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>
43 #include <asm/uaccess.h>
45 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
46 #define MEM_CGROUP_RECLAIM_RETRIES 5
47 struct mem_cgroup *root_mem_cgroup __read_mostly;
49 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
50 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
51 int do_swap_account __read_mostly;
52 static int really_do_swap_account __initdata = 1; /* for remember boot option*/
54 #define do_swap_account (0)
57 static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */
58 #define SOFTLIMIT_EVENTS_THRESH (1000)
61 * Statistics for memory cgroup.
63 enum mem_cgroup_stat_index {
65 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
67 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
68 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
69 MEM_CGROUP_STAT_MAPPED_FILE, /* # of pages charged as file rss */
70 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
71 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
72 MEM_CGROUP_STAT_EVENTS, /* sum of pagein + pageout for internal use */
74 MEM_CGROUP_STAT_NSTATS,
77 struct mem_cgroup_stat_cpu {
78 s64 count[MEM_CGROUP_STAT_NSTATS];
79 } ____cacheline_aligned_in_smp;
81 struct mem_cgroup_stat {
82 struct mem_cgroup_stat_cpu cpustat[0];
86 __mem_cgroup_stat_reset_safe(struct mem_cgroup_stat_cpu *stat,
87 enum mem_cgroup_stat_index idx)
93 __mem_cgroup_stat_read_local(struct mem_cgroup_stat_cpu *stat,
94 enum mem_cgroup_stat_index idx)
96 return stat->count[idx];
100 * For accounting under irq disable, no need for increment preempt count.
102 static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
103 enum mem_cgroup_stat_index idx, int val)
105 stat->count[idx] += val;
108 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
109 enum mem_cgroup_stat_index idx)
113 for_each_possible_cpu(cpu)
114 ret += stat->cpustat[cpu].count[idx];
118 static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat)
122 ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE);
123 ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS);
128 * per-zone information in memory controller.
130 struct mem_cgroup_per_zone {
132 * spin_lock to protect the per cgroup LRU
134 struct list_head lists[NR_LRU_LISTS];
135 unsigned long count[NR_LRU_LISTS];
137 struct zone_reclaim_stat reclaim_stat;
138 struct rb_node tree_node; /* RB tree node */
139 unsigned long long usage_in_excess;/* Set to the value by which */
140 /* the soft limit is exceeded*/
142 struct mem_cgroup *mem; /* Back pointer, we cannot */
143 /* use container_of */
145 /* Macro for accessing counter */
146 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
148 struct mem_cgroup_per_node {
149 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
152 struct mem_cgroup_lru_info {
153 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
157 * Cgroups above their limits are maintained in a RB-Tree, independent of
158 * their hierarchy representation
161 struct mem_cgroup_tree_per_zone {
162 struct rb_root rb_root;
166 struct mem_cgroup_tree_per_node {
167 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
170 struct mem_cgroup_tree {
171 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
174 static struct mem_cgroup_tree soft_limit_tree __read_mostly;
177 * The memory controller data structure. The memory controller controls both
178 * page cache and RSS per cgroup. We would eventually like to provide
179 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
180 * to help the administrator determine what knobs to tune.
182 * TODO: Add a water mark for the memory controller. Reclaim will begin when
183 * we hit the water mark. May be even add a low water mark, such that
184 * no reclaim occurs from a cgroup at it's low water mark, this is
185 * a feature that will be implemented much later in the future.
188 struct cgroup_subsys_state css;
190 * the counter to account for memory usage
192 struct res_counter res;
194 * the counter to account for mem+swap usage.
196 struct res_counter memsw;
198 * Per cgroup active and inactive list, similar to the
199 * per zone LRU lists.
201 struct mem_cgroup_lru_info info;
204 protect against reclaim related member.
206 spinlock_t reclaim_param_lock;
208 int prev_priority; /* for recording reclaim priority */
211 * While reclaiming in a hiearchy, we cache the last child we
214 int last_scanned_child;
216 * Should the accounting and control be hierarchical, per subtree?
219 unsigned long last_oom_jiffies;
222 unsigned int swappiness;
224 /* set when res.limit == memsw.limit */
225 bool memsw_is_minimum;
228 * statistics. This must be placed at the end of memcg.
230 struct mem_cgroup_stat stat;
234 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
235 * limit reclaim to prevent infinite loops, if they ever occur.
237 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
238 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
241 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
242 MEM_CGROUP_CHARGE_TYPE_MAPPED,
243 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
244 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
245 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
246 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
250 /* only for here (for easy reading.) */
251 #define PCGF_CACHE (1UL << PCG_CACHE)
252 #define PCGF_USED (1UL << PCG_USED)
253 #define PCGF_LOCK (1UL << PCG_LOCK)
254 /* Not used, but added here for completeness */
255 #define PCGF_ACCT (1UL << PCG_ACCT)
257 /* for encoding cft->private value on file */
260 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
261 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
262 #define MEMFILE_ATTR(val) ((val) & 0xffff)
265 * Reclaim flags for mem_cgroup_hierarchical_reclaim
267 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
268 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
269 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
270 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
271 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
272 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
274 static void mem_cgroup_get(struct mem_cgroup *mem);
275 static void mem_cgroup_put(struct mem_cgroup *mem);
276 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
278 static struct mem_cgroup_per_zone *
279 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
281 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
284 static struct mem_cgroup_per_zone *
285 page_cgroup_zoneinfo(struct page_cgroup *pc)
287 struct mem_cgroup *mem = pc->mem_cgroup;
288 int nid = page_cgroup_nid(pc);
289 int zid = page_cgroup_zid(pc);
294 return mem_cgroup_zoneinfo(mem, nid, zid);
297 static struct mem_cgroup_tree_per_zone *
298 soft_limit_tree_node_zone(int nid, int zid)
300 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
303 static struct mem_cgroup_tree_per_zone *
304 soft_limit_tree_from_page(struct page *page)
306 int nid = page_to_nid(page);
307 int zid = page_zonenum(page);
309 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
313 __mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
314 struct mem_cgroup_per_zone *mz,
315 struct mem_cgroup_tree_per_zone *mctz)
317 struct rb_node **p = &mctz->rb_root.rb_node;
318 struct rb_node *parent = NULL;
319 struct mem_cgroup_per_zone *mz_node;
324 mz->usage_in_excess = res_counter_soft_limit_excess(&mem->res);
327 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
329 if (mz->usage_in_excess < mz_node->usage_in_excess)
332 * We can't avoid mem cgroups that are over their soft
333 * limit by the same amount
335 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
338 rb_link_node(&mz->tree_node, parent, p);
339 rb_insert_color(&mz->tree_node, &mctz->rb_root);
344 __mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
345 struct mem_cgroup_per_zone *mz,
346 struct mem_cgroup_tree_per_zone *mctz)
350 rb_erase(&mz->tree_node, &mctz->rb_root);
355 mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
356 struct mem_cgroup_per_zone *mz,
357 struct mem_cgroup_tree_per_zone *mctz)
359 spin_lock(&mctz->lock);
360 __mem_cgroup_insert_exceeded(mem, mz, mctz);
361 spin_unlock(&mctz->lock);
365 mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
366 struct mem_cgroup_per_zone *mz,
367 struct mem_cgroup_tree_per_zone *mctz)
369 spin_lock(&mctz->lock);
370 __mem_cgroup_remove_exceeded(mem, mz, mctz);
371 spin_unlock(&mctz->lock);
374 static bool mem_cgroup_soft_limit_check(struct mem_cgroup *mem)
379 struct mem_cgroup_stat_cpu *cpustat;
382 cpustat = &mem->stat.cpustat[cpu];
383 val = __mem_cgroup_stat_read_local(cpustat, MEM_CGROUP_STAT_EVENTS);
384 if (unlikely(val > SOFTLIMIT_EVENTS_THRESH)) {
385 __mem_cgroup_stat_reset_safe(cpustat, MEM_CGROUP_STAT_EVENTS);
392 static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page)
394 unsigned long long prev_usage_in_excess, new_usage_in_excess;
395 bool updated_tree = false;
396 struct mem_cgroup_per_zone *mz;
397 struct mem_cgroup_tree_per_zone *mctz;
399 mz = mem_cgroup_zoneinfo(mem, page_to_nid(page), page_zonenum(page));
400 mctz = soft_limit_tree_from_page(page);
403 * We do updates in lazy mode, mem's are removed
404 * lazily from the per-zone, per-node rb tree
406 prev_usage_in_excess = mz->usage_in_excess;
408 new_usage_in_excess = res_counter_soft_limit_excess(&mem->res);
409 if (prev_usage_in_excess) {
410 mem_cgroup_remove_exceeded(mem, mz, mctz);
413 if (!new_usage_in_excess)
415 mem_cgroup_insert_exceeded(mem, mz, mctz);
419 spin_lock(&mctz->lock);
420 mz->usage_in_excess = new_usage_in_excess;
421 spin_unlock(&mctz->lock);
425 static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem)
428 struct mem_cgroup_per_zone *mz;
429 struct mem_cgroup_tree_per_zone *mctz;
431 for_each_node_state(node, N_POSSIBLE) {
432 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
433 mz = mem_cgroup_zoneinfo(mem, node, zone);
434 mctz = soft_limit_tree_node_zone(node, zone);
435 mem_cgroup_remove_exceeded(mem, mz, mctz);
440 static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem)
442 return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT;
445 static struct mem_cgroup_per_zone *
446 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
448 struct rb_node *rightmost = NULL;
449 struct mem_cgroup_per_zone *mz = NULL;
452 rightmost = rb_last(&mctz->rb_root);
454 goto done; /* Nothing to reclaim from */
456 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
458 * Remove the node now but someone else can add it back,
459 * we will to add it back at the end of reclaim to its correct
460 * position in the tree.
