1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
20 #include <linux/res_counter.h>
21 #include <linux/memcontrol.h>
22 #include <linux/cgroup.h>
24 #include <linux/pagemap.h>
25 #include <linux/smp.h>
26 #include <linux/page-flags.h>
27 #include <linux/backing-dev.h>
28 #include <linux/bit_spinlock.h>
29 #include <linux/rcupdate.h>
30 #include <linux/limits.h>
31 #include <linux/mutex.h>
32 #include <linux/rbtree.h>
33 #include <linux/slab.h>
34 #include <linux/swap.h>
35 #include <linux/spinlock.h>
37 #include <linux/seq_file.h>
38 #include <linux/vmalloc.h>
39 #include <linux/mm_inline.h>
40 #include <linux/page_cgroup.h>
41 #include <linux/cpu.h>
44 #include <asm/uaccess.h>
46 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
47 #define MEM_CGROUP_RECLAIM_RETRIES 5
48 struct mem_cgroup *root_mem_cgroup __read_mostly;
50 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
51 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
52 int do_swap_account __read_mostly;
53 static int really_do_swap_account __initdata = 1; /* for remember boot option*/
55 #define do_swap_account (0)
58 static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */
59 #define SOFTLIMIT_EVENTS_THRESH (1000)
62 * Statistics for memory cgroup.
64 enum mem_cgroup_stat_index {
66 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
68 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
69 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
70 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
71 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
72 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
73 MEM_CGROUP_STAT_EVENTS, /* sum of pagein + pageout for internal use */
74 MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
76 MEM_CGROUP_STAT_NSTATS,
79 struct mem_cgroup_stat_cpu {
80 s64 count[MEM_CGROUP_STAT_NSTATS];
81 } ____cacheline_aligned_in_smp;
83 struct mem_cgroup_stat {
84 struct mem_cgroup_stat_cpu cpustat[0];
88 __mem_cgroup_stat_reset_safe(struct mem_cgroup_stat_cpu *stat,
89 enum mem_cgroup_stat_index idx)
95 __mem_cgroup_stat_read_local(struct mem_cgroup_stat_cpu *stat,
96 enum mem_cgroup_stat_index idx)
98 return stat->count[idx];
102 * For accounting under irq disable, no need for increment preempt count.
104 static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
105 enum mem_cgroup_stat_index idx, int val)
107 stat->count[idx] += val;
110 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
111 enum mem_cgroup_stat_index idx)
115 for_each_possible_cpu(cpu)
116 ret += stat->cpustat[cpu].count[idx];
120 static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat)
124 ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE);
125 ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS);
130 * per-zone information in memory controller.
132 struct mem_cgroup_per_zone {
134 * spin_lock to protect the per cgroup LRU
136 struct list_head lists[NR_LRU_LISTS];
137 unsigned long count[NR_LRU_LISTS];
139 struct zone_reclaim_stat reclaim_stat;
140 struct rb_node tree_node; /* RB tree node */
141 unsigned long long usage_in_excess;/* Set to the value by which */
142 /* the soft limit is exceeded*/
144 struct mem_cgroup *mem; /* Back pointer, we cannot */
145 /* use container_of */
147 /* Macro for accessing counter */
148 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
150 struct mem_cgroup_per_node {
151 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
154 struct mem_cgroup_lru_info {
155 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
159 * Cgroups above their limits are maintained in a RB-Tree, independent of
160 * their hierarchy representation
163 struct mem_cgroup_tree_per_zone {
164 struct rb_root rb_root;
168 struct mem_cgroup_tree_per_node {
169 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
172 struct mem_cgroup_tree {
173 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
176 static struct mem_cgroup_tree soft_limit_tree __read_mostly;
179 * The memory controller data structure. The memory controller controls both
180 * page cache and RSS per cgroup. We would eventually like to provide
181 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
182 * to help the administrator determine what knobs to tune.
184 * TODO: Add a water mark for the memory controller. Reclaim will begin when
185 * we hit the water mark. May be even add a low water mark, such that
186 * no reclaim occurs from a cgroup at it's low water mark, this is
187 * a feature that will be implemented much later in the future.
190 struct cgroup_subsys_state css;
192 * the counter to account for memory usage
194 struct res_counter res;
196 * the counter to account for mem+swap usage.
198 struct res_counter memsw;
200 * Per cgroup active and inactive list, similar to the
201 * per zone LRU lists.
203 struct mem_cgroup_lru_info info;
206 protect against reclaim related member.
208 spinlock_t reclaim_param_lock;
210 int prev_priority; /* for recording reclaim priority */
213 * While reclaiming in a hierarchy, we cache the last child we
216 int last_scanned_child;
218 * Should the accounting and control be hierarchical, per subtree?
221 unsigned long last_oom_jiffies;
224 unsigned int swappiness;
226 /* set when res.limit == memsw.limit */
227 bool memsw_is_minimum;
230 * statistics. This must be placed at the end of memcg.
232 struct mem_cgroup_stat stat;
236 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
237 * limit reclaim to prevent infinite loops, if they ever occur.
239 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
240 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
243 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
244 MEM_CGROUP_CHARGE_TYPE_MAPPED,
245 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
246 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
247 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
248 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
252 /* only for here (for easy reading.) */
253 #define PCGF_CACHE (1UL << PCG_CACHE)
254 #define PCGF_USED (1UL << PCG_USED)
255 #define PCGF_LOCK (1UL << PCG_LOCK)
256 /* Not used, but added here for completeness */
257 #define PCGF_ACCT (1UL << PCG_ACCT)
259 /* for encoding cft->private value on file */
262 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
263 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
264 #define MEMFILE_ATTR(val) ((val) & 0xffff)
267 * Reclaim flags for mem_cgroup_hierarchical_reclaim
269 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
270 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
271 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
272 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
273 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
274 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
276 static void mem_cgroup_get(struct mem_cgroup *mem);
277 static void mem_cgroup_put(struct mem_cgroup *mem);
278 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
279 static void drain_all_stock_async(void);
281 static struct mem_cgroup_per_zone *
282 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
284 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
287 static struct mem_cgroup_per_zone *
288 page_cgroup_zoneinfo(struct page_cgroup *pc)
290 struct mem_cgroup *mem = pc->mem_cgroup;
291 int nid = page_cgroup_nid(pc);
292 int zid = page_cgroup_zid(pc);
297 return mem_cgroup_zoneinfo(mem, nid, zid);
300 static struct mem_cgroup_tree_per_zone *
301 soft_limit_tree_node_zone(int nid, int zid)
303 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
306 static struct mem_cgroup_tree_per_zone *
307 soft_limit_tree_from_page(struct page *page)
309 int nid = page_to_nid(page);
310 int zid = page_zonenum(page);
312 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
316 __mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
317 struct mem_cgroup_per_zone *mz,
318 struct mem_cgroup_tree_per_zone *mctz,
319 unsigned long long new_usage_in_excess)
321 struct rb_node **p = &mctz->rb_root.rb_node;
322 struct rb_node *parent = NULL;
323 struct mem_cgroup_per_zone *mz_node;
328 mz->usage_in_excess = new_usage_in_excess;
329 if (!mz->usage_in_excess)
333 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
335 if (mz->usage_in_excess < mz_node->usage_in_excess)
338 * We can't avoid mem cgroups that are over their soft
339 * limit by the same amount
341 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
344 rb_link_node(&mz->tree_node, parent, p);
345 rb_insert_color(&mz->tree_node, &mctz->rb_root);
350 __mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
351 struct mem_cgroup_per_zone *mz,
352 struct mem_cgroup_tree_per_zone *mctz)
356 rb_erase(&mz->tree_node, &mctz->rb_root);
361 mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
362 struct mem_cgroup_per_zone *mz,
363 struct mem_cgroup_tree_per_zone *mctz)
365 spin_lock(&mctz->lock);
366 __mem_cgroup_remove_exceeded(mem, mz, mctz);
367 spin_unlock(&mctz->lock);
370 static bool mem_cgroup_soft_limit_check(struct mem_cgroup *mem)
375 struct mem_cgroup_stat_cpu *cpustat;
378 cpustat = &mem->stat.cpustat[cpu];
379 val = __mem_cgroup_stat_read_local(cpustat, MEM_CGROUP_STAT_EVENTS);
380 if (unlikely(val > SOFTLIMIT_EVENTS_THRESH)) {
381 __mem_cgroup_stat_reset_safe(cpustat, MEM_CGROUP_STAT_EVENTS);
388 static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page)
390 unsigned long long excess;
391 struct mem_cgroup_per_zone *mz;
392 struct mem_cgroup_tree_per_zone *mctz;
393 int nid = page_to_nid(page);
394 int zid = page_zonenum(page);
395 mctz = soft_limit_tree_from_page(page);
398 * Necessary to update all ancestors when hierarchy is used.
399 * because their event counter is not touched.
401 for (; mem; mem = parent_mem_cgroup(mem)) {
402 mz = mem_cgroup_zoneinfo(mem, nid, zid);
403 excess = res_counter_soft_limit_excess(&mem->res);
405 * We have to update the tree if mz is on RB-tree or
406 * mem is over its softlimit.
408 if (excess || mz->on_tree) {
409 spin_lock(&mctz->lock);
410 /* if on-tree, remove it */
412 __mem_cgroup_remove_exceeded(mem, mz, mctz);
414 * Insert again. mz->usage_in_excess will be updated.
415 * If excess is 0, no tree ops.
