem_cgroup_per_zone *mz;
707 struct page_cgroup *pc;
709 if (mem_cgroup_disabled())
712 pc = lookup_page_cgroup(page);
714 * Used bit is set without atomic ops but after smp_wmb().
715 * For making pc->mem_cgroup visible, insert smp_rmb() here.
718 /* unused or root page is not rotated. */
719 if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup))
721 mz = page_cgroup_zoneinfo(pc);
722 list_move(&pc->lru, &mz->lists[lru]);
725 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
727 struct page_cgroup *pc;
728 struct mem_cgroup_per_zone *mz;
730 if (mem_cgroup_disabled())
732 pc = lookup_page_cgroup(page);
733 VM_BUG_ON(PageCgroupAcctLRU(pc));
735 * Used bit is set without atomic ops but after smp_wmb().
736 * For making pc->mem_cgroup visible, insert smp_rmb() here.
739 if (!PageCgroupUsed(pc))
742 mz = page_cgroup_zoneinfo(pc);
743 MEM_CGROUP_ZSTAT(mz, lru) += 1;
744 SetPageCgroupAcctLRU(pc);
745 if (mem_cgroup_is_root(pc->mem_cgroup))
747 list_add(&pc->lru, &mz->lists[lru]);
751 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
752 * lru because the page may.be reused after it's fully uncharged (because of
753 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
754 * it again. This function is only used to charge SwapCache. It's done under
755 * lock_page and expected that zone->lru_lock is never held.
757 static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
760 struct zone *zone = page_zone(page);
761 struct page_cgroup *pc = lookup_page_cgroup(page);
763 spin_lock_irqsave(&zone->lru_lock, flags);
765 * Forget old LRU when this page_cgroup is *not* used. This Used bit
766 * is guarded by lock_page() because the page is SwapCache.
768 if (!PageCgroupUsed(pc))
769 mem_cgroup_del_lru_list(page, page_lru(page));
770 spin_unlock_irqrestore(&zone->lru_lock, flags);
773 static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
776 struct zone *zone = page_zone(page);
777 struct page_cgroup *pc = lookup_page_cgroup(page);
779 spin_lock_irqsave(&zone->lru_lock, flags);
780 /* link when the page is linked to LRU but page_cgroup isn't */
781 if (PageLRU(page) && !PageCgroupAcctLRU(pc))
782 mem_cgroup_add_lru_list(page, page_lru(page));
783 spin_unlock_irqrestore(&zone->lru_lock, flags);
787 void mem_cgroup_move_lists(struct page *page,
788 enum lru_list from, enum lru_list to)
790 if (mem_cgroup_disabled())
792 mem_cgroup_del_lru_list(page, from);
793 mem_cgroup_add_lru_list(page, to);
796 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
799 struct mem_cgroup *curr = NULL;
803 curr = try_get_mem_cgroup_from_mm(task->mm);
809 * We should check use_hierarchy of "mem" not "curr". Because checking
810 * use_hierarchy of "curr" here make this function true if hierarchy is
811 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
812 * hierarchy(even if use_hierarchy is disabled in "mem").
814 if (mem->use_hierarchy)
815 ret = css_is_ancestor(&curr->css, &mem->css);
823 * prev_priority control...this will be used in memory reclaim path.
825 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
829 spin_lock(&mem->reclaim_param_lock);
830 prev_priority = mem->prev_priority;
831 spin_unlock(&mem->reclaim_param_lock);
833 return prev_priority;
836 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
838 spin_lock(&mem->reclaim_param_lock);
839 if (priority < mem->prev_priority)
840 mem->prev_priority = priority;
841 spin_unlock(&mem->reclaim_param_lock);
844 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
846 spin_lock(&mem->reclaim_param_lock);
847 mem->prev_priority = priority;
848 spin_unlock(&mem->reclaim_param_lock);
851 static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
853 unsigned long active;
854 unsigned long inactive;
856 unsigned long inactive_ratio;
858 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
859 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
861 gb = (inactive + active) >> (30 - PAGE_SHIFT);
863 inactive_ratio = int_sqrt(10 * gb);
868 present_pages[0] = inactive;
869 present_pages[1] = active;
872 return inactive_ratio;
875 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
877 unsigned long active;
878 unsigned long inactive;
879 unsigned long present_pages[2];
880 unsigned long inactive_ratio;
882 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
884 inactive = present_pages[0];
885 active = present_pages[1];
887 if (inactive * inactive_ratio < active)
893 int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
895 unsigned long active;
896 unsigned long inactive;
898 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
899 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
901 return (active > inactive);
904 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
908 int nid = zone->zone_pgdat->node_id;
909 int zid = zone_idx(zone);
910 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
912 return MEM_CGROUP_ZSTAT(mz, lru);
915 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
918 int nid = zone->zone_pgdat->node_id;
919 int zid = zone_idx(zone);
920 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
922 return &mz->reclaim_stat;
925 struct zone_reclaim_stat *
926 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
928 struct page_cgroup *pc;
929 struct mem_cgroup_per_zone *mz;
931 if (mem_cgroup_disabled())
934 pc = lookup_page_cgroup(page);
936 * Used bit is set without atomic ops but after smp_wmb().
937 * For making pc->mem_cgroup visible, insert smp_rmb() here.
940 if (!PageCgroupUsed(pc))
943 mz = page_cgroup_zoneinfo(pc);
947 return &mz->reclaim_stat;
950 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
951 struct list_head *dst,
952 unsigned long *scanned, int order,
953 int mode, struct zone *z,
954 struct mem_cgroup *mem_cont,
955 int active, int file)
957 unsigned long nr_taken = 0;
961 struct list_head *src;
962 struct page_cgroup *pc, *tmp;
963 int nid = z->zone_pgdat->node_id;
964 int zid = zone_idx(z);
965 struct mem_cgroup_per_zone *mz;
966 int lru = LRU_FILE * file + active;
970 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
971 src = &mz->lists[lru];
974 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
975 if (scan >= nr_to_scan)
979 if (unlikely(!PageCgroupUsed(pc)))
981 if (unlikely(!PageLRU(page)))
985 ret = __isolate_lru_page(page, mode, file);
988 list_move(&page->lru, dst);
989 mem_cgroup_del_lru(page);
993 /* we don't affect global LRU but rotate in our LRU */
994 mem_cgroup_rotate_lru_list(page, page_lru(page));
1005 #define mem_cgroup_from_res_counter(counter, member) \
1006 container_of(counter, struct mem_cgroup, member)
1008 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
1010 if (do_swap_account) {
1011 if (res_counter_check_under_limit(&mem->res) &&
1012 res_counter_check_under_limit(&mem->memsw))
1015 if (res_counter_check_under_limit(&mem->res))
1020 static unsigned int get_swappiness(struct mem_cgroup *memcg)
1022 struct cgroup *cgrp = memcg->css.cgroup;
1023 unsigned int swappiness;
1026 if (cgrp->parent == NULL)
1027 return vm_swappiness;
1029 spin_lock(&memcg->reclaim_param_lock);
1030 swappiness = memcg->swappiness;
1031 spin_unlock(&memcg->reclaim_param_lock);
1036 static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
1044 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1045 * @memcg: The memory cgroup that went over limit
1046 * @p: Task that is going to be killed
1048 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1051 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1053 struct cgroup *task_cgrp;
1054 struct cgroup *mem_cgrp;
1056 * Need a buffer in BSS, can't rely on allocations. The code relies
1057 * on the assumption that OOM is serialized for memory controller.
1058 * If this assumption is broken, revisit this code.
1060 static char memcg_name[PATH_MAX];
1069 mem_cgrp = memcg->css.cgroup;
1070 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1072 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1075 * Unfortunately, we are unable to convert to a useful name
1076 * But we'll still print out the usage information
1083 printk(KERN_INFO "Task in %s killed", memcg_name);
1086 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1094 * Continues from above, so we don't need an KERN_ level
1096 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1099 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1100 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1101 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1102 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1103 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1105 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1106 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1107 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1111 * This function returns the number of memcg under hierarchy tree. Returns
1112 * 1(self count) if no children.
1114 static int mem_cgroup_count_children(struct mem_cgroup *mem)
1117 mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
1122 * Visit the first child (need not be the first child as per the ordering
1123 * of the cgroup list, since we track last_scanned_child) of @mem and use
1124 * that to reclaim free pages from.
1126 static struct mem_cgroup *
1127 mem_cgroup_select_victim(struct mem_cgroup *root_mem)
1129 struct mem_cgroup *ret = NULL;
1130 struct cgroup_subsys_state *css;
1133 if (!root_mem->use_hierarchy) {
1134 css_get(&root_mem->css);
1140 nextid = root_mem->last_scanned_child + 1;
1141 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
1143 if (css && css_tryget(css))
1144 ret = container_of(css, struct mem_cgroup, css);
1147 /* Updates scanning parameter */
1148 spin_lock(&root_mem->reclaim_param_lock);
1150 /* this means start scan from ID:1 */
1151 root_mem->last_scanned_child = 0;
1153 root_mem->last_scanned_child = found;
1154 spin_unlock(&root_mem->reclaim_param_lock);
1161 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1162 * we reclaimed from, so that we don't end up penalizing one child extensively
1163 * based on its position in the children list.
1165 * root_mem is the original ancestor that we've been reclaim from.
1167 * We give up and return to the caller when we visit root_mem twice.
1168 * (other groups can be removed while we're walking....)
1170 * If shrink==true, for avoiding to free too much, this returns immedieately.
1172 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
1175 unsigned long reclaim_options)
1177 struct mem_cgroup *victim;
1180 bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
1181 bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
1182 bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT;
1183 unsigned long excess = mem_cgroup_get_excess(root_mem);
1185 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1186 if (root_mem->memsw_is_minimum)
1190 victim = mem_cgroup_select_victim(root_mem);
1191 if (victim == root_mem) {
1194 drain_all_stock_async();
1197 * If we have not been able to reclaim
1198 * anything, it might because there are
1199 * no reclaimable pages under this hierarchy
1201 if (!check_soft || !total) {
1202 css_put(&victim->css);
1206 * We want to do more targetted reclaim.
