1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
24 #include <linux/res_counter.h>
25 #include <linux/memcontrol.h>
26 #include <linux/cgroup.h>
28 #include <linux/hugetlb.h>
29 #include <linux/pagemap.h>
30 #include <linux/smp.h>
31 #include <linux/page-flags.h>
32 #include <linux/backing-dev.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/rcupdate.h>
35 #include <linux/limits.h>
36 #include <linux/mutex.h>
37 #include <linux/rbtree.h>
38 #include <linux/slab.h>
39 #include <linux/swap.h>
40 #include <linux/swapops.h>
41 #include <linux/spinlock.h>
42 #include <linux/eventfd.h>
43 #include <linux/sort.h>
45 #include <linux/seq_file.h>
46 #include <linux/vmalloc.h>
47 #include <linux/mm_inline.h>
48 #include <linux/page_cgroup.h>
49 #include <linux/cpu.h>
50 #include <linux/oom.h>
53 #include <asm/uaccess.h>
55 #include <trace/events/vmscan.h>
57 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
58 #define MEM_CGROUP_RECLAIM_RETRIES 5
59 struct mem_cgroup *root_mem_cgroup __read_mostly;
61 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
62 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
63 int do_swap_account __read_mostly;
65 /* for remember boot option*/
66 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
67 static int really_do_swap_account __initdata = 1;
69 static int really_do_swap_account __initdata = 0;
73 #define do_swap_account (0)
77 * Per memcg event counter is incremented at every pagein/pageout. This counter
78 * is used for trigger some periodic events. This is straightforward and better
79 * than using jiffies etc. to handle periodic memcg event.
81 * These values will be used as !((event) & ((1 <<(thresh)) - 1))
83 #define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */
84 #define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */
87 * Statistics for memory cgroup.
89 enum mem_cgroup_stat_index {
91 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
93 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
94 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
95 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
96 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
97 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
98 MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
99 MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */
100 /* incremented at every pagein/pageout */
101 MEM_CGROUP_EVENTS = MEM_CGROUP_STAT_DATA,
102 MEM_CGROUP_ON_MOVE, /* someone is moving account between groups */
104 MEM_CGROUP_STAT_NSTATS,
107 struct mem_cgroup_stat_cpu {
108 s64 count[MEM_CGROUP_STAT_NSTATS];
112 * per-zone information in memory controller.
114 struct mem_cgroup_per_zone {
116 * spin_lock to protect the per cgroup LRU
118 struct list_head lists[NR_LRU_LISTS];
119 unsigned long count[NR_LRU_LISTS];
121 struct zone_reclaim_stat reclaim_stat;
122 struct rb_node tree_node; /* RB tree node */
123 unsigned long long usage_in_excess;/* Set to the value by which */
124 /* the soft limit is exceeded*/
126 struct mem_cgroup *mem; /* Back pointer, we cannot */
127 /* use container_of */
129 /* Macro for accessing counter */
130 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
132 struct mem_cgroup_per_node {
133 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
136 struct mem_cgroup_lru_info {
137 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
141 * Cgroups above their limits are maintained in a RB-Tree, independent of
142 * their hierarchy representation
145 struct mem_cgroup_tree_per_zone {
146 struct rb_root rb_root;
150 struct mem_cgroup_tree_per_node {
151 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
154 struct mem_cgroup_tree {
155 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
158 static struct mem_cgroup_tree soft_limit_tree __read_mostly;
160 struct mem_cgroup_threshold {
161 struct eventfd_ctx *eventfd;
166 struct mem_cgroup_threshold_ary {
167 /* An array index points to threshold just below usage. */
168 int current_threshold;
169 /* Size of entries[] */
171 /* Array of thresholds */
172 struct mem_cgroup_threshold entries[0];
175 struct mem_cgroup_thresholds {
176 /* Primary thresholds array */
177 struct mem_cgroup_threshold_ary *primary;
179 * Spare threshold array.
180 * This is needed to make mem_cgroup_unregister_event() "never fail".
181 * It must be able to store at least primary->size - 1 entries.
183 struct mem_cgroup_threshold_ary *spare;
187 struct mem_cgroup_eventfd_list {
188 struct list_head list;
189 struct eventfd_ctx *eventfd;
192 static void mem_cgroup_threshold(struct mem_cgroup *mem);
193 static void mem_cgroup_oom_notify(struct mem_cgroup *mem);
196 * The memory controller data structure. The memory controller controls both
197 * page cache and RSS per cgroup. We would eventually like to provide
198 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
199 * to help the administrator determine what knobs to tune.
201 * TODO: Add a water mark for the memory controller. Reclaim will begin when
202 * we hit the water mark. May be even add a low water mark, such that
203 * no reclaim occurs from a cgroup at it's low water mark, this is
204 * a feature that will be implemented much later in the future.
207 struct cgroup_subsys_state css;
209 * the counter to account for memory usage
211 struct res_counter res;
213 * the counter to account for mem+swap usage.
215 struct res_counter memsw;
217 * Per cgroup active and inactive list, similar to the
218 * per zone LRU lists.
220 struct mem_cgroup_lru_info info;
223 protect against reclaim related member.
225 spinlock_t reclaim_param_lock;
228 * While reclaiming in a hierarchy, we cache the last child we
231 int last_scanned_child;
233 * Should the accounting and control be hierarchical, per subtree?
239 unsigned int swappiness;
240 /* OOM-Killer disable */
241 int oom_kill_disable;
243 /* set when res.limit == memsw.limit */
244 bool memsw_is_minimum;
246 /* protect arrays of thresholds */
247 struct mutex thresholds_lock;
249 /* thresholds for memory usage. RCU-protected */
250 struct mem_cgroup_thresholds thresholds;
252 /* thresholds for mem+swap usage. RCU-protected */
253 struct mem_cgroup_thresholds memsw_thresholds;
255 /* For oom notifier event fd */
256 struct list_head oom_notify;
259 * Should we move charges of a task when a task is moved into this
260 * mem_cgroup ? And what type of charges should we move ?
262 unsigned long move_charge_at_immigrate;
266 struct mem_cgroup_stat_cpu *stat;
268 * used when a cpu is offlined or other synchronizations
269 * See mem_cgroup_read_stat().
271 struct mem_cgroup_stat_cpu nocpu_base;
272 spinlock_t pcp_counter_lock;
275 /* Stuffs for move charges at task migration. */
277 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
278 * left-shifted bitmap of these types.
281 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
282 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
286 /* "mc" and its members are protected by cgroup_mutex */
287 static struct move_charge_struct {
288 spinlock_t lock; /* for from, to */
289 struct mem_cgroup *from;
290 struct mem_cgroup *to;
291 unsigned long precharge;
292 unsigned long moved_charge;
293 unsigned long moved_swap;
294 struct task_struct *moving_task; /* a task moving charges */
295 wait_queue_head_t waitq; /* a waitq for other context */
297 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
298 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
301 static bool move_anon(void)
303 return test_bit(MOVE_CHARGE_TYPE_ANON,
304 &mc.to->move_charge_at_immigrate);
307 static bool move_file(void)
309 return test_bit(MOVE_CHARGE_TYPE_FILE,
310 &mc.to->move_charge_at_immigrate);
314 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
315 * limit reclaim to prevent infinite loops, if they ever occur.
317 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
318 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
321 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
322 MEM_CGROUP_CHARGE_TYPE_MAPPED,
323 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
324 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
325 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
326 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
330 /* only for here (for easy reading.) */
331 #define PCGF_CACHE (1UL << PCG_CACHE)
332 #define PCGF_USED (1UL << PCG_USED)
333 #define PCGF_LOCK (1UL << PCG_LOCK)
334 /* Not used, but added here for completeness */
335 #define PCGF_ACCT (1UL << PCG_ACCT)
337 /* for encoding cft->private value on file */
340 #define _OOM_TYPE (2)
341 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
342 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
343 #define MEMFILE_ATTR(val) ((val) & 0xffff)
344 /* Used for OOM nofiier */
345 #define OOM_CONTROL (0)
348 * Reclaim flags for mem_cgroup_hierarchical_reclaim
350 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
351 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
352 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
353 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
354 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
355 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
357 static void mem_cgroup_get(struct mem_cgroup *mem);
358 static void mem_cgroup_put(struct mem_cgroup *mem);
359 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
360 static void drain_all_stock_async(void);
362 static struct mem_cgroup_per_zone *
363 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
365 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
368 struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem)
373 static struct mem_cgroup_per_zone *
374 page_cgroup_zoneinfo(struct page_cgroup *pc)
376 struct mem_cgroup *mem = pc->mem_cgroup;
377 int nid = page_cgroup_nid(pc);
378 int zid = page_cgroup_zid(pc);
383 return mem_cgroup_zoneinfo(mem, nid, zid);
386 static struct mem_cgroup_tree_per_zone *
387 soft_limit_tree_node_zone(int nid, int zid)
389 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
392 static struct mem_cgroup_tree_per_zone *
393 soft_limit_tree_from_page(struct page *page)
395 int nid = page_to_nid(page);
396 int zid = page_zonenum(page);
398 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
402 __mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
403 struct mem_cgroup_per_zone *mz,
404 struct mem_cgroup_tree_per_zone *mctz,
405 unsigned long long new_usage_in_excess)
407 struct rb_node **p = &mctz->rb_root.rb_node;
408 struct rb_node *parent = NULL;
409 struct mem_cgroup_per_zone *mz_node;
414 mz->usage_in_excess = new_usage_in_excess;
415 if (!mz->usage_in_excess)
419 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
421 if (mz->usage_in_excess < mz_node->usage_in_excess)
424 * We can't avoid mem cgroups that are over their soft
425 * limit by the same amount
427 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
430 rb_link_node(&mz->tree_node, parent, p);
431 rb_insert_color(&mz->tree_node, &mctz->rb_root);
436 __mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
437 struct mem_cgroup_per_zone *mz,
438 struct mem_cgroup_tree_per_zone *mctz)
442 rb_erase(&mz->tree_node, &mctz->rb_root);
447 mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
448 struct mem_cgroup_per_zone *mz,
449 struct mem_cgroup_tree_per_zone *mctz)
451 spin_lock(&mctz->lock);
452 __mem_cgroup_remove_exceeded(mem, mz, mctz);
453 spin_unlock(&mctz->lock);
457 static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page)
459 unsigned long long excess;
460 struct mem_cgroup_per_zone *mz;
461 struct mem_cgroup_tree_per_zone *mctz;
462 int nid = page_to_nid(page);
463 int zid = page_zonenum(page);
464 mctz = soft_limit_tree_from_page(page);
467 * Necessary to update all ancestors when hierarchy is used.
468 * because their event counter is not touched.
470 for (; mem; mem = parent_mem_cgroup(mem)) {
471 mz = mem_cgroup_zoneinfo(mem, nid, zid);
472 excess = res_counter_soft_limit_excess(&mem->res);
474 * We have to update the tree if mz is on RB-tree or
475 * mem is over its softlimit.
477 if (excess || mz->on_tree) {
478 spin_lock(&mctz->lock);
479 /* if on-tree, remove it */
481 __mem_cgroup_remove_exceeded(mem, mz, mctz);
483 * Insert again. mz->usage_in_excess will be updated.
484 * If excess is 0, no tree ops.
486 __mem_cgroup_insert_exceeded(mem, mz, mctz, excess);
487 spin_unlock(&mctz->lock);
492 static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem)
495 struct mem_cgroup_per_zone *mz;
496 struct mem_cgroup_tree_per_zone *mctz;
498 for_each_node_state(node, N_POSSIBLE) {
499 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
500 mz = mem_cgroup_zoneinfo(mem, node, zone);
501 mctz = soft_limit_tree_node_zone(node, zone);
502 mem_cgroup_remove_exceeded(mem, mz, mctz);
507 static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem)
509 return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT;
512 static struct mem_cgroup_per_zone *
513 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
515 struct rb_node *rightmost = NULL;
516 struct mem_cgroup_per_zone *mz;
520 rightmost = rb_last(&mctz->rb_root);
522 goto done; /* Nothing to reclaim from */
524 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
526 * Remove the node now but someone else can add it back,
527 * we will to add it back at the end of reclaim to its correct
528 * position in the tree.
530 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
531 if (!res_counter_soft_limit_excess(&mz->mem->res) ||
532 !css_tryget(&mz->mem->css))
538 static struct mem_cgroup_per_zone *
539 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
541 struct mem_cgroup_per_zone *mz;
543 spin_lock(&mctz->lock);
544 mz = __mem_cgroup_largest_soft_limit_node(mctz);
545 spin_unlock(&mctz->lock);
550 * Implementation Note: reading percpu statistics for memcg.
552 * Both of vmstat[] and percpu_counter has threshold and do periodic
553 * synchronization to implement "quick" read. There are trade-off between
554 * reading cost and precision of value. Then, we may have a chance to implement
555 * a periodic synchronizion of counter in memcg's counter.
557 * But this _read() function is used for user interface now. The user accounts
558 * memory usage by memory cgroup and he _always_ requires exact value because
559 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
560 * have to visit all online cpus and make sum. So, for now, unnecessary
561 * synchronization is not implemented. (just implemented for cpu hotplug)
563 * If there are kernel internal actions which can make use of some not-exact
564 * value, and reading all cpu value can be performance bottleneck in some
565 * common workload, threashold and synchonization as vmstat[] should be
568 static s64 mem_cgroup_read_stat(struct mem_cgroup *mem,
569 enum mem_cgroup_stat_index idx)
575 for_each_online_cpu(cpu)
576 val += per_cpu(mem->stat->count[idx], cpu);
577 #ifdef CONFIG_HOTPLUG_CPU
578 spin_lock(&mem->pcp_counter_lock);
579 val += mem->nocpu_base.count[idx];
580 spin_unlock(&mem->pcp_counter_lock);
586 static s64 mem_cgroup_local_usage(struct mem_cgroup *mem)
590 ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
591 ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
595 static void mem_cgroup_swap_statistics(struct mem_cgroup *mem,
598 int val = (charge) ? 1 : -1;
599 this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
602 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
603 bool file, int nr_pages)
608 __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], nr_pages);
610 __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], nr_pages);
612 /* pagein of a big page is an event. So, ignore page size */
614 __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]);
616 __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]);
618 __this_cpu_add(mem->stat->count[MEM_CGROUP_EVENTS], nr_pages);
623 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
627 struct mem_cgroup_per_zone *mz;
630 for_each_online_node(nid)
631 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
632 mz = mem_cgroup_zoneinfo(mem, nid, zid);
633 total += MEM_CGROUP_ZSTAT(mz, idx);
638 static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift)
642 val = this_cpu_read(mem->stat->count[MEM_CGROUP_EVENTS]);
644 return !(val & ((1 << event_mask_shift) - 1));
648 * Check events in order.