417 __mem_cgroup_insert_exceeded(mem, mz, mctz, excess);
418 spin_unlock(&mctz->lock);
423 static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem)
426 struct mem_cgroup_per_zone *mz;
427 struct mem_cgroup_tree_per_zone *mctz;
429 for_each_node_state(node, N_POSSIBLE) {
430 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
431 mz = mem_cgroup_zoneinfo(mem, node, zone);
432 mctz = soft_limit_tree_node_zone(node, zone);
433 mem_cgroup_remove_exceeded(mem, mz, mctz);
438 static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem)
440 return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT;
443 static struct mem_cgroup_per_zone *
444 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
446 struct rb_node *rightmost = NULL;
447 struct mem_cgroup_per_zone *mz;
451 rightmost = rb_last(&mctz->rb_root);
453 goto done; /* Nothing to reclaim from */
455 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
457 * Remove the node now but someone else can add it back,
458 * we will to add it back at the end of reclaim to its correct
459 * position in the tree.
461 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
462 if (!res_counter_soft_limit_excess(&mz->mem->res) ||
463 !css_tryget(&mz->mem->css))
469 static struct mem_cgroup_per_zone *
470 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
472 struct mem_cgroup_per_zone *mz;
474 spin_lock(&mctz->lock);
475 mz = __mem_cgroup_largest_soft_limit_node(mctz);
476 spin_unlock(&mctz->lock);
480 static void mem_cgroup_swap_statistics(struct mem_cgroup *mem,
483 int val = (charge) ? 1 : -1;
484 struct mem_cgroup_stat *stat = &mem->stat;
485 struct mem_cgroup_stat_cpu *cpustat;
488 cpustat = &stat->cpustat[cpu];
489 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_SWAPOUT, val);
493 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
494 struct page_cgroup *pc,
497 int val = (charge) ? 1 : -1;
498 struct mem_cgroup_stat *stat = &mem->stat;
499 struct mem_cgroup_stat_cpu *cpustat;
502 cpustat = &stat->cpustat[cpu];
503 if (PageCgroupCache(pc))
504 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
506 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
509 __mem_cgroup_stat_add_safe(cpustat,
510 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
512 __mem_cgroup_stat_add_safe(cpustat,
513 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
514 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_EVENTS, 1);
518 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
522 struct mem_cgroup_per_zone *mz;
525 for_each_online_node(nid)
526 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
527 mz = mem_cgroup_zoneinfo(mem, nid, zid);
528 total += MEM_CGROUP_ZSTAT(mz, idx);
533 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
535 return container_of(cgroup_subsys_state(cont,
536 mem_cgroup_subsys_id), struct mem_cgroup,
540 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
543 * mm_update_next_owner() may clear mm->owner to NULL
544 * if it races with swapoff, page migration, etc.
545 * So this can be called with p == NULL.
550 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
551 struct mem_cgroup, css);
554 static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
556 struct mem_cgroup *mem = NULL;
561 * Because we have no locks, mm->owner's may be being moved to other
562 * cgroup. We use css_tryget() here even if this looks
563 * pessimistic (rather than adding locks here).
567 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
570 } while (!css_tryget(&mem->css));
576 * Call callback function against all cgroup under hierarchy tree.
578 static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
579 int (*func)(struct mem_cgroup *, void *))
581 int found, ret, nextid;
582 struct cgroup_subsys_state *css;
583 struct mem_cgroup *mem;
585 if (!root->use_hierarchy)
586 return (*func)(root, data);
594 css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
596 if (css && css_tryget(css))
597 mem = container_of(css, struct mem_cgroup, css);
601 ret = (*func)(mem, data);
605 } while (!ret && css);
610 static inline bool mem_cgroup_is_root(struct mem_cgroup *mem)
612 return (mem == root_mem_cgroup);
616 * Following LRU functions are allowed to be used without PCG_LOCK.
617 * Operations are called by routine of global LRU independently from memcg.
618 * What we have to take care of here is validness of pc->mem_cgroup.
620 * Changes to pc->mem_cgroup happens when
623 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
624 * It is added to LRU before charge.
625 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
626 * When moving account, the page is not on LRU. It's isolated.
629 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
631 struct page_cgroup *pc;
632 struct mem_cgroup_per_zone *mz;
634 if (mem_cgroup_disabled())
636 pc = lookup_page_cgroup(page);
637 /* can happen while we handle swapcache. */
638 if (!TestClearPageCgroupAcctLRU(pc))
640 VM_BUG_ON(!pc->mem_cgroup);
642 * We don't check PCG_USED bit. It's cleared when the "page" is finally
643 * removed from global LRU.
645 mz = page_cgroup_zoneinfo(pc);
646 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
647 if (mem_cgroup_is_root(pc->mem_cgroup))
649 VM_BUG_ON(list_empty(&pc->lru));
650 list_del_init(&pc->lru);
654 void mem_cgroup_del_lru(struct page *page)
656 mem_cgroup_del_lru_list(page, page_lru(page));
659 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
661 struct mem_cgroup_per_zone *mz;
662 struct page_cgroup *pc;
664 if (mem_cgroup_disabled())
667 pc = lookup_page_cgroup(page);
669 * Used bit is set without atomic ops but after smp_wmb().
670 * For making pc->mem_cgroup visible, insert smp_rmb() here.
673 /* unused or root page is not rotated. */
674 if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup))
676 mz = page_cgroup_zoneinfo(pc);
677 list_move(&pc->lru, &mz->lists[lru]);
680 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
682 struct page_cgroup *pc;
683 struct mem_cgroup_per_zone *mz;
685 if (mem_cgroup_disabled())
687 pc = lookup_page_cgroup(page);
688 VM_BUG_ON(PageCgroupAcctLRU(pc));
690 * Used bit is set without atomic ops but after smp_wmb().
691 * For making pc->mem_cgroup visible, insert smp_rmb() here.
694 if (!PageCgroupUsed(pc))
697 mz = page_cgroup_zoneinfo(pc);
698 MEM_CGROUP_ZSTAT(mz, lru) += 1;
699 SetPageCgroupAcctLRU(pc);
700 if (mem_cgroup_is_root(pc->mem_cgroup))
702 list_add(&pc->lru, &mz->lists[lru]);
706 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
707 * lru because the page may.be reused after it's fully uncharged (because of
708 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
709 * it again. This function is only used to charge SwapCache. It's done under
710 * lock_page and expected that zone->lru_lock is never held.
712 static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
715 struct zone *zone = page_zone(page);
716 struct page_cgroup *pc = lookup_page_cgroup(page);
718 spin_lock_irqsave(&zone->lru_lock, flags);
720 * Forget old LRU when this page_cgroup is *not* used. This Used bit
721 * is guarded by lock_page() because the page is SwapCache.
723 if (!PageCgroupUsed(pc))
724 mem_cgroup_del_lru_list(page, page_lru(page));
725 spin_unlock_irqrestore(&zone->lru_lock, flags);
728 static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
731 struct zone *zone = page_zone(page);
732 struct page_cgroup *pc = lookup_page_cgroup(page);
734 spin_lock_irqsave(&zone->lru_lock, flags);
735 /* link when the page is linked to LRU but page_cgroup isn't */
736 if (PageLRU(page) && !PageCgroupAcctLRU(pc))
737 mem_cgroup_add_lru_list(page, page_lru(page));
738 spin_unlock_irqrestore(&zone->lru_lock, flags);
742 void mem_cgroup_move_lists(struct page *page,
743 enum lru_list from, enum lru_list to)
745 if (mem_cgroup_disabled())
747 mem_cgroup_del_lru_list(page, from);
748 mem_cgroup_add_lru_list(page, to);
751 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
754 struct mem_cgroup *curr = NULL;
758 curr = try_get_mem_cgroup_from_mm(task->mm);
763 if (curr->use_hierarchy)
764 ret = css_is_ancestor(&curr->css, &mem->css);
772 * prev_priority control...this will be used in memory reclaim path.
774 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
778 spin_lock(&mem->reclaim_param_lock);
779 prev_priority = mem->prev_priority;
780 spin_unlock(&mem->reclaim_param_lock);
782 return prev_priority;
785 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
787 spin_lock(&mem->reclaim_param_lock);
788 if (priority < mem->prev_priority)
789 mem->prev_priority = priority;
790 spin_unlock(&mem->reclaim_param_lock);
793 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
795 spin_lock(&mem->reclaim_param_lock);
796 mem->prev_priority = priority;
797 spin_unlock(&mem->reclaim_param_lock);
800 static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
802 unsigned long active;
803 unsigned long inactive;
805 unsigned long inactive_ratio;
807 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
808 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
810 gb = (inactive + active) >> (30 - PAGE_SHIFT);
812 inactive_ratio = int_sqrt(10 * gb);
817 present_pages[0] = inactive;
818 present_pages[1] = active;
821 return inactive_ratio;
824 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
826 unsigned long active;
827 unsigned long inactive;
828 unsigned long present_pages[2];
829 unsigned long inactive_ratio;
831 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
833 inactive = present_pages[0];
834 active = present_pages[1];
836 if (inactive * inactive_ratio < active)
842 int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
844 unsigned long active;
845 unsigned long inactive;
847 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
848 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
850 return (active > inactive);
853 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
857 int nid = zone->zone_pgdat->node_id;
858 int zid = zone_idx(zone);
859 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
861 return MEM_CGROUP_ZSTAT(mz, lru);
864 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
867 int nid = zone->zone_pgdat->node_id;
868 int zid = zone_idx(zone);
869 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
871 return &mz->reclaim_stat;
874 struct zone_reclaim_stat *
875 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
877 struct page_cgroup *pc;
878 struct mem_cgroup_per_zone *mz;
880 if (mem_cgroup_disabled())
883 pc = lookup_page_cgroup(page);
885 * Used bit is set without atomic ops but after smp_wmb().