1207 * excess >> 2 is not to excessive so as to
1208 * reclaim too much, nor too less that we keep
1209 * coming back to reclaim from this cgroup
1211 if (total >= (excess >> 2) ||
1212 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) {
1213 css_put(&victim->css);
1218 if (!mem_cgroup_local_usage(victim)) {
1219 /* this cgroup's local usage == 0 */
1220 css_put(&victim->css);
1223 /* we use swappiness of local cgroup */
1225 ret = mem_cgroup_shrink_node_zone(victim, gfp_mask,
1226 noswap, get_swappiness(victim), zone,
1227 zone->zone_pgdat->node_id);
1229 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask,
1230 noswap, get_swappiness(victim));
1231 css_put(&victim->css);
1233 * At shrinking usage, we can't check we should stop here or
1234 * reclaim more. It's depends on callers. last_scanned_child
1235 * will work enough for keeping fairness under tree.
1241 if (res_counter_check_under_soft_limit(&root_mem->res))
1243 } else if (mem_cgroup_check_under_limit(root_mem))
1249 static int mem_cgroup_oom_lock_cb(struct mem_cgroup *mem, void *data)
1251 int *val = (int *)data;
1254 * Logically, we can stop scanning immediately when we find
1255 * a memcg is already locked. But condidering unlock ops and
1256 * creation/removal of memcg, scan-all is simple operation.
1258 x = atomic_inc_return(&mem->oom_lock);
1259 *val = max(x, *val);
1263 * Check OOM-Killer is already running under our hierarchy.
1264 * If someone is running, return false.
1266 static bool mem_cgroup_oom_lock(struct mem_cgroup *mem)
1270 mem_cgroup_walk_tree(mem, &lock_count, mem_cgroup_oom_lock_cb);
1272 if (lock_count == 1)
1277 static int mem_cgroup_oom_unlock_cb(struct mem_cgroup *mem, void *data)
1280 * When a new child is created while the hierarchy is under oom,
1281 * mem_cgroup_oom_lock() may not be called. We have to use
1282 * atomic_add_unless() here.
1284 atomic_add_unless(&mem->oom_lock, -1, 0);
1288 static void mem_cgroup_oom_unlock(struct mem_cgroup *mem)
1290 mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_unlock_cb);
1293 static DEFINE_MUTEX(memcg_oom_mutex);
1294 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1296 struct oom_wait_info {
1297 struct mem_cgroup *mem;
1301 static int memcg_oom_wake_function(wait_queue_t *wait,
1302 unsigned mode, int sync, void *arg)
1304 struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg;
1305 struct oom_wait_info *oom_wait_info;
1307 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
1309 if (oom_wait_info->mem == wake_mem)
1311 /* if no hierarchy, no match */
1312 if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy)
1315 * Both of oom_wait_info->mem and wake_mem are stable under us.
1316 * Then we can use css_is_ancestor without taking care of RCU.
1318 if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) &&
1319 !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css))
1323 return autoremove_wake_function(wait, mode, sync, arg);
1326 static void memcg_wakeup_oom(struct mem_cgroup *mem)
1328 /* for filtering, pass "mem" as argument. */
1329 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem);
1333 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1335 bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask)
1337 struct oom_wait_info owait;
1341 owait.wait.flags = 0;
1342 owait.wait.func = memcg_oom_wake_function;
1343 owait.wait.private = current;
1344 INIT_LIST_HEAD(&owait.wait.task_list);
1346 /* At first, try to OOM lock hierarchy under mem.*/
1347 mutex_lock(&memcg_oom_mutex);
1348 locked = mem_cgroup_oom_lock(mem);
1350 * Even if signal_pending(), we can't quit charge() loop without
1351 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1352 * under OOM is always welcomed, use TASK_KILLABLE here.
1355 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
1356 mutex_unlock(&memcg_oom_mutex);
1359 mem_cgroup_out_of_memory(mem, mask);
1362 finish_wait(&memcg_oom_waitq, &owait.wait);
1364 mutex_lock(&memcg_oom_mutex);
1365 mem_cgroup_oom_unlock(mem);
1366 memcg_wakeup_oom(mem);
1367 mutex_unlock(&memcg_oom_mutex);
1369 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1371 /* Give chance to dying process */
1372 schedule_timeout(1);
1377 * Currently used to update mapped file statistics, but the routine can be
1378 * generalized to update other statistics as well.
1380 void mem_cgroup_update_file_mapped(struct page *page, int val)
1382 struct mem_cgroup *mem;
1383 struct page_cgroup *pc;
1385 pc = lookup_page_cgroup(page);
1389 lock_page_cgroup(pc);
1390 mem = pc->mem_cgroup;
1391 if (!mem || !PageCgroupUsed(pc))
1395 * Preemption is already disabled. We can use __this_cpu_xxx
1398 __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
1399 SetPageCgroupFileMapped(pc);
1401 __this_cpu_dec(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
1402 ClearPageCgroupFileMapped(pc);
1406 unlock_page_cgroup(pc);
1410 * size of first charge trial. "32" comes from vmscan.c's magic value.
1411 * TODO: maybe necessary to use big numbers in big irons.
1413 #define CHARGE_SIZE (32 * PAGE_SIZE)
1414 struct memcg_stock_pcp {
1415 struct mem_cgroup *cached; /* this never be root cgroup */
1417 struct work_struct work;
1419 static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
1420 static atomic_t memcg_drain_count;
1423 * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
1424 * from local stock and true is returned. If the stock is 0 or charges from a
1425 * cgroup which is not current target, returns false. This stock will be
1428 static bool consume_stock(struct mem_cgroup *mem)
1430 struct memcg_stock_pcp *stock;
1433 stock = &get_cpu_var(memcg_stock);
1434 if (mem == stock->cached && stock->charge)
1435 stock->charge -= PAGE_SIZE;
1436 else /* need to call res_counter_charge */
1438 put_cpu_var(memcg_stock);
1443 * Returns stocks cached in percpu to res_counter and reset cached information.
1445 static void drain_stock(struct memcg_stock_pcp *stock)
1447 struct mem_cgroup *old = stock->cached;
1449 if (stock->charge) {
1450 res_counter_uncharge(&old->res, stock->charge);
1451 if (do_swap_account)
1452 res_counter_uncharge(&old->memsw, stock->charge);
1454 stock->cached = NULL;
1459 * This must be called under preempt disabled or must be called by
1460 * a thread which is pinned to local cpu.
1462 static void drain_local_stock(struct work_struct *dummy)
1464 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
1469 * Cache charges(val) which is from res_counter, to local per_cpu area.
1470 * This will be consumed by consume_stock() function, later.
1472 static void refill_stock(struct mem_cgroup *mem, int val)
1474 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1476 if (stock->cached != mem) { /* reset if necessary */
1478 stock->cached = mem;
1480 stock->charge += val;
1481 put_cpu_var(memcg_stock);
1485 * Tries to drain stocked charges in other cpus. This function is asynchronous
1486 * and just put a work per cpu for draining localy on each cpu. Caller can
1487 * expects some charges will be back to res_counter later but cannot wait for
1490 static void drain_all_stock_async(void)
1493 /* This function is for scheduling "drain" in asynchronous way.
1494 * The result of "drain" is not directly handled by callers. Then,
1495 * if someone is calling drain, we don't have to call drain more.
1496 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1497 * there is a race. We just do loose check here.
1499 if (atomic_read(&memcg_drain_count))
1501 /* Notify other cpus that system-wide "drain" is running */
1502 atomic_inc(&memcg_drain_count);
1504 for_each_online_cpu(cpu) {
1505 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
1506 schedule_work_on(cpu, &stock->work);
1509 atomic_dec(&memcg_drain_count);
1510 /* We don't wait for flush_work */
1513 /* This is a synchronous drain interface. */
1514 static void drain_all_stock_sync(void)
1516 /* called when force_empty is called */
1517 atomic_inc(&memcg_drain_count);
1518 schedule_on_each_cpu(drain_local_stock);
1519 atomic_dec(&memcg_drain_count);
1522 static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb,
1523 unsigned long action,
1526 int cpu = (unsigned long)hcpu;
1527 struct memcg_stock_pcp *stock;
1529 if (action != CPU_DEAD)
1531 stock = &per_cpu(memcg_stock, cpu);
1537 * Unlike exported interface, "oom" parameter is added. if oom==true,
1538 * oom-killer can be invoked.
1540 static int __mem_cgroup_try_charge(struct mm_struct *mm,
1541 gfp_t gfp_mask, struct mem_cgroup **memcg, bool oom)
1543 struct mem_cgroup *mem, *mem_over_limit;
1544 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1545 struct res_counter *fail_res;
1546 int csize = CHARGE_SIZE;
1549 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
1550 * in system level. So, allow to go ahead dying process in addition to
1553 if (unlikely(test_thread_flag(TIF_MEMDIE)
1554 || fatal_signal_pending(current)))
1558 * We always charge the cgroup the mm_struct belongs to.
1559 * The mm_struct's mem_cgroup changes on task migration if the
1560 * thread group leader migrates. It's possible that mm is not
1561 * set, if so charge the init_mm (happens for pagecache usage).
1565 mem = try_get_mem_cgroup_from_mm(mm);
1573 VM_BUG_ON(css_is_removed(&mem->css));
1574 if (mem_cgroup_is_root(mem))
1579 unsigned long flags = 0;
1581 if (consume_stock(mem))
1584 ret = res_counter_charge(&mem->res, csize, &fail_res);
1586 if (!do_swap_account)
1588 ret = res_counter_charge(&mem->memsw, csize, &fail_res);
1591 /* mem+swap counter fails */
1592 res_counter_uncharge(&mem->res, csize);
1593 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
1594 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1597 /* mem counter fails */
1598 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1601 /* reduce request size and retry */
1602 if (csize > PAGE_SIZE) {
1606 if (!(gfp_mask & __GFP_WAIT))
1609 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
1615 * try_to_free_mem_cgroup_pages() might not give us a full
1616 * picture of reclaim. Some pages are reclaimed and might be
1617 * moved to swap cache or just unmapped from the cgroup.
1618 * Check the limit again to see if the reclaim reduced the
1619 * current usage of the cgroup before giving up
1622 if (mem_cgroup_check_under_limit(mem_over_limit))
1625 /* try to avoid oom while someone is moving charge */
1626 if (mc.moving_task && current != mc.moving_task) {
1627 struct mem_cgroup *from, *to;
1628 bool do_continue = false;
1630 * There is a small race that "from" or "to" can be
1631 * freed by rmdir, so we use css_tryget().