462 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
463 if (!res_counter_soft_limit_excess(&mz->mem->res) ||
464 !css_tryget(&mz->mem->css))
470 static struct mem_cgroup_per_zone *
471 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
473 struct mem_cgroup_per_zone *mz;
475 spin_lock(&mctz->lock);
476 mz = __mem_cgroup_largest_soft_limit_node(mctz);
477 spin_unlock(&mctz->lock);
481 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
482 struct page_cgroup *pc,
485 int val = (charge)? 1 : -1;
486 struct mem_cgroup_stat *stat = &mem->stat;
487 struct mem_cgroup_stat_cpu *cpustat;
490 cpustat = &stat->cpustat[cpu];
491 if (PageCgroupCache(pc))
492 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
494 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
497 __mem_cgroup_stat_add_safe(cpustat,
498 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
500 __mem_cgroup_stat_add_safe(cpustat,
501 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
502 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_EVENTS, 1);
506 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
510 struct mem_cgroup_per_zone *mz;
513 for_each_online_node(nid)
514 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
515 mz = mem_cgroup_zoneinfo(mem, nid, zid);
516 total += MEM_CGROUP_ZSTAT(mz, idx);
521 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
523 return container_of(cgroup_subsys_state(cont,
524 mem_cgroup_subsys_id), struct mem_cgroup,
528 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
531 * mm_update_next_owner() may clear mm->owner to NULL
532 * if it races with swapoff, page migration, etc.
533 * So this can be called with p == NULL.
538 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
539 struct mem_cgroup, css);
542 static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
544 struct mem_cgroup *mem = NULL;
549 * Because we have no locks, mm->owner's may be being moved to other
550 * cgroup. We use css_tryget() here even if this looks
551 * pessimistic (rather than adding locks here).
555 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
558 } while (!css_tryget(&mem->css));
564 * Call callback function against all cgroup under hierarchy tree.
566 static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
567 int (*func)(struct mem_cgroup *, void *))
569 int found, ret, nextid;
570 struct cgroup_subsys_state *css;
571 struct mem_cgroup *mem;
573 if (!root->use_hierarchy)
574 return (*func)(root, data);
582 css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
584 if (css && css_tryget(css))
585 mem = container_of(css, struct mem_cgroup, css);
589 ret = (*func)(mem, data);
593 } while (!ret && css);
598 static inline bool mem_cgroup_is_root(struct mem_cgroup *mem)
600 return (mem == root_mem_cgroup);
604 * Following LRU functions are allowed to be used without PCG_LOCK.
605 * Operations are called by routine of global LRU independently from memcg.
606 * What we have to take care of here is validness of pc->mem_cgroup.
608 * Changes to pc->mem_cgroup happens when
611 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
612 * It is added to LRU before charge.
613 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
614 * When moving account, the page is not on LRU. It's isolated.
617 void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
619 struct page_cgroup *pc;
620 struct mem_cgroup_per_zone *mz;
622 if (mem_cgroup_disabled())
624 pc = lookup_page_cgroup(page);
625 /* can happen while we handle swapcache. */
626 if (!TestClearPageCgroupAcctLRU(pc))
628 VM_BUG_ON(!pc->mem_cgroup);
630 * We don't check PCG_USED bit. It's cleared when the "page" is finally
631 * removed from global LRU.
633 mz = page_cgroup_zoneinfo(pc);
634 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
635 if (mem_cgroup_is_root(pc->mem_cgroup))
637 VM_BUG_ON(list_empty(&pc->lru));
638 list_del_init(&pc->lru);
642 void mem_cgroup_del_lru(struct page *page)
644 mem_cgroup_del_lru_list(page, page_lru(page));
647 void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
649 struct mem_cgroup_per_zone *mz;
650 struct page_cgroup *pc;
652 if (mem_cgroup_disabled())
655 pc = lookup_page_cgroup(page);
657 * Used bit is set without atomic ops but after smp_wmb().
658 * For making pc->mem_cgroup visible, insert smp_rmb() here.
661 /* unused or root page is not rotated. */
662 if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup))
664 mz = page_cgroup_zoneinfo(pc);
665 list_move(&pc->lru, &mz->lists[lru]);
668 void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
670 struct page_cgroup *pc;
671 struct mem_cgroup_per_zone *mz;
673 if (mem_cgroup_disabled())
675 pc = lookup_page_cgroup(page);
676 VM_BUG_ON(PageCgroupAcctLRU(pc));
678 * Used bit is set without atomic ops but after smp_wmb().
679 * For making pc->mem_cgroup visible, insert smp_rmb() here.
682 if (!PageCgroupUsed(pc))
685 mz = page_cgroup_zoneinfo(pc);
686 MEM_CGROUP_ZSTAT(mz, lru) += 1;
687 SetPageCgroupAcctLRU(pc);
688 if (mem_cgroup_is_root(pc->mem_cgroup))
690 list_add(&pc->lru, &mz->lists[lru]);
694 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
695 * lru because the page may.be reused after it's fully uncharged (because of
696 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
697 * it again. This function is only used to charge SwapCache. It's done under
698 * lock_page and expected that zone->lru_lock is never held.
700 static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
703 struct zone *zone = page_zone(page);
704 struct page_cgroup *pc = lookup_page_cgroup(page);
706 spin_lock_irqsave(&zone->lru_lock, flags);
708 * Forget old LRU when this page_cgroup is *not* used. This Used bit
709 * is guarded by lock_page() because the page is SwapCache.
711 if (!PageCgroupUsed(pc))
712 mem_cgroup_del_lru_list(page, page_lru(page));
713 spin_unlock_irqrestore(&zone->lru_lock, flags);
716 static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
719 struct zone *zone = page_zone(page);
720 struct page_cgroup *pc = lookup_page_cgroup(page);
722 spin_lock_irqsave(&zone->lru_lock, flags);
723 /* link when the page is linked to LRU but page_cgroup isn't */
724 if (PageLRU(page) && !PageCgroupAcctLRU(pc))
725 mem_cgroup_add_lru_list(page, page_lru(page));
726 spin_unlock_irqrestore(&zone->lru_lock, flags);
730 void mem_cgroup_move_lists(struct page *page,
731 enum lru_list from, enum lru_list to)
733 if (mem_cgroup_disabled())
735 mem_cgroup_del_lru_list(page, from);
736 mem_cgroup_add_lru_list(page, to);
739 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
742 struct mem_cgroup *curr = NULL;
746 curr = try_get_mem_cgroup_from_mm(task->mm);
751 if (curr->use_hierarchy)
752 ret = css_is_ancestor(&curr->css, &mem->css);
760 * prev_priority control...this will be used in memory reclaim path.