886 * For making pc->mem_cgroup visible, insert smp_rmb() here.
889 if (!PageCgroupUsed(pc))
892 mz = page_cgroup_zoneinfo(pc);
896 return &mz->reclaim_stat;
899 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
900 struct list_head *dst,
901 unsigned long *scanned, int order,
902 int mode, struct zone *z,
903 struct mem_cgroup *mem_cont,
904 int active, int file)
906 unsigned long nr_taken = 0;
910 struct list_head *src;
911 struct page_cgroup *pc, *tmp;
912 int nid = z->zone_pgdat->node_id;
913 int zid = zone_idx(z);
914 struct mem_cgroup_per_zone *mz;
915 int lru = LRU_FILE * file + active;
919 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
920 src = &mz->lists[lru];
923 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
924 if (scan >= nr_to_scan)
928 if (unlikely(!PageCgroupUsed(pc)))
930 if (unlikely(!PageLRU(page)))
934 ret = __isolate_lru_page(page, mode, file);
937 list_move(&page->lru, dst);
938 mem_cgroup_del_lru(page);
942 /* we don't affect global LRU but rotate in our LRU */
943 mem_cgroup_rotate_lru_list(page, page_lru(page));
954 #define mem_cgroup_from_res_counter(counter, member) \
955 container_of(counter, struct mem_cgroup, member)
957 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
959 if (do_swap_account) {
960 if (res_counter_check_under_limit(&mem->res) &&
961 res_counter_check_under_limit(&mem->memsw))
964 if (res_counter_check_under_limit(&mem->res))
969 static unsigned int get_swappiness(struct mem_cgroup *memcg)
971 struct cgroup *cgrp = memcg->css.cgroup;
972 unsigned int swappiness;
975 if (cgrp->parent == NULL)
976 return vm_swappiness;
978 spin_lock(&memcg->reclaim_param_lock);
979 swappiness = memcg->swappiness;
980 spin_unlock(&memcg->reclaim_param_lock);
985 static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
993 * mem_cgroup_print_mem_info: Called from OOM with tasklist_lock held in read mode.
994 * @memcg: The memory cgroup that went over limit
995 * @p: Task that is going to be killed
997 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1000 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1002 struct cgroup *task_cgrp;
1003 struct cgroup *mem_cgrp;
1005 * Need a buffer in BSS, can't rely on allocations. The code relies
1006 * on the assumption that OOM is serialized for memory controller.
1007 * If this assumption is broken, revisit this code.
1009 static char memcg_name[PATH_MAX];
1018 mem_cgrp = memcg->css.cgroup;
1019 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1021 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1024 * Unfortunately, we are unable to convert to a useful name
1025 * But we'll still print out the usage information
1032 printk(KERN_INFO "Task in %s killed", memcg_name);
1035 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1043 * Continues from above, so we don't need an KERN_ level
1045 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1048 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1049 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1050 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1051 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1052 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1054 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1055 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1056 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1060 * This function returns the number of memcg under hierarchy tree. Returns
1061 * 1(self count) if no children.
1063 static int mem_cgroup_count_children(struct mem_cgroup *mem)
1066 mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
1071 * Visit the first child (need not be the first child as per the ordering
1072 * of the cgroup list, since we track last_scanned_child) of @mem and use
1073 * that to reclaim free pages from.
1075 static struct mem_cgroup *
1076 mem_cgroup_select_victim(struct mem_cgroup *root_mem)
1078 struct mem_cgroup *ret = NULL;
1079 struct cgroup_subsys_state *css;
1082 if (!root_mem->use_hierarchy) {
1083 css_get(&root_mem->css);
1089 nextid = root_mem->last_scanned_child + 1;
1090 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
1092 if (css && css_tryget(css))
1093 ret = container_of(css, struct mem_cgroup, css);
1096 /* Updates scanning parameter */
1097 spin_lock(&root_mem->reclaim_param_lock);
1099 /* this means start scan from ID:1 */
1100 root_mem->last_scanned_child = 0;
1102 root_mem->last_scanned_child = found;
1103 spin_unlock(&root_mem->reclaim_param_lock);
1110 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1111 * we reclaimed from, so that we don't end up penalizing one child extensively
1112 * based on its position in the children list.
1114 * root_mem is the original ancestor that we've been reclaim from.
1116 * We give up and return to the caller when we visit root_mem twice.
1117 * (other groups can be removed while we're walking....)
1119 * If shrink==true, for avoiding to free too much, this returns immedieately.
1121 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
1124 unsigned long reclaim_options)
1126 struct mem_cgroup *victim;
1129 bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
1130 bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
1131 bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT;
1132 unsigned long excess = mem_cgroup_get_excess(root_mem);
1134 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1135 if (root_mem->memsw_is_minimum)
1139 victim = mem_cgroup_select_victim(root_mem);
1140 if (victim == root_mem) {
1143 drain_all_stock_async();
1146 * If we have not been able to reclaim
1147 * anything, it might because there are
1148 * no reclaimable pages under this hierarchy
1150 if (!check_soft || !total) {
1151 css_put(&victim->css);
1155 * We want to do more targetted reclaim.
1156 * excess >> 2 is not to excessive so as to
1157 * reclaim too much, nor too less that we keep
1158 * coming back to reclaim from this cgroup
1160 if (total >= (excess >> 2) ||
1161 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) {
1162 css_put(&victim->css);
1167 if (!mem_cgroup_local_usage(&victim->stat)) {
1168 /* this cgroup's local usage == 0 */
1169 css_put(&victim->css);
1172 /* we use swappiness of local cgroup */
1174 ret = mem_cgroup_shrink_node_zone(victim, gfp_mask,
1175 noswap, get_swappiness(victim), zone,
1176 zone->zone_pgdat->node_id);
1178 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask,
1179 noswap, get_swappiness(victim));
1180 css_put(&victim->css);
1182 * At shrinking usage, we can't check we should stop here or
1183 * reclaim more. It's depends on callers. last_scanned_child
1184 * will work enough for keeping fairness under tree.
1190 if (res_counter_check_under_soft_limit(&root_mem->res))
1192 } else if (mem_cgroup_check_under_limit(root_mem))
1198 bool mem_cgroup_oom_called(struct task_struct *task)
1201 struct mem_cgroup *mem;
1202 struct mm_struct *mm;
1208 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
1209 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
1215 static int record_last_oom_cb(struct mem_cgroup *mem, void *data)
1217 mem->last_oom_jiffies = jiffies;
1221 static void record_last_oom(struct mem_cgroup *mem)
1223 mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb);
1227 * Currently used to update mapped file statistics, but the routine can be
1228 * generalized to update other statistics as well.
1230 void mem_cgroup_update_file_mapped(struct page *page, int val)
1232 struct mem_cgroup *mem;
1233 struct mem_cgroup_stat *stat;
1234 struct mem_cgroup_stat_cpu *cpustat;
1236 struct page_cgroup *pc;
1238 pc = lookup_page_cgroup(page);
1242 lock_page_cgroup(pc);
1243 mem = pc->mem_cgroup;
1247 if (!PageCgroupUsed(pc))
1251 * Preemption is already disabled, we don't need get_cpu()
1253 cpu = smp_processor_id();
1255 cpustat = &stat->cpustat[cpu];
1257 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, val);
1259 unlock_page_cgroup(pc);
1263 * size of first charge trial. "32" comes from vmscan.c's magic value.
1264 * TODO: maybe necessary to use big numbers in big irons.
1266 #define CHARGE_SIZE (32 * PAGE_SIZE)
1267 struct memcg_stock_pcp {
1268 struct mem_cgroup *cached; /* this never be root cgroup */
1270 struct work_struct work;
1272 static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
1273 static atomic_t memcg_drain_count;
1276 * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
1277 * from local stock and true is returned. If the stock is 0 or charges from a
1278 * cgroup which is not current target, returns false. This stock will be
1281 static bool consume_stock(struct mem_cgroup *mem)
1283 struct memcg_stock_pcp *stock;
1286 stock = &get_cpu_var(memcg_stock);
1287 if (mem == stock->cached && stock->charge)
1288 stock->charge -= PAGE_SIZE;
1289 else /* need to call res_counter_charge */
1291 put_cpu_var(memcg_stock);
1296 * Returns stocks cached in percpu to res_counter and reset cached information.
1298 static void drain_stock(struct memcg_stock_pcp *stock)
1300 struct mem_cgroup *old = stock->cached;
1302 if (stock->charge) {
1303 res_counter_uncharge(&old->res, stock->charge);
1304 if (do_swap_account)
1305 res_counter_uncharge(&old->memsw, stock->charge);
1307 stock->cached = NULL;
1312 * This must be called under preempt disabled or must be called by
1313 * a thread which is pinned to local cpu.
1315 static void drain_local_stock(struct work_struct *dummy)
1317 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
1322 * Cache charges(val) which is from res_counter, to local per_cpu area.
1323 * This will be consumed by consumt_stock() function, later.