651 static void memcg_check_events(struct mem_cgroup *mem, struct page *page)
653 /* threshold event is triggered in finer grain than soft limit */
654 if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) {
655 mem_cgroup_threshold(mem);
656 if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH)))
657 mem_cgroup_update_tree(mem, page);
661 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
663 return container_of(cgroup_subsys_state(cont,
664 mem_cgroup_subsys_id), struct mem_cgroup,
668 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
671 * mm_update_next_owner() may clear mm->owner to NULL
672 * if it races with swapoff, page migration, etc.
673 * So this can be called with p == NULL.
678 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
679 struct mem_cgroup, css);
682 static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
684 struct mem_cgroup *mem = NULL;
689 * Because we have no locks, mm->owner's may be being moved to other
690 * cgroup. We use css_tryget() here even if this looks
691 * pessimistic (rather than adding locks here).
695 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
698 } while (!css_tryget(&mem->css));
703 /* The caller has to guarantee "mem" exists before calling this */
704 static struct mem_cgroup *mem_cgroup_start_loop(struct mem_cgroup *mem)
706 struct cgroup_subsys_state *css;
709 if (!mem) /* ROOT cgroup has the smallest ID */
710 return root_mem_cgroup; /*css_put/get against root is ignored*/
711 if (!mem->use_hierarchy) {
712 if (css_tryget(&mem->css))
718 * searching a memory cgroup which has the smallest ID under given
719 * ROOT cgroup. (ID >= 1)
721 css = css_get_next(&mem_cgroup_subsys, 1, &mem->css, &found);
722 if (css && css_tryget(css))
723 mem = container_of(css, struct mem_cgroup, css);
730 static struct mem_cgroup *mem_cgroup_get_next(struct mem_cgroup *iter,
731 struct mem_cgroup *root,
734 int nextid = css_id(&iter->css) + 1;
737 struct cgroup_subsys_state *css;
739 hierarchy_used = iter->use_hierarchy;
742 /* If no ROOT, walk all, ignore hierarchy */
743 if (!cond || (root && !hierarchy_used))
747 root = root_mem_cgroup;
753 css = css_get_next(&mem_cgroup_subsys, nextid,
755 if (css && css_tryget(css))
756 iter = container_of(css, struct mem_cgroup, css);
758 /* If css is NULL, no more cgroups will be found */
760 } while (css && !iter);
765 * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please
766 * be careful that "break" loop is not allowed. We have reference count.
767 * Instead of that modify "cond" to be false and "continue" to exit the loop.
769 #define for_each_mem_cgroup_tree_cond(iter, root, cond) \
770 for (iter = mem_cgroup_start_loop(root);\
772 iter = mem_cgroup_get_next(iter, root, cond))
774 #define for_each_mem_cgroup_tree(iter, root) \
775 for_each_mem_cgroup_tree_cond(iter, root, true)
777 #define for_each_mem_cgroup_all(iter) \
778 for_each_mem_cgroup_tree_cond(iter, NULL, true)
781 static inline bool mem_cgroup_is_root(struct mem_cgroup *mem)
783 return (mem == root_mem_cgroup);
787 * Following LRU functions are allowed to be used without PCG_LOCK.
788 * Operations are called by routine of global LRU independently from memcg.
789 * What we have to take care of here is validness of pc->mem_cgroup.
791 * Changes to pc->mem_cgroup happens when
794 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
795 * It is added to LRU before charge.
796 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
797 * When moving account, the page is not on LRU. It's isolated.
800 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
802 struct page_cgroup *pc;
803 struct mem_cgroup_per_zone *mz;
805 if (mem_cgroup_disabled())
807 pc = lookup_page_cgroup(page);
808 /* can happen while we handle swapcache. */
809 if (!TestClearPageCgroupAcctLRU(pc))
811 VM_BUG_ON(!pc->mem_cgroup);
813 * We don't check PCG_USED bit. It's cleared when the "page" is finally
814 * removed from global LRU.
816 mz = page_cgroup_zoneinfo(pc);
817 /* huge page split is done under lru_lock. so, we have no races. */
818 MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page);
819 if (mem_cgroup_is_root(pc->mem_cgroup))
821 VM_BUG_ON(list_empty(&pc->lru));
822 list_del_init(&pc->lru);
825 void mem_cgroup_del_lru(struct page *page)
827 mem_cgroup_del_lru_list(page, page_lru(page));
830 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
832 struct mem_cgroup_per_zone *mz;
833 struct page_cgroup *pc;
835 if (mem_cgroup_disabled())
838 pc = lookup_page_cgroup(page);
839 /* unused or root page is not rotated. */
840 if (!PageCgroupUsed(pc))
842 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
844 if (mem_cgroup_is_root(pc->mem_cgroup))
846 mz = page_cgroup_zoneinfo(pc);
847 list_move(&pc->lru, &mz->lists[lru]);
850 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
852 struct page_cgroup *pc;
853 struct mem_cgroup_per_zone *mz;
855 if (mem_cgroup_disabled())
857 pc = lookup_page_cgroup(page);
858 VM_BUG_ON(PageCgroupAcctLRU(pc));
859 if (!PageCgroupUsed(pc))
861 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
863 mz = page_cgroup_zoneinfo(pc);
864 /* huge page split is done under lru_lock. so, we have no races. */
865 MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page);
866 SetPageCgroupAcctLRU(pc);
867 if (mem_cgroup_is_root(pc->mem_cgroup))
869 list_add(&pc->lru, &mz->lists[lru]);
873 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
874 * lru because the page may.be reused after it's fully uncharged (because of
875 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
876 * it again. This function is only used to charge SwapCache. It's done under
877 * lock_page and expected that zone->lru_lock is never held.
879 static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
882 struct zone *zone = page_zone(page);
883 struct page_cgroup *pc = lookup_page_cgroup(page);
885 spin_lock_irqsave(&zone->lru_lock, flags);
887 * Forget old LRU when this page_cgroup is *not* used. This Used bit
888 * is guarded by lock_page() because the page is SwapCache.
890 if (!PageCgroupUsed(pc))
891 mem_cgroup_del_lru_list(page, page_lru(page));
892 spin_unlock_irqrestore(&zone->lru_lock, flags);
895 static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
898 struct zone *zone = page_zone(page);
899 struct page_cgroup *pc = lookup_page_cgroup(page);
901 spin_lock_irqsave(&zone->lru_lock, flags);
902 /* link when the page is linked to LRU but page_cgroup isn't */
903 if (PageLRU(page) && !PageCgroupAcctLRU(pc))
904 mem_cgroup_add_lru_list(page, page_lru(page));
905 spin_unlock_irqrestore(&zone->lru_lock, flags);
909 void mem_cgroup_move_lists(struct page *page,
910 enum lru_list from, enum lru_list to)
912 if (mem_cgroup_disabled())
914 mem_cgroup_del_lru_list(page, from);
915 mem_cgroup_add_lru_list(page, to);
918 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
921 struct mem_cgroup *curr = NULL;
922 struct task_struct *p;
924 p = find_lock_task_mm(task);
927 curr = try_get_mem_cgroup_from_mm(p->mm);
932 * We should check use_hierarchy of "mem" not "curr". Because checking
933 * use_hierarchy of "curr" here make this function true if hierarchy is
934 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
935 * hierarchy(even if use_hierarchy is disabled in "mem").
937 if (mem->use_hierarchy)
938 ret = css_is_ancestor(&curr->css, &mem->css);
945 static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
947 unsigned long active;
948 unsigned long inactive;
950 unsigned long inactive_ratio;
952 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
953 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
955 gb = (inactive + active) >> (30 - PAGE_SHIFT);
957 inactive_ratio = int_sqrt(10 * gb);
962 present_pages[0] = inactive;
963 present_pages[1] = active;
966 return inactive_ratio;
969 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
971 unsigned long active;
972 unsigned long inactive;
973 unsigned long present_pages[2];
974 unsigned long inactive_ratio;
976 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
978 inactive = present_pages[0];
979 active = present_pages[1];
981 if (inactive * inactive_ratio < active)
987 int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
989 unsigned long active;
990 unsigned long inactive;
992 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
993 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
995 return (active > inactive);
998 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
1002 int nid = zone_to_nid(zone);
1003 int zid = zone_idx(zone);
1004 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
1006 return MEM_CGROUP_ZSTAT(mz, lru);
1009 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
1012 int nid = zone_to_nid(zone);
1013 int zid = zone_idx(zone);
1014 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
1016 return &mz->reclaim_stat;
1019 struct zone_reclaim_stat *
1020 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
1022 struct page_cgroup *pc;
1023 struct mem_cgroup_per_zone *mz;
1025 if (mem_cgroup_disabled())
1028 pc = lookup_page_cgroup(page);
1029 if (!PageCgroupUsed(pc))
1031 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1033 mz = page_cgroup_zoneinfo(pc);
1037 return &mz->reclaim_stat;
1040 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
1041 struct list_head *dst,
1042 unsigned long *scanned, int order,
1043 int mode, struct zone *z,
1044 struct mem_cgroup *mem_cont,
1045 int active, int file)
1047 unsigned long nr_taken = 0;
1051 struct list_head *src;
1052 struct page_cgroup *pc, *tmp;
1053 int nid = zone_to_nid(z);
1054 int zid = zone_idx(z);
1055 struct mem_cgroup_per_zone *mz;
1056 int lru = LRU_FILE * file + active;
1060 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
1061 src = &mz->lists[lru];
1064 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
1065 if (scan >= nr_to_scan)
1069 if (unlikely(!PageCgroupUsed(pc)))
1071 if (unlikely(!PageLRU(page)))
1075 ret = __isolate_lru_page(page, mode, file);
1078 list_move(&page->lru, dst);
1079 mem_cgroup_del_lru(page);
1080 nr_taken += hpage_nr_pages(page);
1083 /* we don't affect global LRU but rotate in our LRU */
1084 mem_cgroup_rotate_lru_list(page, page_lru(page));
1093 trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken,
1099 #define mem_cgroup_from_res_counter(counter, member) \
1100 container_of(counter, struct mem_cgroup, member)
1102 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
1104 if (do_swap_account) {
1105 if (res_counter_check_under_limit(&mem->res) &&
1106 res_counter_check_under_limit(&mem->memsw))
1109 if (res_counter_check_under_limit(&mem->res))
1115 * mem_cgroup_check_margin - check if the memory cgroup allows charging
1116 * @mem: memory cgroup to check
1117 * @bytes: the number of bytes the caller intends to charge
1119 * Returns a boolean value on whether @mem can be charged @bytes or
1120 * whether this would exceed the limit.
1122 static bool mem_cgroup_check_margin(struct mem_cgroup *mem, unsigned long bytes)
1124 if (!res_counter_check_margin(&mem->res, bytes))
1126 if (do_swap_account && !res_counter_check_margin(&mem->memsw, bytes))
1131 static unsigned int get_swappiness(struct mem_cgroup *memcg)
1133 struct cgroup *cgrp = memcg->css.cgroup;
1134 unsigned int swappiness;
1137 if (cgrp->parent == NULL)
1138 return vm_swappiness;
1140 spin_lock(&memcg->reclaim_param_lock);
1141 swappiness = memcg->swappiness;
1142 spin_unlock(&memcg->reclaim_param_lock);
1147 static void mem_cgroup_start_move(struct mem_cgroup *mem)
1152 spin_lock(&mem->pcp_counter_lock);
1153 for_each_online_cpu(cpu)
1154 per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1;
1155 mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1;
1156 spin_unlock(&mem->pcp_counter_lock);
1162 static void mem_cgroup_end_move(struct mem_cgroup *mem)
1169 spin_lock(&mem->pcp_counter_lock);
1170 for_each_online_cpu(cpu)
1171 per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1;
1172 mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1;
1173 spin_unlock(&mem->pcp_counter_lock);
1177 * 2 routines for checking "mem" is under move_account() or not.
1179 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1180 * for avoiding race in accounting. If true,
1181 * pc->mem_cgroup may be overwritten.
1183 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1184 * under hierarchy of moving cgroups. This is for
1185 * waiting at hith-memory prressure caused by "move".
1188 static bool mem_cgroup_stealed(struct mem_cgroup *mem)
1190 VM_BUG_ON(!rcu_read_lock_held());
1191 return this_cpu_read(mem->stat->count[MEM_CGROUP_ON_MOVE]) > 0;
1194 static bool mem_cgroup_under_move(struct mem_cgroup *mem)
1196 struct mem_cgroup *from;
1197 struct mem_cgroup *to;
1200 * Unlike task_move routines, we access mc.to, mc.from not under
1201 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1203 spin_lock(&mc.lock);
1208 if (from == mem || to == mem
1209 || (mem->use_hierarchy && css_is_ancestor(&from->css, &mem->css))
1210 || (mem->use_hierarchy && css_is_ancestor(&to->css, &mem->css)))
1213 spin_unlock(&mc.lock);
1217 static bool mem_cgroup_wait_acct_move(struct mem_cgroup *mem)
1219 if (mc.moving_task && current != mc.moving_task) {
1220 if (mem_cgroup_under_move(mem)) {
1222 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1223 /* moving charge context might have finished. */
1226 finish_wait(&mc.waitq, &wait);
1234 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1235 * @memcg: The memory cgroup that went over limit
1236 * @p: Task that is going to be killed
1238 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1241 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1243 struct cgroup *task_cgrp;
1244 struct cgroup *mem_cgrp;
1246 * Need a buffer in BSS, can't rely on allocations. The code relies
1247 * on the assumption that OOM is serialized for memory controller.
1248 * If this assumption is broken, revisit this code.
1250 static char memcg_name[PATH_MAX];
1259 mem_cgrp = memcg->css.cgroup;
1260 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1262 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1265 * Unfortunately, we are unable to convert to a useful name
1266 * But we'll still print out the usage information
1273 printk(KERN_INFO "Task in %s killed", memcg_name);
1276 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1284 * Continues from above, so we don't need an KERN_ level
1286 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1289 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1290 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1291 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1292 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1293 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1295 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1296 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1297 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1301 * This function returns the number of memcg under hierarchy tree. Returns
1302 * 1(self count) if no children.
1304 static int mem_cgroup_count_children(struct mem_cgroup *mem)
1307 struct mem_cgroup *iter;
1309 for_each_mem_cgroup_tree(iter, mem)
1315 * Return the memory (and swap, if configured) limit for a memcg.
1317 u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1322 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1323 limit += total_swap_pages << PAGE_SHIFT;
1325 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1327 * If memsw is finite and limits the amount of swap space available
1328 * to this memcg, return that limit.