762 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
766 spin_lock(&mem->reclaim_param_lock);
767 prev_priority = mem->prev_priority;
768 spin_unlock(&mem->reclaim_param_lock);
770 return prev_priority;
773 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
775 spin_lock(&mem->reclaim_param_lock);
776 if (priority < mem->prev_priority)
777 mem->prev_priority = priority;
778 spin_unlock(&mem->reclaim_param_lock);
781 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
783 spin_lock(&mem->reclaim_param_lock);
784 mem->prev_priority = priority;
785 spin_unlock(&mem->reclaim_param_lock);
788 static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
790 unsigned long active;
791 unsigned long inactive;
793 unsigned long inactive_ratio;
795 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
796 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
798 gb = (inactive + active) >> (30 - PAGE_SHIFT);
800 inactive_ratio = int_sqrt(10 * gb);
805 present_pages[0] = inactive;
806 present_pages[1] = active;
809 return inactive_ratio;
812 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
814 unsigned long active;
815 unsigned long inactive;
816 unsigned long present_pages[2];
817 unsigned long inactive_ratio;
819 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
821 inactive = present_pages[0];
822 active = present_pages[1];
824 if (inactive * inactive_ratio < active)
830 int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
832 unsigned long active;
833 unsigned long inactive;
835 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
836 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
838 return (active > inactive);
841 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
845 int nid = zone->zone_pgdat->node_id;
846 int zid = zone_idx(zone);
847 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
849 return MEM_CGROUP_ZSTAT(mz, lru);
852 struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
855 int nid = zone->zone_pgdat->node_id;
856 int zid = zone_idx(zone);
857 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
859 return &mz->reclaim_stat;
862 struct zone_reclaim_stat *
863 mem_cgroup_get_reclaim_stat_from_page(struct page *page)
865 struct page_cgroup *pc;
866 struct mem_cgroup_per_zone *mz;
868 if (mem_cgroup_disabled())
871 pc = lookup_page_cgroup(page);
873 * Used bit is set without atomic ops but after smp_wmb().
874 * For making pc->mem_cgroup visible, insert smp_rmb() here.
877 if (!PageCgroupUsed(pc))
880 mz = page_cgroup_zoneinfo(pc);
884 return &mz->reclaim_stat;
887 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
888 struct list_head *dst,
889 unsigned long *scanned, int order,
890 int mode, struct zone *z,
891 struct mem_cgroup *mem_cont,
892 int active, int file)
894 unsigned long nr_taken = 0;
898 struct list_head *src;
899 struct page_cgroup *pc, *tmp;
900 int nid = z->zone_pgdat->node_id;
901 int zid = zone_idx(z);
902 struct mem_cgroup_per_zone *mz;
903 int lru = LRU_FILE * file + active;
907 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
908 src = &mz->lists[lru];
911 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
912 if (scan >= nr_to_scan)
916 if (unlikely(!PageCgroupUsed(pc)))
918 if (unlikely(!PageLRU(page)))
922 ret = __isolate_lru_page(page, mode, file);
925 list_move(&page->lru, dst);
926 mem_cgroup_del_lru(page);
930 /* we don't affect global LRU but rotate in our LRU */
931 mem_cgroup_rotate_lru_list(page, page_lru(page));
942 #define mem_cgroup_from_res_counter(counter, member) \
943 container_of(counter, struct mem_cgroup, member)
945 static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
947 if (do_swap_account) {
948 if (res_counter_check_under_limit(&mem->res) &&
949 res_counter_check_under_limit(&mem->memsw))
952 if (res_counter_check_under_limit(&mem->res))
957 static unsigned int get_swappiness(struct mem_cgroup *memcg)
959 struct cgroup *cgrp = memcg->css.cgroup;
960 unsigned int swappiness;
963 if (cgrp->parent == NULL)
964 return vm_swappiness;
966 spin_lock(&memcg->reclaim_param_lock);
967 swappiness = memcg->swappiness;
968 spin_unlock(&memcg->reclaim_param_lock);
973 static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
981 * mem_cgroup_print_mem_info: Called from OOM with tasklist_lock held in read mode.
982 * @memcg: The memory cgroup that went over limit
983 * @p: Task that is going to be killed
985 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
988 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
990 struct cgroup *task_cgrp;
991 struct cgroup *mem_cgrp;
993 * Need a buffer in BSS, can't rely on allocations. The code relies
994 * on the assumption that OOM is serialized for memory controller.
995 * If this assumption is broken, revisit this code.
997 static char memcg_name[PATH_MAX];
1006 mem_cgrp = memcg->css.cgroup;
1007 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1009 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1012 * Unfortunately, we are unable to convert to a useful name
1013 * But we'll still print out the usage information
1020 printk(KERN_INFO "Task in %s killed", memcg_name);
1023 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1031 * Continues from above, so we don't need an KERN_ level
1033 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1036 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1037 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1038 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1039 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1040 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1042 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1043 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1044 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1048 * This function returns the number of memcg under hierarchy tree. Returns
1049 * 1(self count) if no children.
1051 static int mem_cgroup_count_children(struct mem_cgroup *mem)
1054 mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
1059 * Visit the first child (need not be the first child as per the ordering
1060 * of the cgroup list, since we track last_scanned_child) of @mem and use
1061 * that to reclaim free pages from.
1063 static struct mem_cgroup *
1064 mem_cgroup_select_victim(struct mem_cgroup *root_mem)
1066 struct mem_cgroup *ret = NULL;
1067 struct cgroup_subsys_state *css;
1070 if (!root_mem->use_hierarchy) {
1071 css_get(&root_mem->css);
1077 nextid = root_mem->last_scanned_child + 1;
1078 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
1080 if (css && css_tryget(css))
1081 ret = container_of(css, struct mem_cgroup, css);
1084 /* Updates scanning parameter */
1085 spin_lock(&root_mem->reclaim_param_lock);
1087 /* this means start scan from ID:1 */
1088 root_mem->last_scanned_child = 0;
1090 root_mem->last_scanned_child = found;
1091 spin_unlock(&root_mem->reclaim_param_lock);
1098 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1099 * we reclaimed from, so that we don't end up penalizing one child extensively
1100 * based on its position in the children list.
1102 * root_mem is the original ancestor that we've been reclaim from.
1104 * We give up and return to the caller when we visit root_mem twice.
1105 * (other groups can be removed while we're walking....)
1107 * If shrink==true, for avoiding to free too much, this returns immedieately.
1109 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
1112 unsigned long reclaim_options)
1114 struct mem_cgroup *victim;
1117 bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
1118 bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
1119 bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT;
1120 unsigned long excess = mem_cgroup_get_excess(root_mem);
1122 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1123 if (root_mem->memsw_is_minimum)
1127 victim = mem_cgroup_select_victim(root_mem);
1128 if (victim == root_mem) {
1132 * If we have not been able to reclaim
1133 * anything, it might because there are
1134 * no reclaimable pages under this hierarchy
1136 if (!check_soft || !total) {
1137 css_put(&victim->css);
1141 * We want to do more targetted reclaim.
1142 * excess >> 2 is not to excessive so as to
1143 * reclaim too much, nor too less that we keep
1144 * coming back to reclaim from this cgroup
1146 if (total >= (excess >> 2) ||
1147 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) {
1148 css_put(&victim->css);
1153 if (!mem_cgroup_local_usage(&victim->stat)) {
1154 /* this cgroup's local usage == 0 */
1155 css_put(&victim->css);
1158 /* we use swappiness of local cgroup */
1160 ret = mem_cgroup_shrink_node_zone(victim, gfp_mask,
1161 noswap, get_swappiness(victim), zone,
1162 zone->zone_pgdat->node_id);
1164 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask,
1165 noswap, get_swappiness(victim));
1166 css_put(&victim->css);
1168 * At shrinking usage, we can't check we should stop here or
1169 * reclaim more. It's depends on callers. last_scanned_child
1170 * will work enough for keeping fairness under tree.
1176 if (res_counter_check_under_soft_limit(&root_mem->res))
1178 } else if (mem_cgroup_check_under_limit(root_mem))
1184 bool mem_cgroup_oom_called(struct task_struct *task)
1187 struct mem_cgroup *mem;
1188 struct mm_struct *mm;
1194 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
1195 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
1201 static int record_last_oom_cb(struct mem_cgroup *mem, void *data)
1203 mem->last_oom_jiffies = jiffies;
1207 static void record_last_oom(struct mem_cgroup *mem)
1209 mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb);
1213 * Currently used to update mapped file statistics, but the routine can be
1214 * generalized to update other statistics as well.