1635 if (from && css_tryget(&from->css)) {
1636 if (mem_over_limit->use_hierarchy)
1637 do_continue = css_is_ancestor(
1639 &mem_over_limit->css);
1641 do_continue = (from == mem_over_limit);
1642 css_put(&from->css);
1644 if (!do_continue && to && css_tryget(&to->css)) {
1645 if (mem_over_limit->use_hierarchy)
1646 do_continue = css_is_ancestor(
1648 &mem_over_limit->css);
1650 do_continue = (to == mem_over_limit);
1655 prepare_to_wait(&mc.waitq, &wait,
1656 TASK_INTERRUPTIBLE);
1657 /* moving charge context might have finished. */
1660 finish_wait(&mc.waitq, &wait);
1665 if (!nr_retries--) {
1668 if (mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) {
1669 nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1672 /* When we reach here, current task is dying .*/
1677 if (csize > PAGE_SIZE)
1678 refill_stock(mem, csize - PAGE_SIZE);
1690 * Somemtimes we have to undo a charge we got by try_charge().
1691 * This function is for that and do uncharge, put css's refcnt.
1692 * gotten by try_charge().
1694 static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem,
1695 unsigned long count)
1697 if (!mem_cgroup_is_root(mem)) {
1698 res_counter_uncharge(&mem->res, PAGE_SIZE * count);
1699 if (do_swap_account)
1700 res_counter_uncharge(&mem->memsw, PAGE_SIZE * count);
1701 VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags));
1702 WARN_ON_ONCE(count > INT_MAX);
1703 __css_put(&mem->css, (int)count);
1705 /* we don't need css_put for root */
1708 static void mem_cgroup_cancel_charge(struct mem_cgroup *mem)
1710 __mem_cgroup_cancel_charge(mem, 1);
1714 * A helper function to get mem_cgroup from ID. must be called under
1715 * rcu_read_lock(). The caller must check css_is_removed() or some if
1716 * it's concern. (dropping refcnt from swap can be called against removed
1719 static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
1721 struct cgroup_subsys_state *css;
1723 /* ID 0 is unused ID */
1726 css = css_lookup(&mem_cgroup_subsys, id);
1729 return container_of(css, struct mem_cgroup, css);
1732 struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
1734 struct mem_cgroup *mem = NULL;
1735 struct page_cgroup *pc;
1739 VM_BUG_ON(!PageLocked(page));
1741 pc = lookup_page_cgroup(page);
1742 lock_page_cgroup(pc);
1743 if (PageCgroupUsed(pc)) {
1744 mem = pc->mem_cgroup;
1745 if (mem && !css_tryget(&mem->css))
1747 } else if (PageSwapCache(page)) {
1748 ent.val = page_private(page);
1749 id = lookup_swap_cgroup(ent);
1751 mem = mem_cgroup_lookup(id);
1752 if (mem && !css_tryget(&mem->css))
1756 unlock_page_cgroup(pc);
1761 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
1762 * USED state. If already USED, uncharge and return.
1765 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
1766 struct page_cgroup *pc,
1767 enum charge_type ctype)
1769 /* try_charge() can return NULL to *memcg, taking care of it. */
1773 lock_page_cgroup(pc);
1774 if (unlikely(PageCgroupUsed(pc))) {
1775 unlock_page_cgroup(pc);
1776 mem_cgroup_cancel_charge(mem);
1780 pc->mem_cgroup = mem;
1782 * We access a page_cgroup asynchronously without lock_page_cgroup().
1783 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
1784 * is accessed after testing USED bit. To make pc->mem_cgroup visible
1785 * before USED bit, we need memory barrier here.
1786 * See mem_cgroup_add_lru_list(), etc.
1790 case MEM_CGROUP_CHARGE_TYPE_CACHE:
1791 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
1792 SetPageCgroupCache(pc);
1793 SetPageCgroupUsed(pc);
1795 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1796 ClearPageCgroupCache(pc);
1797 SetPageCgroupUsed(pc);
1803 mem_cgroup_charge_statistics(mem, pc, true);
1805 unlock_page_cgroup(pc);
1807 * "charge_statistics" updated event counter. Then, check it.
1808 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
1809 * if they exceeds softlimit.
1811 memcg_check_events(mem, pc->page);
1815 * __mem_cgroup_move_account - move account of the page
1816 * @pc: page_cgroup of the page.
1817 * @from: mem_cgroup which the page is moved from.
1818 * @to: mem_cgroup which the page is moved to. @from != @to.
1819 * @uncharge: whether we should call uncharge and css_put against @from.
1821 * The caller must confirm following.
1822 * - page is not on LRU (isolate_page() is useful.)
1823 * - the pc is locked, used, and ->mem_cgroup points to @from.
1825 * This function doesn't do "charge" nor css_get to new cgroup. It should be
1826 * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
1827 * true, this function does "uncharge" from old cgroup, but it doesn't if
1828 * @uncharge is false, so a caller should do "uncharge".
1831 static void __mem_cgroup_move_account(struct page_cgroup *pc,
1832 struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge)
1834 VM_BUG_ON(from == to);
1835 VM_BUG_ON(PageLRU(pc->page));
1836 VM_BUG_ON(!PageCgroupLocked(pc));
1837 VM_BUG_ON(!PageCgroupUsed(pc));
1838 VM_BUG_ON(pc->mem_cgroup != from);
1840 if (PageCgroupFileMapped(pc)) {
1841 /* Update mapped_file data for mem_cgroup */
1843 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
1844 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
1847 mem_cgroup_charge_statistics(from, pc, false);
1849 /* This is not "cancel", but cancel_charge does all we need. */
1850 mem_cgroup_cancel_charge(from);
1852 /* caller should have done css_get */
1853 pc->mem_cgroup = to;
1854 mem_cgroup_charge_statistics(to, pc, true);
1856 * We charges against "to" which may not have any tasks. Then, "to"
1857 * can be under rmdir(). But in current implementation, caller of
1858 * this function is just force_empty() and move charge, so it's
1859 * garanteed that "to" is never removed. So, we don't check rmdir
1865 * check whether the @pc is valid for moving account and call
1866 * __mem_cgroup_move_account()
1868 static int mem_cgroup_move_account(struct page_cgroup *pc,
1869 struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge)
1872 lock_page_cgroup(pc);
1873 if (PageCgroupUsed(pc) && pc->mem_cgroup == from) {
1874 __mem_cgroup_move_account(pc, from, to, uncharge);
1877 unlock_page_cgroup(pc);
1881 memcg_check_events(to, pc->page);
1882 memcg_check_events(from, pc->page);
1887 * move charges to its parent.
1890 static int mem_cgroup_move_parent(struct page_cgroup *pc,
1891 struct mem_cgroup *child,
1894 struct page *page = pc->page;
1895 struct cgroup *cg = child->css.cgroup;
1896 struct cgroup *pcg = cg->parent;
1897 struct mem_cgroup *parent;
1905 if (!get_page_unless_zero(page))
1907 if (isolate_lru_page(page))
1910 parent = mem_cgroup_from_cont(pcg);
1911 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
1915 ret = mem_cgroup_move_account(pc, child, parent, true);
1917 mem_cgroup_cancel_charge(parent);
1919 putback_lru_page(page);
1927 * Charge the memory controller for page usage.
1929 * 0 if the charge was successful
1930 * < 0 if the cgroup is over its limit
1932 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1933 gfp_t gfp_mask, enum charge_type ctype,
1934 struct mem_cgroup *memcg)
1936 struct mem_cgroup *mem;
1937 struct page_cgroup *pc;
1940 pc = lookup_page_cgroup(page);
1941 /* can happen at boot */
1947 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
1951 __mem_cgroup_commit_charge(mem, pc, ctype);
1955 int mem_cgroup_newpage_charge(struct page *page,
1956 struct mm_struct *mm, gfp_t gfp_mask)
1958 if (mem_cgroup_disabled())
1960 if (PageCompound(page))
1963 * If already mapped, we don't have to account.
1964 * If page cache, page->mapping has address_space.
1965 * But page->mapping may have out-of-use anon_vma pointer,
1966 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1969 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1973 return mem_cgroup_charge_common(page, mm, gfp_mask,
1974 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1978 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1979 enum charge_type ctype);
1981 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1984 struct mem_cgroup *mem = NULL;
1987 if (mem_cgroup_disabled())
1989 if (PageCompound(page))
1992 * Corner case handling. This is called from add_to_page_cache()
1993 * in usual. But some FS (shmem) precharges this page before calling it
1994 * and call add_to_page_cache() with GFP_NOWAIT.
1996 * For GFP_NOWAIT case, the page may be pre-charged before calling
1997 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1998 * charge twice. (It works but has to pay a bit larger cost.)
1999 * And when the page is SwapCache, it should take swap information
2000 * into account. This is under lock_page() now.
2002 if (!(gfp_mask & __GFP_WAIT)) {
2003 struct page_cgroup *pc;
2006 pc = lookup_page_cgroup(page);
2009 lock_page_cgroup(pc);
2010 if (PageCgroupUsed(pc)) {
2011 unlock_page_cgroup(pc);
2014 unlock_page_cgroup(pc);
2017 if (unlikely(!mm && !mem))
2020 if (page_is_file_cache(page))
2021 return mem_cgroup_charge_common(page, mm, gfp_mask,
2022 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
2025 if (PageSwapCache(page)) {
2026 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2028 __mem_cgroup_commit_charge_swapin(page, mem,
2029 MEM_CGROUP_CHARGE_TYPE_SHMEM);
2031 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
2032 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
2038 * While swap-in, try_charge -> commit or cancel, the page is locked.
2039 * And when try_charge() successfully returns, one refcnt to memcg without
2040 * struct page_cgroup is acquired. This refcnt will be consumed by
2041 * "commit()" or removed by "cancel()"
2043 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2045 gfp_t mask, struct mem_cgroup **ptr)
2047 struct mem_cgroup *mem;
2050 if (mem_cgroup_disabled())
2053 if (!do_swap_account)
2056 * A racing thread's fault, or swapoff, may have already updated
2057 * the pte, and even removed page from swap cache: in those cases
2058 * do_swap_page()'s pte_same() test will fail; but there's also a
2059 * KSM case which does need to charge the page.