1325 static void refill_stock(struct mem_cgroup *mem, int val)
1327 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1329 if (stock->cached != mem) { /* reset if necessary */
1331 stock->cached = mem;
1333 stock->charge += val;
1334 put_cpu_var(memcg_stock);
1338 * Tries to drain stocked charges in other cpus. This function is asynchronous
1339 * and just put a work per cpu for draining localy on each cpu. Caller can
1340 * expects some charges will be back to res_counter later but cannot wait for
1343 static void drain_all_stock_async(void)
1346 /* This function is for scheduling "drain" in asynchronous way.
1347 * The result of "drain" is not directly handled by callers. Then,
1348 * if someone is calling drain, we don't have to call drain more.
1349 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1350 * there is a race. We just do loose check here.
1352 if (atomic_read(&memcg_drain_count))
1354 /* Notify other cpus that system-wide "drain" is running */
1355 atomic_inc(&memcg_drain_count);
1357 for_each_online_cpu(cpu) {
1358 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
1359 schedule_work_on(cpu, &stock->work);
1362 atomic_dec(&memcg_drain_count);
1363 /* We don't wait for flush_work */
1366 /* This is a synchronous drain interface. */
1367 static void drain_all_stock_sync(void)
1369 /* called when force_empty is called */
1370 atomic_inc(&memcg_drain_count);
1371 schedule_on_each_cpu(drain_local_stock);
1372 atomic_dec(&memcg_drain_count);
1375 static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb,
1376 unsigned long action,
1379 int cpu = (unsigned long)hcpu;
1380 struct memcg_stock_pcp *stock;
1382 if (action != CPU_DEAD)
1384 stock = &per_cpu(memcg_stock, cpu);
1390 * Unlike exported interface, "oom" parameter is added. if oom==true,
1391 * oom-killer can be invoked.
1393 static int __mem_cgroup_try_charge(struct mm_struct *mm,
1394 gfp_t gfp_mask, struct mem_cgroup **memcg,
1395 bool oom, struct page *page)
1397 struct mem_cgroup *mem, *mem_over_limit;
1398 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1399 struct res_counter *fail_res;
1400 int csize = CHARGE_SIZE;
1402 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
1403 /* Don't account this! */
1409 * We always charge the cgroup the mm_struct belongs to.
1410 * The mm_struct's mem_cgroup changes on task migration if the
1411 * thread group leader migrates. It's possible that mm is not
1412 * set, if so charge the init_mm (happens for pagecache usage).
1416 mem = try_get_mem_cgroup_from_mm(mm);
1424 VM_BUG_ON(css_is_removed(&mem->css));
1425 if (mem_cgroup_is_root(mem))
1430 unsigned long flags = 0;
1432 if (consume_stock(mem))
1435 ret = res_counter_charge(&mem->res, csize, &fail_res);
1437 if (!do_swap_account)
1439 ret = res_counter_charge(&mem->memsw, csize, &fail_res);
1442 /* mem+swap counter fails */
1443 res_counter_uncharge(&mem->res, csize);
1444 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
1445 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1448 /* mem counter fails */
1449 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1452 /* reduce request size and retry */
1453 if (csize > PAGE_SIZE) {
1457 if (!(gfp_mask & __GFP_WAIT))
1460 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
1466 * try_to_free_mem_cgroup_pages() might not give us a full
1467 * picture of reclaim. Some pages are reclaimed and might be
1468 * moved to swap cache or just unmapped from the cgroup.
1469 * Check the limit again to see if the reclaim reduced the
1470 * current usage of the cgroup before giving up
1473 if (mem_cgroup_check_under_limit(mem_over_limit))
1476 if (!nr_retries--) {
1478 mutex_lock(&memcg_tasklist);
1479 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
1480 mutex_unlock(&memcg_tasklist);
1481 record_last_oom(mem_over_limit);
1486 if (csize > PAGE_SIZE)
1487 refill_stock(mem, csize - PAGE_SIZE);
1490 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
1491 * if they exceeds softlimit.
1493 if (mem_cgroup_soft_limit_check(mem))
1494 mem_cgroup_update_tree(mem, page);
1503 * A helper function to get mem_cgroup from ID. must be called under
1504 * rcu_read_lock(). The caller must check css_is_removed() or some if
1505 * it's concern. (dropping refcnt from swap can be called against removed
1508 static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
1510 struct cgroup_subsys_state *css;
1512 /* ID 0 is unused ID */
1515 css = css_lookup(&mem_cgroup_subsys, id);
1518 return container_of(css, struct mem_cgroup, css);
1521 static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
1523 struct mem_cgroup *mem;
1524 struct page_cgroup *pc;
1528 VM_BUG_ON(!PageLocked(page));
1530 if (!PageSwapCache(page))
1533 pc = lookup_page_cgroup(page);
1534 lock_page_cgroup(pc);
1535 if (PageCgroupUsed(pc)) {
1536 mem = pc->mem_cgroup;
1537 if (mem && !css_tryget(&mem->css))
1540 ent.val = page_private(page);
1541 id = lookup_swap_cgroup(ent);
1543 mem = mem_cgroup_lookup(id);
1544 if (mem && !css_tryget(&mem->css))
1548 unlock_page_cgroup(pc);
1553 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
1554 * USED state. If already USED, uncharge and return.
1557 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
1558 struct page_cgroup *pc,
1559 enum charge_type ctype)
1561 /* try_charge() can return NULL to *memcg, taking care of it. */
1565 lock_page_cgroup(pc);
1566 if (unlikely(PageCgroupUsed(pc))) {
1567 unlock_page_cgroup(pc);
1568 if (!mem_cgroup_is_root(mem)) {
1569 res_counter_uncharge(&mem->res, PAGE_SIZE);
1570 if (do_swap_account)
1571 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1577 pc->mem_cgroup = mem;
1579 * We access a page_cgroup asynchronously without lock_page_cgroup().
1580 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
1581 * is accessed after testing USED bit. To make pc->mem_cgroup visible
1582 * before USED bit, we need memory barrier here.
1583 * See mem_cgroup_add_lru_list(), etc.
1587 case MEM_CGROUP_CHARGE_TYPE_CACHE:
1588 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
1589 SetPageCgroupCache(pc);
1590 SetPageCgroupUsed(pc);
1592 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1593 ClearPageCgroupCache(pc);
1594 SetPageCgroupUsed(pc);
1600 mem_cgroup_charge_statistics(mem, pc, true);
1602 unlock_page_cgroup(pc);
1606 * mem_cgroup_move_account - move account of the page
1607 * @pc: page_cgroup of the page.
1608 * @from: mem_cgroup which the page is moved from.
1609 * @to: mem_cgroup which the page is moved to. @from != @to.
1611 * The caller must confirm following.
1612 * - page is not on LRU (isolate_page() is useful.)
1614 * returns 0 at success,
1615 * returns -EBUSY when lock is busy or "pc" is unstable.
1617 * This function does "uncharge" from old cgroup but doesn't do "charge" to
1618 * new cgroup. It should be done by a caller.
1621 static int mem_cgroup_move_account(struct page_cgroup *pc,
1622 struct mem_cgroup *from, struct mem_cgroup *to)
1624 struct mem_cgroup_per_zone *from_mz, *to_mz;
1629 struct mem_cgroup_stat *stat;
1630 struct mem_cgroup_stat_cpu *cpustat;
1632 VM_BUG_ON(from == to);
1633 VM_BUG_ON(PageLRU(pc->page));
1635 nid = page_cgroup_nid(pc);
1636 zid = page_cgroup_zid(pc);
1637 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
1638 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
1640 if (!trylock_page_cgroup(pc))
1643 if (!PageCgroupUsed(pc))
1646 if (pc->mem_cgroup != from)
1649 if (!mem_cgroup_is_root(from))
1650 res_counter_uncharge(&from->res, PAGE_SIZE);
1651 mem_cgroup_charge_statistics(from, pc, false);
1654 if (page_mapped(page) && !PageAnon(page)) {
1655 cpu = smp_processor_id();
1656 /* Update mapped_file data for mem_cgroup "from" */
1658 cpustat = &stat->cpustat[cpu];
1659 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED,
1662 /* Update mapped_file data for mem_cgroup "to" */
1664 cpustat = &stat->cpustat[cpu];
1665 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED,
1669 if (do_swap_account && !mem_cgroup_is_root(from))
1670 res_counter_uncharge(&from->memsw, PAGE_SIZE);
1671 css_put(&from->css);
1674 pc->mem_cgroup = to;
1675 mem_cgroup_charge_statistics(to, pc, true);
1678 unlock_page_cgroup(pc);
1680 * We charges against "to" which may not have any tasks. Then, "to"
1681 * can be under rmdir(). But in current implementation, caller of
1682 * this function is just force_empty() and it's garanteed that
1683 * "to" is never removed. So, we don't check rmdir status here.
1689 * move charges to its parent.
1692 static int mem_cgroup_move_parent(struct page_cgroup *pc,
1693 struct mem_cgroup *child,
1696 struct page *page = pc->page;
1697 struct cgroup *cg = child->css.cgroup;
1698 struct cgroup *pcg = cg->parent;
1699 struct mem_cgroup *parent;
1707 parent = mem_cgroup_from_cont(pcg);
1710 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false, page);
1714 if (!get_page_unless_zero(page)) {
1719 ret = isolate_lru_page(page);
1724 ret = mem_cgroup_move_account(pc, child, parent);
1726 putback_lru_page(page);
1729 /* drop extra refcnt by try_charge() */
1730 css_put(&parent->css);
1737 /* drop extra refcnt by try_charge() */
1738 css_put(&parent->css);
1739 /* uncharge if move fails */
1740 if (!mem_cgroup_is_root(parent)) {
1741 res_counter_uncharge(&parent->res, PAGE_SIZE);
1742 if (do_swap_account)
1743 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
1749 * Charge the memory controller for page usage.