1216 void mem_cgroup_update_mapped_file_stat(struct page *page, int val)
1218 struct mem_cgroup *mem;
1219 struct mem_cgroup_stat *stat;
1220 struct mem_cgroup_stat_cpu *cpustat;
1222 struct page_cgroup *pc;
1224 if (!page_is_file_cache(page))
1227 pc = lookup_page_cgroup(page);
1231 lock_page_cgroup(pc);
1232 mem = pc->mem_cgroup;
1236 if (!PageCgroupUsed(pc))
1240 * Preemption is already disabled, we don't need get_cpu()
1242 cpu = smp_processor_id();
1244 cpustat = &stat->cpustat[cpu];
1246 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE, val);
1248 unlock_page_cgroup(pc);
1252 * Unlike exported interface, "oom" parameter is added. if oom==true,
1253 * oom-killer can be invoked.
1255 static int __mem_cgroup_try_charge(struct mm_struct *mm,
1256 gfp_t gfp_mask, struct mem_cgroup **memcg,
1257 bool oom, struct page *page)
1259 struct mem_cgroup *mem, *mem_over_limit, *mem_over_soft_limit;
1260 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1261 struct res_counter *fail_res, *soft_fail_res = NULL;
1263 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
1264 /* Don't account this! */
1270 * We always charge the cgroup the mm_struct belongs to.
1271 * The mm_struct's mem_cgroup changes on task migration if the
1272 * thread group leader migrates. It's possible that mm is not
1273 * set, if so charge the init_mm (happens for pagecache usage).
1277 mem = try_get_mem_cgroup_from_mm(mm);
1285 VM_BUG_ON(css_is_removed(&mem->css));
1289 unsigned long flags = 0;
1291 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res,
1294 if (!do_swap_account)
1296 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
1300 /* mem+swap counter fails */
1301 res_counter_uncharge(&mem->res, PAGE_SIZE, NULL);
1302 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
1303 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1306 /* mem counter fails */
1307 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1310 if (!(gfp_mask & __GFP_WAIT))
1313 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
1319 * try_to_free_mem_cgroup_pages() might not give us a full
1320 * picture of reclaim. Some pages are reclaimed and might be
1321 * moved to swap cache or just unmapped from the cgroup.
1322 * Check the limit again to see if the reclaim reduced the
1323 * current usage of the cgroup before giving up
1326 if (mem_cgroup_check_under_limit(mem_over_limit))
1329 if (!nr_retries--) {
1331 mutex_lock(&memcg_tasklist);
1332 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
1333 mutex_unlock(&memcg_tasklist);
1334 record_last_oom(mem_over_limit);
1340 * Insert just the ancestor, we should trickle down to the correct
1341 * cgroup for reclaim, since the other nodes will be below their
1344 if (soft_fail_res) {
1345 mem_over_soft_limit =
1346 mem_cgroup_from_res_counter(soft_fail_res, res);
1347 if (mem_cgroup_soft_limit_check(mem_over_soft_limit))
1348 mem_cgroup_update_tree(mem_over_soft_limit, page);
1357 * A helper function to get mem_cgroup from ID. must be called under
1358 * rcu_read_lock(). The caller must check css_is_removed() or some if
1359 * it's concern. (dropping refcnt from swap can be called against removed
1362 static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
1364 struct cgroup_subsys_state *css;
1366 /* ID 0 is unused ID */
1369 css = css_lookup(&mem_cgroup_subsys, id);
1372 return container_of(css, struct mem_cgroup, css);
1375 static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
1377 struct mem_cgroup *mem;
1378 struct page_cgroup *pc;
1382 VM_BUG_ON(!PageLocked(page));
1384 if (!PageSwapCache(page))
1387 pc = lookup_page_cgroup(page);
1388 lock_page_cgroup(pc);
1389 if (PageCgroupUsed(pc)) {
1390 mem = pc->mem_cgroup;
1391 if (mem && !css_tryget(&mem->css))
1394 ent.val = page_private(page);
1395 id = lookup_swap_cgroup(ent);
1397 mem = mem_cgroup_lookup(id);
1398 if (mem && !css_tryget(&mem->css))
1402 unlock_page_cgroup(pc);
1407 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
1408 * USED state. If already USED, uncharge and return.
1411 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
1412 struct page_cgroup *pc,
1413 enum charge_type ctype)
1415 /* try_charge() can return NULL to *memcg, taking care of it. */
1419 lock_page_cgroup(pc);
1420 if (unlikely(PageCgroupUsed(pc))) {
1421 unlock_page_cgroup(pc);
1422 res_counter_uncharge(&mem->res, PAGE_SIZE, NULL);
1423 if (do_swap_account)
1424 res_counter_uncharge(&mem->memsw, PAGE_SIZE, NULL);
1429 pc->mem_cgroup = mem;
1431 * We access a page_cgroup asynchronously without lock_page_cgroup().
1432 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
1433 * is accessed after testing USED bit. To make pc->mem_cgroup visible
1434 * before USED bit, we need memory barrier here.
1435 * See mem_cgroup_add_lru_list(), etc.
1439 case MEM_CGROUP_CHARGE_TYPE_CACHE:
1440 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
1441 SetPageCgroupCache(pc);
1442 SetPageCgroupUsed(pc);
1444 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1445 ClearPageCgroupCache(pc);
1446 SetPageCgroupUsed(pc);
1452 mem_cgroup_charge_statistics(mem, pc, true);
1454 unlock_page_cgroup(pc);
1458 * mem_cgroup_move_account - move account of the page
1459 * @pc: page_cgroup of the page.
1460 * @from: mem_cgroup which the page is moved from.
1461 * @to: mem_cgroup which the page is moved to. @from != @to.
1463 * The caller must confirm following.
1464 * - page is not on LRU (isolate_page() is useful.)
1466 * returns 0 at success,
1467 * returns -EBUSY when lock is busy or "pc" is unstable.
1469 * This function does "uncharge" from old cgroup but doesn't do "charge" to
1470 * new cgroup. It should be done by a caller.
1473 static int mem_cgroup_move_account(struct page_cgroup *pc,
1474 struct mem_cgroup *from, struct mem_cgroup *to)
1476 struct mem_cgroup_per_zone *from_mz, *to_mz;
1481 struct mem_cgroup_stat *stat;
1482 struct mem_cgroup_stat_cpu *cpustat;
1484 VM_BUG_ON(from == to);
1485 VM_BUG_ON(PageLRU(pc->page));
1487 nid = page_cgroup_nid(pc);
1488 zid = page_cgroup_zid(pc);
1489 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
1490 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
1492 if (!trylock_page_cgroup(pc))
1495 if (!PageCgroupUsed(pc))
1498 if (pc->mem_cgroup != from)
1501 res_counter_uncharge(&from->res, PAGE_SIZE, NULL);
1502 mem_cgroup_charge_statistics(from, pc, false);
1505 if (page_is_file_cache(page) && page_mapped(page)) {
1506 cpu = smp_processor_id();
1507 /* Update mapped_file data for mem_cgroup "from" */
1509 cpustat = &stat->cpustat[cpu];
1510 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE,
1513 /* Update mapped_file data for mem_cgroup "to" */
1515 cpustat = &stat->cpustat[cpu];
1516 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE,
1520 if (do_swap_account)
1521 res_counter_uncharge(&from->memsw, PAGE_SIZE, NULL);
1522 css_put(&from->css);
1525 pc->mem_cgroup = to;
1526 mem_cgroup_charge_statistics(to, pc, true);
1529 unlock_page_cgroup(pc);
1531 * We charges against "to" which may not have any tasks. Then, "to"
1532 * can be under rmdir(). But in current implementation, caller of
1533 * this function is just force_empty() and it's garanteed that
1534 * "to" is never removed. So, we don't check rmdir status here.
1540 * move charges to its parent.
1543 static int mem_cgroup_move_parent(struct page_cgroup *pc,
1544 struct mem_cgroup *child,
1547 struct page *page = pc->page;
1548 struct cgroup *cg = child->css.cgroup;
1549 struct cgroup *pcg = cg->parent;
1550 struct mem_cgroup *parent;
1558 parent = mem_cgroup_from_cont(pcg);
1561 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false, page);
1565 if (!get_page_unless_zero(page)) {
1570 ret = isolate_lru_page(page);
1575 ret = mem_cgroup_move_account(pc, child, parent);
1577 putback_lru_page(page);
1580 /* drop extra refcnt by try_charge() */
1581 css_put(&parent->css);
1588 /* drop extra refcnt by try_charge() */
1589 css_put(&parent->css);
1590 /* uncharge if move fails */
1591 res_counter_uncharge(&parent->res, PAGE_SIZE, NULL);
1592 if (do_swap_account)
1593 res_counter_uncharge(&parent->memsw, PAGE_SIZE, NULL);
1598 * Charge the memory controller for page usage.