2061 if (!PageSwapCache(page))
2063 mem = try_get_mem_cgroup_from_page(page);
2067 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
2068 /* drop extra refcnt from tryget */
2074 return __mem_cgroup_try_charge(mm, mask, ptr, true);
2078 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2079 enum charge_type ctype)
2081 struct page_cgroup *pc;
2083 if (mem_cgroup_disabled())
2087 cgroup_exclude_rmdir(&ptr->css);
2088 pc = lookup_page_cgroup(page);
2089 mem_cgroup_lru_del_before_commit_swapcache(page);
2090 __mem_cgroup_commit_charge(ptr, pc, ctype);
2091 mem_cgroup_lru_add_after_commit_swapcache(page);
2093 * Now swap is on-memory. This means this page may be
2094 * counted both as mem and swap....double count.
2095 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2096 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2097 * may call delete_from_swap_cache() before reach here.
2099 if (do_swap_account && PageSwapCache(page)) {
2100 swp_entry_t ent = {.val = page_private(page)};
2102 struct mem_cgroup *memcg;
2104 id = swap_cgroup_record(ent, 0);
2106 memcg = mem_cgroup_lookup(id);
2109 * This recorded memcg can be obsolete one. So, avoid
2110 * calling css_tryget
2112 if (!mem_cgroup_is_root(memcg))
2113 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
2114 mem_cgroup_swap_statistics(memcg, false);
2115 mem_cgroup_put(memcg);
2120 * At swapin, we may charge account against cgroup which has no tasks.
2121 * So, rmdir()->pre_destroy() can be called while we do this charge.
2122 * In that case, we need to call pre_destroy() again. check it here.
2124 cgroup_release_and_wakeup_rmdir(&ptr->css);
2127 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
2129 __mem_cgroup_commit_charge_swapin(page, ptr,
2130 MEM_CGROUP_CHARGE_TYPE_MAPPED);
2133 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
2135 if (mem_cgroup_disabled())
2139 mem_cgroup_cancel_charge(mem);
2143 __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype)
2145 struct memcg_batch_info *batch = NULL;
2146 bool uncharge_memsw = true;
2147 /* If swapout, usage of swap doesn't decrease */
2148 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2149 uncharge_memsw = false;
2151 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2152 * In those cases, all pages freed continously can be expected to be in
2153 * the same cgroup and we have chance to coalesce uncharges.
2154 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2155 * because we want to do uncharge as soon as possible.
2157 if (!current->memcg_batch.do_batch || test_thread_flag(TIF_MEMDIE))
2158 goto direct_uncharge;
2160 batch = ¤t->memcg_batch;
2162 * In usual, we do css_get() when we remember memcg pointer.
2163 * But in this case, we keep res->usage until end of a series of
2164 * uncharges. Then, it's ok to ignore memcg's refcnt.
2169 * In typical case, batch->memcg == mem. This means we can
2170 * merge a series of uncharges to an uncharge of res_counter.
2171 * If not, we uncharge res_counter ony by one.
2173 if (batch->memcg != mem)
2174 goto direct_uncharge;
2175 /* remember freed charge and uncharge it later */
2176 batch->bytes += PAGE_SIZE;
2178 batch->memsw_bytes += PAGE_SIZE;
2181 res_counter_uncharge(&mem->res, PAGE_SIZE);
2183 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
2188 * uncharge if !page_mapped(page)
2190 static struct mem_cgroup *
2191 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
2193 struct page_cgroup *pc;
2194 struct mem_cgroup *mem = NULL;
2195 struct mem_cgroup_per_zone *mz;
2197 if (mem_cgroup_disabled())
2200 if (PageSwapCache(page))
2204 * Check if our page_cgroup is valid
2206 pc = lookup_page_cgroup(page);
2207 if (unlikely(!pc || !PageCgroupUsed(pc)))
2210 lock_page_cgroup(pc);
2212 mem = pc->mem_cgroup;
2214 if (!PageCgroupUsed(pc))
2218 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2219 case MEM_CGROUP_CHARGE_TYPE_DROP:
2220 if (page_mapped(page))
2223 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
2224 if (!PageAnon(page)) { /* Shared memory */
2225 if (page->mapping && !page_is_file_cache(page))
2227 } else if (page_mapped(page)) /* Anon */
2234 if (!mem_cgroup_is_root(mem))
2235 __do_uncharge(mem, ctype);
2236 if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2237 mem_cgroup_swap_statistics(mem, true);
2238 mem_cgroup_charge_statistics(mem, pc, false);
2240 ClearPageCgroupUsed(pc);
2242 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2243 * freed from LRU. This is safe because uncharged page is expected not
2244 * to be reused (freed soon). Exception is SwapCache, it's handled by
2245 * special functions.
2248 mz = page_cgroup_zoneinfo(pc);
2249 unlock_page_cgroup(pc);
2251 memcg_check_events(mem, page);
2252 /* at swapout, this memcg will be accessed to record to swap */
2253 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2259 unlock_page_cgroup(pc);
2263 void mem_cgroup_uncharge_page(struct page *page)
2266 if (page_mapped(page))
2268 if (page->mapping && !PageAnon(page))
2270 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
2273 void mem_cgroup_uncharge_cache_page(struct page *page)
2275 VM_BUG_ON(page_mapped(page));
2276 VM_BUG_ON(page->mapping);
2277 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
2281 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2282 * In that cases, pages are freed continuously and we can expect pages
2283 * are in the same memcg. All these calls itself limits the number of
2284 * pages freed at once, then uncharge_start/end() is called properly.
2285 * This may be called prural(2) times in a context,
2288 void mem_cgroup_uncharge_start(void)
2290 current->memcg_batch.do_batch++;
2291 /* We can do nest. */
2292 if (current->memcg_batch.do_batch == 1) {
2293 current->memcg_batch.memcg = NULL;
2294 current->memcg_batch.bytes = 0;
2295 current->memcg_batch.memsw_bytes = 0;
2299 void mem_cgroup_uncharge_end(void)
2301 struct memcg_batch_info *batch = ¤t->memcg_batch;
2303 if (!batch->do_batch)
2307 if (batch->do_batch) /* If stacked, do nothing. */
2313 * This "batch->memcg" is valid without any css_get/put etc...
2314 * bacause we hide charges behind us.
2317 res_counter_uncharge(&batch->memcg->res, batch->bytes);
2318 if (batch->memsw_bytes)
2319 res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes);
2320 /* forget this pointer (for sanity check) */
2321 batch->memcg = NULL;
2326 * called after __delete_from_swap_cache() and drop "page" account.
2327 * memcg information is recorded to swap_cgroup of "ent"
2330 mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
2332 struct mem_cgroup *memcg;
2333 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
2335 if (!swapout) /* this was a swap cache but the swap is unused ! */
2336 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
2338 memcg = __mem_cgroup_uncharge_common(page, ctype);
2340 /* record memcg information */
2341 if (do_swap_account && swapout && memcg) {
2342 swap_cgroup_record(ent, css_id(&memcg->css));
2343 mem_cgroup_get(memcg);
2345 if (swapout && memcg)
2346 css_put(&memcg->css);
2350 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2352 * called from swap_entry_free(). remove record in swap_cgroup and
2353 * uncharge "memsw" account.
2355 void mem_cgroup_uncharge_swap(swp_entry_t ent)
2357 struct mem_cgroup *memcg;
2360 if (!do_swap_account)
2363 id = swap_cgroup_record(ent, 0);
2365 memcg = mem_cgroup_lookup(id);
2368 * We uncharge this because swap is freed.
2369 * This memcg can be obsolete one. We avoid calling css_tryget
2371 if (!mem_cgroup_is_root(memcg))
2372 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
2373 mem_cgroup_swap_statistics(memcg, false);
2374 mem_cgroup_put(memcg);
2380 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2381 * @entry: swap entry to be moved
2382 * @from: mem_cgroup which the entry is moved from
2383 * @to: mem_cgroup which the entry is moved to
2384 * @need_fixup: whether we should fixup res_counters and refcounts.
2386 * It succeeds only when the swap_cgroup's record for this entry is the same
2387 * as the mem_cgroup's id of @from.
2389 * Returns 0 on success, -EINVAL on failure.
2391 * The caller must have charged to @to, IOW, called res_counter_charge() about
2392 * both res and memsw, and called css_get().
2394 static int mem_cgroup_move_swap_account(swp_entry_t entry,
2395 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
2397 unsigned short old_id, new_id;
2399 old_id = css_id(&from->css);
2400 new_id = css_id(&to->css);
2402 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
2403 mem_cgroup_swap_statistics(from, false);
2404 mem_cgroup_swap_statistics(to, true);
2406 * This function is only called from task migration context now.
2407 * It postpones res_counter and refcount handling till the end
2408 * of task migration(mem_cgroup_clear_mc()) for performance
2409 * improvement. But we cannot postpone mem_cgroup_get(to)
2410 * because if the process that has been moved to @to does
2411 * swap-in, the refcount of @to might be decreased to 0.
2415 if (!mem_cgroup_is_root(from))
2416 res_counter_uncharge(&from->memsw, PAGE_SIZE);
2417 mem_cgroup_put(from);
2419 * we charged both to->res and to->memsw, so we should
2422 if (!mem_cgroup_is_root(to))
2423 res_counter_uncharge(&to->res, PAGE_SIZE);
2431 static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
2432 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
2439 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2442 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
2444 struct page_cgroup *pc;
2445 struct mem_cgroup *mem = NULL;
2448 if (mem_cgroup_disabled())
2451 pc = lookup_page_cgroup(page);
2452 lock_page_cgroup(pc);
2453 if (PageCgroupUsed(pc)) {
2454 mem = pc->mem_cgroup;
2457 unlock_page_cgroup(pc);
2461 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, ptr, false);
2467 /* remove redundant charge if migration failed*/
2468 void mem_cgroup_end_migration(struct mem_cgroup *mem,
2469 struct page *oldpage, struct page *newpage)
2471 struct page *target, *unused;
2472 struct page_cgroup *pc;
2473 enum charge_type ctype;
2477 cgroup_exclude_rmdir(&mem->css);
2478 /* at migration success, oldpage->mapping is NULL. */
2479 if (oldpage->mapping) {
2487 if (PageAnon(target))
2488 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
2489 else if (page_is_file_cache(target))
2490 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
2492 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
2494 /* unused page is not on radix-tree now. */
2496 __mem_cgroup_uncharge_common(unused, ctype);
2498 pc = lookup_page_cgroup(target);
2500 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
2501 * So, double-counting is effectively avoided.
2503 __mem_cgroup_commit_charge(mem, pc, ctype);
2506 * Both of oldpage and newpage are still under lock_page().