1330 return min(limit, memsw);
1334 * Visit the first child (need not be the first child as per the ordering
1335 * of the cgroup list, since we track last_scanned_child) of @mem and use
1336 * that to reclaim free pages from.
1338 static struct mem_cgroup *
1339 mem_cgroup_select_victim(struct mem_cgroup *root_mem)
1341 struct mem_cgroup *ret = NULL;
1342 struct cgroup_subsys_state *css;
1345 if (!root_mem->use_hierarchy) {
1346 css_get(&root_mem->css);
1352 nextid = root_mem->last_scanned_child + 1;
1353 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
1355 if (css && css_tryget(css))
1356 ret = container_of(css, struct mem_cgroup, css);
1359 /* Updates scanning parameter */
1360 spin_lock(&root_mem->reclaim_param_lock);
1362 /* this means start scan from ID:1 */
1363 root_mem->last_scanned_child = 0;
1365 root_mem->last_scanned_child = found;
1366 spin_unlock(&root_mem->reclaim_param_lock);
1373 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1374 * we reclaimed from, so that we don't end up penalizing one child extensively
1375 * based on its position in the children list.
1377 * root_mem is the original ancestor that we've been reclaim from.
1379 * We give up and return to the caller when we visit root_mem twice.
1380 * (other groups can be removed while we're walking....)
1382 * If shrink==true, for avoiding to free too much, this returns immedieately.
1384 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
1387 unsigned long reclaim_options)
1389 struct mem_cgroup *victim;
1392 bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
1393 bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
1394 bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT;
1395 unsigned long excess = mem_cgroup_get_excess(root_mem);
1397 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1398 if (root_mem->memsw_is_minimum)
1402 victim = mem_cgroup_select_victim(root_mem);
1403 if (victim == root_mem) {
1406 drain_all_stock_async();
1409 * If we have not been able to reclaim
1410 * anything, it might because there are
1411 * no reclaimable pages under this hierarchy
1413 if (!check_soft || !total) {
1414 css_put(&victim->css);
1418 * We want to do more targetted reclaim.
1419 * excess >> 2 is not to excessive so as to
1420 * reclaim too much, nor too less that we keep
1421 * coming back to reclaim from this cgroup
1423 if (total >= (excess >> 2) ||
1424 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) {
1425 css_put(&victim->css);
1430 if (!mem_cgroup_local_usage(victim)) {
1431 /* this cgroup's local usage == 0 */
1432 css_put(&victim->css);
1435 /* we use swappiness of local cgroup */
1437 ret = mem_cgroup_shrink_node_zone(victim, gfp_mask,
1438 noswap, get_swappiness(victim), zone);
1440 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask,
1441 noswap, get_swappiness(victim));
1442 css_put(&victim->css);
1444 * At shrinking usage, we can't check we should stop here or
1445 * reclaim more. It's depends on callers. last_scanned_child
1446 * will work enough for keeping fairness under tree.
1452 if (res_counter_check_under_soft_limit(&root_mem->res))
1454 } else if (mem_cgroup_check_under_limit(root_mem))
1461 * Check OOM-Killer is already running under our hierarchy.
1462 * If someone is running, return false.
1464 static bool mem_cgroup_oom_lock(struct mem_cgroup *mem)
1466 int x, lock_count = 0;
1467 struct mem_cgroup *iter;
1469 for_each_mem_cgroup_tree(iter, mem) {
1470 x = atomic_inc_return(&iter->oom_lock);
1471 lock_count = max(x, lock_count);
1474 if (lock_count == 1)
1479 static int mem_cgroup_oom_unlock(struct mem_cgroup *mem)
1481 struct mem_cgroup *iter;
1484 * When a new child is created while the hierarchy is under oom,
1485 * mem_cgroup_oom_lock() may not be called. We have to use
1486 * atomic_add_unless() here.
1488 for_each_mem_cgroup_tree(iter, mem)
1489 atomic_add_unless(&iter->oom_lock, -1, 0);
1494 static DEFINE_MUTEX(memcg_oom_mutex);
1495 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1497 struct oom_wait_info {
1498 struct mem_cgroup *mem;
1502 static int memcg_oom_wake_function(wait_queue_t *wait,
1503 unsigned mode, int sync, void *arg)
1505 struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg;
1506 struct oom_wait_info *oom_wait_info;
1508 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
1510 if (oom_wait_info->mem == wake_mem)
1512 /* if no hierarchy, no match */
1513 if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy)
1516 * Both of oom_wait_info->mem and wake_mem are stable under us.
1517 * Then we can use css_is_ancestor without taking care of RCU.
1519 if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) &&
1520 !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css))
1524 return autoremove_wake_function(wait, mode, sync, arg);
1527 static void memcg_wakeup_oom(struct mem_cgroup *mem)
1529 /* for filtering, pass "mem" as argument. */
1530 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem);
1533 static void memcg_oom_recover(struct mem_cgroup *mem)
1535 if (mem && atomic_read(&mem->oom_lock))
1536 memcg_wakeup_oom(mem);
1540 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1542 bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask)
1544 struct oom_wait_info owait;
1545 bool locked, need_to_kill;
1548 owait.wait.flags = 0;
1549 owait.wait.func = memcg_oom_wake_function;
1550 owait.wait.private = current;
1551 INIT_LIST_HEAD(&owait.wait.task_list);
1552 need_to_kill = true;
1553 /* At first, try to OOM lock hierarchy under mem.*/
1554 mutex_lock(&memcg_oom_mutex);
1555 locked = mem_cgroup_oom_lock(mem);
1557 * Even if signal_pending(), we can't quit charge() loop without
1558 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1559 * under OOM is always welcomed, use TASK_KILLABLE here.
1561 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
1562 if (!locked || mem->oom_kill_disable)
1563 need_to_kill = false;
1565 mem_cgroup_oom_notify(mem);
1566 mutex_unlock(&memcg_oom_mutex);
1569 finish_wait(&memcg_oom_waitq, &owait.wait);
1570 mem_cgroup_out_of_memory(mem, mask);
1573 finish_wait(&memcg_oom_waitq, &owait.wait);
1575 mutex_lock(&memcg_oom_mutex);
1576 mem_cgroup_oom_unlock(mem);
1577 memcg_wakeup_oom(mem);
1578 mutex_unlock(&memcg_oom_mutex);
1580 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1582 /* Give chance to dying process */
1583 schedule_timeout(1);
1588 * Currently used to update mapped file statistics, but the routine can be
1589 * generalized to update other statistics as well.
1591 * Notes: Race condition
1593 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1594 * it tends to be costly. But considering some conditions, we doesn't need
1595 * to do so _always_.
1597 * Considering "charge", lock_page_cgroup() is not required because all
1598 * file-stat operations happen after a page is attached to radix-tree. There
1599 * are no race with "charge".
1601 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1602 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1603 * if there are race with "uncharge". Statistics itself is properly handled
1606 * Considering "move", this is an only case we see a race. To make the race
1607 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1608 * possibility of race condition. If there is, we take a lock.
1611 void mem_cgroup_update_page_stat(struct page *page,
1612 enum mem_cgroup_page_stat_item idx, int val)
1614 struct mem_cgroup *mem;
1615 struct page_cgroup *pc = lookup_page_cgroup(page);
1616 bool need_unlock = false;
1617 unsigned long uninitialized_var(flags);
1623 mem = pc->mem_cgroup;
1624 if (unlikely(!mem || !PageCgroupUsed(pc)))
1626 /* pc->mem_cgroup is unstable ? */
1627 if (unlikely(mem_cgroup_stealed(mem)) || PageTransHuge(page)) {
1628 /* take a lock against to access pc->mem_cgroup */
1629 move_lock_page_cgroup(pc, &flags);
1631 mem = pc->mem_cgroup;
1632 if (!mem || !PageCgroupUsed(pc))
1637 case MEMCG_NR_FILE_MAPPED:
1639 SetPageCgroupFileMapped(pc);
1640 else if (!page_mapped(page))
1641 ClearPageCgroupFileMapped(pc);
1642 idx = MEM_CGROUP_STAT_FILE_MAPPED;
1648 this_cpu_add(mem->stat->count[idx], val);
1651 if (unlikely(need_unlock))
1652 move_unlock_page_cgroup(pc, &flags);
1656 EXPORT_SYMBOL(mem_cgroup_update_page_stat);
1659 * size of first charge trial. "32" comes from vmscan.c's magic value.
1660 * TODO: maybe necessary to use big numbers in big irons.
1662 #define CHARGE_SIZE (32 * PAGE_SIZE)
1663 struct memcg_stock_pcp {
1664 struct mem_cgroup *cached; /* this never be root cgroup */
1666 struct work_struct work;
1668 static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
1669 static atomic_t memcg_drain_count;
1672 * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
1673 * from local stock and true is returned. If the stock is 0 or charges from a
1674 * cgroup which is not current target, returns false. This stock will be
1677 static bool consume_stock(struct mem_cgroup *mem)
1679 struct memcg_stock_pcp *stock;
1682 stock = &get_cpu_var(memcg_stock);
1683 if (mem == stock->cached && stock->charge)
1684 stock->charge -= PAGE_SIZE;
1685 else /* need to call res_counter_charge */
1687 put_cpu_var(memcg_stock);
1692 * Returns stocks cached in percpu to res_counter and reset cached information.
1694 static void drain_stock(struct memcg_stock_pcp *stock)
1696 struct mem_cgroup *old = stock->cached;
1698 if (stock->charge) {
1699 res_counter_uncharge(&old->res, stock->charge);
1700 if (do_swap_account)
1701 res_counter_uncharge(&old->memsw, stock->charge);
1703 stock->cached = NULL;
1708 * This must be called under preempt disabled or must be called by
1709 * a thread which is pinned to local cpu.
1711 static void drain_local_stock(struct work_struct *dummy)
1713 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
1718 * Cache charges(val) which is from res_counter, to local per_cpu area.
1719 * This will be consumed by consume_stock() function, later.
1721 static void refill_stock(struct mem_cgroup *mem, int val)
1723 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1725 if (stock->cached != mem) { /* reset if necessary */
1727 stock->cached = mem;
1729 stock->charge += val;
1730 put_cpu_var(memcg_stock);
1734 * Tries to drain stocked charges in other cpus. This function is asynchronous
1735 * and just put a work per cpu for draining localy on each cpu. Caller can
1736 * expects some charges will be back to res_counter later but cannot wait for
1739 static void drain_all_stock_async(void)
1742 /* This function is for scheduling "drain" in asynchronous way.
1743 * The result of "drain" is not directly handled by callers. Then,
1744 * if someone is calling drain, we don't have to call drain more.
1745 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1746 * there is a race. We just do loose check here.
1748 if (atomic_read(&memcg_drain_count))
1750 /* Notify other cpus that system-wide "drain" is running */
1751 atomic_inc(&memcg_drain_count);
1753 for_each_online_cpu(cpu) {
1754 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
1755 schedule_work_on(cpu, &stock->work);
1758 atomic_dec(&memcg_drain_count);
1759 /* We don't wait for flush_work */
1762 /* This is a synchronous drain interface. */
1763 static void drain_all_stock_sync(void)
1765 /* called when force_empty is called */
1766 atomic_inc(&memcg_drain_count);
1767 schedule_on_each_cpu(drain_local_stock);
1768 atomic_dec(&memcg_drain_count);
1772 * This function drains percpu counter value from DEAD cpu and
1773 * move it to local cpu. Note that this function can be preempted.
1775 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *mem, int cpu)
1779 spin_lock(&mem->pcp_counter_lock);
1780 for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) {
1781 s64 x = per_cpu(mem->stat->count[i], cpu);
1783 per_cpu(mem->stat->count[i], cpu) = 0;
1784 mem->nocpu_base.count[i] += x;
1786 /* need to clear ON_MOVE value, works as a kind of lock. */
1787 per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0;
1788 spin_unlock(&mem->pcp_counter_lock);
1791 static void synchronize_mem_cgroup_on_move(struct mem_cgroup *mem, int cpu)
1793 int idx = MEM_CGROUP_ON_MOVE;
1795 spin_lock(&mem->pcp_counter_lock);
1796 per_cpu(mem->stat->count[idx], cpu) = mem->nocpu_base.count[idx];
1797 spin_unlock(&mem->pcp_counter_lock);
1800 static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
1801 unsigned long action,
1804 int cpu = (unsigned long)hcpu;
1805 struct memcg_stock_pcp *stock;
1806 struct mem_cgroup *iter;
1808 if ((action == CPU_ONLINE)) {
1809 for_each_mem_cgroup_all(iter)
1810 synchronize_mem_cgroup_on_move(iter, cpu);
1814 if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN)
1817 for_each_mem_cgroup_all(iter)
1818 mem_cgroup_drain_pcp_counter(iter, cpu);
1820 stock = &per_cpu(memcg_stock, cpu);
1826 /* See __mem_cgroup_try_charge() for details */
1828 CHARGE_OK, /* success */
1829 CHARGE_RETRY, /* need to retry but retry is not bad */
1830 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
1831 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
1832 CHARGE_OOM_DIE, /* the current is killed because of OOM */
1835 static int __mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask,
1836 int csize, bool oom_check)
1838 struct mem_cgroup *mem_over_limit;
1839 struct res_counter *fail_res;
1840 unsigned long flags = 0;
1843 ret = res_counter_charge(&mem->res, csize, &fail_res);
1846 if (!do_swap_account)
1848 ret = res_counter_charge(&mem->memsw, csize, &fail_res);
1852 res_counter_uncharge(&mem->res, csize);
1853 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
1854 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
1856 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
1858 * csize can be either a huge page (HPAGE_SIZE), a batch of
1859 * regular pages (CHARGE_SIZE), or a single regular page
1862 * Never reclaim on behalf of optional batching, retry with a
1863 * single page instead.
1865 if (csize == CHARGE_SIZE)
1866 return CHARGE_RETRY;
1868 if (!(gfp_mask & __GFP_WAIT))
1869 return CHARGE_WOULDBLOCK;
1871 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
1873 if (mem_cgroup_check_margin(mem_over_limit, csize))
1874 return CHARGE_RETRY;
1876 * Even though the limit is exceeded at this point, reclaim
1877 * may have been able to free some pages. Retry the charge
1878 * before killing the task.
1880 * Only for regular pages, though: huge pages are rather
1881 * unlikely to succeed so close to the limit, and we fall back
1882 * to regular pages anyway in case of failure.
1884 if (csize == PAGE_SIZE && ret)
1885 return CHARGE_RETRY;
1888 * At task move, charge accounts can be doubly counted. So, it's
1889 * better to wait until the end of task_move if something is going on.