1600 * 0 if the charge was successful
1601 * < 0 if the cgroup is over its limit
1603 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1604 gfp_t gfp_mask, enum charge_type ctype,
1605 struct mem_cgroup *memcg)
1607 struct mem_cgroup *mem;
1608 struct page_cgroup *pc;
1611 pc = lookup_page_cgroup(page);
1612 /* can happen at boot */
1618 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true, page);
1622 __mem_cgroup_commit_charge(mem, pc, ctype);
1626 int mem_cgroup_newpage_charge(struct page *page,
1627 struct mm_struct *mm, gfp_t gfp_mask)
1629 if (mem_cgroup_disabled())
1631 if (PageCompound(page))
1634 * If already mapped, we don't have to account.
1635 * If page cache, page->mapping has address_space.
1636 * But page->mapping may have out-of-use anon_vma pointer,
1637 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1640 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1644 return mem_cgroup_charge_common(page, mm, gfp_mask,
1645 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
1649 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1650 enum charge_type ctype);
1652 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1655 struct mem_cgroup *mem = NULL;
1658 if (mem_cgroup_disabled())
1660 if (PageCompound(page))
1663 * Corner case handling. This is called from add_to_page_cache()
1664 * in usual. But some FS (shmem) precharges this page before calling it
1665 * and call add_to_page_cache() with GFP_NOWAIT.
1667 * For GFP_NOWAIT case, the page may be pre-charged before calling
1668 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1669 * charge twice. (It works but has to pay a bit larger cost.)
1670 * And when the page is SwapCache, it should take swap information
1671 * into account. This is under lock_page() now.
1673 if (!(gfp_mask & __GFP_WAIT)) {
1674 struct page_cgroup *pc;
1677 pc = lookup_page_cgroup(page);
1680 lock_page_cgroup(pc);
1681 if (PageCgroupUsed(pc)) {
1682 unlock_page_cgroup(pc);
1685 unlock_page_cgroup(pc);
1688 if (unlikely(!mm && !mem))
1691 if (page_is_file_cache(page))
1692 return mem_cgroup_charge_common(page, mm, gfp_mask,
1693 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
1696 if (PageSwapCache(page)) {
1697 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
1699 __mem_cgroup_commit_charge_swapin(page, mem,
1700 MEM_CGROUP_CHARGE_TYPE_SHMEM);
1702 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1703 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1709 * While swap-in, try_charge -> commit or cancel, the page is locked.
1710 * And when try_charge() successfully returns, one refcnt to memcg without
1711 * struct page_cgroup is aquired. This refcnt will be cumsumed by
1712 * "commit()" or removed by "cancel()"
1714 int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1716 gfp_t mask, struct mem_cgroup **ptr)
1718 struct mem_cgroup *mem;
1721 if (mem_cgroup_disabled())
1724 if (!do_swap_account)
1727 * A racing thread's fault, or swapoff, may have already updated
1728 * the pte, and even removed page from swap cache: return success
1729 * to go on to do_swap_page()'s pte_same() test, which should fail.
1731 if (!PageSwapCache(page))
1733 mem = try_get_mem_cgroup_from_swapcache(page);
1737 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, page);
1738 /* drop extra refcnt from tryget */
1744 return __mem_cgroup_try_charge(mm, mask, ptr, true, page);
1748 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1749 enum charge_type ctype)
1751 struct page_cgroup *pc;
1753 if (mem_cgroup_disabled())
1757 cgroup_exclude_rmdir(&ptr->css);
1758 pc = lookup_page_cgroup(page);
1759 mem_cgroup_lru_del_before_commit_swapcache(page);
1760 __mem_cgroup_commit_charge(ptr, pc, ctype);
1761 mem_cgroup_lru_add_after_commit_swapcache(page);
1763 * Now swap is on-memory. This means this page may be
1764 * counted both as mem and swap....double count.
1765 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1766 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1767 * may call delete_from_swap_cache() before reach here.
1769 if (do_swap_account && PageSwapCache(page)) {
1770 swp_entry_t ent = {.val = page_private(page)};
1772 struct mem_cgroup *memcg;
1774 id = swap_cgroup_record(ent, 0);
1776 memcg = mem_cgroup_lookup(id);
1779 * This recorded memcg can be obsolete one. So, avoid
1780 * calling css_tryget
1782 res_counter_uncharge(&memcg->memsw, PAGE_SIZE, NULL);
1783 mem_cgroup_put(memcg);
1788 * At swapin, we may charge account against cgroup which has no tasks.
1789 * So, rmdir()->pre_destroy() can be called while we do this charge.
1790 * In that case, we need to call pre_destroy() again. check it here.
1792 cgroup_release_and_wakeup_rmdir(&ptr->css);
1795 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1797 __mem_cgroup_commit_charge_swapin(page, ptr,
1798 MEM_CGROUP_CHARGE_TYPE_MAPPED);
1801 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1803 if (mem_cgroup_disabled())
1807 res_counter_uncharge(&mem->res, PAGE_SIZE, NULL);
1808 if (do_swap_account)
1809 res_counter_uncharge(&mem->memsw, PAGE_SIZE, NULL);
1815 * uncharge if !page_mapped(page)
1817 static struct mem_cgroup *
1818 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
1820 struct page_cgroup *pc;
1821 struct mem_cgroup *mem = NULL;
1822 struct mem_cgroup_per_zone *mz;
1823 bool soft_limit_excess = false;
1825 if (mem_cgroup_disabled())
1828 if (PageSwapCache(page))
1832 * Check if our page_cgroup is valid
1834 pc = lookup_page_cgroup(page);
1835 if (unlikely(!pc || !PageCgroupUsed(pc)))
1838 lock_page_cgroup(pc);
1840 mem = pc->mem_cgroup;
1842 if (!PageCgroupUsed(pc))
1846 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1847 case MEM_CGROUP_CHARGE_TYPE_DROP:
1848 if (page_mapped(page))
1851 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1852 if (!PageAnon(page)) { /* Shared memory */
1853 if (page->mapping && !page_is_file_cache(page))
1855 } else if (page_mapped(page)) /* Anon */
1862 res_counter_uncharge(&mem->res, PAGE_SIZE, &soft_limit_excess);
1863 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1864 res_counter_uncharge(&mem->memsw, PAGE_SIZE, NULL);
1865 mem_cgroup_charge_statistics(mem, pc, false);
1867 ClearPageCgroupUsed(pc);
1869 * pc->mem_cgroup is not cleared here. It will be accessed when it's
1870 * freed from LRU. This is safe because uncharged page is expected not
1871 * to be reused (freed soon). Exception is SwapCache, it's handled by
1872 * special functions.
1875 mz = page_cgroup_zoneinfo(pc);
1876 unlock_page_cgroup(pc);
1878 if (soft_limit_excess && mem_cgroup_soft_limit_check(mem))
1879 mem_cgroup_update_tree(mem, page);
1880 /* at swapout, this memcg will be accessed to record to swap */
1881 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1887 unlock_page_cgroup(pc);
1891 void mem_cgroup_uncharge_page(struct page *page)
1894 if (page_mapped(page))
1896 if (page->mapping && !PageAnon(page))
1898 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1901 void mem_cgroup_uncharge_cache_page(struct page *page)
1903 VM_BUG_ON(page_mapped(page));
1904 VM_BUG_ON(page->mapping);
1905 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1910 * called after __delete_from_swap_cache() and drop "page" account.
1911 * memcg information is recorded to swap_cgroup of "ent"
1914 mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
1916 struct mem_cgroup *memcg;
1917 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
1919 if (!swapout) /* this was a swap cache but the swap is unused ! */
1920 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
1922 memcg = __mem_cgroup_uncharge_common(page, ctype);
1924 /* record memcg information */
1925 if (do_swap_account && swapout && memcg) {
1926 swap_cgroup_record(ent, css_id(&memcg->css));
1927 mem_cgroup_get(memcg);
1929 if (swapout && memcg)
1930 css_put(&memcg->css);
1934 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1936 * called from swap_entry_free(). remove record in swap_cgroup and
1937 * uncharge "memsw" account.