2507 * Then, we don't have to care about race in radix-tree.
2508 * But we have to be careful that this page is unmapped or not.
2510 * There is a case for !page_mapped(). At the start of
2511 * migration, oldpage was mapped. But now, it's zapped.
2512 * But we know *target* page is not freed/reused under us.
2513 * mem_cgroup_uncharge_page() does all necessary checks.
2515 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
2516 mem_cgroup_uncharge_page(target);
2518 * At migration, we may charge account against cgroup which has no tasks
2519 * So, rmdir()->pre_destroy() can be called while we do this charge.
2520 * In that case, we need to call pre_destroy() again. check it here.
2522 cgroup_release_and_wakeup_rmdir(&mem->css);
2526 * A call to try to shrink memory usage on charge failure at shmem's swapin.
2527 * Calling hierarchical_reclaim is not enough because we should update
2528 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
2529 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
2530 * not from the memcg which this page would be charged to.
2531 * try_charge_swapin does all of these works properly.
2533 int mem_cgroup_shmem_charge_fallback(struct page *page,
2534 struct mm_struct *mm,
2537 struct mem_cgroup *mem = NULL;
2540 if (mem_cgroup_disabled())
2543 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2545 mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
2550 static DEFINE_MUTEX(set_limit_mutex);
2552 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
2553 unsigned long long val)
2558 int children = mem_cgroup_count_children(memcg);
2559 u64 curusage, oldusage;
2562 * For keeping hierarchical_reclaim simple, how long we should retry
2563 * is depends on callers. We set our retry-count to be function
2564 * of # of children which we should visit in this loop.
2566 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
2568 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2570 while (retry_count) {
2571 if (signal_pending(current)) {
2576 * Rather than hide all in some function, I do this in
2577 * open coded manner. You see what this really does.
2578 * We have to guarantee mem->res.limit < mem->memsw.limit.
2580 mutex_lock(&set_limit_mutex);
2581 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2582 if (memswlimit < val) {
2584 mutex_unlock(&set_limit_mutex);
2587 ret = res_counter_set_limit(&memcg->res, val);
2589 if (memswlimit == val)
2590 memcg->memsw_is_minimum = true;
2592 memcg->memsw_is_minimum = false;
2594 mutex_unlock(&set_limit_mutex);
2599 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
2600 MEM_CGROUP_RECLAIM_SHRINK);
2601 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2602 /* Usage is reduced ? */
2603 if (curusage >= oldusage)
2606 oldusage = curusage;
2612 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2613 unsigned long long val)
2616 u64 memlimit, oldusage, curusage;
2617 int children = mem_cgroup_count_children(memcg);
2620 /* see mem_cgroup_resize_res_limit */
2621 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
2622 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2623 while (retry_count) {
2624 if (signal_pending(current)) {
2629 * Rather than hide all in some function, I do this in
2630 * open coded manner. You see what this really does.
2631 * We have to guarantee mem->res.limit < mem->memsw.limit.
2633 mutex_lock(&set_limit_mutex);
2634 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2635 if (memlimit > val) {
2637 mutex_unlock(&set_limit_mutex);
2640 ret = res_counter_set_limit(&memcg->memsw, val);
2642 if (memlimit == val)
2643 memcg->memsw_is_minimum = true;
2645 memcg->memsw_is_minimum = false;
2647 mutex_unlock(&set_limit_mutex);
2652 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
2653 MEM_CGROUP_RECLAIM_NOSWAP |
2654 MEM_CGROUP_RECLAIM_SHRINK);
2655 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2656 /* Usage is reduced ? */
2657 if (curusage >= oldusage)
2660 oldusage = curusage;
2665 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2666 gfp_t gfp_mask, int nid,
2669 unsigned long nr_reclaimed = 0;
2670 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2671 unsigned long reclaimed;
2673 struct mem_cgroup_tree_per_zone *mctz;
2674 unsigned long long excess;
2679 mctz = soft_limit_tree_node_zone(nid, zid);
2681 * This loop can run a while, specially if mem_cgroup's continuously
2682 * keep exceeding their soft limit and putting the system under
2689 mz = mem_cgroup_largest_soft_limit_node(mctz);
2693 reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone,
2695 MEM_CGROUP_RECLAIM_SOFT);
2696 nr_reclaimed += reclaimed;
2697 spin_lock(&mctz->lock);
2700 * If we failed to reclaim anything from this memory cgroup
2701 * it is time to move on to the next cgroup
2707 * Loop until we find yet another one.
2709 * By the time we get the soft_limit lock
2710 * again, someone might have aded the
2711 * group back on the RB tree. Iterate to
2712 * make sure we get a different mem.
2713 * mem_cgroup_largest_soft_limit_node returns
2714 * NULL if no other cgroup is present on
2718 __mem_cgroup_largest_soft_limit_node(mctz);
2719 if (next_mz == mz) {
2720 css_put(&next_mz->mem->css);
2722 } else /* next_mz == NULL or other memcg */
2726 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
2727 excess = res_counter_soft_limit_excess(&mz->mem->res);
2729 * One school of thought says that we should not add
2730 * back the node to the tree if reclaim returns 0.
2731 * But our reclaim could return 0, simply because due
2732 * to priority we are exposing a smaller subset of
2733 * memory to reclaim from. Consider this as a longer
2736 /* If excess == 0, no tree ops */
2737 __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
2738 spin_unlock(&mctz->lock);
2739 css_put(&mz->mem->css);
2742 * Could not reclaim anything and there are no more
2743 * mem cgroups to try or we seem to be looping without
2744 * reclaiming anything.
2746 if (!nr_reclaimed &&
2748 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2750 } while (!nr_reclaimed);
2752 css_put(&next_mz->mem->css);
2753 return nr_reclaimed;
2757 * This routine traverse page_cgroup in given list and drop them all.
2758 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
2760 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
2761 int node, int zid, enum lru_list lru)
2764 struct mem_cgroup_per_zone *mz;
2765 struct page_cgroup *pc, *busy;
2766 unsigned long flags, loop;
2767 struct list_head *list;
2770 zone = &NODE_DATA(node)->node_zones[zid];
2771 mz = mem_cgroup_zoneinfo(mem, node, zid);
2772 list = &mz->lists[lru];
2774 loop = MEM_CGROUP_ZSTAT(mz, lru);
2775 /* give some margin against EBUSY etc...*/
2780 spin_lock_irqsave(&zone->lru_lock, flags);
2781 if (list_empty(list)) {
2782 spin_unlock_irqrestore(&zone->lru_lock, flags);
2785 pc = list_entry(list->prev, struct page_cgroup, lru);
2787 list_move(&pc->lru, list);
2789 spin_unlock_irqrestore(&zone->lru_lock, flags);
2792 spin_unlock_irqrestore(&zone->lru_lock, flags);
2794 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
2798 if (ret == -EBUSY || ret == -EINVAL) {
2799 /* found lock contention or "pc" is obsolete. */
2806 if (!ret && !list_empty(list))
2812 * make mem_cgroup's charge to be 0 if there is no task.
2813 * This enables deleting this mem_cgroup.
2815 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
2818 int node, zid, shrink;
2819 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
2820 struct cgroup *cgrp = mem->css.cgroup;
2825 /* should free all ? */
2831 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
2834 if (signal_pending(current))
2836 /* This is for making all *used* pages to be on LRU. */
2837 lru_add_drain_all();
2838 drain_all_stock_sync();
2840 for_each_node_state(node, N_HIGH_MEMORY) {
2841 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
2844 ret = mem_cgroup_force_empty_list(mem,
2853 /* it seems parent cgroup doesn't have enough mem */
2857 /* "ret" should also be checked to ensure all lists are empty. */
2858 } while (mem->res.usage > 0 || ret);
2864 /* returns EBUSY if there is a task or if we come here twice. */
2865 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
2869 /* we call try-to-free pages for make this cgroup empty */
2870 lru_add_drain_all();
2871 /* try to free all pages in this cgroup */
2873 while (nr_retries && mem->res.usage > 0) {
2876 if (signal_pending(current)) {
2880 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
2881 false, get_swappiness(mem));
2884 /* maybe some writeback is necessary */
2885 congestion_wait(BLK_RW_ASYNC, HZ/10);
2890 /* try move_account...there may be some *locked* pages. */
2894 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
2896 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
2900 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
2902 return mem_cgroup_from_cont(cont)->use_hierarchy;
2905 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
2909 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2910 struct cgroup *parent = cont->parent;
2911 struct mem_cgroup *parent_mem = NULL;
2914 parent_mem = mem_cgroup_from_cont(parent);
2918 * If parent's use_hierarchy is set, we can't make any modifications
2919 * in the child subtrees. If it is unset, then the change can
2920 * occur, provided the current cgroup has no children.
2922 * For the root cgroup, parent_mem is NULL, we allow value to be
2923 * set if there are no children.
2925 if ((!parent_mem || !parent_mem->use_hierarchy) &&
2926 (val == 1 || val == 0)) {
2927 if (list_empty(&cont->children))
2928 mem->use_hierarchy = val;
2938 struct mem_cgroup_idx_data {
2940 enum mem_cgroup_stat_index idx;
2944 mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data)
2946 struct mem_cgroup_idx_data *d = data;
2947 d->val += mem_cgroup_read_stat(mem, d->idx);
2952 mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem,
2953 enum mem_cgroup_stat_index idx, s64 *val)
2955 struct mem_cgroup_idx_data d;
2958 mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat);
2962 static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap)
2966 if (!mem_cgroup_is_root(mem)) {
2968 return res_counter_read_u64(&mem->res, RES_USAGE);
2970 return res_counter_read_u64(&mem->memsw, RES_USAGE);
2973 mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE, &idx_val);
2975 mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS, &idx_val);
2979 mem_cgroup_get_recursive_idx_stat(mem,
2980 MEM_CGROUP_STAT_SWAPOUT, &idx_val);
2984 return val << PAGE_SHIFT;
2987 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
2989 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2993 type = MEMFILE_TYPE(cft->private);
2994 name = MEMFILE_ATTR(cft->private);
2997 if (name == RES_USAGE)
2998 val = mem_cgroup_usage(mem, false);
3000 val = res_counter_read_u64(&mem->res, name);
3003 if (name == RES_USAGE)
3004 val = mem_cgroup_usage(mem, true);
3006 val = res_counter_read_u64(&mem->memsw, name);
3015 * The user of this function is...