1751 * 0 if the charge was successful
1752 * < 0 if the cgroup is over its limit
1754 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1755 gfp_t gfp_mask, enum charge_type ctype,
1756 struct mem_cgroup *memcg)
1758 struct mem_cgroup *mem;
1759 struct page_cgroup *pc;
1762 pc = lookup_page_cgroup(page);
1763 /* can happen at boot */
1769 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true, page);
1773 __mem_cgroup_commit_charge(mem, pc, ctype);
1777 int mem_cgroup_newpage_charge(struct page *page,
1778 struct mm_struct *mm, gfp_t gfp_mask)
1780 if (mem_cgroup_disabled())
1782 if (PageCompound(page))
1785 * If already mapped, we don't have to account.
1786 * If page cache, page->mapping has address_space.
1787 * But page->mapping may have out-of-use anon_vma pointer,
1788 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1791 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1795 return mem_cgroup_charge_common(page, mm, gfp_mask,
1796 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1800 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1801 enum charge_type ctype);
1803 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1806 struct mem_cgroup *mem = NULL;
1809 if (mem_cgroup_disabled())
1811 if (PageCompound(page))
1814 * Corner case handling. This is called from add_to_page_cache()
1815 * in usual. But some FS (shmem) precharges this page before calling it
1816 * and call add_to_page_cache() with GFP_NOWAIT.
1818 * For GFP_NOWAIT case, the page may be pre-charged before calling
1819 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1820 * charge twice. (It works but has to pay a bit larger cost.)
1821 * And when the page is SwapCache, it should take swap information
1822 * into account. This is under lock_page() now.
1824 if (!(gfp_mask & __GFP_WAIT)) {
1825 struct page_cgroup *pc;
1828 pc = lookup_page_cgroup(page);
1831 lock_page_cgroup(pc);
1832 if (PageCgroupUsed(pc)) {
1833 unlock_page_cgroup(pc);
1836 unlock_page_cgroup(pc);
1839 if (unlikely(!mm && !mem))
1842 if (page_is_file_cache(page))
1843 return mem_cgroup_charge_common(page, mm, gfp_mask,
1844 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1847 if (PageSwapCache(page)) {
1848 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
1850 __mem_cgroup_commit_charge_swapin(page, mem,
1851 MEM_CGROUP_CHARGE_TYPE_SHMEM);
1853 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1854 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1860 * While swap-in, try_charge -> commit or cancel, the page is locked.
1861 * And when try_charge() successfully returns, one refcnt to memcg without
1862 * struct page_cgroup is acquired. This refcnt will be consumed by
1863 * "commit()" or removed by "cancel()"
1865 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1867 gfp_t mask, struct mem_cgroup **ptr)
1869 struct mem_cgroup *mem;
1872 if (mem_cgroup_disabled())
1875 if (!do_swap_account)
1878 * A racing thread's fault, or swapoff, may have already updated
1879 * the pte, and even removed page from swap cache: in those cases
1880 * do_swap_page()'s pte_same() test will fail; but there's also a
1881 * KSM case which does need to charge the page.
1883 if (!PageSwapCache(page))
1885 mem = try_get_mem_cgroup_from_swapcache(page);
1889 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, page);
1890 /* drop extra refcnt from tryget */
1896 return __mem_cgroup_try_charge(mm, mask, ptr, true, page);
1900 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1901 enum charge_type ctype)
1903 struct page_cgroup *pc;
1905 if (mem_cgroup_disabled())
1909 cgroup_exclude_rmdir(&ptr->css);
1910 pc = lookup_page_cgroup(page);
1911 mem_cgroup_lru_del_before_commit_swapcache(page);
1912 __mem_cgroup_commit_charge(ptr, pc, ctype);
1913 mem_cgroup_lru_add_after_commit_swapcache(page);
1915 * Now swap is on-memory. This means this page may be
1916 * counted both as mem and swap....double count.
1917 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1918 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1919 * may call delete_from_swap_cache() before reach here.
1921 if (do_swap_account && PageSwapCache(page)) {
1922 swp_entry_t ent = {.val = page_private(page)};
1924 struct mem_cgroup *memcg;
1926 id = swap_cgroup_record(ent, 0);
1928 memcg = mem_cgroup_lookup(id);
1931 * This recorded memcg can be obsolete one. So, avoid
1932 * calling css_tryget
1934 if (!mem_cgroup_is_root(memcg))
1935 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1936 mem_cgroup_swap_statistics(memcg, false);
1937 mem_cgroup_put(memcg);
1942 * At swapin, we may charge account against cgroup which has no tasks.
1943 * So, rmdir()->pre_destroy() can be called while we do this charge.
1944 * In that case, we need to call pre_destroy() again. check it here.
1946 cgroup_release_and_wakeup_rmdir(&ptr->css);
1949 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1951 __mem_cgroup_commit_charge_swapin(page, ptr,
1952 MEM_CGROUP_CHARGE_TYPE_MAPPED);
1955 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1957 if (mem_cgroup_disabled())
1961 if (!mem_cgroup_is_root(mem)) {
1962 res_counter_uncharge(&mem->res, PAGE_SIZE);
1963 if (do_swap_account)
1964 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1970 __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype)
1972 struct memcg_batch_info *batch = NULL;
1973 bool uncharge_memsw = true;
1974 /* If swapout, usage of swap doesn't decrease */
1975 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1976 uncharge_memsw = false;
1978 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
1979 * In those cases, all pages freed continously can be expected to be in
1980 * the same cgroup and we have chance to coalesce uncharges.
1981 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
1982 * because we want to do uncharge as soon as possible.
1984 if (!current->memcg_batch.do_batch || test_thread_flag(TIF_MEMDIE))
1985 goto direct_uncharge;
1987 batch = ¤t->memcg_batch;
1989 * In usual, we do css_get() when we remember memcg pointer.
1990 * But in this case, we keep res->usage until end of a series of
1991 * uncharges. Then, it's ok to ignore memcg's refcnt.
1996 * In typical case, batch->memcg == mem. This means we can
1997 * merge a series of uncharges to an uncharge of res_counter.
1998 * If not, we uncharge res_counter ony by one.
2000 if (batch->memcg != mem)
2001 goto direct_uncharge;
2002 /* remember freed charge and uncharge it later */
2003 batch->bytes += PAGE_SIZE;
2005 batch->memsw_bytes += PAGE_SIZE;
2008 res_counter_uncharge(&mem->res, PAGE_SIZE);
2010 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
2015 * uncharge if !page_mapped(page)
2017 static struct mem_cgroup *
2018 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
2020 struct page_cgroup *pc;
2021 struct mem_cgroup *mem = NULL;
2022 struct mem_cgroup_per_zone *mz;
2024 if (mem_cgroup_disabled())
2027 if (PageSwapCache(page))
2031 * Check if our page_cgroup is valid
2033 pc = lookup_page_cgroup(page);
2034 if (unlikely(!pc || !PageCgroupUsed(pc)))
2037 lock_page_cgroup(pc);
2039 mem = pc->mem_cgroup;
2041 if (!PageCgroupUsed(pc))
2045 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2046 case MEM_CGROUP_CHARGE_TYPE_DROP:
2047 if (page_mapped(page))
2050 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
2051 if (!PageAnon(page)) { /* Shared memory */
2052 if (page->mapping && !page_is_file_cache(page))
2054 } else if (page_mapped(page)) /* Anon */
2061 if (!mem_cgroup_is_root(mem))
2062 __do_uncharge(mem, ctype);
2063 if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2064 mem_cgroup_swap_statistics(mem, true);
2065 mem_cgroup_charge_statistics(mem, pc, false);
2067 ClearPageCgroupUsed(pc);
2069 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2070 * freed from LRU. This is safe because uncharged page is expected not
2071 * to be reused (freed soon). Exception is SwapCache, it's handled by
2072 * special functions.
2075 mz = page_cgroup_zoneinfo(pc);
2076 unlock_page_cgroup(pc);
2078 if (mem_cgroup_soft_limit_check(mem))
2079 mem_cgroup_update_tree(mem, page);
2080 /* at swapout, this memcg will be accessed to record to swap */
2081 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2087 unlock_page_cgroup(pc);
2091 void mem_cgroup_uncharge_page(struct page *page)
2094 if (page_mapped(page))
2096 if (page->mapping && !PageAnon(page))
2098 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
2101 void mem_cgroup_uncharge_cache_page(struct page *page)
2103 VM_BUG_ON(page_mapped(page));
2104 VM_BUG_ON(page->mapping);
2105 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
2109 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2110 * In that cases, pages are freed continuously and we can expect pages
2111 * are in the same memcg. All these calls itself limits the number of
2112 * pages freed at once, then uncharge_start/end() is called properly.