1891 if (mem_cgroup_wait_acct_move(mem_over_limit))
1892 return CHARGE_RETRY;
1894 /* If we don't need to call oom-killer at el, return immediately */
1896 return CHARGE_NOMEM;
1898 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask))
1899 return CHARGE_OOM_DIE;
1901 return CHARGE_RETRY;
1905 * Unlike exported interface, "oom" parameter is added. if oom==true,
1906 * oom-killer can be invoked.
1908 static int __mem_cgroup_try_charge(struct mm_struct *mm,
1910 struct mem_cgroup **memcg, bool oom,
1913 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
1914 struct mem_cgroup *mem = NULL;
1916 int csize = max(CHARGE_SIZE, (unsigned long) page_size);
1919 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
1920 * in system level. So, allow to go ahead dying process in addition to
1923 if (unlikely(test_thread_flag(TIF_MEMDIE)
1924 || fatal_signal_pending(current)))
1928 * We always charge the cgroup the mm_struct belongs to.
1929 * The mm_struct's mem_cgroup changes on task migration if the
1930 * thread group leader migrates. It's possible that mm is not
1931 * set, if so charge the init_mm (happens for pagecache usage).
1936 if (*memcg) { /* css should be a valid one */
1938 VM_BUG_ON(css_is_removed(&mem->css));
1939 if (mem_cgroup_is_root(mem))
1941 if (page_size == PAGE_SIZE && consume_stock(mem))
1945 struct task_struct *p;
1948 p = rcu_dereference(mm->owner);
1950 * Because we don't have task_lock(), "p" can exit.
1951 * In that case, "mem" can point to root or p can be NULL with
1952 * race with swapoff. Then, we have small risk of mis-accouning.
1953 * But such kind of mis-account by race always happens because
1954 * we don't have cgroup_mutex(). It's overkill and we allo that
1956 * (*) swapoff at el will charge against mm-struct not against
1957 * task-struct. So, mm->owner can be NULL.
1959 mem = mem_cgroup_from_task(p);
1960 if (!mem || mem_cgroup_is_root(mem)) {
1964 if (page_size == PAGE_SIZE && consume_stock(mem)) {
1966 * It seems dagerous to access memcg without css_get().
1967 * But considering how consume_stok works, it's not
1968 * necessary. If consume_stock success, some charges
1969 * from this memcg are cached on this cpu. So, we
1970 * don't need to call css_get()/css_tryget() before
1971 * calling consume_stock().
1976 /* after here, we may be blocked. we need to get refcnt */
1977 if (!css_tryget(&mem->css)) {
1987 /* If killed, bypass charge */
1988 if (fatal_signal_pending(current)) {
1994 if (oom && !nr_oom_retries) {
1996 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
1999 ret = __mem_cgroup_do_charge(mem, gfp_mask, csize, oom_check);
2004 case CHARGE_RETRY: /* not in OOM situation but retry */
2009 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
2012 case CHARGE_NOMEM: /* OOM routine works */
2017 /* If oom, we never return -ENOMEM */
2020 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
2024 } while (ret != CHARGE_OK);
2026 if (csize > page_size)
2027 refill_stock(mem, csize - page_size);
2041 * Somemtimes we have to undo a charge we got by try_charge().
2042 * This function is for that and do uncharge, put css's refcnt.
2043 * gotten by try_charge().
2045 static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem,
2046 unsigned long count)
2048 if (!mem_cgroup_is_root(mem)) {
2049 res_counter_uncharge(&mem->res, PAGE_SIZE * count);
2050 if (do_swap_account)
2051 res_counter_uncharge(&mem->memsw, PAGE_SIZE * count);
2055 static void mem_cgroup_cancel_charge(struct mem_cgroup *mem,
2058 __mem_cgroup_cancel_charge(mem, page_size >> PAGE_SHIFT);
2062 * A helper function to get mem_cgroup from ID. must be called under
2063 * rcu_read_lock(). The caller must check css_is_removed() or some if
2064 * it's concern. (dropping refcnt from swap can be called against removed
2067 static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2069 struct cgroup_subsys_state *css;
2071 /* ID 0 is unused ID */
2074 css = css_lookup(&mem_cgroup_subsys, id);
2077 return container_of(css, struct mem_cgroup, css);
2080 struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
2082 struct mem_cgroup *mem = NULL;
2083 struct page_cgroup *pc;
2087 VM_BUG_ON(!PageLocked(page));
2089 pc = lookup_page_cgroup(page);
2090 lock_page_cgroup(pc);
2091 if (PageCgroupUsed(pc)) {
2092 mem = pc->mem_cgroup;
2093 if (mem && !css_tryget(&mem->css))
2095 } else if (PageSwapCache(page)) {
2096 ent.val = page_private(page);
2097 id = lookup_swap_cgroup(ent);
2099 mem = mem_cgroup_lookup(id);
2100 if (mem && !css_tryget(&mem->css))
2104 unlock_page_cgroup(pc);
2108 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
2109 struct page_cgroup *pc,
2110 enum charge_type ctype,
2113 int nr_pages = page_size >> PAGE_SHIFT;
2115 /* try_charge() can return NULL to *memcg, taking care of it. */
2119 lock_page_cgroup(pc);
2120 if (unlikely(PageCgroupUsed(pc))) {
2121 unlock_page_cgroup(pc);
2122 mem_cgroup_cancel_charge(mem, page_size);
2126 * we don't need page_cgroup_lock about tail pages, becase they are not
2127 * accessed by any other context at this point.
2129 pc->mem_cgroup = mem;
2131 * We access a page_cgroup asynchronously without lock_page_cgroup().
2132 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2133 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2134 * before USED bit, we need memory barrier here.
2135 * See mem_cgroup_add_lru_list(), etc.
2139 case MEM_CGROUP_CHARGE_TYPE_CACHE:
2140 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
2141 SetPageCgroupCache(pc);
2142 SetPageCgroupUsed(pc);
2144 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2145 ClearPageCgroupCache(pc);
2146 SetPageCgroupUsed(pc);
2152 mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), nr_pages);
2153 unlock_page_cgroup(pc);
2155 * "charge_statistics" updated event counter. Then, check it.
2156 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2157 * if they exceeds softlimit.
2159 memcg_check_events(mem, pc->page);
2162 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2164 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2165 (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION))
2167 * Because tail pages are not marked as "used", set it. We're under
2168 * zone->lru_lock, 'splitting on pmd' and compund_lock.
2170 void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail)
2172 struct page_cgroup *head_pc = lookup_page_cgroup(head);
2173 struct page_cgroup *tail_pc = lookup_page_cgroup(tail);
2174 unsigned long flags;
2176 if (mem_cgroup_disabled())
2179 * We have no races with charge/uncharge but will have races with
2180 * page state accounting.
2182 move_lock_page_cgroup(head_pc, &flags);
2184 tail_pc->mem_cgroup = head_pc->mem_cgroup;
2185 smp_wmb(); /* see __commit_charge() */
2186 if (PageCgroupAcctLRU(head_pc)) {
2188 struct mem_cgroup_per_zone *mz;
2191 * LRU flags cannot be copied because we need to add tail
2192 *.page to LRU by generic call and our hook will be called.
2193 * We hold lru_lock, then, reduce counter directly.
2195 lru = page_lru(head);
2196 mz = page_cgroup_zoneinfo(head_pc);
2197 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
2199 tail_pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2200 move_unlock_page_cgroup(head_pc, &flags);
2205 * __mem_cgroup_move_account - move account of the page
2206 * @pc: page_cgroup of the page.
2207 * @from: mem_cgroup which the page is moved from.
2208 * @to: mem_cgroup which the page is moved to. @from != @to.
2209 * @uncharge: whether we should call uncharge and css_put against @from.
2211 * The caller must confirm following.
2212 * - page is not on LRU (isolate_page() is useful.)
2213 * - the pc is locked, used, and ->mem_cgroup points to @from.
2215 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2216 * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
2217 * true, this function does "uncharge" from old cgroup, but it doesn't if
2218 * @uncharge is false, so a caller should do "uncharge".
2221 static void __mem_cgroup_move_account(struct page_cgroup *pc,
2222 struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge,
2225 int nr_pages = charge_size >> PAGE_SHIFT;
2227 VM_BUG_ON(from == to);
2228 VM_BUG_ON(PageLRU(pc->page));
2229 VM_BUG_ON(!page_is_cgroup_locked(pc));
2230 VM_BUG_ON(!PageCgroupUsed(pc));
2231 VM_BUG_ON(pc->mem_cgroup != from);
2233 if (PageCgroupFileMapped(pc)) {
2234 /* Update mapped_file data for mem_cgroup */
2236 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2237 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2240 mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages);
2242 /* This is not "cancel", but cancel_charge does all we need. */
2243 mem_cgroup_cancel_charge(from, charge_size);
2245 /* caller should have done css_get */
2246 pc->mem_cgroup = to;
2247 mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages);
2249 * We charges against "to" which may not have any tasks. Then, "to"
2250 * can be under rmdir(). But in current implementation, caller of
2251 * this function is just force_empty() and move charge, so it's
2252 * garanteed that "to" is never removed. So, we don't check rmdir
2258 * check whether the @pc is valid for moving account and call
2259 * __mem_cgroup_move_account()
2261 static int mem_cgroup_move_account(struct page_cgroup *pc,
2262 struct mem_cgroup *from, struct mem_cgroup *to,
2263 bool uncharge, int charge_size)
2266 unsigned long flags;
2268 * The page is isolated from LRU. So, collapse function
2269 * will not handle this page. But page splitting can happen.
2270 * Do this check under compound_page_lock(). The caller should
2273 if ((charge_size > PAGE_SIZE) && !PageTransHuge(pc->page))
2276 lock_page_cgroup(pc);
2277 if (PageCgroupUsed(pc) && pc->mem_cgroup == from) {
2278 move_lock_page_cgroup(pc, &flags);
2279 __mem_cgroup_move_account(pc, from, to, uncharge, charge_size);
2280 move_unlock_page_cgroup(pc, &flags);
2283 unlock_page_cgroup(pc);
2287 memcg_check_events(to, pc->page);
2288 memcg_check_events(from, pc->page);
2293 * move charges to its parent.
2296 static int mem_cgroup_move_parent(struct page_cgroup *pc,
2297 struct mem_cgroup *child,
2300 struct page *page = pc->page;
2301 struct cgroup *cg = child->css.cgroup;
2302 struct cgroup *pcg = cg->parent;
2303 struct mem_cgroup *parent;
2304 int page_size = PAGE_SIZE;
2305 unsigned long flags;
2313 if (!get_page_unless_zero(page))
2315 if (isolate_lru_page(page))
2318 if (PageTransHuge(page))
2319 page_size = HPAGE_SIZE;
2321 parent = mem_cgroup_from_cont(pcg);
2322 ret = __mem_cgroup_try_charge(NULL, gfp_mask,
2323 &parent, false, page_size);
2327 if (page_size > PAGE_SIZE)
2328 flags = compound_lock_irqsave(page);
2330 ret = mem_cgroup_move_account(pc, child, parent, true, page_size);
2332 mem_cgroup_cancel_charge(parent, page_size);
2334 if (page_size > PAGE_SIZE)
2335 compound_unlock_irqrestore(page, flags);
2337 putback_lru_page(page);
2345 * Charge the memory controller for page usage.
2347 * 0 if the charge was successful
2348 * < 0 if the cgroup is over its limit
2350 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
2351 gfp_t gfp_mask, enum charge_type ctype)
2353 struct mem_cgroup *mem = NULL;
2354 int page_size = PAGE_SIZE;
2355 struct page_cgroup *pc;
2359 if (PageTransHuge(page)) {
2360 page_size <<= compound_order(page);
2361 VM_BUG_ON(!PageTransHuge(page));
2363 * Never OOM-kill a process for a huge page. The
2364 * fault handler will fall back to regular pages.
2369 pc = lookup_page_cgroup(page);
2370 /* can happen at boot */
2375 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, oom, page_size);
2379 __mem_cgroup_commit_charge(mem, pc, ctype, page_size);
2383 int mem_cgroup_newpage_charge(struct page *page,
2384 struct mm_struct *mm, gfp_t gfp_mask)
2386 if (mem_cgroup_disabled())
2389 * If already mapped, we don't have to account.
2390 * If page cache, page->mapping has address_space.
2391 * But page->mapping may have out-of-use anon_vma pointer,
2392 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2395 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
2399 return mem_cgroup_charge_common(page, mm, gfp_mask,
2400 MEM_CGROUP_CHARGE_TYPE_MAPPED);
2404 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2405 enum charge_type ctype);
2407 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2412 if (mem_cgroup_disabled())
2414 if (PageCompound(page))
2417 * Corner case handling. This is called from add_to_page_cache()
2418 * in usual. But some FS (shmem) precharges this page before calling it
2419 * and call add_to_page_cache() with GFP_NOWAIT.
2421 * For GFP_NOWAIT case, the page may be pre-charged before calling
2422 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
2423 * charge twice. (It works but has to pay a bit larger cost.)
2424 * And when the page is SwapCache, it should take swap information
2425 * into account. This is under lock_page() now.
2427 if (!(gfp_mask & __GFP_WAIT)) {
2428 struct page_cgroup *pc;
2430 pc = lookup_page_cgroup(page);
2433 lock_page_cgroup(pc);
2434 if (PageCgroupUsed(pc)) {
2435 unlock_page_cgroup(pc);
2438 unlock_page_cgroup(pc);
2444 if (page_is_file_cache(page))
2445 return mem_cgroup_charge_common(page, mm, gfp_mask,
2446 MEM_CGROUP_CHARGE_TYPE_CACHE);
2449 if (PageSwapCache(page)) {
2450 struct mem_cgroup *mem = NULL;
2452 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2454 __mem_cgroup_commit_charge_swapin(page, mem,
2455 MEM_CGROUP_CHARGE_TYPE_SHMEM);
2457 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
2458 MEM_CGROUP_CHARGE_TYPE_SHMEM);
2464 * While swap-in, try_charge -> commit or cancel, the page is locked.
2465 * And when try_charge() successfully returns, one refcnt to memcg without
2466 * struct page_cgroup is acquired. This refcnt will be consumed by
2467 * "commit()" or removed by "cancel()"
2469 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2471 gfp_t mask, struct mem_cgroup **ptr)
2473 struct mem_cgroup *mem;
2476 if (mem_cgroup_disabled())
2479 if (!do_swap_account)
2482 * A racing thread's fault, or swapoff, may have already updated
2483 * the pte, and even removed page from swap cache: in those cases
2484 * do_swap_page()'s pte_same() test will fail; but there's also a
2485 * KSM case which does need to charge the page.