1939 void mem_cgroup_uncharge_swap(swp_entry_t ent)
1941 struct mem_cgroup *memcg;
1944 if (!do_swap_account)
1947 id = swap_cgroup_record(ent, 0);
1949 memcg = mem_cgroup_lookup(id);
1952 * We uncharge this because swap is freed.
1953 * This memcg can be obsolete one. We avoid calling css_tryget
1955 res_counter_uncharge(&memcg->memsw, PAGE_SIZE, NULL);
1956 mem_cgroup_put(memcg);
1963 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1966 int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
1968 struct page_cgroup *pc;
1969 struct mem_cgroup *mem = NULL;
1972 if (mem_cgroup_disabled())
1975 pc = lookup_page_cgroup(page);
1976 lock_page_cgroup(pc);
1977 if (PageCgroupUsed(pc)) {
1978 mem = pc->mem_cgroup;
1981 unlock_page_cgroup(pc);
1984 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false,
1992 /* remove redundant charge if migration failed*/
1993 void mem_cgroup_end_migration(struct mem_cgroup *mem,
1994 struct page *oldpage, struct page *newpage)
1996 struct page *target, *unused;
1997 struct page_cgroup *pc;
1998 enum charge_type ctype;
2002 cgroup_exclude_rmdir(&mem->css);
2003 /* at migration success, oldpage->mapping is NULL. */
2004 if (oldpage->mapping) {
2012 if (PageAnon(target))
2013 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
2014 else if (page_is_file_cache(target))
2015 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
2017 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
2019 /* unused page is not on radix-tree now. */
2021 __mem_cgroup_uncharge_common(unused, ctype);
2023 pc = lookup_page_cgroup(target);
2025 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
2026 * So, double-counting is effectively avoided.
2028 __mem_cgroup_commit_charge(mem, pc, ctype);
2031 * Both of oldpage and newpage are still under lock_page().
2032 * Then, we don't have to care about race in radix-tree.
2033 * But we have to be careful that this page is unmapped or not.
2035 * There is a case for !page_mapped(). At the start of
2036 * migration, oldpage was mapped. But now, it's zapped.
2037 * But we know *target* page is not freed/reused under us.
2038 * mem_cgroup_uncharge_page() does all necessary checks.
2040 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
2041 mem_cgroup_uncharge_page(target);
2043 * At migration, we may charge account against cgroup which has no tasks
2044 * So, rmdir()->pre_destroy() can be called while we do this charge.
2045 * In that case, we need to call pre_destroy() again. check it here.
2047 cgroup_release_and_wakeup_rmdir(&mem->css);
2051 * A call to try to shrink memory usage on charge failure at shmem's swapin.
2052 * Calling hierarchical_reclaim is not enough because we should update
2053 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
2054 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
2055 * not from the memcg which this page would be charged to.
2056 * try_charge_swapin does all of these works properly.
2058 int mem_cgroup_shmem_charge_fallback(struct page *page,
2059 struct mm_struct *mm,
2062 struct mem_cgroup *mem = NULL;
2065 if (mem_cgroup_disabled())
2068 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2070 mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
2075 static DEFINE_MUTEX(set_limit_mutex);
2077 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
2078 unsigned long long val)
2084 int children = mem_cgroup_count_children(memcg);
2085 u64 curusage, oldusage;
2088 * For keeping hierarchical_reclaim simple, how long we should retry
2089 * is depends on callers. We set our retry-count to be function
2090 * of # of children which we should visit in this loop.
2092 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
2094 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2096 while (retry_count) {
2097 if (signal_pending(current)) {
2102 * Rather than hide all in some function, I do this in
2103 * open coded manner. You see what this really does.
2104 * We have to guarantee mem->res.limit < mem->memsw.limit.
2106 mutex_lock(&set_limit_mutex);
2107 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2108 if (memswlimit < val) {
2110 mutex_unlock(&set_limit_mutex);
2113 ret = res_counter_set_limit(&memcg->res, val);
2115 if (memswlimit == val)
2116 memcg->memsw_is_minimum = true;
2118 memcg->memsw_is_minimum = false;
2120 mutex_unlock(&set_limit_mutex);
2125 progress = mem_cgroup_hierarchical_reclaim(memcg, NULL,
2127 MEM_CGROUP_RECLAIM_SHRINK);
2128 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2129 /* Usage is reduced ? */
2130 if (curusage >= oldusage)
2133 oldusage = curusage;
2139 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2140 unsigned long long val)
2143 u64 memlimit, oldusage, curusage;
2144 int children = mem_cgroup_count_children(memcg);
2147 /* see mem_cgroup_resize_res_limit */
2148 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
2149 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2150 while (retry_count) {
2151 if (signal_pending(current)) {
2156 * Rather than hide all in some function, I do this in
2157 * open coded manner. You see what this really does.
2158 * We have to guarantee mem->res.limit < mem->memsw.limit.
2160 mutex_lock(&set_limit_mutex);
2161 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2162 if (memlimit > val) {
2164 mutex_unlock(&set_limit_mutex);
2167 ret = res_counter_set_limit(&memcg->memsw, val);
2169 if (memlimit == val)
2170 memcg->memsw_is_minimum = true;
2172 memcg->memsw_is_minimum = false;
2174 mutex_unlock(&set_limit_mutex);
2179 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
2180 MEM_CGROUP_RECLAIM_NOSWAP |
2181 MEM_CGROUP_RECLAIM_SHRINK);
2182 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2183 /* Usage is reduced ? */
2184 if (curusage >= oldusage)
2187 oldusage = curusage;
2192 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2193 gfp_t gfp_mask, int nid,
2196 unsigned long nr_reclaimed = 0;
2197 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2198 unsigned long reclaimed;
2200 struct mem_cgroup_tree_per_zone *mctz;
2205 mctz = soft_limit_tree_node_zone(nid, zid);
2207 * This loop can run a while, specially if mem_cgroup's continuously
2208 * keep exceeding their soft limit and putting the system under
2215 mz = mem_cgroup_largest_soft_limit_node(mctz);
2219 reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone,
2221 MEM_CGROUP_RECLAIM_SOFT);
2222 nr_reclaimed += reclaimed;
2223 spin_lock(&mctz->lock);
2226 * If we failed to reclaim anything from this memory cgroup
2227 * it is time to move on to the next cgroup
2233 * Loop until we find yet another one.
2235 * By the time we get the soft_limit lock
2236 * again, someone might have aded the
2237 * group back on the RB tree. Iterate to
2238 * make sure we get a different mem.
2239 * mem_cgroup_largest_soft_limit_node returns
2240 * NULL if no other cgroup is present on
2244 __mem_cgroup_largest_soft_limit_node(mctz);
2245 if (next_mz == mz) {
2246 css_put(&next_mz->mem->css);
2248 } else /* next_mz == NULL or other memcg */
2252 mz->usage_in_excess =
2253 res_counter_soft_limit_excess(&mz->mem->res);
2254 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
2256 * One school of thought says that we should not add
2257 * back the node to the tree if reclaim returns 0.
2258 * But our reclaim could return 0, simply because due
2259 * to priority we are exposing a smaller subset of
2260 * memory to reclaim from. Consider this as a longer
2263 if (mz->usage_in_excess)
2264 __mem_cgroup_insert_exceeded(mz->mem, mz, mctz);
2265 spin_unlock(&mctz->lock);
2266 css_put(&mz->mem->css);
2269 * Could not reclaim anything and there are no more
2270 * mem cgroups to try or we seem to be looping without
2271 * reclaiming anything.
2273 if (!nr_reclaimed &&
2275 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2277 } while (!nr_reclaimed);
2279 css_put(&next_mz->mem->css);
2280 return nr_reclaimed;
2284 * This routine traverse page_cgroup in given list and drop them all.
2285 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
2287 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
2288 int node, int zid, enum lru_list lru)
2291 struct mem_cgroup_per_zone *mz;
2292 struct page_cgroup *pc, *busy;
2293 unsigned long flags, loop;
2294 struct list_head *list;
2297 zone = &NODE_DATA(node)->node_zones[zid];
2298 mz = mem_cgroup_zoneinfo(mem, node, zid);
2299 list = &mz->lists[lru];
2301 loop = MEM_CGROUP_ZSTAT(mz, lru);
2302 /* give some margin against EBUSY etc...*/
2307 spin_lock_irqsave(&zone->lru_lock, flags);
2308 if (list_empty(list)) {
2309 spin_unlock_irqrestore(&zone->lru_lock, flags);
2312 pc = list_entry(list->prev, struct page_cgroup, lru);
2314 list_move(&pc->lru, list);
2316 spin_unlock_irqrestore(&zone->lru_lock, flags);
2319 spin_unlock_irqrestore(&zone->lru_lock, flags);
2321 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
2325 if (ret == -EBUSY || ret == -EINVAL) {
2326 /* found lock contention or "pc" is obsolete. */
2333 if (!ret && !list_empty(list))
2339 * make mem_cgroup's charge to be 0 if there is no task.