2113 * This may be called prural(2) times in a context,
2116 void mem_cgroup_uncharge_start(void)
2118 current->memcg_batch.do_batch++;
2119 /* We can do nest. */
2120 if (current->memcg_batch.do_batch == 1) {
2121 current->memcg_batch.memcg = NULL;
2122 current->memcg_batch.bytes = 0;
2123 current->memcg_batch.memsw_bytes = 0;
2127 void mem_cgroup_uncharge_end(void)
2129 struct memcg_batch_info *batch = ¤t->memcg_batch;
2131 if (!batch->do_batch)
2135 if (batch->do_batch) /* If stacked, do nothing. */
2141 * This "batch->memcg" is valid without any css_get/put etc...
2142 * bacause we hide charges behind us.
2145 res_counter_uncharge(&batch->memcg->res, batch->bytes);
2146 if (batch->memsw_bytes)
2147 res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes);
2148 /* forget this pointer (for sanity check) */
2149 batch->memcg = NULL;
2154 * called after __delete_from_swap_cache() and drop "page" account.
2155 * memcg information is recorded to swap_cgroup of "ent"
2158 mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
2160 struct mem_cgroup *memcg;
2161 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
2163 if (!swapout) /* this was a swap cache but the swap is unused ! */
2164 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
2166 memcg = __mem_cgroup_uncharge_common(page, ctype);
2168 /* record memcg information */
2169 if (do_swap_account && swapout && memcg) {
2170 swap_cgroup_record(ent, css_id(&memcg->css));
2171 mem_cgroup_get(memcg);
2173 if (swapout && memcg)
2174 css_put(&memcg->css);
2178 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2180 * called from swap_entry_free(). remove record in swap_cgroup and
2181 * uncharge "memsw" account.
2183 void mem_cgroup_uncharge_swap(swp_entry_t ent)
2185 struct mem_cgroup *memcg;
2188 if (!do_swap_account)
2191 id = swap_cgroup_record(ent, 0);
2193 memcg = mem_cgroup_lookup(id);
2196 * We uncharge this because swap is freed.
2197 * This memcg can be obsolete one. We avoid calling css_tryget
2199 if (!mem_cgroup_is_root(memcg))
2200 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
2201 mem_cgroup_swap_statistics(memcg, false);
2202 mem_cgroup_put(memcg);
2209 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2212 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
2214 struct page_cgroup *pc;
2215 struct mem_cgroup *mem = NULL;
2218 if (mem_cgroup_disabled())
2221 pc = lookup_page_cgroup(page);
2222 lock_page_cgroup(pc);
2223 if (PageCgroupUsed(pc)) {
2224 mem = pc->mem_cgroup;
2227 unlock_page_cgroup(pc);
2230 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false,
2238 /* remove redundant charge if migration failed*/
2239 void mem_cgroup_end_migration(struct mem_cgroup *mem,
2240 struct page *oldpage, struct page *newpage)
2242 struct page *target, *unused;
2243 struct page_cgroup *pc;
2244 enum charge_type ctype;
2248 cgroup_exclude_rmdir(&mem->css);
2249 /* at migration success, oldpage->mapping is NULL. */
2250 if (oldpage->mapping) {
2258 if (PageAnon(target))
2259 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
2260 else if (page_is_file_cache(target))
2261 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
2263 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
2265 /* unused page is not on radix-tree now. */
2267 __mem_cgroup_uncharge_common(unused, ctype);
2269 pc = lookup_page_cgroup(target);
2271 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
2272 * So, double-counting is effectively avoided.
2274 __mem_cgroup_commit_charge(mem, pc, ctype);
2277 * Both of oldpage and newpage are still under lock_page().
2278 * Then, we don't have to care about race in radix-tree.
2279 * But we have to be careful that this page is unmapped or not.
2281 * There is a case for !page_mapped(). At the start of
2282 * migration, oldpage was mapped. But now, it's zapped.
2283 * But we know *target* page is not freed/reused under us.
2284 * mem_cgroup_uncharge_page() does all necessary checks.
2286 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
2287 mem_cgroup_uncharge_page(target);
2289 * At migration, we may charge account against cgroup which has no tasks
2290 * So, rmdir()->pre_destroy() can be called while we do this charge.
2291 * In that case, we need to call pre_destroy() again. check it here.
2293 cgroup_release_and_wakeup_rmdir(&mem->css);
2297 * A call to try to shrink memory usage on charge failure at shmem's swapin.
2298 * Calling hierarchical_reclaim is not enough because we should update
2299 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
2300 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
2301 * not from the memcg which this page would be charged to.
2302 * try_charge_swapin does all of these works properly.
2304 int mem_cgroup_shmem_charge_fallback(struct page *page,
2305 struct mm_struct *mm,
2308 struct mem_cgroup *mem = NULL;
2311 if (mem_cgroup_disabled())
2314 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2316 mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
2321 static DEFINE_MUTEX(set_limit_mutex);
2323 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
2324 unsigned long long val)
2330 int children = mem_cgroup_count_children(memcg);
2331 u64 curusage, oldusage;
2334 * For keeping hierarchical_reclaim simple, how long we should retry
2335 * is depends on callers. We set our retry-count to be function
2336 * of # of children which we should visit in this loop.
2338 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
2340 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2342 while (retry_count) {
2343 if (signal_pending(current)) {
2348 * Rather than hide all in some function, I do this in
2349 * open coded manner. You see what this really does.
2350 * We have to guarantee mem->res.limit < mem->memsw.limit.
2352 mutex_lock(&set_limit_mutex);
2353 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2354 if (memswlimit < val) {
2356 mutex_unlock(&set_limit_mutex);
2359 ret = res_counter_set_limit(&memcg->res, val);
2361 if (memswlimit == val)
2362 memcg->memsw_is_minimum = true;
2364 memcg->memsw_is_minimum = false;
2366 mutex_unlock(&set_limit_mutex);
2371 progress = mem_cgroup_hierarchical_reclaim(memcg, NULL,
2373 MEM_CGROUP_RECLAIM_SHRINK);
2374 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2375 /* Usage is reduced ? */
2376 if (curusage >= oldusage)
2379 oldusage = curusage;
2385 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2386 unsigned long long val)
2389 u64 memlimit, oldusage, curusage;
2390 int children = mem_cgroup_count_children(memcg);
2393 /* see mem_cgroup_resize_res_limit */
2394 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
2395 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2396 while (retry_count) {
2397 if (signal_pending(current)) {
2402 * Rather than hide all in some function, I do this in
2403 * open coded manner. You see what this really does.
2404 * We have to guarantee mem->res.limit < mem->memsw.limit.
2406 mutex_lock(&set_limit_mutex);
2407 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2408 if (memlimit > val) {
2410 mutex_unlock(&set_limit_mutex);
2413 ret = res_counter_set_limit(&memcg->memsw, val);
2415 if (memlimit == val)
2416 memcg->memsw_is_minimum = true;
2418 memcg->memsw_is_minimum = false;
2420 mutex_unlock(&set_limit_mutex);
2425 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
2426 MEM_CGROUP_RECLAIM_NOSWAP |
2427 MEM_CGROUP_RECLAIM_SHRINK);
2428 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2429 /* Usage is reduced ? */
2430 if (curusage >= oldusage)
2433 oldusage = curusage;
2438 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2439 gfp_t gfp_mask, int nid,
2442 unsigned long nr_reclaimed = 0;
2443 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2444 unsigned long reclaimed;
2446 struct mem_cgroup_tree_per_zone *mctz;
2447 unsigned long long excess;
2452 mctz = soft_limit_tree_node_zone(nid, zid);
2454 * This loop can run a while, specially if mem_cgroup's continuously
2455 * keep exceeding their soft limit and putting the system under
2462 mz = mem_cgroup_largest_soft_limit_node(mctz);
2466 reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone,
2468 MEM_CGROUP_RECLAIM_SOFT);
2469 nr_reclaimed += reclaimed;
2470 spin_lock(&mctz->lock);
2473 * If we failed to reclaim anything from this memory cgroup
2474 * it is time to move on to the next cgroup
2480 * Loop until we find yet another one.
2482 * By the time we get the soft_limit lock
2483 * again, someone might have aded the
2484 * group back on the RB tree. Iterate to
2485 * make sure we get a different mem.
2486 * mem_cgroup_largest_soft_limit_node returns
2487 * NULL if no other cgroup is present on
2491 __mem_cgroup_largest_soft_limit_node(mctz);
2492 if (next_mz == mz) {
2493 css_put(&next_mz->mem->css);
2495 } else /* next_mz == NULL or other memcg */
2499 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
2500 excess = res_counter_soft_limit_excess(&mz->mem->res);
2502 * One school of thought says that we should not add
2503 * back the node to the tree if reclaim returns 0.
2504 * But our reclaim could return 0, simply because due
2505 * to priority we are exposing a smaller subset of
2506 * memory to reclaim from. Consider this as a longer
2509 /* If excess == 0, no tree ops */
2510 __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
2511 spin_unlock(&mctz->lock);
2512 css_put(&mz->mem->css);
2515 * Could not reclaim anything and there are no more
2516 * mem cgroups to try or we seem to be looping without
2517 * reclaiming anything.
2519 if (!nr_reclaimed &&
2521 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2523 } while (!nr_reclaimed);
2525 css_put(&next_mz->mem->css);
2526 return nr_reclaimed;
2530 * This routine traverse page_cgroup in given list and drop them all.