3018 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3021 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3023 unsigned long long val;
3026 type = MEMFILE_TYPE(cft->private);
3027 name = MEMFILE_ATTR(cft->private);
3030 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3034 /* This function does all necessary parse...reuse it */
3035 ret = res_counter_memparse_write_strategy(buffer, &val);
3039 ret = mem_cgroup_resize_limit(memcg, val);
3041 ret = mem_cgroup_resize_memsw_limit(memcg, val);
3043 case RES_SOFT_LIMIT:
3044 ret = res_counter_memparse_write_strategy(buffer, &val);
3048 * For memsw, soft limits are hard to implement in terms
3049 * of semantics, for now, we support soft limits for
3050 * control without swap
3053 ret = res_counter_set_soft_limit(&memcg->res, val);
3058 ret = -EINVAL; /* should be BUG() ? */
3064 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3065 unsigned long long *mem_limit, unsigned long long *memsw_limit)
3067 struct cgroup *cgroup;
3068 unsigned long long min_limit, min_memsw_limit, tmp;
3070 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3071 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3072 cgroup = memcg->css.cgroup;
3073 if (!memcg->use_hierarchy)
3076 while (cgroup->parent) {
3077 cgroup = cgroup->parent;
3078 memcg = mem_cgroup_from_cont(cgroup);
3079 if (!memcg->use_hierarchy)
3081 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
3082 min_limit = min(min_limit, tmp);
3083 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3084 min_memsw_limit = min(min_memsw_limit, tmp);
3087 *mem_limit = min_limit;
3088 *memsw_limit = min_memsw_limit;
3092 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
3094 struct mem_cgroup *mem;
3097 mem = mem_cgroup_from_cont(cont);
3098 type = MEMFILE_TYPE(event);
3099 name = MEMFILE_ATTR(event);
3103 res_counter_reset_max(&mem->res);
3105 res_counter_reset_max(&mem->memsw);
3109 res_counter_reset_failcnt(&mem->res);
3111 res_counter_reset_failcnt(&mem->memsw);
3118 static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
3121 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
3125 static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3126 struct cftype *cft, u64 val)
3128 struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
3130 if (val >= (1 << NR_MOVE_TYPE))
3133 * We check this value several times in both in can_attach() and
3134 * attach(), so we need cgroup lock to prevent this value from being
3138 mem->move_charge_at_immigrate = val;
3144 static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3145 struct cftype *cft, u64 val)
3152 /* For read statistics */
3168 struct mcs_total_stat {
3169 s64 stat[NR_MCS_STAT];
3175 } memcg_stat_strings[NR_MCS_STAT] = {
3176 {"cache", "total_cache"},
3177 {"rss", "total_rss"},
3178 {"mapped_file", "total_mapped_file"},
3179 {"pgpgin", "total_pgpgin"},
3180 {"pgpgout", "total_pgpgout"},
3181 {"swap", "total_swap"},
3182 {"inactive_anon", "total_inactive_anon"},
3183 {"active_anon", "total_active_anon"},
3184 {"inactive_file", "total_inactive_file"},
3185 {"active_file", "total_active_file"},
3186 {"unevictable", "total_unevictable"}
3190 static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
3192 struct mcs_total_stat *s = data;
3196 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
3197 s->stat[MCS_CACHE] += val * PAGE_SIZE;
3198 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
3199 s->stat[MCS_RSS] += val * PAGE_SIZE;
3200 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED);
3201 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
3202 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT);
3203 s->stat[MCS_PGPGIN] += val;
3204 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT);
3205 s->stat[MCS_PGPGOUT] += val;
3206 if (do_swap_account) {
3207 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT);
3208 s->stat[MCS_SWAP] += val * PAGE_SIZE;
3212 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
3213 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
3214 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
3215 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
3216 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
3217 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
3218 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
3219 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
3220 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
3221 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
3226 mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
3228 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
3231 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
3232 struct cgroup_map_cb *cb)
3234 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
3235 struct mcs_total_stat mystat;
3238 memset(&mystat, 0, sizeof(mystat));
3239 mem_cgroup_get_local_stat(mem_cont, &mystat);
3241 for (i = 0; i < NR_MCS_STAT; i++) {
3242 if (i == MCS_SWAP && !do_swap_account)
3244 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
3247 /* Hierarchical information */
3249 unsigned long long limit, memsw_limit;
3250 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
3251 cb->fill(cb, "hierarchical_memory_limit", limit);
3252 if (do_swap_account)
3253 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
3256 memset(&mystat, 0, sizeof(mystat));
3257 mem_cgroup_get_total_stat(mem_cont, &mystat);
3258 for (i = 0; i < NR_MCS_STAT; i++) {
3259 if (i == MCS_SWAP && !do_swap_account)
3261 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
3264 #ifdef CONFIG_DEBUG_VM
3265 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
3269 struct mem_cgroup_per_zone *mz;
3270 unsigned long recent_rotated[2] = {0, 0};
3271 unsigned long recent_scanned[2] = {0, 0};
3273 for_each_online_node(nid)
3274 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
3275 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
3277 recent_rotated[0] +=
3278 mz->reclaim_stat.recent_rotated[0];
3279 recent_rotated[1] +=
3280 mz->reclaim_stat.recent_rotated[1];
3281 recent_scanned[0] +=
3282 mz->reclaim_stat.recent_scanned[0];
3283 recent_scanned[1] +=
3284 mz->reclaim_stat.recent_scanned[1];
3286 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
3287 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
3288 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
3289 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
3296 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
3298 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3300 return get_swappiness(memcg);
3303 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
3306 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3307 struct mem_cgroup *parent;
3312 if (cgrp->parent == NULL)
3315 parent = mem_cgroup_from_cont(cgrp->parent);
3319 /* If under hierarchy, only empty-root can set this value */
3320 if ((parent->use_hierarchy) ||
3321 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
3326 spin_lock(&memcg->reclaim_param_lock);
3327 memcg->swappiness = val;
3328 spin_unlock(&memcg->reclaim_param_lock);
3335 static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3337 struct mem_cgroup_threshold_ary *t;
3343 t = rcu_dereference(memcg->thresholds);
3345 t = rcu_dereference(memcg->memsw_thresholds);
3350 usage = mem_cgroup_usage(memcg, swap);
3353 * current_threshold points to threshold just below usage.
3354 * If it's not true, a threshold was crossed after last
3355 * call of __mem_cgroup_threshold().
3357 i = atomic_read(&t->current_threshold);
3360 * Iterate backward over array of thresholds starting from
3361 * current_threshold and check if a threshold is crossed.
3362 * If none of thresholds below usage is crossed, we read
3363 * only one element of the array here.
3365 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3366 eventfd_signal(t->entries[i].eventfd, 1);
3368 /* i = current_threshold + 1 */
3372 * Iterate forward over array of thresholds starting from
3373 * current_threshold+1 and check if a threshold is crossed.
3374 * If none of thresholds above usage is crossed, we read
3375 * only one element of the array here.
3377 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3378 eventfd_signal(t->entries[i].eventfd, 1);
3380 /* Update current_threshold */
3381 atomic_set(&t->current_threshold, i - 1);
3386 static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3388 __mem_cgroup_threshold(memcg, false);
3389 if (do_swap_account)
3390 __mem_cgroup_threshold(memcg, true);
3393 static int compare_thresholds(const void *a, const void *b)
3395 const struct mem_cgroup_threshold *_a = a;
3396 const struct mem_cgroup_threshold *_b = b;
3398 return _a->threshold - _b->threshold;
3401 static int mem_cgroup_register_event(struct cgroup *cgrp, struct cftype *cft,
3402 struct eventfd_ctx *eventfd, const char *args)
3404 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3405 struct mem_cgroup_threshold_ary *thresholds, *thresholds_new;
3406 int type = MEMFILE_TYPE(cft->private);
3407 u64 threshold, usage;
3411 ret = res_counter_memparse_write_strategy(args, &threshold);
3415 mutex_lock(&memcg->thresholds_lock);
3417 thresholds = memcg->thresholds;
3418 else if (type == _MEMSWAP)
3419 thresholds = memcg->memsw_thresholds;
3423 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
3425 /* Check if a threshold crossed before adding a new one */
3427 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3430 size = thresholds->size + 1;
3434 /* Allocate memory for new array of thresholds */
3435 thresholds_new = kmalloc(sizeof(*thresholds_new) +
3436 size * sizeof(struct mem_cgroup_threshold),
3438 if (!thresholds_new) {
3442 thresholds_new->size = size;
3444 /* Copy thresholds (if any) to new array */
3446 memcpy(thresholds_new->entries, thresholds->entries,