2487 if (!PageSwapCache(page))
2489 mem = try_get_mem_cgroup_from_page(page);
2493 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, PAGE_SIZE);
2499 return __mem_cgroup_try_charge(mm, mask, ptr, true, PAGE_SIZE);
2503 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2504 enum charge_type ctype)
2506 struct page_cgroup *pc;
2508 if (mem_cgroup_disabled())
2512 cgroup_exclude_rmdir(&ptr->css);
2513 pc = lookup_page_cgroup(page);
2514 mem_cgroup_lru_del_before_commit_swapcache(page);
2515 __mem_cgroup_commit_charge(ptr, pc, ctype, PAGE_SIZE);
2516 mem_cgroup_lru_add_after_commit_swapcache(page);
2518 * Now swap is on-memory. This means this page may be
2519 * counted both as mem and swap....double count.
2520 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2521 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2522 * may call delete_from_swap_cache() before reach here.
2524 if (do_swap_account && PageSwapCache(page)) {
2525 swp_entry_t ent = {.val = page_private(page)};
2527 struct mem_cgroup *memcg;
2529 id = swap_cgroup_record(ent, 0);
2531 memcg = mem_cgroup_lookup(id);
2534 * This recorded memcg can be obsolete one. So, avoid
2535 * calling css_tryget
2537 if (!mem_cgroup_is_root(memcg))
2538 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
2539 mem_cgroup_swap_statistics(memcg, false);
2540 mem_cgroup_put(memcg);
2545 * At swapin, we may charge account against cgroup which has no tasks.
2546 * So, rmdir()->pre_destroy() can be called while we do this charge.
2547 * In that case, we need to call pre_destroy() again. check it here.
2549 cgroup_release_and_wakeup_rmdir(&ptr->css);
2552 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
2554 __mem_cgroup_commit_charge_swapin(page, ptr,
2555 MEM_CGROUP_CHARGE_TYPE_MAPPED);
2558 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
2560 if (mem_cgroup_disabled())
2564 mem_cgroup_cancel_charge(mem, PAGE_SIZE);
2568 __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype,
2571 struct memcg_batch_info *batch = NULL;
2572 bool uncharge_memsw = true;
2573 /* If swapout, usage of swap doesn't decrease */
2574 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2575 uncharge_memsw = false;
2577 batch = ¤t->memcg_batch;
2579 * In usual, we do css_get() when we remember memcg pointer.
2580 * But in this case, we keep res->usage until end of a series of
2581 * uncharges. Then, it's ok to ignore memcg's refcnt.
2586 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2587 * In those cases, all pages freed continously can be expected to be in
2588 * the same cgroup and we have chance to coalesce uncharges.
2589 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2590 * because we want to do uncharge as soon as possible.
2593 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2594 goto direct_uncharge;
2596 if (page_size != PAGE_SIZE)
2597 goto direct_uncharge;
2600 * In typical case, batch->memcg == mem. This means we can
2601 * merge a series of uncharges to an uncharge of res_counter.
2602 * If not, we uncharge res_counter ony by one.
2604 if (batch->memcg != mem)
2605 goto direct_uncharge;
2606 /* remember freed charge and uncharge it later */
2607 batch->bytes += PAGE_SIZE;
2609 batch->memsw_bytes += PAGE_SIZE;
2612 res_counter_uncharge(&mem->res, page_size);
2614 res_counter_uncharge(&mem->memsw, page_size);
2615 if (unlikely(batch->memcg != mem))
2616 memcg_oom_recover(mem);
2621 * uncharge if !page_mapped(page)
2623 static struct mem_cgroup *
2624 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
2627 struct page_cgroup *pc;
2628 struct mem_cgroup *mem = NULL;
2629 int page_size = PAGE_SIZE;
2631 if (mem_cgroup_disabled())
2634 if (PageSwapCache(page))
2637 if (PageTransHuge(page)) {
2638 page_size <<= compound_order(page);
2639 VM_BUG_ON(!PageTransHuge(page));
2642 count = page_size >> PAGE_SHIFT;
2644 * Check if our page_cgroup is valid
2646 pc = lookup_page_cgroup(page);
2647 if (unlikely(!pc || !PageCgroupUsed(pc)))
2650 lock_page_cgroup(pc);
2652 mem = pc->mem_cgroup;
2654 if (!PageCgroupUsed(pc))
2658 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2659 case MEM_CGROUP_CHARGE_TYPE_DROP:
2660 /* See mem_cgroup_prepare_migration() */
2661 if (page_mapped(page) || PageCgroupMigration(pc))
2664 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
2665 if (!PageAnon(page)) { /* Shared memory */
2666 if (page->mapping && !page_is_file_cache(page))
2668 } else if (page_mapped(page)) /* Anon */
2675 mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), -count);
2677 ClearPageCgroupUsed(pc);
2679 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2680 * freed from LRU. This is safe because uncharged page is expected not
2681 * to be reused (freed soon). Exception is SwapCache, it's handled by
2682 * special functions.
2685 unlock_page_cgroup(pc);
2687 * even after unlock, we have mem->res.usage here and this memcg
2688 * will never be freed.
2690 memcg_check_events(mem, page);
2691 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
2692 mem_cgroup_swap_statistics(mem, true);
2693 mem_cgroup_get(mem);
2695 if (!mem_cgroup_is_root(mem))
2696 __do_uncharge(mem, ctype, page_size);
2701 unlock_page_cgroup(pc);
2705 void mem_cgroup_uncharge_page(struct page *page)
2708 if (page_mapped(page))
2710 if (page->mapping && !PageAnon(page))
2712 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
2715 void mem_cgroup_uncharge_cache_page(struct page *page)
2717 VM_BUG_ON(page_mapped(page));
2718 VM_BUG_ON(page->mapping);
2719 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
2723 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2724 * In that cases, pages are freed continuously and we can expect pages
2725 * are in the same memcg. All these calls itself limits the number of
2726 * pages freed at once, then uncharge_start/end() is called properly.
2727 * This may be called prural(2) times in a context,
2730 void mem_cgroup_uncharge_start(void)
2732 current->memcg_batch.do_batch++;
2733 /* We can do nest. */
2734 if (current->memcg_batch.do_batch == 1) {
2735 current->memcg_batch.memcg = NULL;
2736 current->memcg_batch.bytes = 0;
2737 current->memcg_batch.memsw_bytes = 0;
2741 void mem_cgroup_uncharge_end(void)
2743 struct memcg_batch_info *batch = ¤t->memcg_batch;
2745 if (!batch->do_batch)
2749 if (batch->do_batch) /* If stacked, do nothing. */
2755 * This "batch->memcg" is valid without any css_get/put etc...
2756 * bacause we hide charges behind us.
2759 res_counter_uncharge(&batch->memcg->res, batch->bytes);
2760 if (batch->memsw_bytes)
2761 res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes);
2762 memcg_oom_recover(batch->memcg);
2763 /* forget this pointer (for sanity check) */
2764 batch->memcg = NULL;
2769 * called after __delete_from_swap_cache() and drop "page" account.
2770 * memcg information is recorded to swap_cgroup of "ent"
2773 mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
2775 struct mem_cgroup *memcg;
2776 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
2778 if (!swapout) /* this was a swap cache but the swap is unused ! */
2779 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
2781 memcg = __mem_cgroup_uncharge_common(page, ctype);
2784 * record memcg information, if swapout && memcg != NULL,
2785 * mem_cgroup_get() was called in uncharge().
2787 if (do_swap_account && swapout && memcg)
2788 swap_cgroup_record(ent, css_id(&memcg->css));
2792 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2794 * called from swap_entry_free(). remove record in swap_cgroup and
2795 * uncharge "memsw" account.
2797 void mem_cgroup_uncharge_swap(swp_entry_t ent)
2799 struct mem_cgroup *memcg;
2802 if (!do_swap_account)
2805 id = swap_cgroup_record(ent, 0);
2807 memcg = mem_cgroup_lookup(id);
2810 * We uncharge this because swap is freed.
2811 * This memcg can be obsolete one. We avoid calling css_tryget
2813 if (!mem_cgroup_is_root(memcg))
2814 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
2815 mem_cgroup_swap_statistics(memcg, false);
2816 mem_cgroup_put(memcg);
2822 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2823 * @entry: swap entry to be moved
2824 * @from: mem_cgroup which the entry is moved from
2825 * @to: mem_cgroup which the entry is moved to
2826 * @need_fixup: whether we should fixup res_counters and refcounts.
2828 * It succeeds only when the swap_cgroup's record for this entry is the same
2829 * as the mem_cgroup's id of @from.
2831 * Returns 0 on success, -EINVAL on failure.
2833 * The caller must have charged to @to, IOW, called res_counter_charge() about
2834 * both res and memsw, and called css_get().
2836 static int mem_cgroup_move_swap_account(swp_entry_t entry,
2837 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
2839 unsigned short old_id, new_id;
2841 old_id = css_id(&from->css);
2842 new_id = css_id(&to->css);
2844 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
2845 mem_cgroup_swap_statistics(from, false);
2846 mem_cgroup_swap_statistics(to, true);
2848 * This function is only called from task migration context now.
2849 * It postpones res_counter and refcount handling till the end
2850 * of task migration(mem_cgroup_clear_mc()) for performance
2851 * improvement. But we cannot postpone mem_cgroup_get(to)
2852 * because if the process that has been moved to @to does
2853 * swap-in, the refcount of @to might be decreased to 0.
2857 if (!mem_cgroup_is_root(from))
2858 res_counter_uncharge(&from->memsw, PAGE_SIZE);
2859 mem_cgroup_put(from);
2861 * we charged both to->res and to->memsw, so we should
2864 if (!mem_cgroup_is_root(to))
2865 res_counter_uncharge(&to->res, PAGE_SIZE);
2872 static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
2873 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
2880 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2883 int mem_cgroup_prepare_migration(struct page *page,
2884 struct page *newpage, struct mem_cgroup **ptr)
2886 struct page_cgroup *pc;
2887 struct mem_cgroup *mem = NULL;
2888 enum charge_type ctype;
2891 VM_BUG_ON(PageTransHuge(page));
2892 if (mem_cgroup_disabled())
2895 pc = lookup_page_cgroup(page);
2896 lock_page_cgroup(pc);
2897 if (PageCgroupUsed(pc)) {
2898 mem = pc->mem_cgroup;
2901 * At migrating an anonymous page, its mapcount goes down
2902 * to 0 and uncharge() will be called. But, even if it's fully
2903 * unmapped, migration may fail and this page has to be
2904 * charged again. We set MIGRATION flag here and delay uncharge
2905 * until end_migration() is called
2907 * Corner Case Thinking
2909 * When the old page was mapped as Anon and it's unmap-and-freed
2910 * while migration was ongoing.
2911 * If unmap finds the old page, uncharge() of it will be delayed
2912 * until end_migration(). If unmap finds a new page, it's
2913 * uncharged when it make mapcount to be 1->0. If unmap code
2914 * finds swap_migration_entry, the new page will not be mapped
2915 * and end_migration() will find it(mapcount==0).
2918 * When the old page was mapped but migraion fails, the kernel
2919 * remaps it. A charge for it is kept by MIGRATION flag even
2920 * if mapcount goes down to 0. We can do remap successfully
2921 * without charging it again.
2924 * The "old" page is under lock_page() until the end of
2925 * migration, so, the old page itself will not be swapped-out.
2926 * If the new page is swapped out before end_migraton, our
2927 * hook to usual swap-out path will catch the event.
2930 SetPageCgroupMigration(pc);
2932 unlock_page_cgroup(pc);
2934 * If the page is not charged at this point,
2941 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, ptr, false, PAGE_SIZE);
2942 css_put(&mem->css);/* drop extra refcnt */
2943 if (ret || *ptr == NULL) {
2944 if (PageAnon(page)) {
2945 lock_page_cgroup(pc);
2946 ClearPageCgroupMigration(pc);
2947 unlock_page_cgroup(pc);
2949 * The old page may be fully unmapped while we kept it.
2951 mem_cgroup_uncharge_page(page);
2956 * We charge new page before it's used/mapped. So, even if unlock_page()
2957 * is called before end_migration, we can catch all events on this new
2958 * page. In the case new page is migrated but not remapped, new page's
2959 * mapcount will be finally 0 and we call uncharge in end_migration().
2961 pc = lookup_page_cgroup(newpage);
2963 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
2964 else if (page_is_file_cache(page))
2965 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
2967 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
2968 __mem_cgroup_commit_charge(mem, pc, ctype, PAGE_SIZE);
2972 /* remove redundant charge if migration failed*/
2973 void mem_cgroup_end_migration(struct mem_cgroup *mem,
2974 struct page *oldpage, struct page *newpage, bool migration_ok)
2976 struct page *used, *unused;
2977 struct page_cgroup *pc;
2981 /* blocks rmdir() */
2982 cgroup_exclude_rmdir(&mem->css);
2983 if (!migration_ok) {
2991 * We disallowed uncharge of pages under migration because mapcount
2992 * of the page goes down to zero, temporarly.
2993 * Clear the flag and check the page should be charged.
2995 pc = lookup_page_cgroup(oldpage);
2996 lock_page_cgroup(pc);
2997 ClearPageCgroupMigration(pc);
2998 unlock_page_cgroup(pc);
3000 __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE);
3003 * If a page is a file cache, radix-tree replacement is very atomic
3004 * and we can skip this check. When it was an Anon page, its mapcount
3005 * goes down to 0. But because we added MIGRATION flage, it's not
3006 * uncharged yet. There are several case but page->mapcount check
3007 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3008 * check. (see prepare_charge() also)
3011 mem_cgroup_uncharge_page(used);
3013 * At migration, we may charge account against cgroup which has no
3015 * So, rmdir()->pre_destroy() can be called while we do this charge.
3016 * In that case, we need to call pre_destroy() again. check it here.
3018 cgroup_release_and_wakeup_rmdir(&mem->css);
3022 * A call to try to shrink memory usage on charge failure at shmem's swapin.
3023 * Calling hierarchical_reclaim is not enough because we should update
3024 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
3025 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
3026 * not from the memcg which this page would be charged to.
3027 * try_charge_swapin does all of these works properly.
3029 int mem_cgroup_shmem_charge_fallback(struct page *page,
3030 struct mm_struct *mm,
3033 struct mem_cgroup *mem = NULL;
3036 if (mem_cgroup_disabled())
3039 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
3041 mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
3046 static DEFINE_MUTEX(set_limit_mutex);
3048 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
3049 unsigned long long val)
3052 u64 memswlimit, memlimit;
3054 int children = mem_cgroup_count_children(memcg);
3055 u64 curusage, oldusage;
3059 * For keeping hierarchical_reclaim simple, how long we should retry
3060 * is depends on callers. We set our retry-count to be function
3061 * of # of children which we should visit in this loop.