2340 * This enables deleting this mem_cgroup.
2342 static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
2345 int node, zid, shrink;
2346 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
2347 struct cgroup *cgrp = mem->css.cgroup;
2352 /* should free all ? */
2356 while (mem->res.usage > 0) {
2358 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
2361 if (signal_pending(current))
2363 /* This is for making all *used* pages to be on LRU. */
2364 lru_add_drain_all();
2366 for_each_node_state(node, N_HIGH_MEMORY) {
2367 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
2370 ret = mem_cgroup_force_empty_list(mem,
2379 /* it seems parent cgroup doesn't have enough mem */
2390 /* returns EBUSY if there is a task or if we come here twice. */
2391 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
2395 /* we call try-to-free pages for make this cgroup empty */
2396 lru_add_drain_all();
2397 /* try to free all pages in this cgroup */
2399 while (nr_retries && mem->res.usage > 0) {
2402 if (signal_pending(current)) {
2406 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
2407 false, get_swappiness(mem));
2410 /* maybe some writeback is necessary */
2411 congestion_wait(BLK_RW_ASYNC, HZ/10);
2416 /* try move_account...there may be some *locked* pages. */
2423 int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
2425 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
2429 static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
2431 return mem_cgroup_from_cont(cont)->use_hierarchy;
2434 static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
2438 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2439 struct cgroup *parent = cont->parent;
2440 struct mem_cgroup *parent_mem = NULL;
2443 parent_mem = mem_cgroup_from_cont(parent);
2447 * If parent's use_hiearchy is set, we can't make any modifications
2448 * in the child subtrees. If it is unset, then the change can
2449 * occur, provided the current cgroup has no children.
2451 * For the root cgroup, parent_mem is NULL, we allow value to be
2452 * set if there are no children.
2454 if ((!parent_mem || !parent_mem->use_hierarchy) &&
2455 (val == 1 || val == 0)) {
2456 if (list_empty(&cont->children))
2457 mem->use_hierarchy = val;
2467 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
2469 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2473 type = MEMFILE_TYPE(cft->private);
2474 name = MEMFILE_ATTR(cft->private);
2477 val = res_counter_read_u64(&mem->res, name);
2480 val = res_counter_read_u64(&mem->memsw, name);
2489 * The user of this function is...
2492 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
2495 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
2497 unsigned long long val;
2500 type = MEMFILE_TYPE(cft->private);
2501 name = MEMFILE_ATTR(cft->private);
2504 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2508 /* This function does all necessary parse...reuse it */
2509 ret = res_counter_memparse_write_strategy(buffer, &val);
2513 ret = mem_cgroup_resize_limit(memcg, val);
2515 ret = mem_cgroup_resize_memsw_limit(memcg, val);
2517 case RES_SOFT_LIMIT:
2518 ret = res_counter_memparse_write_strategy(buffer, &val);
2522 * For memsw, soft limits are hard to implement in terms
2523 * of semantics, for now, we support soft limits for
2524 * control without swap
2527 ret = res_counter_set_soft_limit(&memcg->res, val);
2532 ret = -EINVAL; /* should be BUG() ? */
2538 static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
2539 unsigned long long *mem_limit, unsigned long long *memsw_limit)
2541 struct cgroup *cgroup;
2542 unsigned long long min_limit, min_memsw_limit, tmp;
2544 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2545 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2546 cgroup = memcg->css.cgroup;
2547 if (!memcg->use_hierarchy)
2550 while (cgroup->parent) {
2551 cgroup = cgroup->parent;
2552 memcg = mem_cgroup_from_cont(cgroup);
2553 if (!memcg->use_hierarchy)
2555 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
2556 min_limit = min(min_limit, tmp);
2557 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2558 min_memsw_limit = min(min_memsw_limit, tmp);
2561 *mem_limit = min_limit;
2562 *memsw_limit = min_memsw_limit;
2566 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
2568 struct mem_cgroup *mem;
2571 mem = mem_cgroup_from_cont(cont);
2572 type = MEMFILE_TYPE(event);
2573 name = MEMFILE_ATTR(event);
2577 res_counter_reset_max(&mem->res);
2579 res_counter_reset_max(&mem->memsw);
2583 res_counter_reset_failcnt(&mem->res);
2585 res_counter_reset_failcnt(&mem->memsw);
2593 /* For read statistics */
2608 struct mcs_total_stat {
2609 s64 stat[NR_MCS_STAT];
2615 } memcg_stat_strings[NR_MCS_STAT] = {
2616 {"cache", "total_cache"},
2617 {"rss", "total_rss"},
2618 {"mapped_file", "total_mapped_file"},
2619 {"pgpgin", "total_pgpgin"},
2620 {"pgpgout", "total_pgpgout"},
2621 {"inactive_anon", "total_inactive_anon"},
2622 {"active_anon", "total_active_anon"},
2623 {"inactive_file", "total_inactive_file"},
2624 {"active_file", "total_active_file"},
2625 {"unevictable", "total_unevictable"}
2629 static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
2631 struct mcs_total_stat *s = data;
2635 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE);
2636 s->stat[MCS_CACHE] += val * PAGE_SIZE;
2637 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
2638 s->stat[MCS_RSS] += val * PAGE_SIZE;
2639 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_MAPPED_FILE);
2640 s->stat[MCS_MAPPED_FILE] += val * PAGE_SIZE;
2641 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT);
2642 s->stat[MCS_PGPGIN] += val;
2643 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT);
2644 s->stat[MCS_PGPGOUT] += val;
2647 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
2648 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
2649 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
2650 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
2651 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
2652 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
2653 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
2654 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
2655 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
2656 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
2661 mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
2663 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
2666 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
2667 struct cgroup_map_cb *cb)
2669 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
2670 struct mcs_total_stat mystat;
2673 memset(&mystat, 0, sizeof(mystat));
2674 mem_cgroup_get_local_stat(mem_cont, &mystat);
2676 for (i = 0; i < NR_MCS_STAT; i++)
2677 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
2679 /* Hierarchical information */
2681 unsigned long long limit, memsw_limit;
2682 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
2683 cb->fill(cb, "hierarchical_memory_limit", limit);
2684 if (do_swap_account)
2685 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
2688 memset(&mystat, 0, sizeof(mystat));
2689 mem_cgroup_get_total_stat(mem_cont, &mystat);
2690 for (i = 0; i < NR_MCS_STAT; i++)
2691 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
2694 #ifdef CONFIG_DEBUG_VM
2695 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
2699 struct mem_cgroup_per_zone *mz;
2700 unsigned long recent_rotated[2] = {0, 0};
2701 unsigned long recent_scanned[2] = {0, 0};
2703 for_each_online_node(nid)
2704 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2705 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
2707 recent_rotated[0] +=
2708 mz->reclaim_stat.recent_rotated[0];
2709 recent_rotated[1] +=
2710 mz->reclaim_stat.recent_rotated[1];
2711 recent_scanned[0] +=
2712 mz->reclaim_stat.recent_scanned[0];
2713 recent_scanned[1] +=
2714 mz->reclaim_stat.recent_scanned[1];
2716 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
2717 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
2718 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
2719 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
2726 static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
2728 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2730 return get_swappiness(memcg);
2733 static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
2736 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2737 struct mem_cgroup *parent;
2742 if (cgrp->parent == NULL)
2745 parent = mem_cgroup_from_cont(cgrp->parent);
2749 /* If under hierarchy, only empty-root can set this value */
2750 if ((parent->use_hierarchy) ||
2751 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
2756 spin_lock(&memcg->reclaim_param_lock);
2757 memcg->swappiness = val;
2758 spin_unlock(&memcg->reclaim_param_lock);
2766 static struct cftype mem_cgroup_files[] = {
2768 .name = "usage_in_bytes",
2769 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
2770 .read_u64 = mem_cgroup_read,
2773 .name = "max_usage_in_bytes",
2774 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
2775 .trigger = mem_cgroup_reset,
2776 .read_u64 = mem_cgroup_read,
2779 .name = "limit_in_bytes",
2780 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
2781 .write_string = mem_cgroup_write,
2782 .read_u64 = mem_cgroup_read,
2785 .name = "soft_limit_in_bytes",
2786 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
2787 .write_string = mem_cgroup_write,
2788 .read_u64 = mem_cgroup_read,
2792 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
2793 .trigger = mem_cgroup_reset,
2794 .read_u64 = mem_cgroup_read,
2798 .read_map = mem_control_stat_show,
2801 .name = "force_empty",
2802 .trigger = mem_cgroup_force_empty_write,
2805 .name = "use_hierarchy",
2806 .write_u64 = mem_cgroup_hierarchy_write,
2807 .read_u64 = mem_cgroup_hierarchy_read,
2810 .name = "swappiness",
2811 .read_u64 = mem_cgroup_swappiness_read,
2812 .write_u64 = mem_cgroup_swappiness_write,
2816 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2817 static struct cftype memsw_cgroup_files[] = {
2819 .name = "memsw.usage_in_bytes",
2820 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
2821 .read_u64 = mem_cgroup_read,
2824 .name = "memsw.max_usage_in_bytes",
2825 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
2826 .trigger = mem_cgroup_reset,
2827 .read_u64 = mem_cgroup_read,
2830 .name = "memsw.limit_in_bytes",
2831 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
2832 .write_string = mem_cgroup_write,
2833 .read_u64 = mem_cgroup_read,
2836 .name = "memsw.failcnt",
2837 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
2838 .trigger = mem_cgroup_reset,
2839 .read_u64 = mem_cgroup_read,
2843 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2845 if (!do_swap_account)
2847 return cgroup_add_files(cont, ss, memsw_cgroup_files,
2848 ARRAY_SIZE(memsw_cgroup_files));
2851 static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2857 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2859 struct mem_cgroup_per_node *pn;
2860 struct mem_cgroup_per_zone *mz;
2862 int zone, tmp = node;
2864 * This routine is called against possible nodes.