3448 sizeof(struct mem_cgroup_threshold));
3449 /* Add new threshold */
3450 thresholds_new->entries[size - 1].eventfd = eventfd;
3451 thresholds_new->entries[size - 1].threshold = threshold;
3453 /* Sort thresholds. Registering of new threshold isn't time-critical */
3454 sort(thresholds_new->entries, size,
3455 sizeof(struct mem_cgroup_threshold),
3456 compare_thresholds, NULL);
3458 /* Find current threshold */
3459 atomic_set(&thresholds_new->current_threshold, -1);
3460 for (i = 0; i < size; i++) {
3461 if (thresholds_new->entries[i].threshold < usage) {
3463 * thresholds_new->current_threshold will not be used
3464 * until rcu_assign_pointer(), so it's safe to increment
3467 atomic_inc(&thresholds_new->current_threshold);
3472 rcu_assign_pointer(memcg->thresholds, thresholds_new);
3474 rcu_assign_pointer(memcg->memsw_thresholds, thresholds_new);
3476 /* To be sure that nobody uses thresholds before freeing it */
3481 mutex_unlock(&memcg->thresholds_lock);
3486 static int mem_cgroup_unregister_event(struct cgroup *cgrp, struct cftype *cft,
3487 struct eventfd_ctx *eventfd)
3489 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3490 struct mem_cgroup_threshold_ary *thresholds, *thresholds_new;
3491 int type = MEMFILE_TYPE(cft->private);
3496 mutex_lock(&memcg->thresholds_lock);
3498 thresholds = memcg->thresholds;
3499 else if (type == _MEMSWAP)
3500 thresholds = memcg->memsw_thresholds;
3505 * Something went wrong if we trying to unregister a threshold
3506 * if we don't have thresholds
3508 BUG_ON(!thresholds);
3510 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
3512 /* Check if a threshold crossed before removing */
3513 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3515 /* Calculate new number of threshold */
3516 for (i = 0; i < thresholds->size; i++) {
3517 if (thresholds->entries[i].eventfd != eventfd)
3521 /* Set thresholds array to NULL if we don't have thresholds */
3523 thresholds_new = NULL;
3527 /* Allocate memory for new array of thresholds */
3528 thresholds_new = kmalloc(sizeof(*thresholds_new) +
3529 size * sizeof(struct mem_cgroup_threshold),
3531 if (!thresholds_new) {
3535 thresholds_new->size = size;
3537 /* Copy thresholds and find current threshold */
3538 atomic_set(&thresholds_new->current_threshold, -1);
3539 for (i = 0, j = 0; i < thresholds->size; i++) {
3540 if (thresholds->entries[i].eventfd == eventfd)
3543 thresholds_new->entries[j] = thresholds->entries[i];
3544 if (thresholds_new->entries[j].threshold < usage) {
3546 * thresholds_new->current_threshold will not be used
3547 * until rcu_assign_pointer(), so it's safe to increment
3550 atomic_inc(&thresholds_new->current_threshold);
3557 rcu_assign_pointer(memcg->thresholds, thresholds_new);
3559 rcu_assign_pointer(memcg->memsw_thresholds, thresholds_new);
3561 /* To be sure that nobody uses thresholds before freeing it */
3566 mutex_unlock(&memcg->thresholds_lock);
3571 static struct cftype mem_cgroup_files[] = {
3573 .name = "usage_in_bytes",
3574 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
3575 .read_u64 = mem_cgroup_read,
3576 .register_event = mem_cgroup_register_event,
3577 .unregister_event = mem_cgroup_unregister_event,
3580 .name = "max_usage_in_bytes",
3581 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
3582 .trigger = mem_cgroup_reset,
3583 .read_u64 = mem_cgroup_read,
3586 .name = "limit_in_bytes",
3587 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
3588 .write_string = mem_cgroup_write,
3589 .read_u64 = mem_cgroup_read,
3592 .name = "soft_limit_in_bytes",
3593 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
3594 .write_string = mem_cgroup_write,
3595 .read_u64 = mem_cgroup_read,
3599 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
3600 .trigger = mem_cgroup_reset,
3601 .read_u64 = mem_cgroup_read,
3605 .read_map = mem_control_stat_show,
3608 .name = "force_empty",
3609 .trigger = mem_cgroup_force_empty_write,
3612 .name = "use_hierarchy",
3613 .write_u64 = mem_cgroup_hierarchy_write,
3614 .read_u64 = mem_cgroup_hierarchy_read,
3617 .name = "swappiness",
3618 .read_u64 = mem_cgroup_swappiness_read,
3619 .write_u64 = mem_cgroup_swappiness_write,
3622 .name = "move_charge_at_immigrate",
3623 .read_u64 = mem_cgroup_move_charge_read,
3624 .write_u64 = mem_cgroup_move_charge_write,
3628 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3629 static struct cftype memsw_cgroup_files[] = {
3631 .name = "memsw.usage_in_bytes",
3632 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
3633 .read_u64 = mem_cgroup_read,
3634 .register_event = mem_cgroup_register_event,
3635 .unregister_event = mem_cgroup_unregister_event,
3638 .name = "memsw.max_usage_in_bytes",
3639 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
3640 .trigger = mem_cgroup_reset,
3641 .read_u64 = mem_cgroup_read,
3644 .name = "memsw.limit_in_bytes",
3645 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
3646 .write_string = mem_cgroup_write,
3647 .read_u64 = mem_cgroup_read,
3650 .name = "memsw.failcnt",
3651 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
3652 .trigger = mem_cgroup_reset,
3653 .read_u64 = mem_cgroup_read,
3657 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
3659 if (!do_swap_account)
3661 return cgroup_add_files(cont, ss, memsw_cgroup_files,
3662 ARRAY_SIZE(memsw_cgroup_files));
3665 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
3671 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
3673 struct mem_cgroup_per_node *pn;
3674 struct mem_cgroup_per_zone *mz;
3676 int zone, tmp = node;
3678 * This routine is called against possible nodes.
3679 * But it's BUG to call kmalloc() against offline node.
3681 * TODO: this routine can waste much memory for nodes which will
3682 * never be onlined. It's better to use memory hotplug callback
3685 if (!node_state(node, N_NORMAL_MEMORY))
3687 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
3691 mem->info.nodeinfo[node] = pn;
3692 memset(pn, 0, sizeof(*pn));
3694 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
3695 mz = &pn->zoneinfo[zone];
3697 INIT_LIST_HEAD(&mz->lists[l]);
3698 mz->usage_in_excess = 0;
3699 mz->on_tree = false;
3705 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
3707 kfree(mem->info.nodeinfo[node]);
3710 static struct mem_cgroup *mem_cgroup_alloc(void)
3712 struct mem_cgroup *mem;
3713 int size = sizeof(struct mem_cgroup);
3715 /* Can be very big if MAX_NUMNODES is very big */
3716 if (size < PAGE_SIZE)
3717 mem = kmalloc(size, GFP_KERNEL);
3719 mem = vmalloc(size);
3724 memset(mem, 0, size);
3725 mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
3727 if (size < PAGE_SIZE)
3737 * At destroying mem_cgroup, references from swap_cgroup can remain.
3738 * (scanning all at force_empty is too costly...)
3740 * Instead of clearing all references at force_empty, we remember
3741 * the number of reference from swap_cgroup and free mem_cgroup when
3742 * it goes down to 0.
3744 * Removal of cgroup itself succeeds regardless of refs from swap.
3747 static void __mem_cgroup_free(struct mem_cgroup *mem)
3751 mem_cgroup_remove_from_trees(mem);
3752 free_css_id(&mem_cgroup_subsys, &mem->css);
3754 for_each_node_state(node, N_POSSIBLE)
3755 free_mem_cgroup_per_zone_info(mem, node);
3757 free_percpu(mem->stat);
3758 if (sizeof(struct mem_cgroup) < PAGE_SIZE)
3764 static void mem_cgroup_get(struct mem_cgroup *mem)
3766 atomic_inc(&mem->refcnt);
3769 static void __mem_cgroup_put(struct mem_cgroup *mem, int count)
3771 if (atomic_sub_and_test(count, &mem->refcnt)) {
3772 struct mem_cgroup *parent = parent_mem_cgroup(mem);
3773 __mem_cgroup_free(mem);
3775 mem_cgroup_put(parent);
3779 static void mem_cgroup_put(struct mem_cgroup *mem)
3781 __mem_cgroup_put(mem, 1);
3785 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
3787 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
3789 if (!mem->res.parent)
3791 return mem_cgroup_from_res_counter(mem->res.parent, res);
3794 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3795 static void __init enable_swap_cgroup(void)
3797 if (!mem_cgroup_disabled() && really_do_swap_account)
3798 do_swap_account = 1;
3801 static void __init enable_swap_cgroup(void)
3806 static int mem_cgroup_soft_limit_tree_init(void)
3808 struct mem_cgroup_tree_per_node *rtpn;
3809 struct mem_cgroup_tree_per_zone *rtpz;
3810 int tmp, node, zone;
3812 for_each_node_state(node, N_POSSIBLE) {
3814 if (!node_state(node, N_NORMAL_MEMORY))
3816 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
3820 soft_limit_tree.rb_tree_per_node[node] = rtpn;
3822 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
3823 rtpz = &rtpn->rb_tree_per_zone[zone];
3824 rtpz->rb_root = RB_ROOT;
3825 spin_lock_init(&rtpz->lock);
3831 static struct cgroup_subsys_state * __ref
3832 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
3834 struct mem_cgroup *mem, *parent;
3835 long error = -ENOMEM;
3838 mem = mem_cgroup_alloc();
3840 return ERR_PTR(error);
3842 for_each_node_state(node, N_POSSIBLE)
3843 if (alloc_mem_cgroup_per_zone_info(mem, node))
3847 if (cont->parent == NULL) {
3849 enable_swap_cgroup();
3851 root_mem_cgroup = mem;
3852 if (mem_cgroup_soft_limit_tree_init())
3854 for_each_possible_cpu(cpu) {
3855 struct memcg_stock_pcp *stock =
3856 &per_cpu(memcg_stock, cpu);
3857 INIT_WORK(&stock->work, drain_local_stock);
3859 hotcpu_notifier(memcg_stock_cpu_callback, 0);
3861 parent = mem_cgroup_from_cont(cont->parent);
3862 mem->use_hierarchy = parent->use_hierarchy;
3865 if (parent && parent->use_hierarchy) {
3866 res_counter_init(&mem->res, &parent->res);
3867 res_counter_init(&mem->memsw, &parent->memsw);
3869 * We increment refcnt of the parent to ensure that we can
3870 * safely access it on res_counter_charge/uncharge.
3871 * This refcnt will be decremented when freeing this
3872 * mem_cgroup(see mem_cgroup_put).