3063 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3065 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3068 while (retry_count) {
3069 if (signal_pending(current)) {
3074 * Rather than hide all in some function, I do this in
3075 * open coded manner. You see what this really does.
3076 * We have to guarantee mem->res.limit < mem->memsw.limit.
3078 mutex_lock(&set_limit_mutex);
3079 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3080 if (memswlimit < val) {
3082 mutex_unlock(&set_limit_mutex);
3086 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3090 ret = res_counter_set_limit(&memcg->res, val);
3092 if (memswlimit == val)
3093 memcg->memsw_is_minimum = true;
3095 memcg->memsw_is_minimum = false;
3097 mutex_unlock(&set_limit_mutex);
3102 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
3103 MEM_CGROUP_RECLAIM_SHRINK);
3104 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3105 /* Usage is reduced ? */
3106 if (curusage >= oldusage)
3109 oldusage = curusage;
3111 if (!ret && enlarge)
3112 memcg_oom_recover(memcg);
3117 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3118 unsigned long long val)
3121 u64 memlimit, memswlimit, oldusage, curusage;
3122 int children = mem_cgroup_count_children(memcg);
3126 /* see mem_cgroup_resize_res_limit */
3127 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3128 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
3129 while (retry_count) {
3130 if (signal_pending(current)) {
3135 * Rather than hide all in some function, I do this in
3136 * open coded manner. You see what this really does.
3137 * We have to guarantee mem->res.limit < mem->memsw.limit.
3139 mutex_lock(&set_limit_mutex);
3140 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3141 if (memlimit > val) {
3143 mutex_unlock(&set_limit_mutex);
3146 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3147 if (memswlimit < val)
3149 ret = res_counter_set_limit(&memcg->memsw, val);
3151 if (memlimit == val)
3152 memcg->memsw_is_minimum = true;
3154 memcg->memsw_is_minimum = false;
3156 mutex_unlock(&set_limit_mutex);
3161 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
3162 MEM_CGROUP_RECLAIM_NOSWAP |
3163 MEM_CGROUP_RECLAIM_SHRINK);
3164 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
3165 /* Usage is reduced ? */
3166 if (curusage >= oldusage)
3169 oldusage = curusage;
3171 if (!ret && enlarge)
3172 memcg_oom_recover(memcg);
3176 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
3179 unsigned long nr_reclaimed = 0;
3180 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3181 unsigned long reclaimed;
3183 struct mem_cgroup_tree_per_zone *mctz;
3184 unsigned long long excess;
3189 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
3191 * This loop can run a while, specially if mem_cgroup's continuously
3192 * keep exceeding their soft limit and putting the system under
3199 mz = mem_cgroup_largest_soft_limit_node(mctz);
3203 reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone,
3205 MEM_CGROUP_RECLAIM_SOFT);
3206 nr_reclaimed += reclaimed;
3207 spin_lock(&mctz->lock);
3210 * If we failed to reclaim anything from this memory cgroup
3211 * it is time to move on to the next cgroup
3217 * Loop until we find yet another one.
3219 * By the time we get the soft_limit lock
3220 * again, someone might have aded the
3221 * group back on the RB tree. Iterate to
3222 * make sure we get a different mem.
3223 * mem_cgroup_largest_soft_limit_node returns
3224 * NULL if no other cgroup is present on
3228 __mem_cgroup_largest_soft_limit_node(mctz);
3229 if (next_mz == mz) {
3230 css_put(&next_mz->mem->css);
3232 } else /* next_mz == NULL or other memcg */
3236 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
3237 excess = res_counter_soft_limit_excess(&mz->mem->res);
3239 * One school of thought says that we should not add
3240 * back the node to the tree if reclaim returns 0.
3241 * But our reclaim could return 0, simply because due
3242 * to priority we are exposing a smaller subset of
3243 * memory to reclaim from. Consider this as a longer
3246 /* If excess == 0, no tree ops */
3247 __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
3248 spin_unlock(&mctz->lock);
3249 css_put(&mz->mem->css);
3252 * Could not reclaim anything and there are no more
3253 * mem cgroups to try or we seem to be looping without
3254 * reclaiming anything.
3256 if (!nr_reclaimed &&
3258 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3260 } while (!nr_reclaimed);
3262 css_put(&next_mz->mem->css);
3263 return nr_reclaimed;
3267 * This routine traverse page_cgroup in given list and drop them all.
3268 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3270 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
3271 int node, int zid, enum lru_list lru)
3274 struct mem_cgroup_per_zone *mz;
3275 struct page_cgroup *pc, *busy;
3276 unsigned long flags, loop;
3277 struct list_head *list;
3280 zone = &NODE_DATA(node)->node_zones[zid];
3281 mz = mem_cgroup_zoneinfo(mem, node, zid);
3282 list = &mz->lists[lru];
3284 loop = MEM_CGROUP_ZSTAT(mz, lru);
3285 /* give some margin against EBUSY etc...*/
3290 spin_lock_irqsave(&zone->lru_lock, flags);
3291 if (list_empty(list)) {
3292 spin_unlock_irqrestore(&zone->lru_lock, flags);
3295 pc = list_entry(list->prev, struct page_cgroup, lru);
3297 list_move(&pc->lru, list);
3299 spin_unlock_irqrestore(&zone->lru_lock, flags);
3302 spin_unlock_irqrestore(&zone->lru_lock, flags);
3304 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
3308 if (ret == -EBUSY || ret == -EINVAL) {
3309 /* found lock contention or "pc" is obsolete. */
3316 if (!ret && !list_empty(list))
3322 * make mem_cgroup's charge to be 0 if there is no task.
3323 * This enables deleting this mem_cgroup.
3325 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
3328 int node, zid, shrink;
3329 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
3330 struct cgroup *cgrp = mem->css.cgroup;
3335 /* should free all ? */
3341 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
3344 if (signal_pending(current))
3346 /* This is for making all *used* pages to be on LRU. */
3347 lru_add_drain_all();
3348 drain_all_stock_sync();
3350 mem_cgroup_start_move(mem);
3351 for_each_node_state(node, N_HIGH_MEMORY) {
3352 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
3355 ret = mem_cgroup_force_empty_list(mem,
3364 mem_cgroup_end_move(mem);
3365 memcg_oom_recover(mem);
3366 /* it seems parent cgroup doesn't have enough mem */
3370 /* "ret" should also be checked to ensure all lists are empty. */
3371 } while (mem->res.usage > 0 || ret);
3377 /* returns EBUSY if there is a task or if we come here twice. */
3378 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
3382 /* we call try-to-free pages for make this cgroup empty */
3383 lru_add_drain_all();
3384 /* try to free all pages in this cgroup */
3386 while (nr_retries && mem->res.usage > 0) {
3389 if (signal_pending(current)) {
3393 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
3394 false, get_swappiness(mem));
3397 /* maybe some writeback is necessary */
3398 congestion_wait(BLK_RW_ASYNC, HZ/10);
3403 /* try move_account...there may be some *locked* pages. */
3407 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3409 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3413 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3415 return mem_cgroup_from_cont(cont)->use_hierarchy;
3418 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3422 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3423 struct cgroup *parent = cont->parent;
3424 struct mem_cgroup *parent_mem = NULL;
3427 parent_mem = mem_cgroup_from_cont(parent);
3431 * If parent's use_hierarchy is set, we can't make any modifications
3432 * in the child subtrees. If it is unset, then the change can
3433 * occur, provided the current cgroup has no children.
3435 * For the root cgroup, parent_mem is NULL, we allow value to be
3436 * set if there are no children.
3438 if ((!parent_mem || !parent_mem->use_hierarchy) &&
3439 (val == 1 || val == 0)) {
3440 if (list_empty(&cont->children))
3441 mem->use_hierarchy = val;
3452 static u64 mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem,
3453 enum mem_cgroup_stat_index idx)
3455 struct mem_cgroup *iter;
3458 /* each per cpu's value can be minus.Then, use s64 */
3459 for_each_mem_cgroup_tree(iter, mem)
3460 val += mem_cgroup_read_stat(iter, idx);
3462 if (val < 0) /* race ? */
3467 static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap)
3471 if (!mem_cgroup_is_root(mem)) {
3473 return res_counter_read_u64(&mem->res, RES_USAGE);
3475 return res_counter_read_u64(&mem->memsw, RES_USAGE);
3478 val = mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE);
3479 val += mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS);
3482 val += mem_cgroup_get_recursive_idx_stat(mem,
3483 MEM_CGROUP_STAT_SWAPOUT);
3485 return val << PAGE_SHIFT;
3488 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
3490 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3494 type = MEMFILE_TYPE(cft->private);
3495 name = MEMFILE_ATTR(cft->private);
3498 if (name == RES_USAGE)
3499 val = mem_cgroup_usage(mem, false);
3501 val = res_counter_read_u64(&mem->res, name);
3504 if (name == RES_USAGE)
3505 val = mem_cgroup_usage(mem, true);
3507 val = res_counter_read_u64(&mem->memsw, name);
3516 * The user of this function is...
3519 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3522 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3524 unsigned long long val;
3527 type = MEMFILE_TYPE(cft->private);
3528 name = MEMFILE_ATTR(cft->private);
3531 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3535 /* This function does all necessary parse...reuse it */
3536 ret = res_counter_memparse_write_strategy(buffer, &val);
3540 ret = mem_cgroup_resize_limit(memcg, val);
3542 ret = mem_cgroup_resize_memsw_limit(memcg, val);
3544 case RES_SOFT_LIMIT:
3545 ret = res_counter_memparse_write_strategy(buffer, &val);
3549 * For memsw, soft limits are hard to implement in terms
3550 * of semantics, for now, we support soft limits for
3551 * control without swap
3554 ret = res_counter_set_soft_limit(&memcg->res, val);
3559 ret = -EINVAL; /* should be BUG() ? */
3565 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3566 unsigned long long *mem_limit, unsigned long long *memsw_limit)
3568 struct cgroup *cgroup;
3569 unsigned long long min_limit, min_memsw_limit, tmp;
3571 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3572 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3573 cgroup = memcg->css.cgroup;
3574 if (!memcg->use_hierarchy)
3577 while (cgroup->parent) {
3578 cgroup = cgroup->parent;
3579 memcg = mem_cgroup_from_cont(cgroup);
3580 if (!memcg->use_hierarchy)
3582 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
3583 min_limit = min(min_limit, tmp);
3584 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3585 min_memsw_limit = min(min_memsw_limit, tmp);
3588 *mem_limit = min_limit;
3589 *memsw_limit = min_memsw_limit;
3593 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
3595 struct mem_cgroup *mem;
3598 mem = mem_cgroup_from_cont(cont);
3599 type = MEMFILE_TYPE(event);
3600 name = MEMFILE_ATTR(event);
3604 res_counter_reset_max(&mem->res);
3606 res_counter_reset_max(&mem->memsw);
3610 res_counter_reset_failcnt(&mem->res);
3612 res_counter_reset_failcnt(&mem->memsw);
3619 static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
3622 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
3626 static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3627 struct cftype *cft, u64 val)
3629 struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
3631 if (val >= (1 << NR_MOVE_TYPE))
3634 * We check this value several times in both in can_attach() and
3635 * attach(), so we need cgroup lock to prevent this value from being
3639 mem->move_charge_at_immigrate = val;
3645 static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3646 struct cftype *cft, u64 val)
3653 /* For read statistics */
3669 struct mcs_total_stat {
3670 s64 stat[NR_MCS_STAT];
3676 } memcg_stat_strings[NR_MCS_STAT] = {
3677 {"cache", "total_cache"},
3678 {"rss", "total_rss"},
3679 {"mapped_file", "total_mapped_file"},
3680 {"pgpgin", "total_pgpgin"},
3681 {"pgpgout", "total_pgpgout"},
3682 {"swap", "total_swap"},
3683 {"inactive_anon", "total_inactive_anon"},
3684 {"active_anon", "total_active_anon"},
3685 {"inactive_file", "total_inactive_file"},
3686 {"active_file", "total_active_file"},
3687 {"unevictable", "total_unevictable"}
3692 mem_cgroup_get_local_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
3697 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
3698 s->stat[MCS_CACHE] += val * PAGE_SIZE;
3699 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
3700 s->stat[MCS_RSS] += val * PAGE_SIZE;
3701 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED);
3702 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
3703 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT);
3704 s->stat[MCS_PGPGIN] += val;
3705 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT);
3706 s->stat[MCS_PGPGOUT] += val;
3707 if (do_swap_account) {
3708 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT);
3709 s->stat[MCS_SWAP] += val * PAGE_SIZE;
3713 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
3714 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
3715 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
3716 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
3717 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
3718 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
3719 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
3720 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
3721 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
3722 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
3726 mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
3728 struct mem_cgroup *iter;
3730 for_each_mem_cgroup_tree(iter, mem)
3731 mem_cgroup_get_local_stat(iter, s);
3734 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
3735 struct cgroup_map_cb *cb)
3737 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
3738 struct mcs_total_stat mystat;
3741 memset(&mystat, 0, sizeof(mystat));
3742 mem_cgroup_get_local_stat(mem_cont, &mystat);
3744 for (i = 0; i < NR_MCS_STAT; i++) {
3745 if (i == MCS_SWAP && !do_swap_account)
3747 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
3750 /* Hierarchical information */
3752 unsigned long long limit, memsw_limit;
3753 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
3754 cb->fill(cb, "hierarchical_memory_limit", limit);
3755 if (do_swap_account)
3756 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
3759 memset(&mystat, 0, sizeof(mystat));
3760 mem_cgroup_get_total_stat(mem_cont, &mystat);
3761 for (i = 0; i < NR_MCS_STAT; i++) {
3762 if (i == MCS_SWAP && !do_swap_account)
3764 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
3767 #ifdef CONFIG_DEBUG_VM
3768 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
3772 struct mem_cgroup_per_zone *mz;
3773 unsigned long recent_rotated[2] = {0, 0};
3774 unsigned long recent_scanned[2] = {0, 0};
3776 for_each_online_node(nid)
3777 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
3778 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
3780 recent_rotated[0] +=
3781 mz->reclaim_stat.recent_rotated[0];
3782 recent_rotated[1] +=
3783 mz->reclaim_stat.recent_rotated[1];
3784 recent_scanned[0] +=
3785 mz->reclaim_stat.recent_scanned[0];
3786 recent_scanned[1] +=
3787 mz->reclaim_stat.recent_scanned[1];
3789 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
3790 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
3791 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
3792 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
3799 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
3801 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3803 return get_swappiness(memcg);
3806 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
3809 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3810 struct mem_cgroup *parent;
3815 if (cgrp->parent == NULL)
3818 parent = mem_cgroup_from_cont(cgrp->parent);
3822 /* If under hierarchy, only empty-root can set this value */
3823 if ((parent->use_hierarchy) ||
3824 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
3829 spin_lock(&memcg->reclaim_param_lock);
3830 memcg->swappiness = val;
3831 spin_unlock(&memcg->reclaim_param_lock);
3838 static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3840 struct mem_cgroup_threshold_ary *t;
3846 t = rcu_dereference(memcg->thresholds.primary);
3848 t = rcu_dereference(memcg->memsw_thresholds.primary);
3853 usage = mem_cgroup_usage(memcg, swap);
3856 * current_threshold points to threshold just below usage.