2531 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
2533 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
2534 int node, int zid, enum lru_list lru)
2537 struct mem_cgroup_per_zone *mz;
2538 struct page_cgroup *pc, *busy;
2539 unsigned long flags, loop;
2540 struct list_head *list;
2543 zone = &NODE_DATA(node)->node_zones[zid];
2544 mz = mem_cgroup_zoneinfo(mem, node, zid);
2545 list = &mz->lists[lru];
2547 loop = MEM_CGROUP_ZSTAT(mz, lru);
2548 /* give some margin against EBUSY etc...*/
2553 spin_lock_irqsave(&zone->lru_lock, flags);
2554 if (list_empty(list)) {
2555 spin_unlock_irqrestore(&zone->lru_lock, flags);
2558 pc = list_entry(list->prev, struct page_cgroup, lru);
2560 list_move(&pc->lru, list);
2562 spin_unlock_irqrestore(&zone->lru_lock, flags);
2565 spin_unlock_irqrestore(&zone->lru_lock, flags);
2567 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
2571 if (ret == -EBUSY || ret == -EINVAL) {
2572 /* found lock contention or "pc" is obsolete. */
2579 if (!ret && !list_empty(list))
2585 * make mem_cgroup's charge to be 0 if there is no task.
2586 * This enables deleting this mem_cgroup.
2588 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
2591 int node, zid, shrink;
2592 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
2593 struct cgroup *cgrp = mem->css.cgroup;
2598 /* should free all ? */
2602 while (mem->res.usage > 0) {
2604 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
2607 if (signal_pending(current))
2609 /* This is for making all *used* pages to be on LRU. */
2610 lru_add_drain_all();
2611 drain_all_stock_sync();
2613 for_each_node_state(node, N_HIGH_MEMORY) {
2614 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
2617 ret = mem_cgroup_force_empty_list(mem,
2626 /* it seems parent cgroup doesn't have enough mem */
2637 /* returns EBUSY if there is a task or if we come here twice. */
2638 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
2642 /* we call try-to-free pages for make this cgroup empty */
2643 lru_add_drain_all();
2644 /* try to free all pages in this cgroup */
2646 while (nr_retries && mem->res.usage > 0) {
2649 if (signal_pending(current)) {
2653 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
2654 false, get_swappiness(mem));
2657 /* maybe some writeback is necessary */
2658 congestion_wait(BLK_RW_ASYNC, HZ/10);
2663 /* try move_account...there may be some *locked* pages. */
2670 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
2672 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
2676 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
2678 return mem_cgroup_from_cont(cont)->use_hierarchy;
2681 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
2685 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2686 struct cgroup *parent = cont->parent;
2687 struct mem_cgroup *parent_mem = NULL;
2690 parent_mem = mem_cgroup_from_cont(parent);
2694 * If parent's use_hierarchy is set, we can't make any modifications
2695 * in the child subtrees. If it is unset, then the change can
2696 * occur, provided the current cgroup has no children.
2698 * For the root cgroup, parent_mem is NULL, we allow value to be
2699 * set if there are no children.
2701 if ((!parent_mem || !parent_mem->use_hierarchy) &&
2702 (val == 1 || val == 0)) {
2703 if (list_empty(&cont->children))
2704 mem->use_hierarchy = val;
2714 struct mem_cgroup_idx_data {
2716 enum mem_cgroup_stat_index idx;
2720 mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data)
2722 struct mem_cgroup_idx_data *d = data;
2723 d->val += mem_cgroup_read_stat(&mem->stat, d->idx);
2728 mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem,
2729 enum mem_cgroup_stat_index idx, s64 *val)
2731 struct mem_cgroup_idx_data d;
2734 mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat);
2738 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
2740 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2744 type = MEMFILE_TYPE(cft->private);
2745 name = MEMFILE_ATTR(cft->private);
2748 if (name == RES_USAGE && mem_cgroup_is_root(mem)) {
2749 mem_cgroup_get_recursive_idx_stat(mem,
2750 MEM_CGROUP_STAT_CACHE, &idx_val);
2752 mem_cgroup_get_recursive_idx_stat(mem,
2753 MEM_CGROUP_STAT_RSS, &idx_val);
2757 val = res_counter_read_u64(&mem->res, name);
2760 if (name == RES_USAGE && mem_cgroup_is_root(mem)) {
2761 mem_cgroup_get_recursive_idx_stat(mem,
2762 MEM_CGROUP_STAT_CACHE, &idx_val);
2764 mem_cgroup_get_recursive_idx_stat(mem,
2765 MEM_CGROUP_STAT_RSS, &idx_val);
2767 mem_cgroup_get_recursive_idx_stat(mem,
2768 MEM_CGROUP_STAT_SWAPOUT, &idx_val);
2772 val = res_counter_read_u64(&mem->memsw, name);
2781 * The user of this function is...
2784 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
2787 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
2789 unsigned long long val;
2792 type = MEMFILE_TYPE(cft->private);
2793 name = MEMFILE_ATTR(cft->private);
2796 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2800 /* This function does all necessary parse...reuse it */
2801 ret = res_counter_memparse_write_strategy(buffer, &val);
2805 ret = mem_cgroup_resize_limit(memcg, val);
2807 ret = mem_cgroup_resize_memsw_limit(memcg, val);
2809 case RES_SOFT_LIMIT:
2810 ret = res_counter_memparse_write_strategy(buffer, &val);
2814 * For memsw, soft limits are hard to implement in terms
2815 * of semantics, for now, we support soft limits for
2816 * control without swap
2819 ret = res_counter_set_soft_limit(&memcg->res, val);
2824 ret = -EINVAL; /* should be BUG() ? */
2830 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
2831 unsigned long long *mem_limit, unsigned long long *memsw_limit)
2833 struct cgroup *cgroup;
2834 unsigned long long min_limit, min_memsw_limit, tmp;
2836 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2837 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2838 cgroup = memcg->css.cgroup;
2839 if (!memcg->use_hierarchy)
2842 while (cgroup->parent) {
2843 cgroup = cgroup->parent;
2844 memcg = mem_cgroup_from_cont(cgroup);
2845 if (!memcg->use_hierarchy)
2847 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
2848 min_limit = min(min_limit, tmp);
2849 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2850 min_memsw_limit = min(min_memsw_limit, tmp);
2853 *mem_limit = min_limit;
2854 *memsw_limit = min_memsw_limit;
2858 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
2860 struct mem_cgroup *mem;
2863 mem = mem_cgroup_from_cont(cont);
2864 type = MEMFILE_TYPE(event);
2865 name = MEMFILE_ATTR(event);
2869 res_counter_reset_max(&mem->res);
2871 res_counter_reset_max(&mem->memsw);
2875 res_counter_reset_failcnt(&mem->res);
2877 res_counter_reset_failcnt(&mem->memsw);
2885 /* For read statistics */
2901 struct mcs_total_stat {
2902 s64 stat[NR_MCS_STAT];
2908 } memcg_stat_strings[NR_MCS_STAT] = {
2909 {"cache", "total_cache"},
2910 {"rss", "total_rss"},
2911 {"mapped_file", "total_mapped_file"},
2912 {"pgpgin", "total_pgpgin"},
2913 {"pgpgout", "total_pgpgout"},
2914 {"swap", "total_swap"},
2915 {"inactive_anon", "total_inactive_anon"},
2916 {"active_anon", "total_active_anon"},
2917 {"inactive_file", "total_inactive_file"},
2918 {"active_file", "total_active_file"},
2919 {"unevictable", "total_unevictable"}
2923 static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
2925 struct mcs_total_stat *s = data;
2929 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE);
2930 s->stat[MCS_CACHE] += val * PAGE_SIZE;
2931 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
2932 s->stat[MCS_RSS] += val * PAGE_SIZE;
2933 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_FILE_MAPPED);
2934 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
2935 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT);
2936 s->stat[MCS_PGPGIN] += val;
2937 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT);
2938 s->stat[MCS_PGPGOUT] += val;
2939 if (do_swap_account) {
2940 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_SWAPOUT);
2941 s->stat[MCS_SWAP] += val * PAGE_SIZE;
2945 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
2946 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
2947 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
2948 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
2949 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
2950 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
2951 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
2952 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
2953 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
2954 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
2959 mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
2961 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
2964 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
2965 struct cgroup_map_cb *cb)
2967 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
2968 struct mcs_total_stat mystat;
2971 memset(&mystat, 0, sizeof(mystat));
2972 mem_cgroup_get_local_stat(mem_cont, &mystat);
2974 for (i = 0; i < NR_MCS_STAT; i++) {
2975 if (i == MCS_SWAP && !do_swap_account)
2977 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
2980 /* Hierarchical information */
2982 unsigned long long limit, memsw_limit;
2983 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
2984 cb->fill(cb, "hierarchical_memory_limit", limit);
2985 if (do_swap_account)
2986 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
2989 memset(&mystat, 0, sizeof(mystat));
2990 mem_cgroup_get_total_stat(mem_cont, &mystat);
2991 for (i = 0; i < NR_MCS_STAT; i++) {
2992 if (i == MCS_SWAP && !do_swap_account)
2994 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
2997 #ifdef CONFIG_DEBUG_VM