3874 mem_cgroup_get(parent);
3876 res_counter_init(&mem->res, NULL);
3877 res_counter_init(&mem->memsw, NULL);
3879 mem->last_scanned_child = 0;
3880 spin_lock_init(&mem->reclaim_param_lock);
3883 mem->swappiness = get_swappiness(parent);
3884 atomic_set(&mem->refcnt, 1);
3885 mem->move_charge_at_immigrate = 0;
3886 mutex_init(&mem->thresholds_lock);
3889 __mem_cgroup_free(mem);
3890 root_mem_cgroup = NULL;
3891 return ERR_PTR(error);
3894 static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
3895 struct cgroup *cont)
3897 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3899 return mem_cgroup_force_empty(mem, false);
3902 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
3903 struct cgroup *cont)
3905 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3907 mem_cgroup_put(mem);
3910 static int mem_cgroup_populate(struct cgroup_subsys *ss,
3911 struct cgroup *cont)
3915 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
3916 ARRAY_SIZE(mem_cgroup_files));
3919 ret = register_memsw_files(cont, ss);
3924 /* Handlers for move charge at task migration. */
3925 #define PRECHARGE_COUNT_AT_ONCE 256
3926 static int mem_cgroup_do_precharge(unsigned long count)
3929 int batch_count = PRECHARGE_COUNT_AT_ONCE;
3930 struct mem_cgroup *mem = mc.to;
3932 if (mem_cgroup_is_root(mem)) {
3933 mc.precharge += count;
3934 /* we don't need css_get for root */
3937 /* try to charge at once */
3939 struct res_counter *dummy;
3941 * "mem" cannot be under rmdir() because we've already checked
3942 * by cgroup_lock_live_cgroup() that it is not removed and we
3943 * are still under the same cgroup_mutex. So we can postpone
3946 if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy))
3948 if (do_swap_account && res_counter_charge(&mem->memsw,
3949 PAGE_SIZE * count, &dummy)) {
3950 res_counter_uncharge(&mem->res, PAGE_SIZE * count);
3953 mc.precharge += count;
3954 VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags));
3955 WARN_ON_ONCE(count > INT_MAX);
3956 __css_get(&mem->css, (int)count);
3960 /* fall back to one by one charge */
3962 if (signal_pending(current)) {
3966 if (!batch_count--) {
3967 batch_count = PRECHARGE_COUNT_AT_ONCE;
3970 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
3972 /* mem_cgroup_clear_mc() will do uncharge later */
3980 * is_target_pte_for_mc - check a pte whether it is valid for move charge
3981 * @vma: the vma the pte to be checked belongs
3982 * @addr: the address corresponding to the pte to be checked
3983 * @ptent: the pte to be checked
3984 * @target: the pointer the target page or swap ent will be stored(can be NULL)
3987 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
3988 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
3989 * move charge. if @target is not NULL, the page is stored in target->page
3990 * with extra refcnt got(Callers should handle it).
3991 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
3992 * target for charge migration. if @target is not NULL, the entry is stored
3995 * Called with pte lock held.
4002 enum mc_target_type {
4003 MC_TARGET_NONE, /* not used */
4008 static int is_target_pte_for_mc(struct vm_area_struct *vma,
4009 unsigned long addr, pte_t ptent, union mc_target *target)
4011 struct page *page = NULL;
4012 struct page_cgroup *pc;
4014 swp_entry_t ent = { .val = 0 };
4015 int usage_count = 0;
4016 bool move_anon = test_bit(MOVE_CHARGE_TYPE_ANON,
4017 &mc.to->move_charge_at_immigrate);
4019 if (!pte_present(ptent)) {
4020 /* TODO: handle swap of shmes/tmpfs */
4021 if (pte_none(ptent) || pte_file(ptent))
4023 else if (is_swap_pte(ptent)) {
4024 ent = pte_to_swp_entry(ptent);
4025 if (!move_anon || non_swap_entry(ent))
4027 usage_count = mem_cgroup_count_swap_user(ent, &page);
4030 page = vm_normal_page(vma, addr, ptent);
4031 if (!page || !page_mapped(page))
4034 * TODO: We don't move charges of file(including shmem/tmpfs)
4037 if (!move_anon || !PageAnon(page))
4039 if (!get_page_unless_zero(page))
4041 usage_count = page_mapcount(page);
4043 if (usage_count > 1) {
4045 * TODO: We don't move charges of shared(used by multiple
4046 * processes) pages for now.
4053 pc = lookup_page_cgroup(page);
4055 * Do only loose check w/o page_cgroup lock.
4056 * mem_cgroup_move_account() checks the pc is valid or not under
4059 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
4060 ret = MC_TARGET_PAGE;
4062 target->page = page;
4064 if (!ret || !target)
4068 if (ent.val && do_swap_account && !ret &&
4069 css_id(&mc.from->css) == lookup_swap_cgroup(ent)) {
4070 ret = MC_TARGET_SWAP;
4077 static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4078 unsigned long addr, unsigned long end,
4079 struct mm_walk *walk)
4081 struct vm_area_struct *vma = walk->private;
4085 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4086 for (; addr != end; pte++, addr += PAGE_SIZE)
4087 if (is_target_pte_for_mc(vma, addr, *pte, NULL))
4088 mc.precharge++; /* increment precharge temporarily */
4089 pte_unmap_unlock(pte - 1, ptl);
4095 static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4097 unsigned long precharge;
4098 struct vm_area_struct *vma;
4100 down_read(&mm->mmap_sem);
4101 for (vma = mm->mmap; vma; vma = vma->vm_next) {
4102 struct mm_walk mem_cgroup_count_precharge_walk = {
4103 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4107 if (is_vm_hugetlb_page(vma))
4109 /* TODO: We don't move charges of shmem/tmpfs pages for now. */
4110 if (vma->vm_flags & VM_SHARED)
4112 walk_page_range(vma->vm_start, vma->vm_end,
4113 &mem_cgroup_count_precharge_walk);
4115 up_read(&mm->mmap_sem);
4117 precharge = mc.precharge;
4123 static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4125 return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm));
4128 static void mem_cgroup_clear_mc(void)
4130 /* we must uncharge all the leftover precharges from mc.to */
4132 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
4136 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4137 * we must uncharge here.
4139 if (mc.moved_charge) {
4140 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
4141 mc.moved_charge = 0;
4143 /* we must fixup refcnts and charges */
4144 if (mc.moved_swap) {
4145 WARN_ON_ONCE(mc.moved_swap > INT_MAX);
4146 /* uncharge swap account from the old cgroup */
4147 if (!mem_cgroup_is_root(mc.from))
4148 res_counter_uncharge(&mc.from->memsw,
4149 PAGE_SIZE * mc.moved_swap);
4150 __mem_cgroup_put(mc.from, mc.moved_swap);
4152 if (!mem_cgroup_is_root(mc.to)) {
4154 * we charged both to->res and to->memsw, so we should
4157 res_counter_uncharge(&mc.to->res,
4158 PAGE_SIZE * mc.moved_swap);
4159 VM_BUG_ON(test_bit(CSS_ROOT, &mc.to->css.flags));
4160 __css_put(&mc.to->css, mc.moved_swap);
4162 /* we've already done mem_cgroup_get(mc.to) */
4168 mc.moving_task = NULL;
4169 wake_up_all(&mc.waitq);
4172 static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
4173 struct cgroup *cgroup,
4174 struct task_struct *p,
4178 struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup);
4180 if (mem->move_charge_at_immigrate) {
4181 struct mm_struct *mm;
4182 struct mem_cgroup *from = mem_cgroup_from_task(p);
4184 VM_BUG_ON(from == mem);
4186 mm = get_task_mm(p);
4189 /* We move charges only when we move a owner of the mm */
4190 if (mm->owner == p) {
4193 VM_BUG_ON(mc.precharge);
4194 VM_BUG_ON(mc.moved_charge);
4195 VM_BUG_ON(mc.moved_swap);
4196 VM_BUG_ON(mc.moving_task);
4200 mc.moved_charge = 0;
4202 mc.moving_task = current;
4204 ret = mem_cgroup_precharge_mc(mm);
4206 mem_cgroup_clear_mc();
4213 static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
4214 struct cgroup *cgroup,
4215 struct task_struct *p,
4218 mem_cgroup_clear_mc();
4221 static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4222 unsigned long addr, unsigned long end,
4223 struct mm_walk *walk)
4226 struct vm_area_struct *vma = walk->private;
4231 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4232 for (; addr != end; addr += PAGE_SIZE) {
4233 pte_t ptent = *(pte++);
4234 union mc_target target;
4237 struct page_cgroup *pc;
4243 type = is_target_pte_for_mc(vma, addr, ptent, &target);
4245 case MC_TARGET_PAGE:
4247 if (isolate_lru_page(page))
4249 pc = lookup_page_cgroup(page);
4250 if (!mem_cgroup_move_account(pc,
4251 mc.from, mc.to, false)) {
4253 /* we uncharge from mc.from later. */
4256 putback_lru_page(page);
4257 put: /* is_target_pte_for_mc() gets the page */
4260 case MC_TARGET_SWAP:
4262 if (!mem_cgroup_move_swap_account(ent,
4263 mc.from, mc.to, false)) {
4265 /* we fixup refcnts and charges later. */
4273 pte_unmap_unlock(pte - 1, ptl);
4278 * We have consumed all precharges we got in can_attach().
4279 * We try charge one by one, but don't do any additional
4280 * charges to mc.to if we have failed in charge once in attach()
4283 ret = mem_cgroup_do_precharge(1);
4291 static void mem_cgroup_move_charge(struct mm_struct *mm)
4293 struct vm_area_struct *vma;
4295 lru_add_drain_all();
4296 down_read(&mm->mmap_sem);
4297 for (vma = mm->mmap; vma; vma = vma->vm_next) {
4299 struct mm_walk mem_cgroup_move_charge_walk = {
4300 .pmd_entry = mem_cgroup_move_charge_pte_range,
4304 if (is_vm_hugetlb_page(vma))
4306 /* TODO: We don't move charges of shmem/tmpfs pages for now. */
4307 if (vma->vm_flags & VM_SHARED)
4309 ret = walk_page_range(vma->vm_start, vma->vm_end,
4310 &mem_cgroup_move_charge_walk);
4313 * means we have consumed all precharges and failed in
4314 * doing additional charge. Just abandon here.
4318 up_read(&mm->mmap_sem);
4321 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
4322 struct cgroup *cont,
4323 struct cgroup *old_cont,
4324 struct task_struct *p,
4327 struct mm_struct *mm;
4330 /* no need to move charge */
4333 mm = get_task_mm(p);
4335 mem_cgroup_move_charge(mm);
4338 mem_cgroup_clear_mc();
4340 #else /* !CONFIG_MMU */
4341 static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
4342 struct cgroup *cgroup,
4343 struct task_struct *p,
4348 static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
4349 struct cgroup *cgroup,
4350 struct task_struct *p,
4354 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
4355 struct cgroup *cont,
4356 struct cgroup *old_cont,
4357 struct task_struct *p,
4363 struct cgroup_subsys mem_cgroup_subsys = {
4365 .subsys_id = mem_cgroup_subsys_id,
4366 .create = mem_cgroup_create,
4367 .pre_destroy = mem_cgroup_pre_destroy,
4368 .destroy = mem_cgroup_destroy,
4369 .populate = mem_cgroup_populate,
4370 .can_attach = mem_cgroup_can_attach,
4371 .cancel_attach = mem_cgroup_cancel_attach,
4372 .attach = mem_cgroup_move_task,
4377 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4379 static int __init disable_swap_account(char *s)
4381 really_do_swap_account = 0;
4384 __setup("noswapaccount", disable_swap_account);