3857 * If it's not true, a threshold was crossed after last
3858 * call of __mem_cgroup_threshold().
3860 i = t->current_threshold;
3863 * Iterate backward over array of thresholds starting from
3864 * current_threshold and check if a threshold is crossed.
3865 * If none of thresholds below usage is crossed, we read
3866 * only one element of the array here.
3868 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3869 eventfd_signal(t->entries[i].eventfd, 1);
3871 /* i = current_threshold + 1 */
3875 * Iterate forward over array of thresholds starting from
3876 * current_threshold+1 and check if a threshold is crossed.
3877 * If none of thresholds above usage is crossed, we read
3878 * only one element of the array here.
3880 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3881 eventfd_signal(t->entries[i].eventfd, 1);
3883 /* Update current_threshold */
3884 t->current_threshold = i - 1;
3889 static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3892 __mem_cgroup_threshold(memcg, false);
3893 if (do_swap_account)
3894 __mem_cgroup_threshold(memcg, true);
3896 memcg = parent_mem_cgroup(memcg);
3900 static int compare_thresholds(const void *a, const void *b)
3902 const struct mem_cgroup_threshold *_a = a;
3903 const struct mem_cgroup_threshold *_b = b;
3905 return _a->threshold - _b->threshold;
3908 static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem)
3910 struct mem_cgroup_eventfd_list *ev;
3912 list_for_each_entry(ev, &mem->oom_notify, list)
3913 eventfd_signal(ev->eventfd, 1);
3917 static void mem_cgroup_oom_notify(struct mem_cgroup *mem)
3919 struct mem_cgroup *iter;
3921 for_each_mem_cgroup_tree(iter, mem)
3922 mem_cgroup_oom_notify_cb(iter);
3925 static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
3926 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
3928 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3929 struct mem_cgroup_thresholds *thresholds;
3930 struct mem_cgroup_threshold_ary *new;
3931 int type = MEMFILE_TYPE(cft->private);
3932 u64 threshold, usage;
3935 ret = res_counter_memparse_write_strategy(args, &threshold);
3939 mutex_lock(&memcg->thresholds_lock);
3942 thresholds = &memcg->thresholds;
3943 else if (type == _MEMSWAP)
3944 thresholds = &memcg->memsw_thresholds;
3948 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
3950 /* Check if a threshold crossed before adding a new one */
3951 if (thresholds->primary)
3952 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3954 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
3956 /* Allocate memory for new array of thresholds */
3957 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
3965 /* Copy thresholds (if any) to new array */
3966 if (thresholds->primary) {
3967 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
3968 sizeof(struct mem_cgroup_threshold));
3971 /* Add new threshold */
3972 new->entries[size - 1].eventfd = eventfd;
3973 new->entries[size - 1].threshold = threshold;
3975 /* Sort thresholds. Registering of new threshold isn't time-critical */
3976 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
3977 compare_thresholds, NULL);
3979 /* Find current threshold */
3980 new->current_threshold = -1;
3981 for (i = 0; i < size; i++) {
3982 if (new->entries[i].threshold < usage) {
3984 * new->current_threshold will not be used until
3985 * rcu_assign_pointer(), so it's safe to increment
3988 ++new->current_threshold;
3992 /* Free old spare buffer and save old primary buffer as spare */
3993 kfree(thresholds->spare);
3994 thresholds->spare = thresholds->primary;
3996 rcu_assign_pointer(thresholds->primary, new);
3998 /* To be sure that nobody uses thresholds */
4002 mutex_unlock(&memcg->thresholds_lock);
4007 static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
4008 struct cftype *cft, struct eventfd_ctx *eventfd)
4010 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4011 struct mem_cgroup_thresholds *thresholds;
4012 struct mem_cgroup_threshold_ary *new;
4013 int type = MEMFILE_TYPE(cft->private);
4017 mutex_lock(&memcg->thresholds_lock);
4019 thresholds = &memcg->thresholds;
4020 else if (type == _MEMSWAP)
4021 thresholds = &memcg->memsw_thresholds;
4026 * Something went wrong if we trying to unregister a threshold
4027 * if we don't have thresholds
4029 BUG_ON(!thresholds);
4031 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4033 /* Check if a threshold crossed before removing */
4034 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4036 /* Calculate new number of threshold */
4038 for (i = 0; i < thresholds->primary->size; i++) {
4039 if (thresholds->primary->entries[i].eventfd != eventfd)
4043 new = thresholds->spare;
4045 /* Set thresholds array to NULL if we don't have thresholds */
4054 /* Copy thresholds and find current threshold */
4055 new->current_threshold = -1;
4056 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4057 if (thresholds->primary->entries[i].eventfd == eventfd)
4060 new->entries[j] = thresholds->primary->entries[i];
4061 if (new->entries[j].threshold < usage) {
4063 * new->current_threshold will not be used
4064 * until rcu_assign_pointer(), so it's safe to increment
4067 ++new->current_threshold;
4073 /* Swap primary and spare array */
4074 thresholds->spare = thresholds->primary;
4075 rcu_assign_pointer(thresholds->primary, new);
4077 /* To be sure that nobody uses thresholds */
4080 mutex_unlock(&memcg->thresholds_lock);
4083 static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4084 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4086 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4087 struct mem_cgroup_eventfd_list *event;
4088 int type = MEMFILE_TYPE(cft->private);
4090 BUG_ON(type != _OOM_TYPE);
4091 event = kmalloc(sizeof(*event), GFP_KERNEL);
4095 mutex_lock(&memcg_oom_mutex);
4097 event->eventfd = eventfd;
4098 list_add(&event->list, &memcg->oom_notify);
4100 /* already in OOM ? */
4101 if (atomic_read(&memcg->oom_lock))
4102 eventfd_signal(eventfd, 1);
4103 mutex_unlock(&memcg_oom_mutex);
4108 static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
4109 struct cftype *cft, struct eventfd_ctx *eventfd)
4111 struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
4112 struct mem_cgroup_eventfd_list *ev, *tmp;
4113 int type = MEMFILE_TYPE(cft->private);
4115 BUG_ON(type != _OOM_TYPE);
4117 mutex_lock(&memcg_oom_mutex);
4119 list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) {
4120 if (ev->eventfd == eventfd) {
4121 list_del(&ev->list);
4126 mutex_unlock(&memcg_oom_mutex);
4129 static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4130 struct cftype *cft, struct cgroup_map_cb *cb)
4132 struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
4134 cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable);
4136 if (atomic_read(&mem->oom_lock))
4137 cb->fill(cb, "under_oom", 1);
4139 cb->fill(cb, "under_oom", 0);
4143 static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4144 struct cftype *cft, u64 val)
4146 struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
4147 struct mem_cgroup *parent;
4149 /* cannot set to root cgroup and only 0 and 1 are allowed */
4150 if (!cgrp->parent || !((val == 0) || (val == 1)))
4153 parent = mem_cgroup_from_cont(cgrp->parent);
4156 /* oom-kill-disable is a flag for subhierarchy. */
4157 if ((parent->use_hierarchy) ||
4158 (mem->use_hierarchy && !list_empty(&cgrp->children))) {
4162 mem->oom_kill_disable = val;
4164 memcg_oom_recover(mem);
4169 static struct cftype mem_cgroup_files[] = {
4171 .name = "usage_in_bytes",
4172 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
4173 .read_u64 = mem_cgroup_read,
4174 .register_event = mem_cgroup_usage_register_event,
4175 .unregister_event = mem_cgroup_usage_unregister_event,
4178 .name = "max_usage_in_bytes",
4179 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
4180 .trigger = mem_cgroup_reset,
4181 .read_u64 = mem_cgroup_read,
4184 .name = "limit_in_bytes",
4185 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
4186 .write_string = mem_cgroup_write,
4187 .read_u64 = mem_cgroup_read,
4190 .name = "soft_limit_in_bytes",
4191 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4192 .write_string = mem_cgroup_write,
4193 .read_u64 = mem_cgroup_read,
4197 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
4198 .trigger = mem_cgroup_reset,
4199 .read_u64 = mem_cgroup_read,
4203 .read_map = mem_control_stat_show,
4206 .name = "force_empty",
4207 .trigger = mem_cgroup_force_empty_write,
4210 .name = "use_hierarchy",
4211 .write_u64 = mem_cgroup_hierarchy_write,
4212 .read_u64 = mem_cgroup_hierarchy_read,
4215 .name = "swappiness",
4216 .read_u64 = mem_cgroup_swappiness_read,
4217 .write_u64 = mem_cgroup_swappiness_write,
4220 .name = "move_charge_at_immigrate",
4221 .read_u64 = mem_cgroup_move_charge_read,
4222 .write_u64 = mem_cgroup_move_charge_write,
4225 .name = "oom_control",
4226 .read_map = mem_cgroup_oom_control_read,
4227 .write_u64 = mem_cgroup_oom_control_write,
4228 .register_event = mem_cgroup_oom_register_event,
4229 .unregister_event = mem_cgroup_oom_unregister_event,
4230 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4234 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4235 static struct cftype memsw_cgroup_files[] = {
4237 .name = "memsw.usage_in_bytes",
4238 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
4239 .read_u64 = mem_cgroup_read,
4240 .register_event = mem_cgroup_usage_register_event,
4241 .unregister_event = mem_cgroup_usage_unregister_event,
4244 .name = "memsw.max_usage_in_bytes",
4245 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4246 .trigger = mem_cgroup_reset,
4247 .read_u64 = mem_cgroup_read,
4250 .name = "memsw.limit_in_bytes",
4251 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4252 .write_string = mem_cgroup_write,
4253 .read_u64 = mem_cgroup_read,
4256 .name = "memsw.failcnt",
4257 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4258 .trigger = mem_cgroup_reset,
4259 .read_u64 = mem_cgroup_read,
4263 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4265 if (!do_swap_account)
4267 return cgroup_add_files(cont, ss, memsw_cgroup_files,
4268 ARRAY_SIZE(memsw_cgroup_files));
4271 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4277 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
4279 struct mem_cgroup_per_node *pn;
4280 struct mem_cgroup_per_zone *mz;
4282 int zone, tmp = node;
4284 * This routine is called against possible nodes.
4285 * But it's BUG to call kmalloc() against offline node.
4287 * TODO: this routine can waste much memory for nodes which will
4288 * never be onlined. It's better to use memory hotplug callback
4291 if (!node_state(node, N_NORMAL_MEMORY))
4293 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
4297 mem->info.nodeinfo[node] = pn;
4298 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4299 mz = &pn->zoneinfo[zone];
4301 INIT_LIST_HEAD(&mz->lists[l]);
4302 mz->usage_in_excess = 0;
4303 mz->on_tree = false;
4309 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
4311 kfree(mem->info.nodeinfo[node]);
4314 static struct mem_cgroup *mem_cgroup_alloc(void)
4316 struct mem_cgroup *mem;
4317 int size = sizeof(struct mem_cgroup);
4319 /* Can be very big if MAX_NUMNODES is very big */
4320 if (size < PAGE_SIZE)
4321 mem = kzalloc(size, GFP_KERNEL);
4323 mem = vzalloc(size);
4328 mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4331 spin_lock_init(&mem->pcp_counter_lock);
4335 if (size < PAGE_SIZE)
4343 * At destroying mem_cgroup, references from swap_cgroup can remain.
4344 * (scanning all at force_empty is too costly...)
4346 * Instead of clearing all references at force_empty, we remember
4347 * the number of reference from swap_cgroup and free mem_cgroup when
4348 * it goes down to 0.
4350 * Removal of cgroup itself succeeds regardless of refs from swap.
4353 static void __mem_cgroup_free(struct mem_cgroup *mem)
4357 mem_cgroup_remove_from_trees(mem);
4358 free_css_id(&mem_cgroup_subsys, &mem->css);
4360 for_each_node_state(node, N_POSSIBLE)
4361 free_mem_cgroup_per_zone_info(mem, node);
4363 free_percpu(mem->stat);
4364 if (sizeof(struct mem_cgroup) < PAGE_SIZE)
4370 static void mem_cgroup_get(struct mem_cgroup *mem)
4372 atomic_inc(&mem->refcnt);
4375 static void __mem_cgroup_put(struct mem_cgroup *mem, int count)
4377 if (atomic_sub_and_test(count, &mem->refcnt)) {
4378 struct mem_cgroup *parent = parent_mem_cgroup(mem);
4379 __mem_cgroup_free(mem);
4381 mem_cgroup_put(parent);
4385 static void mem_cgroup_put(struct mem_cgroup *mem)
4387 __mem_cgroup_put(mem, 1);
4391 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4393 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
4395 if (!mem->res.parent)
4397 return mem_cgroup_from_res_counter(mem->res.parent, res);
4400 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4401 static void __init enable_swap_cgroup(void)
4403 if (!mem_cgroup_disabled() && really_do_swap_account)
4404 do_swap_account = 1;
4407 static void __init enable_swap_cgroup(void)
4412 static int mem_cgroup_soft_limit_tree_init(void)
4414 struct mem_cgroup_tree_per_node *rtpn;
4415 struct mem_cgroup_tree_per_zone *rtpz;
4416 int tmp, node, zone;
4418 for_each_node_state(node, N_POSSIBLE) {
4420 if (!node_state(node, N_NORMAL_MEMORY))
4422 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4426 soft_limit_tree.rb_tree_per_node[node] = rtpn;
4428 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4429 rtpz = &rtpn->rb_tree_per_zone[zone];
4430 rtpz->rb_root = RB_ROOT;
4431 spin_lock_init(&rtpz->lock);
4437 static struct cgroup_subsys_state * __ref
4438 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
4440 struct mem_cgroup *mem, *parent;
4441 long error = -ENOMEM;
4444 mem = mem_cgroup_alloc();
4446 return ERR_PTR(error);
4448 for_each_node_state(node, N_POSSIBLE)
4449 if (alloc_mem_cgroup_per_zone_info(mem, node))
4453 if (cont->parent == NULL) {
4455 enable_swap_cgroup();
4457 root_mem_cgroup = mem;
4458 if (mem_cgroup_soft_limit_tree_init())
4460 for_each_possible_cpu(cpu) {
4461 struct memcg_stock_pcp *stock =
4462 &per_cpu(memcg_stock, cpu);
4463 INIT_WORK(&stock->work, drain_local_stock);
4465 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
4467 parent = mem_cgroup_from_cont(cont->parent);
4468 mem->use_hierarchy = parent->use_hierarchy;
4469 mem->oom_kill_disable = parent->oom_kill_disable;
4472 if (parent && parent->use_hierarchy) {
4473 res_counter_init(&mem->res, &parent->res);
4474 res_counter_init(&mem->memsw, &parent->memsw);
4476 * We increment refcnt of the parent to ensure that we can
4477 * safely access it on res_counter_charge/uncharge.