2998 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
3002 struct mem_cgroup_per_zone *mz;
3003 unsigned long recent_rotated[2] = {0, 0};
3004 unsigned long recent_scanned[2] = {0, 0};
3006 for_each_online_node(nid)
3007 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
3008 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
3010 recent_rotated[0] +=
3011 mz->reclaim_stat.recent_rotated[0];
3012 recent_rotated[1] +=
3013 mz->reclaim_stat.recent_rotated[1];
3014 recent_scanned[0] +=
3015 mz->reclaim_stat.recent_scanned[0];
3016 recent_scanned[1] +=
3017 mz->reclaim_stat.recent_scanned[1];
3019 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
3020 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
3021 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
3022 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
3029 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
3031 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3033 return get_swappiness(memcg);
3036 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
3039 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3040 struct mem_cgroup *parent;
3045 if (cgrp->parent == NULL)
3048 parent = mem_cgroup_from_cont(cgrp->parent);
3052 /* If under hierarchy, only empty-root can set this value */
3053 if ((parent->use_hierarchy) ||
3054 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
3059 spin_lock(&memcg->reclaim_param_lock);
3060 memcg->swappiness = val;
3061 spin_unlock(&memcg->reclaim_param_lock);
3069 static struct cftype mem_cgroup_files[] = {
3071 .name = "usage_in_bytes",
3072 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
3073 .read_u64 = mem_cgroup_read,
3076 .name = "max_usage_in_bytes",
3077 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
3078 .trigger = mem_cgroup_reset,
3079 .read_u64 = mem_cgroup_read,
3082 .name = "limit_in_bytes",
3083 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
3084 .write_string = mem_cgroup_write,
3085 .read_u64 = mem_cgroup_read,
3088 .name = "soft_limit_in_bytes",
3089 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
3090 .write_string = mem_cgroup_write,
3091 .read_u64 = mem_cgroup_read,
3095 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
3096 .trigger = mem_cgroup_reset,
3097 .read_u64 = mem_cgroup_read,
3101 .read_map = mem_control_stat_show,
3104 .name = "force_empty",
3105 .trigger = mem_cgroup_force_empty_write,
3108 .name = "use_hierarchy",
3109 .write_u64 = mem_cgroup_hierarchy_write,
3110 .read_u64 = mem_cgroup_hierarchy_read,
3113 .name = "swappiness",
3114 .read_u64 = mem_cgroup_swappiness_read,
3115 .write_u64 = mem_cgroup_swappiness_write,
3119 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3120 static struct cftype memsw_cgroup_files[] = {
3122 .name = "memsw.usage_in_bytes",
3123 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
3124 .read_u64 = mem_cgroup_read,
3127 .name = "memsw.max_usage_in_bytes",
3128 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
3129 .trigger = mem_cgroup_reset,
3130 .read_u64 = mem_cgroup_read,
3133 .name = "memsw.limit_in_bytes",
3134 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
3135 .write_string = mem_cgroup_write,
3136 .read_u64 = mem_cgroup_read,
3139 .name = "memsw.failcnt",
3140 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
3141 .trigger = mem_cgroup_reset,
3142 .read_u64 = mem_cgroup_read,
3146 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
3148 if (!do_swap_account)
3150 return cgroup_add_files(cont, ss, memsw_cgroup_files,
3151 ARRAY_SIZE(memsw_cgroup_files));
3154 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
3160 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
3162 struct mem_cgroup_per_node *pn;
3163 struct mem_cgroup_per_zone *mz;
3165 int zone, tmp = node;
3167 * This routine is called against possible nodes.
3168 * But it's BUG to call kmalloc() against offline node.
3170 * TODO: this routine can waste much memory for nodes which will
3171 * never be onlined. It's better to use memory hotplug callback
3174 if (!node_state(node, N_NORMAL_MEMORY))
3176 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
3180 mem->info.nodeinfo[node] = pn;
3181 memset(pn, 0, sizeof(*pn));
3183 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
3184 mz = &pn->zoneinfo[zone];
3186 INIT_LIST_HEAD(&mz->lists[l]);
3187 mz->usage_in_excess = 0;
3188 mz->on_tree = false;
3194 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
3196 kfree(mem->info.nodeinfo[node]);
3199 static int mem_cgroup_size(void)
3201 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
3202 return sizeof(struct mem_cgroup) + cpustat_size;
3205 static struct mem_cgroup *mem_cgroup_alloc(void)
3207 struct mem_cgroup *mem;
3208 int size = mem_cgroup_size();
3210 if (size < PAGE_SIZE)
3211 mem = kmalloc(size, GFP_KERNEL);
3213 mem = vmalloc(size);
3216 memset(mem, 0, size);
3221 * At destroying mem_cgroup, references from swap_cgroup can remain.
3222 * (scanning all at force_empty is too costly...)
3224 * Instead of clearing all references at force_empty, we remember
3225 * the number of reference from swap_cgroup and free mem_cgroup when
3226 * it goes down to 0.
3228 * Removal of cgroup itself succeeds regardless of refs from swap.
3231 static void __mem_cgroup_free(struct mem_cgroup *mem)
3235 mem_cgroup_remove_from_trees(mem);
3236 free_css_id(&mem_cgroup_subsys, &mem->css);
3238 for_each_node_state(node, N_POSSIBLE)
3239 free_mem_cgroup_per_zone_info(mem, node);
3241 if (mem_cgroup_size() < PAGE_SIZE)
3247 static void mem_cgroup_get(struct mem_cgroup *mem)
3249 atomic_inc(&mem->refcnt);
3252 static void mem_cgroup_put(struct mem_cgroup *mem)
3254 if (atomic_dec_and_test(&mem->refcnt)) {
3255 struct mem_cgroup *parent = parent_mem_cgroup(mem);
3256 __mem_cgroup_free(mem);
3258 mem_cgroup_put(parent);
3263 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
3265 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
3267 if (!mem->res.parent)
3269 return mem_cgroup_from_res_counter(mem->res.parent, res);
3272 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3273 static void __init enable_swap_cgroup(void)
3275 if (!mem_cgroup_disabled() && really_do_swap_account)
3276 do_swap_account = 1;
3279 static void __init enable_swap_cgroup(void)
3284 static int mem_cgroup_soft_limit_tree_init(void)
3286 struct mem_cgroup_tree_per_node *rtpn;
3287 struct mem_cgroup_tree_per_zone *rtpz;
3288 int tmp, node, zone;
3290 for_each_node_state(node, N_POSSIBLE) {
3292 if (!node_state(node, N_NORMAL_MEMORY))
3294 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
3298 soft_limit_tree.rb_tree_per_node[node] = rtpn;
3300 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
3301 rtpz = &rtpn->rb_tree_per_zone[zone];
3302 rtpz->rb_root = RB_ROOT;
3303 spin_lock_init(&rtpz->lock);
3309 static struct cgroup_subsys_state * __ref
3310 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
3312 struct mem_cgroup *mem, *parent;
3313 long error = -ENOMEM;
3316 mem = mem_cgroup_alloc();
3318 return ERR_PTR(error);
3320 for_each_node_state(node, N_POSSIBLE)
3321 if (alloc_mem_cgroup_per_zone_info(mem, node))
3325 if (cont->parent == NULL) {
3327 enable_swap_cgroup();
3329 root_mem_cgroup = mem;
3330 if (mem_cgroup_soft_limit_tree_init())
3332 for_each_possible_cpu(cpu) {
3333 struct memcg_stock_pcp *stock =
3334 &per_cpu(memcg_stock, cpu);
3335 INIT_WORK(&stock->work, drain_local_stock);
3337 hotcpu_notifier(memcg_stock_cpu_callback, 0);
3340 parent = mem_cgroup_from_cont(cont->parent);
3341 mem->use_hierarchy = parent->use_hierarchy;
3344 if (parent && parent->use_hierarchy) {
3345 res_counter_init(&mem->res, &parent->res);
3346 res_counter_init(&mem->memsw, &parent->memsw);
3348 * We increment refcnt of the parent to ensure that we can
3349 * safely access it on res_counter_charge/uncharge.
3350 * This refcnt will be decremented when freeing this
3351 * mem_cgroup(see mem_cgroup_put).
3353 mem_cgroup_get(parent);
3355 res_counter_init(&mem->res, NULL);
3356 res_counter_init(&mem->memsw, NULL);
3358 mem->last_scanned_child = 0;
3359 spin_lock_init(&mem->reclaim_param_lock);
3362 mem->swappiness = get_swappiness(parent);
3363 atomic_set(&mem->refcnt, 1);
3366 __mem_cgroup_free(mem);
3367 root_mem_cgroup = NULL;
3368 return ERR_PTR(error);
3371 static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
3372 struct cgroup *cont)
3374 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3376 return mem_cgroup_force_empty(mem, false);
3379 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
3380 struct cgroup *cont)
3382 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3384 mem_cgroup_put(mem);
3387 static int mem_cgroup_populate(struct cgroup_subsys *ss,
3388 struct cgroup *cont)
3392 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
3393 ARRAY_SIZE(mem_cgroup_files));
3396 ret = register_memsw_files(cont, ss);
3400 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
3401 struct cgroup *cont,
3402 struct cgroup *old_cont,
3403 struct task_struct *p,
3406 mutex_lock(&memcg_tasklist);
3408 * FIXME: It's better to move charges of this process from old
3409 * memcg to new memcg. But it's just on TODO-List now.
3411 mutex_unlock(&memcg_tasklist);
3414 struct cgroup_subsys mem_cgroup_subsys = {
3416 .subsys_id = mem_cgroup_subsys_id,
3417 .create = mem_cgroup_create,
3418 .pre_destroy = mem_cgroup_pre_destroy,
3419 .destroy = mem_cgroup_destroy,
3420 .populate = mem_cgroup_populate,
3421 .attach = mem_cgroup_move_task,
3426 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3428 static int __init disable_swap_account(char *s)
3430 really_do_swap_account = 0;
3433 __setup("noswapaccount", disable_swap_account);