2865 * But it's BUG to call kmalloc() against offline node.
2867 * TODO: this routine can waste much memory for nodes which will
2868 * never be onlined. It's better to use memory hotplug callback
2871 if (!node_state(node, N_NORMAL_MEMORY))
2873 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
2877 mem->info.nodeinfo[node] = pn;
2878 memset(pn, 0, sizeof(*pn));
2880 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2881 mz = &pn->zoneinfo[zone];
2883 INIT_LIST_HEAD(&mz->lists[l]);
2884 mz->usage_in_excess = 0;
2885 mz->on_tree = false;
2891 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2893 kfree(mem->info.nodeinfo[node]);
2896 static int mem_cgroup_size(void)
2898 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
2899 return sizeof(struct mem_cgroup) + cpustat_size;
2902 static struct mem_cgroup *mem_cgroup_alloc(void)
2904 struct mem_cgroup *mem;
2905 int size = mem_cgroup_size();
2907 if (size < PAGE_SIZE)
2908 mem = kmalloc(size, GFP_KERNEL);
2910 mem = vmalloc(size);
2913 memset(mem, 0, size);
2918 * At destroying mem_cgroup, references from swap_cgroup can remain.
2919 * (scanning all at force_empty is too costly...)
2921 * Instead of clearing all references at force_empty, we remember
2922 * the number of reference from swap_cgroup and free mem_cgroup when
2923 * it goes down to 0.
2925 * Removal of cgroup itself succeeds regardless of refs from swap.
2928 static void __mem_cgroup_free(struct mem_cgroup *mem)
2932 mem_cgroup_remove_from_trees(mem);
2933 free_css_id(&mem_cgroup_subsys, &mem->css);
2935 for_each_node_state(node, N_POSSIBLE)
2936 free_mem_cgroup_per_zone_info(mem, node);
2938 if (mem_cgroup_size() < PAGE_SIZE)
2944 static void mem_cgroup_get(struct mem_cgroup *mem)
2946 atomic_inc(&mem->refcnt);
2949 static void mem_cgroup_put(struct mem_cgroup *mem)
2951 if (atomic_dec_and_test(&mem->refcnt)) {
2952 struct mem_cgroup *parent = parent_mem_cgroup(mem);
2953 __mem_cgroup_free(mem);
2955 mem_cgroup_put(parent);
2960 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
2962 static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
2964 if (!mem->res.parent)
2966 return mem_cgroup_from_res_counter(mem->res.parent, res);
2969 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2970 static void __init enable_swap_cgroup(void)
2972 if (!mem_cgroup_disabled() && really_do_swap_account)
2973 do_swap_account = 1;
2976 static void __init enable_swap_cgroup(void)
2981 static int mem_cgroup_soft_limit_tree_init(void)
2983 struct mem_cgroup_tree_per_node *rtpn;
2984 struct mem_cgroup_tree_per_zone *rtpz;
2985 int tmp, node, zone;
2987 for_each_node_state(node, N_POSSIBLE) {
2989 if (!node_state(node, N_NORMAL_MEMORY))
2991 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
2995 soft_limit_tree.rb_tree_per_node[node] = rtpn;
2997 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2998 rtpz = &rtpn->rb_tree_per_zone[zone];
2999 rtpz->rb_root = RB_ROOT;
3000 spin_lock_init(&rtpz->lock);
3006 static struct cgroup_subsys_state * __ref
3007 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
3009 struct mem_cgroup *mem, *parent;
3010 long error = -ENOMEM;
3013 mem = mem_cgroup_alloc();
3015 return ERR_PTR(error);
3017 for_each_node_state(node, N_POSSIBLE)
3018 if (alloc_mem_cgroup_per_zone_info(mem, node))
3022 if (cont->parent == NULL) {
3023 enable_swap_cgroup();
3025 root_mem_cgroup = mem;
3026 if (mem_cgroup_soft_limit_tree_init())
3030 parent = mem_cgroup_from_cont(cont->parent);
3031 mem->use_hierarchy = parent->use_hierarchy;
3034 if (parent && parent->use_hierarchy) {
3035 res_counter_init(&mem->res, &parent->res);
3036 res_counter_init(&mem->memsw, &parent->memsw);
3038 * We increment refcnt of the parent to ensure that we can
3039 * safely access it on res_counter_charge/uncharge.
3040 * This refcnt will be decremented when freeing this
3041 * mem_cgroup(see mem_cgroup_put).
3043 mem_cgroup_get(parent);
3045 res_counter_init(&mem->res, NULL);
3046 res_counter_init(&mem->memsw, NULL);
3048 mem->last_scanned_child = 0;
3049 spin_lock_init(&mem->reclaim_param_lock);
3052 mem->swappiness = get_swappiness(parent);
3053 atomic_set(&mem->refcnt, 1);
3056 __mem_cgroup_free(mem);
3057 root_mem_cgroup = NULL;
3058 return ERR_PTR(error);
3061 static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
3062 struct cgroup *cont)
3064 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3066 return mem_cgroup_force_empty(mem, false);
3069 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
3070 struct cgroup *cont)
3072 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3074 mem_cgroup_put(mem);
3077 static int mem_cgroup_populate(struct cgroup_subsys *ss,
3078 struct cgroup *cont)
3082 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
3083 ARRAY_SIZE(mem_cgroup_files));
3086 ret = register_memsw_files(cont, ss);
3090 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
3091 struct cgroup *cont,
3092 struct cgroup *old_cont,
3093 struct task_struct *p,
3096 mutex_lock(&memcg_tasklist);
3098 * FIXME: It's better to move charges of this process from old
3099 * memcg to new memcg. But it's just on TODO-List now.
3101 mutex_unlock(&memcg_tasklist);
3104 struct cgroup_subsys mem_cgroup_subsys = {
3106 .subsys_id = mem_cgroup_subsys_id,
3107 .create = mem_cgroup_create,
3108 .pre_destroy = mem_cgroup_pre_destroy,
3109 .destroy = mem_cgroup_destroy,
3110 .populate = mem_cgroup_populate,
3111 .attach = mem_cgroup_move_task,
3116 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3118 static int __init disable_swap_account(char *s)
3120 really_do_swap_account = 0;
3123 __setup("noswapaccount", disable_swap_account);