4478 * This refcnt will be decremented when freeing this
4479 * mem_cgroup(see mem_cgroup_put).
4481 mem_cgroup_get(parent);
4483 res_counter_init(&mem->res, NULL);
4484 res_counter_init(&mem->memsw, NULL);
4486 mem->last_scanned_child = 0;
4487 spin_lock_init(&mem->reclaim_param_lock);
4488 INIT_LIST_HEAD(&mem->oom_notify);
4491 mem->swappiness = get_swappiness(parent);
4492 atomic_set(&mem->refcnt, 1);
4493 mem->move_charge_at_immigrate = 0;
4494 mutex_init(&mem->thresholds_lock);
4497 __mem_cgroup_free(mem);
4498 root_mem_cgroup = NULL;
4499 return ERR_PTR(error);
4502 static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
4503 struct cgroup *cont)
4505 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
4507 return mem_cgroup_force_empty(mem, false);
4510 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
4511 struct cgroup *cont)
4513 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
4515 mem_cgroup_put(mem);
4518 static int mem_cgroup_populate(struct cgroup_subsys *ss,
4519 struct cgroup *cont)
4523 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
4524 ARRAY_SIZE(mem_cgroup_files));
4527 ret = register_memsw_files(cont, ss);
4532 /* Handlers for move charge at task migration. */
4533 #define PRECHARGE_COUNT_AT_ONCE 256
4534 static int mem_cgroup_do_precharge(unsigned long count)
4537 int batch_count = PRECHARGE_COUNT_AT_ONCE;
4538 struct mem_cgroup *mem = mc.to;
4540 if (mem_cgroup_is_root(mem)) {
4541 mc.precharge += count;
4542 /* we don't need css_get for root */
4545 /* try to charge at once */
4547 struct res_counter *dummy;
4549 * "mem" cannot be under rmdir() because we've already checked
4550 * by cgroup_lock_live_cgroup() that it is not removed and we
4551 * are still under the same cgroup_mutex. So we can postpone
4554 if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy))
4556 if (do_swap_account && res_counter_charge(&mem->memsw,
4557 PAGE_SIZE * count, &dummy)) {
4558 res_counter_uncharge(&mem->res, PAGE_SIZE * count);
4561 mc.precharge += count;
4565 /* fall back to one by one charge */
4567 if (signal_pending(current)) {
4571 if (!batch_count--) {
4572 batch_count = PRECHARGE_COUNT_AT_ONCE;
4575 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false,
4578 /* mem_cgroup_clear_mc() will do uncharge later */
4586 * is_target_pte_for_mc - check a pte whether it is valid for move charge
4587 * @vma: the vma the pte to be checked belongs
4588 * @addr: the address corresponding to the pte to be checked
4589 * @ptent: the pte to be checked
4590 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4593 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4594 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4595 * move charge. if @target is not NULL, the page is stored in target->page
4596 * with extra refcnt got(Callers should handle it).
4597 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4598 * target for charge migration. if @target is not NULL, the entry is stored
4601 * Called with pte lock held.
4608 enum mc_target_type {
4609 MC_TARGET_NONE, /* not used */
4614 static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4615 unsigned long addr, pte_t ptent)
4617 struct page *page = vm_normal_page(vma, addr, ptent);
4619 if (!page || !page_mapped(page))
4621 if (PageAnon(page)) {
4622 /* we don't move shared anon */
4623 if (!move_anon() || page_mapcount(page) > 2)
4625 } else if (!move_file())
4626 /* we ignore mapcount for file pages */
4628 if (!get_page_unless_zero(page))
4634 static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4635 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4638 struct page *page = NULL;
4639 swp_entry_t ent = pte_to_swp_entry(ptent);
4641 if (!move_anon() || non_swap_entry(ent))
4643 usage_count = mem_cgroup_count_swap_user(ent, &page);
4644 if (usage_count > 1) { /* we don't move shared anon */
4649 if (do_swap_account)
4650 entry->val = ent.val;
4655 static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4656 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4658 struct page *page = NULL;
4659 struct inode *inode;
4660 struct address_space *mapping;
4663 if (!vma->vm_file) /* anonymous vma */
4668 inode = vma->vm_file->f_path.dentry->d_inode;
4669 mapping = vma->vm_file->f_mapping;
4670 if (pte_none(ptent))
4671 pgoff = linear_page_index(vma, addr);
4672 else /* pte_file(ptent) is true */
4673 pgoff = pte_to_pgoff(ptent);
4675 /* page is moved even if it's not RSS of this task(page-faulted). */
4676 if (!mapping_cap_swap_backed(mapping)) { /* normal file */
4677 page = find_get_page(mapping, pgoff);
4678 } else { /* shmem/tmpfs file. we should take account of swap too. */
4680 mem_cgroup_get_shmem_target(inode, pgoff, &page, &ent);
4681 if (do_swap_account)
4682 entry->val = ent.val;
4688 static int is_target_pte_for_mc(struct vm_area_struct *vma,
4689 unsigned long addr, pte_t ptent, union mc_target *target)
4691 struct page *page = NULL;
4692 struct page_cgroup *pc;
4694 swp_entry_t ent = { .val = 0 };
4696 if (pte_present(ptent))
4697 page = mc_handle_present_pte(vma, addr, ptent);
4698 else if (is_swap_pte(ptent))
4699 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
4700 else if (pte_none(ptent) || pte_file(ptent))
4701 page = mc_handle_file_pte(vma, addr, ptent, &ent);
4703 if (!page && !ent.val)
4706 pc = lookup_page_cgroup(page);
4708 * Do only loose check w/o page_cgroup lock.
4709 * mem_cgroup_move_account() checks the pc is valid or not under
4712 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
4713 ret = MC_TARGET_PAGE;
4715 target->page = page;
4717 if (!ret || !target)
4720 /* There is a swap entry and a page doesn't exist or isn't charged */
4721 if (ent.val && !ret &&
4722 css_id(&mc.from->css) == lookup_swap_cgroup(ent)) {
4723 ret = MC_TARGET_SWAP;
4730 static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4731 unsigned long addr, unsigned long end,
4732 struct mm_walk *walk)
4734 struct vm_area_struct *vma = walk->private;
4738 VM_BUG_ON(pmd_trans_huge(*pmd));
4739 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4740 for (; addr != end; pte++, addr += PAGE_SIZE)
4741 if (is_target_pte_for_mc(vma, addr, *pte, NULL))
4742 mc.precharge++; /* increment precharge temporarily */
4743 pte_unmap_unlock(pte - 1, ptl);
4749 static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4751 unsigned long precharge;
4752 struct vm_area_struct *vma;
4754 down_read(&mm->mmap_sem);
4755 for (vma = mm->mmap; vma; vma = vma->vm_next) {
4756 struct mm_walk mem_cgroup_count_precharge_walk = {
4757 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4761 if (is_vm_hugetlb_page(vma))
4763 walk_page_range(vma->vm_start, vma->vm_end,
4764 &mem_cgroup_count_precharge_walk);
4766 up_read(&mm->mmap_sem);
4768 precharge = mc.precharge;
4774 static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4776 unsigned long precharge = mem_cgroup_count_precharge(mm);
4778 VM_BUG_ON(mc.moving_task);
4779 mc.moving_task = current;
4780 return mem_cgroup_do_precharge(precharge);
4783 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4784 static void __mem_cgroup_clear_mc(void)
4786 struct mem_cgroup *from = mc.from;
4787 struct mem_cgroup *to = mc.to;
4789 /* we must uncharge all the leftover precharges from mc.to */
4791 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
4795 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4796 * we must uncharge here.
4798 if (mc.moved_charge) {
4799 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
4800 mc.moved_charge = 0;
4802 /* we must fixup refcnts and charges */
4803 if (mc.moved_swap) {
4804 /* uncharge swap account from the old cgroup */
4805 if (!mem_cgroup_is_root(mc.from))
4806 res_counter_uncharge(&mc.from->memsw,
4807 PAGE_SIZE * mc.moved_swap);
4808 __mem_cgroup_put(mc.from, mc.moved_swap);
4810 if (!mem_cgroup_is_root(mc.to)) {
4812 * we charged both to->res and to->memsw, so we should
4815 res_counter_uncharge(&mc.to->res,
4816 PAGE_SIZE * mc.moved_swap);
4818 /* we've already done mem_cgroup_get(mc.to) */
4821 memcg_oom_recover(from);
4822 memcg_oom_recover(to);
4823 wake_up_all(&mc.waitq);
4826 static void mem_cgroup_clear_mc(void)
4828 struct mem_cgroup *from = mc.from;
4831 * we must clear moving_task before waking up waiters at the end of
4834 mc.moving_task = NULL;
4835 __mem_cgroup_clear_mc();
4836 spin_lock(&mc.lock);
4839 spin_unlock(&mc.lock);
4840 mem_cgroup_end_move(from);
4843 static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
4844 struct cgroup *cgroup,
4845 struct task_struct *p,
4849 struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup);
4851 if (mem->move_charge_at_immigrate) {
4852 struct mm_struct *mm;
4853 struct mem_cgroup *from = mem_cgroup_from_task(p);
4855 VM_BUG_ON(from == mem);
4857 mm = get_task_mm(p);
4860 /* We move charges only when we move a owner of the mm */
4861 if (mm->owner == p) {
4864 VM_BUG_ON(mc.precharge);
4865 VM_BUG_ON(mc.moved_charge);
4866 VM_BUG_ON(mc.moved_swap);
4867 mem_cgroup_start_move(from);
4868 spin_lock(&mc.lock);
4871 spin_unlock(&mc.lock);
4872 /* We set mc.moving_task later */
4874 ret = mem_cgroup_precharge_mc(mm);
4876 mem_cgroup_clear_mc();
4883 static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
4884 struct cgroup *cgroup,
4885 struct task_struct *p,
4888 mem_cgroup_clear_mc();
4891 static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4892 unsigned long addr, unsigned long end,
4893 struct mm_walk *walk)
4896 struct vm_area_struct *vma = walk->private;
4901 VM_BUG_ON(pmd_trans_huge(*pmd));
4902 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4903 for (; addr != end; addr += PAGE_SIZE) {
4904 pte_t ptent = *(pte++);
4905 union mc_target target;
4908 struct page_cgroup *pc;
4914 type = is_target_pte_for_mc(vma, addr, ptent, &target);
4916 case MC_TARGET_PAGE:
4918 if (isolate_lru_page(page))
4920 pc = lookup_page_cgroup(page);
4921 if (!mem_cgroup_move_account(pc,
4922 mc.from, mc.to, false, PAGE_SIZE)) {
4924 /* we uncharge from mc.from later. */
4927 putback_lru_page(page);
4928 put: /* is_target_pte_for_mc() gets the page */
4931 case MC_TARGET_SWAP:
4933 if (!mem_cgroup_move_swap_account(ent,
4934 mc.from, mc.to, false)) {
4936 /* we fixup refcnts and charges later. */
4944 pte_unmap_unlock(pte - 1, ptl);
4949 * We have consumed all precharges we got in can_attach().
4950 * We try charge one by one, but don't do any additional
4951 * charges to mc.to if we have failed in charge once in attach()
4954 ret = mem_cgroup_do_precharge(1);
4962 static void mem_cgroup_move_charge(struct mm_struct *mm)
4964 struct vm_area_struct *vma;
4966 lru_add_drain_all();
4968 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
4970 * Someone who are holding the mmap_sem might be waiting in
4971 * waitq. So we cancel all extra charges, wake up all waiters,
4972 * and retry. Because we cancel precharges, we might not be able
4973 * to move enough charges, but moving charge is a best-effort
4974 * feature anyway, so it wouldn't be a big problem.
4976 __mem_cgroup_clear_mc();
4980 for (vma = mm->mmap; vma; vma = vma->vm_next) {
4982 struct mm_walk mem_cgroup_move_charge_walk = {
4983 .pmd_entry = mem_cgroup_move_charge_pte_range,
4987 if (is_vm_hugetlb_page(vma))
4989 ret = walk_page_range(vma->vm_start, vma->vm_end,
4990 &mem_cgroup_move_charge_walk);
4993 * means we have consumed all precharges and failed in
4994 * doing additional charge. Just abandon here.
4998 up_read(&mm->mmap_sem);
5001 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
5002 struct cgroup *cont,
5003 struct cgroup *old_cont,
5004 struct task_struct *p,
5007 struct mm_struct *mm;
5010 /* no need to move charge */
5013 mm = get_task_mm(p);
5015 mem_cgroup_move_charge(mm);
5018 mem_cgroup_clear_mc();
5020 #else /* !CONFIG_MMU */
5021 static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
5022 struct cgroup *cgroup,
5023 struct task_struct *p,
5028 static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
5029 struct cgroup *cgroup,
5030 struct task_struct *p,
5034 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
5035 struct cgroup *cont,
5036 struct cgroup *old_cont,
5037 struct task_struct *p,
5043 struct cgroup_subsys mem_cgroup_subsys = {
5045 .subsys_id = mem_cgroup_subsys_id,
5046 .create = mem_cgroup_create,
5047 .pre_destroy = mem_cgroup_pre_destroy,
5048 .destroy = mem_cgroup_destroy,
5049 .populate = mem_cgroup_populate,
5050 .can_attach = mem_cgroup_can_attach,
5051 .cancel_attach = mem_cgroup_cancel_attach,
5052 .attach = mem_cgroup_move_task,
5057 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5058 static int __init enable_swap_account(char *s)
5060 /* consider enabled if no parameter or 1 is given */
5061 if (!(*s) || !strcmp(s, "=1"))
5062 really_do_swap_account = 1;
5063 else if (!strcmp(s, "=0"))
5064 really_do_swap_account = 0;
5067 __setup("swapaccount", enable_swap_account);
5069 static int __init disable_swap_account(char *s)
5071 printk_once("noswapaccount is deprecated and will be removed in 2.6.40. Use swapaccount=0 instead\n");
5072 enable_swap_account("=0");
5075 __setup("noswapaccount", disable_swap_account);